Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * alloc.c | |
5 | * | |
6 | * Extent allocs and frees | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
60b11392 | 30 | #include <linux/swap.h> |
a90714c1 | 31 | #include <linux/quotaops.h> |
ccd979bd MF |
32 | |
33 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
34 | #include <cluster/masklog.h> | |
35 | ||
36 | #include "ocfs2.h" | |
37 | ||
38 | #include "alloc.h" | |
60b11392 | 39 | #include "aops.h" |
d6b32bbb | 40 | #include "blockcheck.h" |
ccd979bd MF |
41 | #include "dlmglue.h" |
42 | #include "extent_map.h" | |
43 | #include "inode.h" | |
44 | #include "journal.h" | |
45 | #include "localalloc.h" | |
46 | #include "suballoc.h" | |
47 | #include "sysfile.h" | |
48 | #include "file.h" | |
49 | #include "super.h" | |
50 | #include "uptodate.h" | |
2a50a743 | 51 | #include "xattr.h" |
bcbbb24a | 52 | #include "refcounttree.h" |
ccd979bd MF |
53 | |
54 | #include "buffer_head_io.h" | |
55 | ||
853a3a14 TM |
56 | enum ocfs2_contig_type { |
57 | CONTIG_NONE = 0, | |
58 | CONTIG_LEFT, | |
59 | CONTIG_RIGHT, | |
60 | CONTIG_LEFTRIGHT, | |
61 | }; | |
e7d4cb6b | 62 | |
853a3a14 TM |
63 | static enum ocfs2_contig_type |
64 | ocfs2_extent_rec_contig(struct super_block *sb, | |
65 | struct ocfs2_extent_rec *ext, | |
66 | struct ocfs2_extent_rec *insert_rec); | |
1625f8ac JB |
67 | /* |
68 | * Operations for a specific extent tree type. | |
69 | * | |
70 | * To implement an on-disk btree (extent tree) type in ocfs2, add | |
71 | * an ocfs2_extent_tree_operations structure and the matching | |
8d6220d6 | 72 | * ocfs2_init_<thingy>_extent_tree() function. That's pretty much it |
1625f8ac JB |
73 | * for the allocation portion of the extent tree. |
74 | */ | |
e7d4cb6b | 75 | struct ocfs2_extent_tree_operations { |
1625f8ac JB |
76 | /* |
77 | * last_eb_blk is the block number of the right most leaf extent | |
78 | * block. Most on-disk structures containing an extent tree store | |
79 | * this value for fast access. The ->eo_set_last_eb_blk() and | |
80 | * ->eo_get_last_eb_blk() operations access this value. They are | |
81 | * both required. | |
82 | */ | |
35dc0aa3 JB |
83 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
84 | u64 blkno); | |
85 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
1625f8ac JB |
86 | |
87 | /* | |
88 | * The on-disk structure usually keeps track of how many total | |
89 | * clusters are stored in this extent tree. This function updates | |
90 | * that value. new_clusters is the delta, and must be | |
91 | * added to the total. Required. | |
92 | */ | |
6136ca5f | 93 | void (*eo_update_clusters)(struct ocfs2_extent_tree *et, |
35dc0aa3 | 94 | u32 new_clusters); |
1625f8ac | 95 | |
92ba470c JB |
96 | /* |
97 | * If this extent tree is supported by an extent map, insert | |
98 | * a record into the map. | |
99 | */ | |
100 | void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et, | |
101 | struct ocfs2_extent_rec *rec); | |
102 | ||
4c911eef JB |
103 | /* |
104 | * If this extent tree is supported by an extent map, truncate the | |
105 | * map to clusters, | |
106 | */ | |
107 | void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et, | |
108 | u32 clusters); | |
109 | ||
1625f8ac JB |
110 | /* |
111 | * If ->eo_insert_check() exists, it is called before rec is | |
112 | * inserted into the extent tree. It is optional. | |
113 | */ | |
6136ca5f | 114 | int (*eo_insert_check)(struct ocfs2_extent_tree *et, |
1e61ee79 | 115 | struct ocfs2_extent_rec *rec); |
6136ca5f | 116 | int (*eo_sanity_check)(struct ocfs2_extent_tree *et); |
0ce1010f | 117 | |
1625f8ac JB |
118 | /* |
119 | * -------------------------------------------------------------- | |
120 | * The remaining are internal to ocfs2_extent_tree and don't have | |
121 | * accessor functions | |
122 | */ | |
123 | ||
124 | /* | |
125 | * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el. | |
126 | * It is required. | |
127 | */ | |
0ce1010f | 128 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); |
1625f8ac JB |
129 | |
130 | /* | |
131 | * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if | |
132 | * it exists. If it does not, et->et_max_leaf_clusters is set | |
133 | * to 0 (unlimited). Optional. | |
134 | */ | |
6136ca5f | 135 | void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et); |
853a3a14 TM |
136 | |
137 | /* | |
138 | * ->eo_extent_contig test whether the 2 ocfs2_extent_rec | |
139 | * are contiguous or not. Optional. Don't need to set it if use | |
140 | * ocfs2_extent_rec as the tree leaf. | |
141 | */ | |
142 | enum ocfs2_contig_type | |
143 | (*eo_extent_contig)(struct ocfs2_extent_tree *et, | |
144 | struct ocfs2_extent_rec *ext, | |
145 | struct ocfs2_extent_rec *insert_rec); | |
e7d4cb6b TM |
146 | }; |
147 | ||
e7d4cb6b | 148 | |
f99b9b7c JB |
149 | /* |
150 | * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check | |
151 | * in the methods. | |
152 | */ | |
153 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et); | |
154 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
155 | u64 blkno); | |
6136ca5f | 156 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
f99b9b7c | 157 | u32 clusters); |
92ba470c JB |
158 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
159 | struct ocfs2_extent_rec *rec); | |
4c911eef JB |
160 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
161 | u32 clusters); | |
6136ca5f | 162 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
f99b9b7c | 163 | struct ocfs2_extent_rec *rec); |
6136ca5f | 164 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et); |
f99b9b7c JB |
165 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et); |
166 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
167 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, | |
168 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
169 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
92ba470c | 170 | .eo_extent_map_insert = ocfs2_dinode_extent_map_insert, |
4c911eef | 171 | .eo_extent_map_truncate = ocfs2_dinode_extent_map_truncate, |
f99b9b7c JB |
172 | .eo_insert_check = ocfs2_dinode_insert_check, |
173 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
174 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, | |
175 | }; | |
0ce1010f | 176 | |
e7d4cb6b TM |
177 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
178 | u64 blkno) | |
179 | { | |
ea5efa15 | 180 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 181 | |
f99b9b7c | 182 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
183 | di->i_last_eb_blk = cpu_to_le64(blkno); |
184 | } | |
185 | ||
186 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
187 | { | |
ea5efa15 | 188 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 189 | |
f99b9b7c | 190 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
e7d4cb6b TM |
191 | return le64_to_cpu(di->i_last_eb_blk); |
192 | } | |
193 | ||
6136ca5f | 194 | static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et, |
e7d4cb6b TM |
195 | u32 clusters) |
196 | { | |
6136ca5f | 197 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
ea5efa15 | 198 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
199 | |
200 | le32_add_cpu(&di->i_clusters, clusters); | |
6136ca5f JB |
201 | spin_lock(&oi->ip_lock); |
202 | oi->ip_clusters = le32_to_cpu(di->i_clusters); | |
203 | spin_unlock(&oi->ip_lock); | |
e7d4cb6b TM |
204 | } |
205 | ||
92ba470c JB |
206 | static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et, |
207 | struct ocfs2_extent_rec *rec) | |
208 | { | |
209 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
210 | ||
211 | ocfs2_extent_map_insert_rec(inode, rec); | |
212 | } | |
213 | ||
4c911eef JB |
214 | static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et, |
215 | u32 clusters) | |
216 | { | |
217 | struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode; | |
218 | ||
219 | ocfs2_extent_map_trunc(inode, clusters); | |
220 | } | |
221 | ||
6136ca5f | 222 | static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
223 | struct ocfs2_extent_rec *rec) |
224 | { | |
6136ca5f JB |
225 | struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci); |
226 | struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb); | |
1e61ee79 | 227 | |
6136ca5f | 228 | BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
1e61ee79 | 229 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && |
6136ca5f | 230 | (oi->ip_clusters != le32_to_cpu(rec->e_cpos)), |
1e61ee79 JB |
231 | "Device %s, asking for sparse allocation: inode %llu, " |
232 | "cpos %u, clusters %u\n", | |
233 | osb->dev_str, | |
6136ca5f JB |
234 | (unsigned long long)oi->ip_blkno, |
235 | rec->e_cpos, oi->ip_clusters); | |
1e61ee79 JB |
236 | |
237 | return 0; | |
238 | } | |
239 | ||
6136ca5f | 240 | static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 241 | { |
10995aa2 | 242 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 243 | |
f99b9b7c | 244 | BUG_ON(et->et_ops != &ocfs2_dinode_et_ops); |
10995aa2 | 245 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); |
e7d4cb6b | 246 | |
10995aa2 | 247 | return 0; |
e7d4cb6b TM |
248 | } |
249 | ||
f99b9b7c JB |
250 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
251 | { | |
252 | struct ocfs2_dinode *di = et->et_object; | |
253 | ||
254 | et->et_root_el = &di->id2.i_list; | |
255 | } | |
256 | ||
e7d4cb6b | 257 | |
0ce1010f JB |
258 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
259 | { | |
2a50a743 | 260 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
0ce1010f | 261 | |
2a50a743 | 262 | et->et_root_el = &vb->vb_xv->xr_list; |
0ce1010f JB |
263 | } |
264 | ||
f56654c4 TM |
265 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
266 | u64 blkno) | |
267 | { | |
2a50a743 | 268 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 269 | |
2a50a743 | 270 | vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno); |
f56654c4 TM |
271 | } |
272 | ||
273 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
274 | { | |
2a50a743 | 275 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 276 | |
2a50a743 | 277 | return le64_to_cpu(vb->vb_xv->xr_last_eb_blk); |
f56654c4 TM |
278 | } |
279 | ||
6136ca5f | 280 | static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et, |
f56654c4 TM |
281 | u32 clusters) |
282 | { | |
2a50a743 | 283 | struct ocfs2_xattr_value_buf *vb = et->et_object; |
f56654c4 | 284 | |
2a50a743 | 285 | le32_add_cpu(&vb->vb_xv->xr_clusters, clusters); |
f56654c4 TM |
286 | } |
287 | ||
1a09f556 | 288 | static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = { |
35dc0aa3 JB |
289 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
290 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
291 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
0ce1010f | 292 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
293 | }; |
294 | ||
0ce1010f JB |
295 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
296 | { | |
297 | struct ocfs2_xattr_block *xb = et->et_object; | |
298 | ||
299 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
300 | } | |
301 | ||
6136ca5f | 302 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et) |
943cced3 | 303 | { |
6136ca5f | 304 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
943cced3 | 305 | et->et_max_leaf_clusters = |
6136ca5f | 306 | ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE); |
943cced3 JB |
307 | } |
308 | ||
ba492615 TM |
309 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
310 | u64 blkno) | |
311 | { | |
ea5efa15 | 312 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
313 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
314 | ||
315 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
316 | } | |
317 | ||
318 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
319 | { | |
ea5efa15 | 320 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
321 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
322 | ||
323 | return le64_to_cpu(xt->xt_last_eb_blk); | |
324 | } | |
325 | ||
6136ca5f | 326 | static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et, |
ba492615 TM |
327 | u32 clusters) |
328 | { | |
ea5efa15 | 329 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
330 | |
331 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
332 | } | |
333 | ||
ba492615 | 334 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { |
35dc0aa3 JB |
335 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
336 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
337 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
0ce1010f | 338 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 339 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
340 | }; |
341 | ||
9b7895ef MF |
342 | static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et, |
343 | u64 blkno) | |
344 | { | |
345 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
346 | ||
347 | dx_root->dr_last_eb_blk = cpu_to_le64(blkno); | |
348 | } | |
349 | ||
350 | static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
351 | { | |
352 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
353 | ||
354 | return le64_to_cpu(dx_root->dr_last_eb_blk); | |
355 | } | |
356 | ||
6136ca5f | 357 | static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et, |
9b7895ef MF |
358 | u32 clusters) |
359 | { | |
360 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
361 | ||
362 | le32_add_cpu(&dx_root->dr_clusters, clusters); | |
363 | } | |
364 | ||
6136ca5f | 365 | static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et) |
9b7895ef MF |
366 | { |
367 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
368 | ||
369 | BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root)); | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et) | |
375 | { | |
376 | struct ocfs2_dx_root_block *dx_root = et->et_object; | |
377 | ||
378 | et->et_root_el = &dx_root->dr_list; | |
379 | } | |
380 | ||
381 | static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = { | |
382 | .eo_set_last_eb_blk = ocfs2_dx_root_set_last_eb_blk, | |
383 | .eo_get_last_eb_blk = ocfs2_dx_root_get_last_eb_blk, | |
384 | .eo_update_clusters = ocfs2_dx_root_update_clusters, | |
385 | .eo_sanity_check = ocfs2_dx_root_sanity_check, | |
386 | .eo_fill_root_el = ocfs2_dx_root_fill_root_el, | |
387 | }; | |
388 | ||
fe924415 TM |
389 | static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et) |
390 | { | |
391 | struct ocfs2_refcount_block *rb = et->et_object; | |
392 | ||
393 | et->et_root_el = &rb->rf_list; | |
394 | } | |
395 | ||
396 | static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, | |
397 | u64 blkno) | |
398 | { | |
399 | struct ocfs2_refcount_block *rb = et->et_object; | |
400 | ||
401 | rb->rf_last_eb_blk = cpu_to_le64(blkno); | |
402 | } | |
403 | ||
404 | static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
405 | { | |
406 | struct ocfs2_refcount_block *rb = et->et_object; | |
407 | ||
408 | return le64_to_cpu(rb->rf_last_eb_blk); | |
409 | } | |
410 | ||
411 | static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et, | |
412 | u32 clusters) | |
413 | { | |
414 | struct ocfs2_refcount_block *rb = et->et_object; | |
415 | ||
416 | le32_add_cpu(&rb->rf_clusters, clusters); | |
417 | } | |
418 | ||
419 | static enum ocfs2_contig_type | |
420 | ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et, | |
421 | struct ocfs2_extent_rec *ext, | |
422 | struct ocfs2_extent_rec *insert_rec) | |
423 | { | |
424 | return CONTIG_NONE; | |
425 | } | |
426 | ||
427 | static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = { | |
428 | .eo_set_last_eb_blk = ocfs2_refcount_tree_set_last_eb_blk, | |
429 | .eo_get_last_eb_blk = ocfs2_refcount_tree_get_last_eb_blk, | |
430 | .eo_update_clusters = ocfs2_refcount_tree_update_clusters, | |
431 | .eo_fill_root_el = ocfs2_refcount_tree_fill_root_el, | |
432 | .eo_extent_contig = ocfs2_refcount_tree_extent_contig, | |
433 | }; | |
434 | ||
8d6220d6 | 435 | static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 436 | struct ocfs2_caching_info *ci, |
8d6220d6 | 437 | struct buffer_head *bh, |
13723d00 | 438 | ocfs2_journal_access_func access, |
8d6220d6 JB |
439 | void *obj, |
440 | struct ocfs2_extent_tree_operations *ops) | |
e7d4cb6b | 441 | { |
1a09f556 | 442 | et->et_ops = ops; |
ce1d9ea6 | 443 | et->et_root_bh = bh; |
5e404e9e | 444 | et->et_ci = ci; |
13723d00 | 445 | et->et_root_journal_access = access; |
ea5efa15 JB |
446 | if (!obj) |
447 | obj = (void *)bh->b_data; | |
448 | et->et_object = obj; | |
e7d4cb6b | 449 | |
0ce1010f | 450 | et->et_ops->eo_fill_root_el(et); |
943cced3 JB |
451 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
452 | et->et_max_leaf_clusters = 0; | |
453 | else | |
6136ca5f | 454 | et->et_ops->eo_fill_max_leaf_clusters(et); |
e7d4cb6b TM |
455 | } |
456 | ||
8d6220d6 | 457 | void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 458 | struct ocfs2_caching_info *ci, |
8d6220d6 | 459 | struct buffer_head *bh) |
1a09f556 | 460 | { |
5e404e9e | 461 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di, |
13723d00 | 462 | NULL, &ocfs2_dinode_et_ops); |
1a09f556 JB |
463 | } |
464 | ||
8d6220d6 | 465 | void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 466 | struct ocfs2_caching_info *ci, |
8d6220d6 | 467 | struct buffer_head *bh) |
1a09f556 | 468 | { |
5e404e9e | 469 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb, |
13723d00 | 470 | NULL, &ocfs2_xattr_tree_et_ops); |
1a09f556 JB |
471 | } |
472 | ||
8d6220d6 | 473 | void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 474 | struct ocfs2_caching_info *ci, |
2a50a743 | 475 | struct ocfs2_xattr_value_buf *vb) |
e7d4cb6b | 476 | { |
5e404e9e | 477 | __ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb, |
8d6220d6 | 478 | &ocfs2_xattr_value_et_ops); |
e7d4cb6b TM |
479 | } |
480 | ||
9b7895ef | 481 | void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et, |
5e404e9e | 482 | struct ocfs2_caching_info *ci, |
9b7895ef MF |
483 | struct buffer_head *bh) |
484 | { | |
5e404e9e | 485 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr, |
9b7895ef MF |
486 | NULL, &ocfs2_dx_root_et_ops); |
487 | } | |
488 | ||
fe924415 TM |
489 | void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et, |
490 | struct ocfs2_caching_info *ci, | |
491 | struct buffer_head *bh) | |
492 | { | |
493 | __ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb, | |
494 | NULL, &ocfs2_refcount_tree_et_ops); | |
495 | } | |
496 | ||
35dc0aa3 JB |
497 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
498 | u64 new_last_eb_blk) | |
e7d4cb6b | 499 | { |
ce1d9ea6 | 500 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
501 | } |
502 | ||
35dc0aa3 | 503 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 504 | { |
ce1d9ea6 | 505 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
506 | } |
507 | ||
6136ca5f | 508 | static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et, |
35dc0aa3 JB |
509 | u32 clusters) |
510 | { | |
6136ca5f | 511 | et->et_ops->eo_update_clusters(et, clusters); |
35dc0aa3 JB |
512 | } |
513 | ||
92ba470c JB |
514 | static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et, |
515 | struct ocfs2_extent_rec *rec) | |
516 | { | |
517 | if (et->et_ops->eo_extent_map_insert) | |
518 | et->et_ops->eo_extent_map_insert(et, rec); | |
519 | } | |
520 | ||
4c911eef JB |
521 | static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et, |
522 | u32 clusters) | |
523 | { | |
524 | if (et->et_ops->eo_extent_map_truncate) | |
525 | et->et_ops->eo_extent_map_truncate(et, clusters); | |
526 | } | |
527 | ||
13723d00 | 528 | static inline int ocfs2_et_root_journal_access(handle_t *handle, |
13723d00 JB |
529 | struct ocfs2_extent_tree *et, |
530 | int type) | |
531 | { | |
d9a0a1f8 | 532 | return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh, |
13723d00 JB |
533 | type); |
534 | } | |
535 | ||
853a3a14 TM |
536 | static inline enum ocfs2_contig_type |
537 | ocfs2_et_extent_contig(struct ocfs2_extent_tree *et, | |
538 | struct ocfs2_extent_rec *rec, | |
539 | struct ocfs2_extent_rec *insert_rec) | |
540 | { | |
541 | if (et->et_ops->eo_extent_contig) | |
542 | return et->et_ops->eo_extent_contig(et, rec, insert_rec); | |
543 | ||
544 | return ocfs2_extent_rec_contig( | |
545 | ocfs2_metadata_cache_get_super(et->et_ci), | |
546 | rec, insert_rec); | |
547 | } | |
548 | ||
6136ca5f | 549 | static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et, |
1e61ee79 JB |
550 | struct ocfs2_extent_rec *rec) |
551 | { | |
552 | int ret = 0; | |
553 | ||
554 | if (et->et_ops->eo_insert_check) | |
6136ca5f | 555 | ret = et->et_ops->eo_insert_check(et, rec); |
1e61ee79 JB |
556 | return ret; |
557 | } | |
558 | ||
6136ca5f | 559 | static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et) |
e7d4cb6b | 560 | { |
1e61ee79 JB |
561 | int ret = 0; |
562 | ||
563 | if (et->et_ops->eo_sanity_check) | |
6136ca5f | 564 | ret = et->et_ops->eo_sanity_check(et); |
1e61ee79 | 565 | return ret; |
e7d4cb6b TM |
566 | } |
567 | ||
dcd0538f | 568 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
569 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
570 | struct ocfs2_extent_block *eb); | |
d401dc12 JB |
571 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
572 | struct ocfs2_extent_tree *et, | |
6b791bcc TM |
573 | struct ocfs2_path *path, |
574 | struct ocfs2_extent_rec *insert_rec); | |
dcd0538f MF |
575 | /* |
576 | * Reset the actual path elements so that we can re-use the structure | |
577 | * to build another path. Generally, this involves freeing the buffer | |
578 | * heads. | |
579 | */ | |
e2e9f608 | 580 | void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) |
dcd0538f MF |
581 | { |
582 | int i, start = 0, depth = 0; | |
583 | struct ocfs2_path_item *node; | |
ccd979bd | 584 | |
dcd0538f MF |
585 | if (keep_root) |
586 | start = 1; | |
ccd979bd | 587 | |
dcd0538f MF |
588 | for(i = start; i < path_num_items(path); i++) { |
589 | node = &path->p_node[i]; | |
590 | ||
591 | brelse(node->bh); | |
592 | node->bh = NULL; | |
593 | node->el = NULL; | |
594 | } | |
595 | ||
596 | /* | |
597 | * Tree depth may change during truncate, or insert. If we're | |
598 | * keeping the root extent list, then make sure that our path | |
599 | * structure reflects the proper depth. | |
600 | */ | |
601 | if (keep_root) | |
602 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
13723d00 JB |
603 | else |
604 | path_root_access(path) = NULL; | |
dcd0538f MF |
605 | |
606 | path->p_tree_depth = depth; | |
607 | } | |
608 | ||
e2e9f608 | 609 | void ocfs2_free_path(struct ocfs2_path *path) |
dcd0538f MF |
610 | { |
611 | if (path) { | |
612 | ocfs2_reinit_path(path, 0); | |
613 | kfree(path); | |
614 | } | |
615 | } | |
616 | ||
328d5752 MF |
617 | /* |
618 | * All the elements of src into dest. After this call, src could be freed | |
619 | * without affecting dest. | |
620 | * | |
621 | * Both paths should have the same root. Any non-root elements of dest | |
622 | * will be freed. | |
623 | */ | |
624 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
625 | { | |
626 | int i; | |
627 | ||
628 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
629 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
13723d00 | 630 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
328d5752 MF |
631 | |
632 | ocfs2_reinit_path(dest, 1); | |
633 | ||
634 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
635 | dest->p_node[i].bh = src->p_node[i].bh; | |
636 | dest->p_node[i].el = src->p_node[i].el; | |
637 | ||
638 | if (dest->p_node[i].bh) | |
639 | get_bh(dest->p_node[i].bh); | |
640 | } | |
641 | } | |
642 | ||
dcd0538f MF |
643 | /* |
644 | * Make the *dest path the same as src and re-initialize src path to | |
645 | * have a root only. | |
646 | */ | |
647 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
648 | { | |
649 | int i; | |
650 | ||
651 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
13723d00 | 652 | BUG_ON(path_root_access(dest) != path_root_access(src)); |
dcd0538f MF |
653 | |
654 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
655 | brelse(dest->p_node[i].bh); | |
656 | ||
657 | dest->p_node[i].bh = src->p_node[i].bh; | |
658 | dest->p_node[i].el = src->p_node[i].el; | |
659 | ||
660 | src->p_node[i].bh = NULL; | |
661 | src->p_node[i].el = NULL; | |
662 | } | |
663 | } | |
664 | ||
665 | /* | |
666 | * Insert an extent block at given index. | |
667 | * | |
668 | * This will not take an additional reference on eb_bh. | |
669 | */ | |
670 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
671 | struct buffer_head *eb_bh) | |
672 | { | |
673 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
674 | ||
675 | /* | |
676 | * Right now, no root bh is an extent block, so this helps | |
677 | * catch code errors with dinode trees. The assertion can be | |
678 | * safely removed if we ever need to insert extent block | |
679 | * structures at the root. | |
680 | */ | |
681 | BUG_ON(index == 0); | |
682 | ||
683 | path->p_node[index].bh = eb_bh; | |
684 | path->p_node[index].el = &eb->h_list; | |
685 | } | |
686 | ||
687 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
13723d00 JB |
688 | struct ocfs2_extent_list *root_el, |
689 | ocfs2_journal_access_func access) | |
dcd0538f MF |
690 | { |
691 | struct ocfs2_path *path; | |
ccd979bd | 692 | |
dcd0538f MF |
693 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
694 | ||
695 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
696 | if (path) { | |
697 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
698 | get_bh(root_bh); | |
699 | path_root_bh(path) = root_bh; | |
700 | path_root_el(path) = root_el; | |
13723d00 | 701 | path_root_access(path) = access; |
dcd0538f MF |
702 | } |
703 | ||
704 | return path; | |
705 | } | |
706 | ||
e2e9f608 | 707 | struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path) |
ffdd7a54 | 708 | { |
13723d00 JB |
709 | return ocfs2_new_path(path_root_bh(path), path_root_el(path), |
710 | path_root_access(path)); | |
ffdd7a54 JB |
711 | } |
712 | ||
e2e9f608 | 713 | struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et) |
ffdd7a54 | 714 | { |
13723d00 JB |
715 | return ocfs2_new_path(et->et_root_bh, et->et_root_el, |
716 | et->et_root_journal_access); | |
717 | } | |
718 | ||
719 | /* | |
720 | * Journal the buffer at depth idx. All idx>0 are extent_blocks, | |
721 | * otherwise it's the root_access function. | |
722 | * | |
723 | * I don't like the way this function's name looks next to | |
724 | * ocfs2_journal_access_path(), but I don't have a better one. | |
725 | */ | |
e2e9f608 TM |
726 | int ocfs2_path_bh_journal_access(handle_t *handle, |
727 | struct ocfs2_caching_info *ci, | |
728 | struct ocfs2_path *path, | |
729 | int idx) | |
13723d00 JB |
730 | { |
731 | ocfs2_journal_access_func access = path_root_access(path); | |
732 | ||
733 | if (!access) | |
734 | access = ocfs2_journal_access; | |
735 | ||
736 | if (idx) | |
737 | access = ocfs2_journal_access_eb; | |
738 | ||
0cf2f763 | 739 | return access(handle, ci, path->p_node[idx].bh, |
13723d00 | 740 | OCFS2_JOURNAL_ACCESS_WRITE); |
ffdd7a54 JB |
741 | } |
742 | ||
dcd0538f MF |
743 | /* |
744 | * Convenience function to journal all components in a path. | |
745 | */ | |
e2e9f608 TM |
746 | int ocfs2_journal_access_path(struct ocfs2_caching_info *ci, |
747 | handle_t *handle, | |
748 | struct ocfs2_path *path) | |
dcd0538f MF |
749 | { |
750 | int i, ret = 0; | |
751 | ||
752 | if (!path) | |
753 | goto out; | |
754 | ||
755 | for(i = 0; i < path_num_items(path); i++) { | |
0cf2f763 | 756 | ret = ocfs2_path_bh_journal_access(handle, ci, path, i); |
dcd0538f MF |
757 | if (ret < 0) { |
758 | mlog_errno(ret); | |
759 | goto out; | |
760 | } | |
761 | } | |
762 | ||
763 | out: | |
764 | return ret; | |
765 | } | |
766 | ||
328d5752 MF |
767 | /* |
768 | * Return the index of the extent record which contains cluster #v_cluster. | |
769 | * -1 is returned if it was not found. | |
770 | * | |
771 | * Should work fine on interior and exterior nodes. | |
772 | */ | |
773 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
774 | { | |
775 | int ret = -1; | |
776 | int i; | |
777 | struct ocfs2_extent_rec *rec; | |
778 | u32 rec_end, rec_start, clusters; | |
779 | ||
780 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
781 | rec = &el->l_recs[i]; | |
782 | ||
783 | rec_start = le32_to_cpu(rec->e_cpos); | |
784 | clusters = ocfs2_rec_clusters(el, rec); | |
785 | ||
786 | rec_end = rec_start + clusters; | |
787 | ||
788 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
789 | ret = i; | |
790 | break; | |
791 | } | |
792 | } | |
793 | ||
794 | return ret; | |
795 | } | |
796 | ||
e48edee2 MF |
797 | /* |
798 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
853a3a14 | 799 | * ocfs2_extent_rec_contig only work properly against leaf nodes! |
e48edee2 | 800 | */ |
dcd0538f MF |
801 | static int ocfs2_block_extent_contig(struct super_block *sb, |
802 | struct ocfs2_extent_rec *ext, | |
803 | u64 blkno) | |
ccd979bd | 804 | { |
e48edee2 MF |
805 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
806 | ||
807 | blk_end += ocfs2_clusters_to_blocks(sb, | |
808 | le16_to_cpu(ext->e_leaf_clusters)); | |
809 | ||
810 | return blkno == blk_end; | |
ccd979bd MF |
811 | } |
812 | ||
dcd0538f MF |
813 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
814 | struct ocfs2_extent_rec *right) | |
815 | { | |
e48edee2 MF |
816 | u32 left_range; |
817 | ||
818 | left_range = le32_to_cpu(left->e_cpos) + | |
819 | le16_to_cpu(left->e_leaf_clusters); | |
820 | ||
821 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
822 | } |
823 | ||
824 | static enum ocfs2_contig_type | |
853a3a14 TM |
825 | ocfs2_extent_rec_contig(struct super_block *sb, |
826 | struct ocfs2_extent_rec *ext, | |
827 | struct ocfs2_extent_rec *insert_rec) | |
dcd0538f MF |
828 | { |
829 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
830 | ||
328d5752 MF |
831 | /* |
832 | * Refuse to coalesce extent records with different flag | |
833 | * fields - we don't want to mix unwritten extents with user | |
834 | * data. | |
835 | */ | |
836 | if (ext->e_flags != insert_rec->e_flags) | |
837 | return CONTIG_NONE; | |
838 | ||
dcd0538f | 839 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
b4a17651 | 840 | ocfs2_block_extent_contig(sb, ext, blkno)) |
dcd0538f MF |
841 | return CONTIG_RIGHT; |
842 | ||
843 | blkno = le64_to_cpu(ext->e_blkno); | |
844 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
b4a17651 | 845 | ocfs2_block_extent_contig(sb, insert_rec, blkno)) |
dcd0538f MF |
846 | return CONTIG_LEFT; |
847 | ||
848 | return CONTIG_NONE; | |
849 | } | |
850 | ||
851 | /* | |
852 | * NOTE: We can have pretty much any combination of contiguousness and | |
853 | * appending. | |
854 | * | |
855 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
856 | * we'll have to update the path to that leaf. | |
857 | */ | |
858 | enum ocfs2_append_type { | |
859 | APPEND_NONE = 0, | |
860 | APPEND_TAIL, | |
861 | }; | |
862 | ||
328d5752 MF |
863 | enum ocfs2_split_type { |
864 | SPLIT_NONE = 0, | |
865 | SPLIT_LEFT, | |
866 | SPLIT_RIGHT, | |
867 | }; | |
868 | ||
dcd0538f | 869 | struct ocfs2_insert_type { |
328d5752 | 870 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
871 | enum ocfs2_append_type ins_appending; |
872 | enum ocfs2_contig_type ins_contig; | |
873 | int ins_contig_index; | |
dcd0538f MF |
874 | int ins_tree_depth; |
875 | }; | |
876 | ||
328d5752 MF |
877 | struct ocfs2_merge_ctxt { |
878 | enum ocfs2_contig_type c_contig_type; | |
879 | int c_has_empty_extent; | |
880 | int c_split_covers_rec; | |
328d5752 MF |
881 | }; |
882 | ||
5e96581a JB |
883 | static int ocfs2_validate_extent_block(struct super_block *sb, |
884 | struct buffer_head *bh) | |
885 | { | |
d6b32bbb | 886 | int rc; |
5e96581a JB |
887 | struct ocfs2_extent_block *eb = |
888 | (struct ocfs2_extent_block *)bh->b_data; | |
889 | ||
970e4936 JB |
890 | mlog(0, "Validating extent block %llu\n", |
891 | (unsigned long long)bh->b_blocknr); | |
892 | ||
d6b32bbb JB |
893 | BUG_ON(!buffer_uptodate(bh)); |
894 | ||
895 | /* | |
896 | * If the ecc fails, we return the error but otherwise | |
897 | * leave the filesystem running. We know any error is | |
898 | * local to this block. | |
899 | */ | |
900 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check); | |
13723d00 JB |
901 | if (rc) { |
902 | mlog(ML_ERROR, "Checksum failed for extent block %llu\n", | |
903 | (unsigned long long)bh->b_blocknr); | |
d6b32bbb | 904 | return rc; |
13723d00 | 905 | } |
d6b32bbb JB |
906 | |
907 | /* | |
908 | * Errors after here are fatal. | |
909 | */ | |
910 | ||
5e96581a JB |
911 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { |
912 | ocfs2_error(sb, | |
913 | "Extent block #%llu has bad signature %.*s", | |
914 | (unsigned long long)bh->b_blocknr, 7, | |
915 | eb->h_signature); | |
916 | return -EINVAL; | |
917 | } | |
918 | ||
919 | if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) { | |
920 | ocfs2_error(sb, | |
921 | "Extent block #%llu has an invalid h_blkno " | |
922 | "of %llu", | |
923 | (unsigned long long)bh->b_blocknr, | |
924 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
925 | return -EINVAL; | |
926 | } | |
927 | ||
928 | if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) { | |
929 | ocfs2_error(sb, | |
930 | "Extent block #%llu has an invalid " | |
931 | "h_fs_generation of #%u", | |
932 | (unsigned long long)bh->b_blocknr, | |
933 | le32_to_cpu(eb->h_fs_generation)); | |
934 | return -EINVAL; | |
935 | } | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
3d03a305 | 940 | int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno, |
5e96581a JB |
941 | struct buffer_head **bh) |
942 | { | |
943 | int rc; | |
944 | struct buffer_head *tmp = *bh; | |
945 | ||
3d03a305 | 946 | rc = ocfs2_read_block(ci, eb_blkno, &tmp, |
970e4936 | 947 | ocfs2_validate_extent_block); |
5e96581a JB |
948 | |
949 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
970e4936 | 950 | if (!rc && !*bh) |
5e96581a JB |
951 | *bh = tmp; |
952 | ||
5e96581a JB |
953 | return rc; |
954 | } | |
955 | ||
956 | ||
ccd979bd MF |
957 | /* |
958 | * How many free extents have we got before we need more meta data? | |
959 | */ | |
960 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
f99b9b7c | 961 | struct ocfs2_extent_tree *et) |
ccd979bd MF |
962 | { |
963 | int retval; | |
e7d4cb6b | 964 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
965 | struct ocfs2_extent_block *eb; |
966 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 967 | u64 last_eb_blk = 0; |
ccd979bd MF |
968 | |
969 | mlog_entry_void(); | |
970 | ||
f99b9b7c JB |
971 | el = et->et_root_el; |
972 | last_eb_blk = ocfs2_et_get_last_eb_blk(et); | |
ccd979bd | 973 | |
e7d4cb6b | 974 | if (last_eb_blk) { |
3d03a305 JB |
975 | retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk, |
976 | &eb_bh); | |
ccd979bd MF |
977 | if (retval < 0) { |
978 | mlog_errno(retval); | |
979 | goto bail; | |
980 | } | |
981 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
982 | el = &eb->h_list; | |
e7d4cb6b | 983 | } |
ccd979bd MF |
984 | |
985 | BUG_ON(el->l_tree_depth != 0); | |
986 | ||
987 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
988 | bail: | |
a81cb88b | 989 | brelse(eb_bh); |
ccd979bd MF |
990 | |
991 | mlog_exit(retval); | |
992 | return retval; | |
993 | } | |
994 | ||
995 | /* expects array to already be allocated | |
996 | * | |
997 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
998 | * l_count for you | |
999 | */ | |
42a5a7a9 JB |
1000 | static int ocfs2_create_new_meta_bhs(handle_t *handle, |
1001 | struct ocfs2_extent_tree *et, | |
ccd979bd MF |
1002 | int wanted, |
1003 | struct ocfs2_alloc_context *meta_ac, | |
1004 | struct buffer_head *bhs[]) | |
1005 | { | |
1006 | int count, status, i; | |
1007 | u16 suballoc_bit_start; | |
1008 | u32 num_got; | |
1009 | u64 first_blkno; | |
42a5a7a9 JB |
1010 | struct ocfs2_super *osb = |
1011 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
ccd979bd MF |
1012 | struct ocfs2_extent_block *eb; |
1013 | ||
1014 | mlog_entry_void(); | |
1015 | ||
1016 | count = 0; | |
1017 | while (count < wanted) { | |
1018 | status = ocfs2_claim_metadata(osb, | |
1019 | handle, | |
1020 | meta_ac, | |
1021 | wanted - count, | |
1022 | &suballoc_bit_start, | |
1023 | &num_got, | |
1024 | &first_blkno); | |
1025 | if (status < 0) { | |
1026 | mlog_errno(status); | |
1027 | goto bail; | |
1028 | } | |
1029 | ||
1030 | for(i = count; i < (num_got + count); i++) { | |
1031 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
1032 | if (bhs[i] == NULL) { | |
1033 | status = -EIO; | |
1034 | mlog_errno(status); | |
1035 | goto bail; | |
1036 | } | |
42a5a7a9 | 1037 | ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]); |
ccd979bd | 1038 | |
42a5a7a9 JB |
1039 | status = ocfs2_journal_access_eb(handle, et->et_ci, |
1040 | bhs[i], | |
13723d00 | 1041 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1042 | if (status < 0) { |
1043 | mlog_errno(status); | |
1044 | goto bail; | |
1045 | } | |
1046 | ||
1047 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
1048 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
1049 | /* Ok, setup the minimal stuff here. */ | |
1050 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
1051 | eb->h_blkno = cpu_to_le64(first_blkno); | |
1052 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
b89c5428 TY |
1053 | eb->h_suballoc_slot = |
1054 | cpu_to_le16(meta_ac->ac_alloc_slot); | |
ccd979bd MF |
1055 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
1056 | eb->h_list.l_count = | |
1057 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
1058 | ||
1059 | suballoc_bit_start++; | |
1060 | first_blkno++; | |
1061 | ||
1062 | /* We'll also be dirtied by the caller, so | |
1063 | * this isn't absolutely necessary. */ | |
ec20cec7 | 1064 | ocfs2_journal_dirty(handle, bhs[i]); |
ccd979bd MF |
1065 | } |
1066 | ||
1067 | count += num_got; | |
1068 | } | |
1069 | ||
1070 | status = 0; | |
1071 | bail: | |
1072 | if (status < 0) { | |
1073 | for(i = 0; i < wanted; i++) { | |
a81cb88b | 1074 | brelse(bhs[i]); |
ccd979bd MF |
1075 | bhs[i] = NULL; |
1076 | } | |
1077 | } | |
1078 | mlog_exit(status); | |
1079 | return status; | |
1080 | } | |
1081 | ||
dcd0538f MF |
1082 | /* |
1083 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
1084 | * | |
1085 | * Returns the sum of the rightmost extent rec logical offset and | |
1086 | * cluster count. | |
1087 | * | |
1088 | * ocfs2_add_branch() uses this to determine what logical cluster | |
1089 | * value should be populated into the leftmost new branch records. | |
1090 | * | |
1091 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
1092 | * value for the new topmost tree record. | |
1093 | */ | |
1094 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
1095 | { | |
1096 | int i; | |
1097 | ||
1098 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1099 | ||
1100 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 1101 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
1102 | } |
1103 | ||
6b791bcc TM |
1104 | /* |
1105 | * Change range of the branches in the right most path according to the leaf | |
1106 | * extent block's rightmost record. | |
1107 | */ | |
1108 | static int ocfs2_adjust_rightmost_branch(handle_t *handle, | |
6b791bcc TM |
1109 | struct ocfs2_extent_tree *et) |
1110 | { | |
1111 | int status; | |
1112 | struct ocfs2_path *path = NULL; | |
1113 | struct ocfs2_extent_list *el; | |
1114 | struct ocfs2_extent_rec *rec; | |
1115 | ||
1116 | path = ocfs2_new_path_from_et(et); | |
1117 | if (!path) { | |
1118 | status = -ENOMEM; | |
1119 | return status; | |
1120 | } | |
1121 | ||
facdb77f | 1122 | status = ocfs2_find_path(et->et_ci, path, UINT_MAX); |
6b791bcc TM |
1123 | if (status < 0) { |
1124 | mlog_errno(status); | |
1125 | goto out; | |
1126 | } | |
1127 | ||
c901fb00 | 1128 | status = ocfs2_extend_trans(handle, path_num_items(path)); |
6b791bcc TM |
1129 | if (status < 0) { |
1130 | mlog_errno(status); | |
1131 | goto out; | |
1132 | } | |
1133 | ||
d401dc12 | 1134 | status = ocfs2_journal_access_path(et->et_ci, handle, path); |
6b791bcc TM |
1135 | if (status < 0) { |
1136 | mlog_errno(status); | |
1137 | goto out; | |
1138 | } | |
1139 | ||
1140 | el = path_leaf_el(path); | |
1141 | rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1]; | |
1142 | ||
d401dc12 | 1143 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
6b791bcc TM |
1144 | |
1145 | out: | |
1146 | ocfs2_free_path(path); | |
1147 | return status; | |
1148 | } | |
1149 | ||
ccd979bd MF |
1150 | /* |
1151 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
d401dc12 | 1152 | * to start at, if we don't want to start the branch at the root |
ccd979bd MF |
1153 | * structure. |
1154 | * | |
1155 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
1156 | * for the new last extent block. | |
1157 | * | |
1158 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 1159 | * contain a single record with cluster count == 0. |
ccd979bd | 1160 | */ |
d401dc12 | 1161 | static int ocfs2_add_branch(handle_t *handle, |
e7d4cb6b | 1162 | struct ocfs2_extent_tree *et, |
ccd979bd | 1163 | struct buffer_head *eb_bh, |
328d5752 | 1164 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
1165 | struct ocfs2_alloc_context *meta_ac) |
1166 | { | |
1167 | int status, new_blocks, i; | |
1168 | u64 next_blkno, new_last_eb_blk; | |
1169 | struct buffer_head *bh; | |
1170 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
1171 | struct ocfs2_extent_block *eb; |
1172 | struct ocfs2_extent_list *eb_el; | |
1173 | struct ocfs2_extent_list *el; | |
6b791bcc | 1174 | u32 new_cpos, root_end; |
ccd979bd MF |
1175 | |
1176 | mlog_entry_void(); | |
1177 | ||
328d5752 | 1178 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 1179 | |
ccd979bd MF |
1180 | if (eb_bh) { |
1181 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
1182 | el = &eb->h_list; | |
1183 | } else | |
ce1d9ea6 | 1184 | el = et->et_root_el; |
ccd979bd MF |
1185 | |
1186 | /* we never add a branch to a leaf. */ | |
1187 | BUG_ON(!el->l_tree_depth); | |
1188 | ||
1189 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
1190 | ||
6b791bcc TM |
1191 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
1192 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); | |
1193 | root_end = ocfs2_sum_rightmost_rec(et->et_root_el); | |
1194 | ||
1195 | /* | |
1196 | * If there is a gap before the root end and the real end | |
1197 | * of the righmost leaf block, we need to remove the gap | |
1198 | * between new_cpos and root_end first so that the tree | |
1199 | * is consistent after we add a new branch(it will start | |
1200 | * from new_cpos). | |
1201 | */ | |
1202 | if (root_end > new_cpos) { | |
1203 | mlog(0, "adjust the cluster end from %u to %u\n", | |
1204 | root_end, new_cpos); | |
d401dc12 | 1205 | status = ocfs2_adjust_rightmost_branch(handle, et); |
6b791bcc TM |
1206 | if (status) { |
1207 | mlog_errno(status); | |
1208 | goto bail; | |
1209 | } | |
1210 | } | |
1211 | ||
ccd979bd MF |
1212 | /* allocate the number of new eb blocks we need */ |
1213 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
1214 | GFP_KERNEL); | |
1215 | if (!new_eb_bhs) { | |
1216 | status = -ENOMEM; | |
1217 | mlog_errno(status); | |
1218 | goto bail; | |
1219 | } | |
1220 | ||
42a5a7a9 | 1221 | status = ocfs2_create_new_meta_bhs(handle, et, new_blocks, |
ccd979bd MF |
1222 | meta_ac, new_eb_bhs); |
1223 | if (status < 0) { | |
1224 | mlog_errno(status); | |
1225 | goto bail; | |
1226 | } | |
1227 | ||
1228 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be | |
1229 | * linked with the rest of the tree. | |
1230 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
1231 | * | |
1232 | * when we leave the loop, new_last_eb_blk will point to the | |
1233 | * newest leaf, and next_blkno will point to the topmost extent | |
1234 | * block. */ | |
1235 | next_blkno = new_last_eb_blk = 0; | |
1236 | for(i = 0; i < new_blocks; i++) { | |
1237 | bh = new_eb_bhs[i]; | |
1238 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
5e96581a JB |
1239 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1240 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1241 | eb_el = &eb->h_list; |
1242 | ||
d401dc12 | 1243 | status = ocfs2_journal_access_eb(handle, et->et_ci, bh, |
13723d00 | 1244 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1245 | if (status < 0) { |
1246 | mlog_errno(status); | |
1247 | goto bail; | |
1248 | } | |
1249 | ||
1250 | eb->h_next_leaf_blk = 0; | |
1251 | eb_el->l_tree_depth = cpu_to_le16(i); | |
1252 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
1253 | /* |
1254 | * This actually counts as an empty extent as | |
1255 | * c_clusters == 0 | |
1256 | */ | |
1257 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 1258 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
1259 | /* |
1260 | * eb_el isn't always an interior node, but even leaf | |
1261 | * nodes want a zero'd flags and reserved field so | |
1262 | * this gets the whole 32 bits regardless of use. | |
1263 | */ | |
1264 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
1265 | if (!eb_el->l_tree_depth) |
1266 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
1267 | ||
ec20cec7 | 1268 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
1269 | next_blkno = le64_to_cpu(eb->h_blkno); |
1270 | } | |
1271 | ||
1272 | /* This is a bit hairy. We want to update up to three blocks | |
1273 | * here without leaving any of them in an inconsistent state | |
1274 | * in case of error. We don't have to worry about | |
1275 | * journal_dirty erroring as it won't unless we've aborted the | |
1276 | * handle (in which case we would never be here) so reserving | |
1277 | * the write with journal_access is all we need to do. */ | |
d401dc12 | 1278 | status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh, |
13723d00 | 1279 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1280 | if (status < 0) { |
1281 | mlog_errno(status); | |
1282 | goto bail; | |
1283 | } | |
d9a0a1f8 | 1284 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1285 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1286 | if (status < 0) { |
1287 | mlog_errno(status); | |
1288 | goto bail; | |
1289 | } | |
1290 | if (eb_bh) { | |
d401dc12 | 1291 | status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh, |
13723d00 | 1292 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1293 | if (status < 0) { |
1294 | mlog_errno(status); | |
1295 | goto bail; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 1300 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
1301 | i = le16_to_cpu(el->l_next_free_rec); |
1302 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 1303 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 1304 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
1305 | le16_add_cpu(&el->l_next_free_rec, 1); |
1306 | ||
1307 | /* fe needs a new last extent block pointer, as does the | |
1308 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 1309 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 1310 | |
328d5752 | 1311 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
1312 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
1313 | ||
ec20cec7 JB |
1314 | ocfs2_journal_dirty(handle, *last_eb_bh); |
1315 | ocfs2_journal_dirty(handle, et->et_root_bh); | |
1316 | if (eb_bh) | |
1317 | ocfs2_journal_dirty(handle, eb_bh); | |
ccd979bd | 1318 | |
328d5752 MF |
1319 | /* |
1320 | * Some callers want to track the rightmost leaf so pass it | |
1321 | * back here. | |
1322 | */ | |
1323 | brelse(*last_eb_bh); | |
1324 | get_bh(new_eb_bhs[0]); | |
1325 | *last_eb_bh = new_eb_bhs[0]; | |
1326 | ||
ccd979bd MF |
1327 | status = 0; |
1328 | bail: | |
1329 | if (new_eb_bhs) { | |
1330 | for (i = 0; i < new_blocks; i++) | |
a81cb88b | 1331 | brelse(new_eb_bhs[i]); |
ccd979bd MF |
1332 | kfree(new_eb_bhs); |
1333 | } | |
1334 | ||
1335 | mlog_exit(status); | |
1336 | return status; | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * adds another level to the allocation tree. | |
1341 | * returns back the new extent block so you can add a branch to it | |
1342 | * after this call. | |
1343 | */ | |
d401dc12 | 1344 | static int ocfs2_shift_tree_depth(handle_t *handle, |
e7d4cb6b | 1345 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1346 | struct ocfs2_alloc_context *meta_ac, |
1347 | struct buffer_head **ret_new_eb_bh) | |
1348 | { | |
1349 | int status, i; | |
dcd0538f | 1350 | u32 new_clusters; |
ccd979bd | 1351 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 1352 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 1353 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
1354 | struct ocfs2_extent_list *eb_el; |
1355 | ||
1356 | mlog_entry_void(); | |
1357 | ||
42a5a7a9 | 1358 | status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac, |
ccd979bd MF |
1359 | &new_eb_bh); |
1360 | if (status < 0) { | |
1361 | mlog_errno(status); | |
1362 | goto bail; | |
1363 | } | |
1364 | ||
1365 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
5e96581a JB |
1366 | /* ocfs2_create_new_meta_bhs() should create it right! */ |
1367 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
1368 | |
1369 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1370 | root_el = et->et_root_el; |
ccd979bd | 1371 | |
d401dc12 | 1372 | status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh, |
13723d00 | 1373 | OCFS2_JOURNAL_ACCESS_CREATE); |
ccd979bd MF |
1374 | if (status < 0) { |
1375 | mlog_errno(status); | |
1376 | goto bail; | |
1377 | } | |
1378 | ||
e7d4cb6b TM |
1379 | /* copy the root extent list data into the new extent block */ |
1380 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1381 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1382 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1383 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd | 1384 | |
ec20cec7 | 1385 | ocfs2_journal_dirty(handle, new_eb_bh); |
ccd979bd | 1386 | |
d9a0a1f8 | 1387 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 1388 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1389 | if (status < 0) { |
1390 | mlog_errno(status); | |
1391 | goto bail; | |
1392 | } | |
1393 | ||
dcd0538f MF |
1394 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1395 | ||
e7d4cb6b TM |
1396 | /* update root_bh now */ |
1397 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1398 | root_el->l_recs[0].e_cpos = 0; | |
1399 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1400 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1401 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1402 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1403 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1404 | |
1405 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1406 | * becomes the allocated extent block */ | |
e7d4cb6b | 1407 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1408 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1409 | |
ec20cec7 | 1410 | ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1411 | |
1412 | *ret_new_eb_bh = new_eb_bh; | |
1413 | new_eb_bh = NULL; | |
1414 | status = 0; | |
1415 | bail: | |
a81cb88b | 1416 | brelse(new_eb_bh); |
ccd979bd MF |
1417 | |
1418 | mlog_exit(status); | |
1419 | return status; | |
1420 | } | |
1421 | ||
ccd979bd MF |
1422 | /* |
1423 | * Should only be called when there is no space left in any of the | |
1424 | * leaf nodes. What we want to do is find the lowest tree depth | |
1425 | * non-leaf extent block with room for new records. There are three | |
1426 | * valid results of this search: | |
1427 | * | |
1428 | * 1) a lowest extent block is found, then we pass it back in | |
1429 | * *lowest_eb_bh and return '0' | |
1430 | * | |
e7d4cb6b | 1431 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1432 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1433 | * | |
e7d4cb6b | 1434 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1435 | * which case we return > 0 |
1436 | * | |
1437 | * return status < 0 indicates an error. | |
1438 | */ | |
d401dc12 | 1439 | static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et, |
ccd979bd MF |
1440 | struct buffer_head **target_bh) |
1441 | { | |
1442 | int status = 0, i; | |
1443 | u64 blkno; | |
ccd979bd MF |
1444 | struct ocfs2_extent_block *eb; |
1445 | struct ocfs2_extent_list *el; | |
1446 | struct buffer_head *bh = NULL; | |
1447 | struct buffer_head *lowest_bh = NULL; | |
1448 | ||
1449 | mlog_entry_void(); | |
1450 | ||
1451 | *target_bh = NULL; | |
1452 | ||
ce1d9ea6 | 1453 | el = et->et_root_el; |
ccd979bd MF |
1454 | |
1455 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1456 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3d03a305 JB |
1457 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1458 | "Owner %llu has empty " | |
ccd979bd | 1459 | "extent list (next_free_rec == 0)", |
3d03a305 | 1460 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); |
ccd979bd MF |
1461 | status = -EIO; |
1462 | goto bail; | |
1463 | } | |
1464 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1465 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1466 | if (!blkno) { | |
3d03a305 JB |
1467 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
1468 | "Owner %llu has extent " | |
ccd979bd MF |
1469 | "list where extent # %d has no physical " |
1470 | "block start", | |
3d03a305 | 1471 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i); |
ccd979bd MF |
1472 | status = -EIO; |
1473 | goto bail; | |
1474 | } | |
1475 | ||
a81cb88b MF |
1476 | brelse(bh); |
1477 | bh = NULL; | |
ccd979bd | 1478 | |
3d03a305 | 1479 | status = ocfs2_read_extent_block(et->et_ci, blkno, &bh); |
ccd979bd MF |
1480 | if (status < 0) { |
1481 | mlog_errno(status); | |
1482 | goto bail; | |
1483 | } | |
dcd0538f MF |
1484 | |
1485 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
dcd0538f MF |
1486 | el = &eb->h_list; |
1487 | ||
1488 | if (le16_to_cpu(el->l_next_free_rec) < | |
1489 | le16_to_cpu(el->l_count)) { | |
a81cb88b | 1490 | brelse(lowest_bh); |
dcd0538f MF |
1491 | lowest_bh = bh; |
1492 | get_bh(lowest_bh); | |
1493 | } | |
1494 | } | |
1495 | ||
1496 | /* If we didn't find one and the fe doesn't have any room, | |
1497 | * then return '1' */ | |
ce1d9ea6 | 1498 | el = et->et_root_el; |
e7d4cb6b | 1499 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1500 | status = 1; |
1501 | ||
1502 | *target_bh = lowest_bh; | |
1503 | bail: | |
a81cb88b | 1504 | brelse(bh); |
dcd0538f MF |
1505 | |
1506 | mlog_exit(status); | |
1507 | return status; | |
1508 | } | |
1509 | ||
c3afcbb3 MF |
1510 | /* |
1511 | * Grow a b-tree so that it has more records. | |
1512 | * | |
1513 | * We might shift the tree depth in which case existing paths should | |
1514 | * be considered invalid. | |
1515 | * | |
1516 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1517 | * |
1518 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 | 1519 | */ |
d401dc12 JB |
1520 | static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
1521 | int *final_depth, struct buffer_head **last_eb_bh, | |
c3afcbb3 MF |
1522 | struct ocfs2_alloc_context *meta_ac) |
1523 | { | |
1524 | int ret, shift; | |
ce1d9ea6 | 1525 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1526 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1527 | struct buffer_head *bh = NULL; |
1528 | ||
1529 | BUG_ON(meta_ac == NULL); | |
1530 | ||
d401dc12 | 1531 | shift = ocfs2_find_branch_target(et, &bh); |
c3afcbb3 MF |
1532 | if (shift < 0) { |
1533 | ret = shift; | |
1534 | mlog_errno(ret); | |
1535 | goto out; | |
1536 | } | |
1537 | ||
1538 | /* We traveled all the way to the bottom of the allocation tree | |
1539 | * and didn't find room for any more extents - we need to add | |
1540 | * another tree level */ | |
1541 | if (shift) { | |
1542 | BUG_ON(bh); | |
1543 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1544 | ||
1545 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1546 | * the new extent block (so we can pass that to | |
1547 | * ocfs2_add_branch). */ | |
d401dc12 | 1548 | ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh); |
c3afcbb3 MF |
1549 | if (ret < 0) { |
1550 | mlog_errno(ret); | |
1551 | goto out; | |
1552 | } | |
1553 | depth++; | |
328d5752 MF |
1554 | if (depth == 1) { |
1555 | /* | |
1556 | * Special case: we have room now if we shifted from | |
1557 | * tree_depth 0, so no more work needs to be done. | |
1558 | * | |
1559 | * We won't be calling add_branch, so pass | |
1560 | * back *last_eb_bh as the new leaf. At depth | |
1561 | * zero, it should always be null so there's | |
1562 | * no reason to brelse. | |
1563 | */ | |
1564 | BUG_ON(*last_eb_bh); | |
1565 | get_bh(bh); | |
1566 | *last_eb_bh = bh; | |
c3afcbb3 | 1567 | goto out; |
328d5752 | 1568 | } |
c3afcbb3 MF |
1569 | } |
1570 | ||
1571 | /* call ocfs2_add_branch to add the final part of the tree with | |
1572 | * the new data. */ | |
1573 | mlog(0, "add branch. bh = %p\n", bh); | |
d401dc12 | 1574 | ret = ocfs2_add_branch(handle, et, bh, last_eb_bh, |
c3afcbb3 MF |
1575 | meta_ac); |
1576 | if (ret < 0) { | |
1577 | mlog_errno(ret); | |
1578 | goto out; | |
1579 | } | |
1580 | ||
1581 | out: | |
1582 | if (final_depth) | |
1583 | *final_depth = depth; | |
1584 | brelse(bh); | |
1585 | return ret; | |
1586 | } | |
1587 | ||
dcd0538f MF |
1588 | /* |
1589 | * This function will discard the rightmost extent record. | |
1590 | */ | |
1591 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1592 | { | |
1593 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1594 | int count = le16_to_cpu(el->l_count); | |
1595 | unsigned int num_bytes; | |
1596 | ||
1597 | BUG_ON(!next_free); | |
1598 | /* This will cause us to go off the end of our extent list. */ | |
1599 | BUG_ON(next_free >= count); | |
1600 | ||
1601 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1602 | ||
1603 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1604 | } | |
1605 | ||
1606 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1607 | struct ocfs2_extent_rec *insert_rec) | |
1608 | { | |
1609 | int i, insert_index, next_free, has_empty, num_bytes; | |
1610 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1611 | struct ocfs2_extent_rec *rec; | |
1612 | ||
1613 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1614 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1615 | ||
1616 | BUG_ON(!next_free); | |
1617 | ||
1618 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1619 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1620 | |
1621 | /* | |
1622 | * The easiest way to approach this is to just remove the | |
1623 | * empty extent and temporarily decrement next_free. | |
1624 | */ | |
1625 | if (has_empty) { | |
1626 | /* | |
1627 | * If next_free was 1 (only an empty extent), this | |
1628 | * loop won't execute, which is fine. We still want | |
1629 | * the decrement above to happen. | |
1630 | */ | |
1631 | for(i = 0; i < (next_free - 1); i++) | |
1632 | el->l_recs[i] = el->l_recs[i+1]; | |
1633 | ||
1634 | next_free--; | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * Figure out what the new record index should be. | |
1639 | */ | |
1640 | for(i = 0; i < next_free; i++) { | |
1641 | rec = &el->l_recs[i]; | |
1642 | ||
1643 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1644 | break; | |
1645 | } | |
1646 | insert_index = i; | |
1647 | ||
1648 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1649 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1650 | ||
1651 | BUG_ON(insert_index < 0); | |
1652 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1653 | BUG_ON(insert_index > next_free); | |
1654 | ||
1655 | /* | |
1656 | * No need to memmove if we're just adding to the tail. | |
1657 | */ | |
1658 | if (insert_index != next_free) { | |
1659 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1660 | ||
1661 | num_bytes = next_free - insert_index; | |
1662 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1663 | memmove(&el->l_recs[insert_index + 1], | |
1664 | &el->l_recs[insert_index], | |
1665 | num_bytes); | |
1666 | } | |
1667 | ||
1668 | /* | |
1669 | * Either we had an empty extent, and need to re-increment or | |
1670 | * there was no empty extent on a non full rightmost leaf node, | |
1671 | * in which case we still need to increment. | |
1672 | */ | |
1673 | next_free++; | |
1674 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1675 | /* | |
1676 | * Make sure none of the math above just messed up our tree. | |
1677 | */ | |
1678 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1679 | ||
1680 | el->l_recs[insert_index] = *insert_rec; | |
1681 | ||
1682 | } | |
1683 | ||
328d5752 MF |
1684 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1685 | { | |
1686 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1687 | ||
1688 | BUG_ON(num_recs == 0); | |
1689 | ||
1690 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1691 | num_recs--; | |
1692 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1693 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1694 | memset(&el->l_recs[num_recs], 0, | |
1695 | sizeof(struct ocfs2_extent_rec)); | |
1696 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1697 | } | |
1698 | } | |
1699 | ||
dcd0538f MF |
1700 | /* |
1701 | * Create an empty extent record . | |
1702 | * | |
1703 | * l_next_free_rec may be updated. | |
1704 | * | |
1705 | * If an empty extent already exists do nothing. | |
1706 | */ | |
1707 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1708 | { | |
1709 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1710 | ||
e48edee2 MF |
1711 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1712 | ||
dcd0538f MF |
1713 | if (next_free == 0) |
1714 | goto set_and_inc; | |
1715 | ||
1716 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1717 | return; | |
1718 | ||
1719 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1720 | "Asked to create an empty extent in a full list:\n" | |
1721 | "count = %u, tree depth = %u", | |
1722 | le16_to_cpu(el->l_count), | |
1723 | le16_to_cpu(el->l_tree_depth)); | |
1724 | ||
1725 | ocfs2_shift_records_right(el); | |
1726 | ||
1727 | set_and_inc: | |
1728 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1729 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * For a rotation which involves two leaf nodes, the "root node" is | |
1734 | * the lowest level tree node which contains a path to both leafs. This | |
1735 | * resulting set of information can be used to form a complete "subtree" | |
1736 | * | |
1737 | * This function is passed two full paths from the dinode down to a | |
1738 | * pair of adjacent leaves. It's task is to figure out which path | |
1739 | * index contains the subtree root - this can be the root index itself | |
1740 | * in a worst-case rotation. | |
1741 | * | |
1742 | * The array index of the subtree root is passed back. | |
1743 | */ | |
38a04e43 TM |
1744 | int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et, |
1745 | struct ocfs2_path *left, | |
1746 | struct ocfs2_path *right) | |
dcd0538f MF |
1747 | { |
1748 | int i = 0; | |
1749 | ||
1750 | /* | |
1751 | * Check that the caller passed in two paths from the same tree. | |
1752 | */ | |
1753 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1754 | ||
1755 | do { | |
1756 | i++; | |
1757 | ||
1758 | /* | |
1759 | * The caller didn't pass two adjacent paths. | |
1760 | */ | |
1761 | mlog_bug_on_msg(i > left->p_tree_depth, | |
7dc02805 | 1762 | "Owner %llu, left depth %u, right depth %u\n" |
dcd0538f | 1763 | "left leaf blk %llu, right leaf blk %llu\n", |
7dc02805 JB |
1764 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
1765 | left->p_tree_depth, right->p_tree_depth, | |
dcd0538f MF |
1766 | (unsigned long long)path_leaf_bh(left)->b_blocknr, |
1767 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1768 | } while (left->p_node[i].bh->b_blocknr == | |
1769 | right->p_node[i].bh->b_blocknr); | |
1770 | ||
1771 | return i - 1; | |
1772 | } | |
1773 | ||
1774 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1775 | ||
1776 | /* | |
1777 | * Traverse a btree path in search of cpos, starting at root_el. | |
1778 | * | |
1779 | * This code can be called with a cpos larger than the tree, in which | |
1780 | * case it will return the rightmost path. | |
1781 | */ | |
facdb77f | 1782 | static int __ocfs2_find_path(struct ocfs2_caching_info *ci, |
dcd0538f MF |
1783 | struct ocfs2_extent_list *root_el, u32 cpos, |
1784 | path_insert_t *func, void *data) | |
1785 | { | |
1786 | int i, ret = 0; | |
1787 | u32 range; | |
1788 | u64 blkno; | |
1789 | struct buffer_head *bh = NULL; | |
1790 | struct ocfs2_extent_block *eb; | |
1791 | struct ocfs2_extent_list *el; | |
1792 | struct ocfs2_extent_rec *rec; | |
dcd0538f MF |
1793 | |
1794 | el = root_el; | |
1795 | while (el->l_tree_depth) { | |
1796 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
facdb77f JB |
1797 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1798 | "Owner %llu has empty extent list at " | |
dcd0538f | 1799 | "depth %u\n", |
facdb77f | 1800 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1801 | le16_to_cpu(el->l_tree_depth)); |
1802 | ret = -EROFS; | |
1803 | goto out; | |
1804 | ||
1805 | } | |
1806 | ||
1807 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1808 | rec = &el->l_recs[i]; | |
1809 | ||
1810 | /* | |
1811 | * In the case that cpos is off the allocation | |
1812 | * tree, this should just wind up returning the | |
1813 | * rightmost record. | |
1814 | */ | |
1815 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1816 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1817 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1818 | break; | |
1819 | } | |
1820 | ||
1821 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1822 | if (blkno == 0) { | |
facdb77f JB |
1823 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1824 | "Owner %llu has bad blkno in extent list " | |
dcd0538f | 1825 | "at depth %u (index %d)\n", |
facdb77f | 1826 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1827 | le16_to_cpu(el->l_tree_depth), i); |
1828 | ret = -EROFS; | |
1829 | goto out; | |
1830 | } | |
1831 | ||
1832 | brelse(bh); | |
1833 | bh = NULL; | |
facdb77f | 1834 | ret = ocfs2_read_extent_block(ci, blkno, &bh); |
dcd0538f MF |
1835 | if (ret) { |
1836 | mlog_errno(ret); | |
1837 | goto out; | |
1838 | } | |
1839 | ||
1840 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1841 | el = &eb->h_list; | |
dcd0538f MF |
1842 | |
1843 | if (le16_to_cpu(el->l_next_free_rec) > | |
1844 | le16_to_cpu(el->l_count)) { | |
facdb77f JB |
1845 | ocfs2_error(ocfs2_metadata_cache_get_super(ci), |
1846 | "Owner %llu has bad count in extent list " | |
dcd0538f | 1847 | "at block %llu (next free=%u, count=%u)\n", |
facdb77f | 1848 | (unsigned long long)ocfs2_metadata_cache_owner(ci), |
dcd0538f MF |
1849 | (unsigned long long)bh->b_blocknr, |
1850 | le16_to_cpu(el->l_next_free_rec), | |
1851 | le16_to_cpu(el->l_count)); | |
1852 | ret = -EROFS; | |
1853 | goto out; | |
1854 | } | |
1855 | ||
1856 | if (func) | |
1857 | func(data, bh); | |
1858 | } | |
1859 | ||
1860 | out: | |
1861 | /* | |
1862 | * Catch any trailing bh that the loop didn't handle. | |
1863 | */ | |
1864 | brelse(bh); | |
1865 | ||
1866 | return ret; | |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * Given an initialized path (that is, it has a valid root extent | |
1871 | * list), this function will traverse the btree in search of the path | |
1872 | * which would contain cpos. | |
1873 | * | |
1874 | * The path traveled is recorded in the path structure. | |
1875 | * | |
1876 | * Note that this will not do any comparisons on leaf node extent | |
1877 | * records, so it will work fine in the case that we just added a tree | |
1878 | * branch. | |
1879 | */ | |
1880 | struct find_path_data { | |
1881 | int index; | |
1882 | struct ocfs2_path *path; | |
1883 | }; | |
1884 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1885 | { | |
1886 | struct find_path_data *fp = data; | |
1887 | ||
1888 | get_bh(bh); | |
1889 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1890 | fp->index++; | |
1891 | } | |
e2e9f608 TM |
1892 | int ocfs2_find_path(struct ocfs2_caching_info *ci, |
1893 | struct ocfs2_path *path, u32 cpos) | |
dcd0538f MF |
1894 | { |
1895 | struct find_path_data data; | |
1896 | ||
1897 | data.index = 1; | |
1898 | data.path = path; | |
facdb77f | 1899 | return __ocfs2_find_path(ci, path_root_el(path), cpos, |
dcd0538f MF |
1900 | find_path_ins, &data); |
1901 | } | |
1902 | ||
1903 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1904 | { | |
1905 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1906 | struct ocfs2_extent_list *el = &eb->h_list; | |
1907 | struct buffer_head **ret = data; | |
1908 | ||
1909 | /* We want to retain only the leaf block. */ | |
1910 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1911 | get_bh(bh); | |
1912 | *ret = bh; | |
1913 | } | |
1914 | } | |
1915 | /* | |
1916 | * Find the leaf block in the tree which would contain cpos. No | |
1917 | * checking of the actual leaf is done. | |
1918 | * | |
1919 | * Some paths want to call this instead of allocating a path structure | |
1920 | * and calling ocfs2_find_path(). | |
1921 | * | |
1922 | * This function doesn't handle non btree extent lists. | |
1923 | */ | |
facdb77f JB |
1924 | int ocfs2_find_leaf(struct ocfs2_caching_info *ci, |
1925 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1926 | struct buffer_head **leaf_bh) | |
dcd0538f MF |
1927 | { |
1928 | int ret; | |
1929 | struct buffer_head *bh = NULL; | |
1930 | ||
facdb77f | 1931 | ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh); |
dcd0538f MF |
1932 | if (ret) { |
1933 | mlog_errno(ret); | |
1934 | goto out; | |
1935 | } | |
1936 | ||
1937 | *leaf_bh = bh; | |
1938 | out: | |
1939 | return ret; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1944 | * | |
1945 | * Basically, we've moved stuff around at the bottom of the tree and | |
1946 | * we need to fix up the extent records above the changes to reflect | |
1947 | * the new changes. | |
1948 | * | |
1949 | * left_rec: the record on the left. | |
1950 | * left_child_el: is the child list pointed to by left_rec | |
1951 | * right_rec: the record to the right of left_rec | |
1952 | * right_child_el: is the child list pointed to by right_rec | |
1953 | * | |
1954 | * By definition, this only works on interior nodes. | |
1955 | */ | |
1956 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1957 | struct ocfs2_extent_list *left_child_el, | |
1958 | struct ocfs2_extent_rec *right_rec, | |
1959 | struct ocfs2_extent_list *right_child_el) | |
1960 | { | |
1961 | u32 left_clusters, right_end; | |
1962 | ||
1963 | /* | |
1964 | * Interior nodes never have holes. Their cpos is the cpos of | |
1965 | * the leftmost record in their child list. Their cluster | |
1966 | * count covers the full theoretical range of their child list | |
1967 | * - the range between their cpos and the cpos of the record | |
1968 | * immediately to their right. | |
1969 | */ | |
1970 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
82e12644 TM |
1971 | if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) { |
1972 | BUG_ON(right_child_el->l_tree_depth); | |
328d5752 MF |
1973 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); |
1974 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1975 | } | |
dcd0538f | 1976 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1977 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1978 | |
1979 | /* | |
1980 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1981 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1982 | * updating e_cpos to keep the same highest cluster count. |
1983 | */ | |
1984 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1985 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1986 | |
1987 | right_rec->e_cpos = left_rec->e_cpos; | |
1988 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1989 | ||
1990 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1991 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1992 | } |
1993 | ||
1994 | /* | |
1995 | * Adjust the adjacent root node records involved in a | |
1996 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1997 | * find it's index in the root list. | |
1998 | */ | |
1999 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
2000 | struct ocfs2_extent_list *left_el, | |
2001 | struct ocfs2_extent_list *right_el, | |
2002 | u64 left_el_blkno) | |
2003 | { | |
2004 | int i; | |
2005 | ||
2006 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
2007 | le16_to_cpu(left_el->l_tree_depth)); | |
2008 | ||
2009 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
2010 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
2011 | break; | |
2012 | } | |
2013 | ||
2014 | /* | |
2015 | * The path walking code should have never returned a root and | |
2016 | * two paths which are not adjacent. | |
2017 | */ | |
2018 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
2019 | ||
2020 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
2021 | &root_el->l_recs[i + 1], right_el); | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * We've changed a leaf block (in right_path) and need to reflect that | |
2026 | * change back up the subtree. | |
2027 | * | |
2028 | * This happens in multiple places: | |
2029 | * - When we've moved an extent record from the left path leaf to the right | |
2030 | * path leaf to make room for an empty extent in the left path leaf. | |
2031 | * - When our insert into the right path leaf is at the leftmost edge | |
2032 | * and requires an update of the path immediately to it's left. This | |
2033 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
2034 | * - When we've adjusted the last extent record in the left path leaf and the |
2035 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f | 2036 | */ |
4619c73e | 2037 | static void ocfs2_complete_edge_insert(handle_t *handle, |
dcd0538f MF |
2038 | struct ocfs2_path *left_path, |
2039 | struct ocfs2_path *right_path, | |
2040 | int subtree_index) | |
2041 | { | |
ec20cec7 | 2042 | int i, idx; |
dcd0538f MF |
2043 | struct ocfs2_extent_list *el, *left_el, *right_el; |
2044 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
2045 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2046 | ||
2047 | /* | |
2048 | * Update the counts and position values within all the | |
2049 | * interior nodes to reflect the leaf rotation we just did. | |
2050 | * | |
2051 | * The root node is handled below the loop. | |
2052 | * | |
2053 | * We begin the loop with right_el and left_el pointing to the | |
2054 | * leaf lists and work our way up. | |
2055 | * | |
2056 | * NOTE: within this loop, left_el and right_el always refer | |
2057 | * to the *child* lists. | |
2058 | */ | |
2059 | left_el = path_leaf_el(left_path); | |
2060 | right_el = path_leaf_el(right_path); | |
2061 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
2062 | mlog(0, "Adjust records at index %u\n", i); | |
2063 | ||
2064 | /* | |
2065 | * One nice property of knowing that all of these | |
2066 | * nodes are below the root is that we only deal with | |
2067 | * the leftmost right node record and the rightmost | |
2068 | * left node record. | |
2069 | */ | |
2070 | el = left_path->p_node[i].el; | |
2071 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2072 | left_rec = &el->l_recs[idx]; | |
2073 | ||
2074 | el = right_path->p_node[i].el; | |
2075 | right_rec = &el->l_recs[0]; | |
2076 | ||
2077 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
2078 | right_el); | |
2079 | ||
ec20cec7 JB |
2080 | ocfs2_journal_dirty(handle, left_path->p_node[i].bh); |
2081 | ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
dcd0538f MF |
2082 | |
2083 | /* | |
2084 | * Setup our list pointers now so that the current | |
2085 | * parents become children in the next iteration. | |
2086 | */ | |
2087 | left_el = left_path->p_node[i].el; | |
2088 | right_el = right_path->p_node[i].el; | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * At the root node, adjust the two adjacent records which | |
2093 | * begin our path to the leaves. | |
2094 | */ | |
2095 | ||
2096 | el = left_path->p_node[subtree_index].el; | |
2097 | left_el = left_path->p_node[subtree_index + 1].el; | |
2098 | right_el = right_path->p_node[subtree_index + 1].el; | |
2099 | ||
2100 | ocfs2_adjust_root_records(el, left_el, right_el, | |
2101 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
2102 | ||
2103 | root_bh = left_path->p_node[subtree_index].bh; | |
2104 | ||
ec20cec7 | 2105 | ocfs2_journal_dirty(handle, root_bh); |
dcd0538f MF |
2106 | } |
2107 | ||
5c601aba JB |
2108 | static int ocfs2_rotate_subtree_right(handle_t *handle, |
2109 | struct ocfs2_extent_tree *et, | |
dcd0538f MF |
2110 | struct ocfs2_path *left_path, |
2111 | struct ocfs2_path *right_path, | |
2112 | int subtree_index) | |
2113 | { | |
2114 | int ret, i; | |
2115 | struct buffer_head *right_leaf_bh; | |
2116 | struct buffer_head *left_leaf_bh = NULL; | |
2117 | struct buffer_head *root_bh; | |
2118 | struct ocfs2_extent_list *right_el, *left_el; | |
2119 | struct ocfs2_extent_rec move_rec; | |
2120 | ||
2121 | left_leaf_bh = path_leaf_bh(left_path); | |
2122 | left_el = path_leaf_el(left_path); | |
2123 | ||
2124 | if (left_el->l_next_free_rec != left_el->l_count) { | |
5c601aba | 2125 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
dcd0538f MF |
2126 | "Inode %llu has non-full interior leaf node %llu" |
2127 | "(next free = %u)", | |
5c601aba | 2128 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2129 | (unsigned long long)left_leaf_bh->b_blocknr, |
2130 | le16_to_cpu(left_el->l_next_free_rec)); | |
2131 | return -EROFS; | |
2132 | } | |
2133 | ||
2134 | /* | |
2135 | * This extent block may already have an empty record, so we | |
2136 | * return early if so. | |
2137 | */ | |
2138 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
2139 | return 0; | |
2140 | ||
2141 | root_bh = left_path->p_node[subtree_index].bh; | |
2142 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2143 | ||
5c601aba | 2144 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2145 | subtree_index); |
dcd0538f MF |
2146 | if (ret) { |
2147 | mlog_errno(ret); | |
2148 | goto out; | |
2149 | } | |
2150 | ||
2151 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
5c601aba | 2152 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2153 | right_path, i); |
dcd0538f MF |
2154 | if (ret) { |
2155 | mlog_errno(ret); | |
2156 | goto out; | |
2157 | } | |
2158 | ||
5c601aba | 2159 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2160 | left_path, i); |
dcd0538f MF |
2161 | if (ret) { |
2162 | mlog_errno(ret); | |
2163 | goto out; | |
2164 | } | |
2165 | } | |
2166 | ||
2167 | right_leaf_bh = path_leaf_bh(right_path); | |
2168 | right_el = path_leaf_el(right_path); | |
2169 | ||
2170 | /* This is a code error, not a disk corruption. */ | |
2171 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
2172 | "because rightmost leaf block %llu is empty\n", | |
5c601aba | 2173 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
dcd0538f MF |
2174 | (unsigned long long)right_leaf_bh->b_blocknr); |
2175 | ||
2176 | ocfs2_create_empty_extent(right_el); | |
2177 | ||
ec20cec7 | 2178 | ocfs2_journal_dirty(handle, right_leaf_bh); |
dcd0538f MF |
2179 | |
2180 | /* Do the copy now. */ | |
2181 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
2182 | move_rec = left_el->l_recs[i]; | |
2183 | right_el->l_recs[0] = move_rec; | |
2184 | ||
2185 | /* | |
2186 | * Clear out the record we just copied and shift everything | |
2187 | * over, leaving an empty extent in the left leaf. | |
2188 | * | |
2189 | * We temporarily subtract from next_free_rec so that the | |
2190 | * shift will lose the tail record (which is now defunct). | |
2191 | */ | |
2192 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
2193 | ocfs2_shift_records_right(left_el); | |
2194 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2195 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
2196 | ||
ec20cec7 | 2197 | ocfs2_journal_dirty(handle, left_leaf_bh); |
dcd0538f | 2198 | |
4619c73e JB |
2199 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
2200 | subtree_index); | |
dcd0538f MF |
2201 | |
2202 | out: | |
2203 | return ret; | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * Given a full path, determine what cpos value would return us a path | |
2208 | * containing the leaf immediately to the left of the current one. | |
2209 | * | |
2210 | * Will return zero if the path passed in is already the leftmost path. | |
2211 | */ | |
2212 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
2213 | struct ocfs2_path *path, u32 *cpos) | |
2214 | { | |
2215 | int i, j, ret = 0; | |
2216 | u64 blkno; | |
2217 | struct ocfs2_extent_list *el; | |
2218 | ||
e48edee2 MF |
2219 | BUG_ON(path->p_tree_depth == 0); |
2220 | ||
dcd0538f MF |
2221 | *cpos = 0; |
2222 | ||
2223 | blkno = path_leaf_bh(path)->b_blocknr; | |
2224 | ||
2225 | /* Start at the tree node just above the leaf and work our way up. */ | |
2226 | i = path->p_tree_depth - 1; | |
2227 | while (i >= 0) { | |
2228 | el = path->p_node[i].el; | |
2229 | ||
2230 | /* | |
2231 | * Find the extent record just before the one in our | |
2232 | * path. | |
2233 | */ | |
2234 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2235 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2236 | if (j == 0) { | |
2237 | if (i == 0) { | |
2238 | /* | |
2239 | * We've determined that the | |
2240 | * path specified is already | |
2241 | * the leftmost one - return a | |
2242 | * cpos of zero. | |
2243 | */ | |
2244 | goto out; | |
2245 | } | |
2246 | /* | |
2247 | * The leftmost record points to our | |
2248 | * leaf - we need to travel up the | |
2249 | * tree one level. | |
2250 | */ | |
2251 | goto next_node; | |
2252 | } | |
2253 | ||
2254 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
2255 | *cpos = *cpos + ocfs2_rec_clusters(el, |
2256 | &el->l_recs[j - 1]); | |
2257 | *cpos = *cpos - 1; | |
dcd0538f MF |
2258 | goto out; |
2259 | } | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * If we got here, we never found a valid node where | |
2264 | * the tree indicated one should be. | |
2265 | */ | |
2266 | ocfs2_error(sb, | |
2267 | "Invalid extent tree at extent block %llu\n", | |
2268 | (unsigned long long)blkno); | |
2269 | ret = -EROFS; | |
2270 | goto out; | |
2271 | ||
2272 | next_node: | |
2273 | blkno = path->p_node[i].bh->b_blocknr; | |
2274 | i--; | |
2275 | } | |
2276 | ||
2277 | out: | |
2278 | return ret; | |
2279 | } | |
2280 | ||
328d5752 MF |
2281 | /* |
2282 | * Extend the transaction by enough credits to complete the rotation, | |
2283 | * and still leave at least the original number of credits allocated | |
2284 | * to this transaction. | |
2285 | */ | |
dcd0538f | 2286 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 2287 | int op_credits, |
dcd0538f MF |
2288 | struct ocfs2_path *path) |
2289 | { | |
c901fb00 | 2290 | int ret = 0; |
328d5752 | 2291 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f | 2292 | |
c901fb00 | 2293 | if (handle->h_buffer_credits < credits) |
c18b812d TM |
2294 | ret = ocfs2_extend_trans(handle, |
2295 | credits - handle->h_buffer_credits); | |
dcd0538f | 2296 | |
c901fb00 | 2297 | return ret; |
dcd0538f MF |
2298 | } |
2299 | ||
2300 | /* | |
2301 | * Trap the case where we're inserting into the theoretical range past | |
2302 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
2303 | * whose cpos is less than ours into the right leaf. | |
2304 | * | |
2305 | * It's only necessary to look at the rightmost record of the left | |
2306 | * leaf because the logic that calls us should ensure that the | |
2307 | * theoretical ranges in the path components above the leaves are | |
2308 | * correct. | |
2309 | */ | |
2310 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
2311 | u32 insert_cpos) | |
2312 | { | |
2313 | struct ocfs2_extent_list *left_el; | |
2314 | struct ocfs2_extent_rec *rec; | |
2315 | int next_free; | |
2316 | ||
2317 | left_el = path_leaf_el(left_path); | |
2318 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
2319 | rec = &left_el->l_recs[next_free - 1]; | |
2320 | ||
2321 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
2322 | return 1; | |
2323 | return 0; | |
2324 | } | |
2325 | ||
328d5752 MF |
2326 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
2327 | { | |
2328 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2329 | unsigned int range; | |
2330 | struct ocfs2_extent_rec *rec; | |
2331 | ||
2332 | if (next_free == 0) | |
2333 | return 0; | |
2334 | ||
2335 | rec = &el->l_recs[0]; | |
2336 | if (ocfs2_is_empty_extent(rec)) { | |
2337 | /* Empty list. */ | |
2338 | if (next_free == 1) | |
2339 | return 0; | |
2340 | rec = &el->l_recs[1]; | |
2341 | } | |
2342 | ||
2343 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2344 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2345 | return 1; | |
2346 | return 0; | |
2347 | } | |
2348 | ||
dcd0538f MF |
2349 | /* |
2350 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2351 | * | |
2352 | * The path to the rightmost leaf should be passed in. | |
2353 | * | |
2354 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2355 | * | |
af901ca1 | 2356 | * Upon successful return from this function: |
dcd0538f MF |
2357 | * |
2358 | * - The 'right_path' array will contain a path to the leaf block | |
2359 | * whose range contains e_cpos. | |
2360 | * - That leaf block will have a single empty extent in list index 0. | |
2361 | * - In the case that the rotation requires a post-insert update, | |
2362 | * *ret_left_path will contain a valid path which can be passed to | |
2363 | * ocfs2_insert_path(). | |
2364 | */ | |
1bbf0b8d | 2365 | static int ocfs2_rotate_tree_right(handle_t *handle, |
5c601aba | 2366 | struct ocfs2_extent_tree *et, |
328d5752 | 2367 | enum ocfs2_split_type split, |
dcd0538f MF |
2368 | u32 insert_cpos, |
2369 | struct ocfs2_path *right_path, | |
2370 | struct ocfs2_path **ret_left_path) | |
2371 | { | |
328d5752 | 2372 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2373 | u32 cpos; |
2374 | struct ocfs2_path *left_path = NULL; | |
5c601aba | 2375 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
dcd0538f MF |
2376 | |
2377 | *ret_left_path = NULL; | |
2378 | ||
ffdd7a54 | 2379 | left_path = ocfs2_new_path_from_path(right_path); |
dcd0538f MF |
2380 | if (!left_path) { |
2381 | ret = -ENOMEM; | |
2382 | mlog_errno(ret); | |
2383 | goto out; | |
2384 | } | |
2385 | ||
5c601aba | 2386 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2387 | if (ret) { |
2388 | mlog_errno(ret); | |
2389 | goto out; | |
2390 | } | |
2391 | ||
2392 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2393 | ||
2394 | /* | |
2395 | * What we want to do here is: | |
2396 | * | |
2397 | * 1) Start with the rightmost path. | |
2398 | * | |
2399 | * 2) Determine a path to the leaf block directly to the left | |
2400 | * of that leaf. | |
2401 | * | |
2402 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2403 | * which contains a path to both leaves. | |
2404 | * | |
2405 | * 4) Rotate the subtree. | |
2406 | * | |
2407 | * 5) Find the next subtree by considering the left path to be | |
2408 | * the new right path. | |
2409 | * | |
2410 | * The check at the top of this while loop also accepts | |
2411 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2412 | * value to get us the left path - insert_cpos might very well | |
2413 | * be filling that hole. | |
2414 | * | |
2415 | * Stop at a cpos of '0' because we either started at the | |
2416 | * leftmost branch (i.e., a tree with one branch and a | |
2417 | * rotation inside of it), or we've gone as far as we can in | |
2418 | * rotating subtrees. | |
2419 | */ | |
2420 | while (cpos && insert_cpos <= cpos) { | |
2421 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2422 | insert_cpos, cpos); | |
2423 | ||
5c601aba | 2424 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
dcd0538f MF |
2425 | if (ret) { |
2426 | mlog_errno(ret); | |
2427 | goto out; | |
2428 | } | |
2429 | ||
2430 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2431 | path_leaf_bh(right_path), | |
5c601aba | 2432 | "Owner %llu: error during insert of %u " |
dcd0538f MF |
2433 | "(left path cpos %u) results in two identical " |
2434 | "paths ending at %llu\n", | |
5c601aba JB |
2435 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
2436 | insert_cpos, cpos, | |
dcd0538f MF |
2437 | (unsigned long long) |
2438 | path_leaf_bh(left_path)->b_blocknr); | |
2439 | ||
328d5752 MF |
2440 | if (split == SPLIT_NONE && |
2441 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2442 | insert_cpos)) { |
dcd0538f MF |
2443 | |
2444 | /* | |
2445 | * We've rotated the tree as much as we | |
2446 | * should. The rest is up to | |
2447 | * ocfs2_insert_path() to complete, after the | |
2448 | * record insertion. We indicate this | |
2449 | * situation by returning the left path. | |
2450 | * | |
2451 | * The reason we don't adjust the records here | |
2452 | * before the record insert is that an error | |
2453 | * later might break the rule where a parent | |
2454 | * record e_cpos will reflect the actual | |
2455 | * e_cpos of the 1st nonempty record of the | |
2456 | * child list. | |
2457 | */ | |
2458 | *ret_left_path = left_path; | |
2459 | goto out_ret_path; | |
2460 | } | |
2461 | ||
7dc02805 | 2462 | start = ocfs2_find_subtree_root(et, left_path, right_path); |
dcd0538f MF |
2463 | |
2464 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2465 | start, | |
2466 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2467 | right_path->p_tree_depth); | |
2468 | ||
2469 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2470 | orig_credits, right_path); |
dcd0538f MF |
2471 | if (ret) { |
2472 | mlog_errno(ret); | |
2473 | goto out; | |
2474 | } | |
2475 | ||
5c601aba | 2476 | ret = ocfs2_rotate_subtree_right(handle, et, left_path, |
dcd0538f MF |
2477 | right_path, start); |
2478 | if (ret) { | |
2479 | mlog_errno(ret); | |
2480 | goto out; | |
2481 | } | |
2482 | ||
328d5752 MF |
2483 | if (split != SPLIT_NONE && |
2484 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2485 | insert_cpos)) { | |
2486 | /* | |
2487 | * A rotate moves the rightmost left leaf | |
2488 | * record over to the leftmost right leaf | |
2489 | * slot. If we're doing an extent split | |
2490 | * instead of a real insert, then we have to | |
2491 | * check that the extent to be split wasn't | |
2492 | * just moved over. If it was, then we can | |
2493 | * exit here, passing left_path back - | |
2494 | * ocfs2_split_extent() is smart enough to | |
2495 | * search both leaves. | |
2496 | */ | |
2497 | *ret_left_path = left_path; | |
2498 | goto out_ret_path; | |
2499 | } | |
2500 | ||
dcd0538f MF |
2501 | /* |
2502 | * There is no need to re-read the next right path | |
2503 | * as we know that it'll be our current left | |
2504 | * path. Optimize by copying values instead. | |
2505 | */ | |
2506 | ocfs2_mv_path(right_path, left_path); | |
2507 | ||
5c601aba | 2508 | ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos); |
dcd0538f MF |
2509 | if (ret) { |
2510 | mlog_errno(ret); | |
2511 | goto out; | |
2512 | } | |
2513 | } | |
2514 | ||
2515 | out: | |
2516 | ocfs2_free_path(left_path); | |
2517 | ||
2518 | out_ret_path: | |
2519 | return ret; | |
2520 | } | |
2521 | ||
09106bae JB |
2522 | static int ocfs2_update_edge_lengths(handle_t *handle, |
2523 | struct ocfs2_extent_tree *et, | |
3c5e1068 | 2524 | int subtree_index, struct ocfs2_path *path) |
dcd0538f | 2525 | { |
3c5e1068 | 2526 | int i, idx, ret; |
dcd0538f | 2527 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2528 | struct ocfs2_extent_list *el; |
2529 | struct ocfs2_extent_block *eb; | |
2530 | u32 range; | |
dcd0538f | 2531 | |
3c5e1068 TM |
2532 | /* |
2533 | * In normal tree rotation process, we will never touch the | |
2534 | * tree branch above subtree_index and ocfs2_extend_rotate_transaction | |
2535 | * doesn't reserve the credits for them either. | |
2536 | * | |
2537 | * But we do have a special case here which will update the rightmost | |
2538 | * records for all the bh in the path. | |
2539 | * So we have to allocate extra credits and access them. | |
2540 | */ | |
c901fb00 | 2541 | ret = ocfs2_extend_trans(handle, subtree_index); |
3c5e1068 TM |
2542 | if (ret) { |
2543 | mlog_errno(ret); | |
2544 | goto out; | |
2545 | } | |
2546 | ||
09106bae | 2547 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
3c5e1068 TM |
2548 | if (ret) { |
2549 | mlog_errno(ret); | |
2550 | goto out; | |
2551 | } | |
2552 | ||
328d5752 MF |
2553 | /* Path should always be rightmost. */ |
2554 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2555 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2556 | |
328d5752 MF |
2557 | el = &eb->h_list; |
2558 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2559 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2560 | rec = &el->l_recs[idx]; | |
2561 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2562 | |
328d5752 MF |
2563 | for (i = 0; i < path->p_tree_depth; i++) { |
2564 | el = path->p_node[i].el; | |
2565 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2566 | rec = &el->l_recs[idx]; | |
dcd0538f | 2567 | |
328d5752 MF |
2568 | rec->e_int_clusters = cpu_to_le32(range); |
2569 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2570 | |
328d5752 | 2571 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2572 | } |
3c5e1068 TM |
2573 | out: |
2574 | return ret; | |
dcd0538f MF |
2575 | } |
2576 | ||
6641b0ce JB |
2577 | static void ocfs2_unlink_path(handle_t *handle, |
2578 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2579 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2580 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2581 | { |
328d5752 MF |
2582 | int ret, i; |
2583 | struct ocfs2_extent_block *eb; | |
2584 | struct ocfs2_extent_list *el; | |
2585 | struct buffer_head *bh; | |
2586 | ||
2587 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2588 | bh = path->p_node[i].bh; | |
2589 | ||
2590 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2591 | /* | |
2592 | * Not all nodes might have had their final count | |
2593 | * decremented by the caller - handle this here. | |
2594 | */ | |
2595 | el = &eb->h_list; | |
2596 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2597 | mlog(ML_ERROR, | |
2598 | "Inode %llu, attempted to remove extent block " | |
2599 | "%llu with %u records\n", | |
6641b0ce | 2600 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
2601 | (unsigned long long)le64_to_cpu(eb->h_blkno), |
2602 | le16_to_cpu(el->l_next_free_rec)); | |
2603 | ||
2604 | ocfs2_journal_dirty(handle, bh); | |
6641b0ce | 2605 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 MF |
2606 | continue; |
2607 | } | |
2608 | ||
2609 | el->l_next_free_rec = 0; | |
2610 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2611 | ||
2612 | ocfs2_journal_dirty(handle, bh); | |
2613 | ||
2614 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2615 | if (ret) | |
2616 | mlog_errno(ret); | |
2617 | ||
6641b0ce | 2618 | ocfs2_remove_from_cache(et->et_ci, bh); |
328d5752 | 2619 | } |
dcd0538f MF |
2620 | } |
2621 | ||
6641b0ce JB |
2622 | static void ocfs2_unlink_subtree(handle_t *handle, |
2623 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2624 | struct ocfs2_path *left_path, |
2625 | struct ocfs2_path *right_path, | |
2626 | int subtree_index, | |
2627 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2628 | { |
328d5752 MF |
2629 | int i; |
2630 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2631 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2632 | struct ocfs2_extent_list *el; |
328d5752 | 2633 | struct ocfs2_extent_block *eb; |
dcd0538f | 2634 | |
328d5752 | 2635 | el = path_leaf_el(left_path); |
dcd0538f | 2636 | |
328d5752 | 2637 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2638 | |
328d5752 MF |
2639 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2640 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2641 | break; | |
dcd0538f | 2642 | |
328d5752 | 2643 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2644 | |
328d5752 MF |
2645 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2646 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2647 | ||
2648 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2649 | eb->h_next_leaf_blk = 0; | |
2650 | ||
2651 | ocfs2_journal_dirty(handle, root_bh); | |
2652 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2653 | ||
6641b0ce | 2654 | ocfs2_unlink_path(handle, et, dealloc, right_path, |
328d5752 MF |
2655 | subtree_index + 1); |
2656 | } | |
2657 | ||
1e2dd63f JB |
2658 | static int ocfs2_rotate_subtree_left(handle_t *handle, |
2659 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
2660 | struct ocfs2_path *left_path, |
2661 | struct ocfs2_path *right_path, | |
2662 | int subtree_index, | |
2663 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
1e2dd63f | 2664 | int *deleted) |
328d5752 MF |
2665 | { |
2666 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2667 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2668 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2669 | struct ocfs2_extent_block *eb; | |
2670 | ||
2671 | *deleted = 0; | |
2672 | ||
2673 | right_leaf_el = path_leaf_el(right_path); | |
2674 | left_leaf_el = path_leaf_el(left_path); | |
2675 | root_bh = left_path->p_node[subtree_index].bh; | |
2676 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2677 | ||
2678 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2679 | return 0; | |
dcd0538f | 2680 | |
328d5752 MF |
2681 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2682 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2683 | /* |
328d5752 MF |
2684 | * It's legal for us to proceed if the right leaf is |
2685 | * the rightmost one and it has an empty extent. There | |
2686 | * are two cases to handle - whether the leaf will be | |
2687 | * empty after removal or not. If the leaf isn't empty | |
2688 | * then just remove the empty extent up front. The | |
2689 | * next block will handle empty leaves by flagging | |
2690 | * them for unlink. | |
2691 | * | |
2692 | * Non rightmost leaves will throw -EAGAIN and the | |
2693 | * caller can manually move the subtree and retry. | |
dcd0538f | 2694 | */ |
dcd0538f | 2695 | |
328d5752 MF |
2696 | if (eb->h_next_leaf_blk != 0ULL) |
2697 | return -EAGAIN; | |
2698 | ||
2699 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
1e2dd63f | 2700 | ret = ocfs2_journal_access_eb(handle, et->et_ci, |
13723d00 JB |
2701 | path_leaf_bh(right_path), |
2702 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2703 | if (ret) { |
2704 | mlog_errno(ret); | |
2705 | goto out; | |
2706 | } | |
2707 | ||
328d5752 MF |
2708 | ocfs2_remove_empty_extent(right_leaf_el); |
2709 | } else | |
2710 | right_has_empty = 1; | |
dcd0538f MF |
2711 | } |
2712 | ||
328d5752 MF |
2713 | if (eb->h_next_leaf_blk == 0ULL && |
2714 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2715 | /* | |
2716 | * We have to update i_last_eb_blk during the meta | |
2717 | * data delete. | |
2718 | */ | |
d9a0a1f8 | 2719 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 2720 | OCFS2_JOURNAL_ACCESS_WRITE); |
328d5752 MF |
2721 | if (ret) { |
2722 | mlog_errno(ret); | |
2723 | goto out; | |
2724 | } | |
2725 | ||
2726 | del_right_subtree = 1; | |
2727 | } | |
2728 | ||
2729 | /* | |
2730 | * Getting here with an empty extent in the right path implies | |
2731 | * that it's the rightmost path and will be deleted. | |
2732 | */ | |
2733 | BUG_ON(right_has_empty && !del_right_subtree); | |
2734 | ||
1e2dd63f | 2735 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 2736 | subtree_index); |
328d5752 MF |
2737 | if (ret) { |
2738 | mlog_errno(ret); | |
2739 | goto out; | |
2740 | } | |
2741 | ||
2742 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1e2dd63f | 2743 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2744 | right_path, i); |
328d5752 MF |
2745 | if (ret) { |
2746 | mlog_errno(ret); | |
2747 | goto out; | |
2748 | } | |
2749 | ||
1e2dd63f | 2750 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2751 | left_path, i); |
328d5752 MF |
2752 | if (ret) { |
2753 | mlog_errno(ret); | |
2754 | goto out; | |
2755 | } | |
2756 | } | |
2757 | ||
2758 | if (!right_has_empty) { | |
2759 | /* | |
2760 | * Only do this if we're moving a real | |
2761 | * record. Otherwise, the action is delayed until | |
2762 | * after removal of the right path in which case we | |
2763 | * can do a simple shift to remove the empty extent. | |
2764 | */ | |
2765 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2766 | memset(&right_leaf_el->l_recs[0], 0, | |
2767 | sizeof(struct ocfs2_extent_rec)); | |
2768 | } | |
2769 | if (eb->h_next_leaf_blk == 0ULL) { | |
2770 | /* | |
2771 | * Move recs over to get rid of empty extent, decrease | |
2772 | * next_free. This is allowed to remove the last | |
2773 | * extent in our leaf (setting l_next_free_rec to | |
2774 | * zero) - the delete code below won't care. | |
2775 | */ | |
2776 | ocfs2_remove_empty_extent(right_leaf_el); | |
2777 | } | |
2778 | ||
ec20cec7 JB |
2779 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
2780 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
328d5752 MF |
2781 | |
2782 | if (del_right_subtree) { | |
6641b0ce | 2783 | ocfs2_unlink_subtree(handle, et, left_path, right_path, |
328d5752 | 2784 | subtree_index, dealloc); |
09106bae | 2785 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
2786 | left_path); |
2787 | if (ret) { | |
2788 | mlog_errno(ret); | |
2789 | goto out; | |
2790 | } | |
328d5752 MF |
2791 | |
2792 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2793 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2794 | |
2795 | /* | |
2796 | * Removal of the extent in the left leaf was skipped | |
2797 | * above so we could delete the right path | |
2798 | * 1st. | |
2799 | */ | |
2800 | if (right_has_empty) | |
2801 | ocfs2_remove_empty_extent(left_leaf_el); | |
2802 | ||
ec20cec7 | 2803 | ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2804 | |
2805 | *deleted = 1; | |
2806 | } else | |
4619c73e | 2807 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
328d5752 MF |
2808 | subtree_index); |
2809 | ||
2810 | out: | |
2811 | return ret; | |
2812 | } | |
2813 | ||
2814 | /* | |
2815 | * Given a full path, determine what cpos value would return us a path | |
2816 | * containing the leaf immediately to the right of the current one. | |
2817 | * | |
2818 | * Will return zero if the path passed in is already the rightmost path. | |
2819 | * | |
2820 | * This looks similar, but is subtly different to | |
2821 | * ocfs2_find_cpos_for_left_leaf(). | |
2822 | */ | |
38a04e43 TM |
2823 | int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, |
2824 | struct ocfs2_path *path, u32 *cpos) | |
328d5752 MF |
2825 | { |
2826 | int i, j, ret = 0; | |
2827 | u64 blkno; | |
2828 | struct ocfs2_extent_list *el; | |
2829 | ||
2830 | *cpos = 0; | |
2831 | ||
2832 | if (path->p_tree_depth == 0) | |
2833 | return 0; | |
2834 | ||
2835 | blkno = path_leaf_bh(path)->b_blocknr; | |
2836 | ||
2837 | /* Start at the tree node just above the leaf and work our way up. */ | |
2838 | i = path->p_tree_depth - 1; | |
2839 | while (i >= 0) { | |
2840 | int next_free; | |
2841 | ||
2842 | el = path->p_node[i].el; | |
2843 | ||
2844 | /* | |
2845 | * Find the extent record just after the one in our | |
2846 | * path. | |
2847 | */ | |
2848 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2849 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2850 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2851 | if (j == (next_free - 1)) { | |
2852 | if (i == 0) { | |
2853 | /* | |
2854 | * We've determined that the | |
2855 | * path specified is already | |
2856 | * the rightmost one - return a | |
2857 | * cpos of zero. | |
2858 | */ | |
2859 | goto out; | |
2860 | } | |
2861 | /* | |
2862 | * The rightmost record points to our | |
2863 | * leaf - we need to travel up the | |
2864 | * tree one level. | |
2865 | */ | |
2866 | goto next_node; | |
2867 | } | |
2868 | ||
2869 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2870 | goto out; | |
2871 | } | |
2872 | } | |
2873 | ||
2874 | /* | |
2875 | * If we got here, we never found a valid node where | |
2876 | * the tree indicated one should be. | |
2877 | */ | |
2878 | ocfs2_error(sb, | |
2879 | "Invalid extent tree at extent block %llu\n", | |
2880 | (unsigned long long)blkno); | |
2881 | ret = -EROFS; | |
2882 | goto out; | |
2883 | ||
2884 | next_node: | |
2885 | blkno = path->p_node[i].bh->b_blocknr; | |
2886 | i--; | |
2887 | } | |
2888 | ||
2889 | out: | |
2890 | return ret; | |
2891 | } | |
2892 | ||
70f18c08 JB |
2893 | static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle, |
2894 | struct ocfs2_extent_tree *et, | |
13723d00 | 2895 | struct ocfs2_path *path) |
328d5752 MF |
2896 | { |
2897 | int ret; | |
13723d00 JB |
2898 | struct buffer_head *bh = path_leaf_bh(path); |
2899 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
328d5752 MF |
2900 | |
2901 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2902 | return 0; | |
2903 | ||
70f18c08 | 2904 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
13723d00 | 2905 | path_num_items(path) - 1); |
328d5752 MF |
2906 | if (ret) { |
2907 | mlog_errno(ret); | |
2908 | goto out; | |
2909 | } | |
2910 | ||
2911 | ocfs2_remove_empty_extent(el); | |
ec20cec7 | 2912 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
2913 | |
2914 | out: | |
2915 | return ret; | |
2916 | } | |
2917 | ||
e46f74dc JB |
2918 | static int __ocfs2_rotate_tree_left(handle_t *handle, |
2919 | struct ocfs2_extent_tree *et, | |
2920 | int orig_credits, | |
328d5752 MF |
2921 | struct ocfs2_path *path, |
2922 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e46f74dc | 2923 | struct ocfs2_path **empty_extent_path) |
328d5752 MF |
2924 | { |
2925 | int ret, subtree_root, deleted; | |
2926 | u32 right_cpos; | |
2927 | struct ocfs2_path *left_path = NULL; | |
2928 | struct ocfs2_path *right_path = NULL; | |
e46f74dc | 2929 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
328d5752 MF |
2930 | |
2931 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2932 | ||
2933 | *empty_extent_path = NULL; | |
2934 | ||
e46f74dc | 2935 | ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
328d5752 MF |
2936 | if (ret) { |
2937 | mlog_errno(ret); | |
2938 | goto out; | |
2939 | } | |
2940 | ||
ffdd7a54 | 2941 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2942 | if (!left_path) { |
2943 | ret = -ENOMEM; | |
2944 | mlog_errno(ret); | |
2945 | goto out; | |
2946 | } | |
2947 | ||
2948 | ocfs2_cp_path(left_path, path); | |
2949 | ||
ffdd7a54 | 2950 | right_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
2951 | if (!right_path) { |
2952 | ret = -ENOMEM; | |
2953 | mlog_errno(ret); | |
2954 | goto out; | |
2955 | } | |
2956 | ||
2957 | while (right_cpos) { | |
facdb77f | 2958 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
328d5752 MF |
2959 | if (ret) { |
2960 | mlog_errno(ret); | |
2961 | goto out; | |
2962 | } | |
2963 | ||
7dc02805 | 2964 | subtree_root = ocfs2_find_subtree_root(et, left_path, |
328d5752 MF |
2965 | right_path); |
2966 | ||
2967 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2968 | subtree_root, | |
2969 | (unsigned long long) | |
2970 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2971 | right_path->p_tree_depth); | |
2972 | ||
2973 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2974 | orig_credits, left_path); | |
2975 | if (ret) { | |
2976 | mlog_errno(ret); | |
2977 | goto out; | |
2978 | } | |
2979 | ||
e8aed345 MF |
2980 | /* |
2981 | * Caller might still want to make changes to the | |
2982 | * tree root, so re-add it to the journal here. | |
2983 | */ | |
e46f74dc | 2984 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 2985 | left_path, 0); |
e8aed345 MF |
2986 | if (ret) { |
2987 | mlog_errno(ret); | |
2988 | goto out; | |
2989 | } | |
2990 | ||
1e2dd63f | 2991 | ret = ocfs2_rotate_subtree_left(handle, et, left_path, |
328d5752 | 2992 | right_path, subtree_root, |
1e2dd63f | 2993 | dealloc, &deleted); |
328d5752 MF |
2994 | if (ret == -EAGAIN) { |
2995 | /* | |
2996 | * The rotation has to temporarily stop due to | |
2997 | * the right subtree having an empty | |
2998 | * extent. Pass it back to the caller for a | |
2999 | * fixup. | |
3000 | */ | |
3001 | *empty_extent_path = right_path; | |
3002 | right_path = NULL; | |
3003 | goto out; | |
3004 | } | |
3005 | if (ret) { | |
3006 | mlog_errno(ret); | |
3007 | goto out; | |
3008 | } | |
3009 | ||
3010 | /* | |
3011 | * The subtree rotate might have removed records on | |
3012 | * the rightmost edge. If so, then rotation is | |
3013 | * complete. | |
3014 | */ | |
3015 | if (deleted) | |
3016 | break; | |
3017 | ||
3018 | ocfs2_mv_path(left_path, right_path); | |
3019 | ||
e46f74dc | 3020 | ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, |
328d5752 MF |
3021 | &right_cpos); |
3022 | if (ret) { | |
3023 | mlog_errno(ret); | |
3024 | goto out; | |
3025 | } | |
3026 | } | |
3027 | ||
3028 | out: | |
3029 | ocfs2_free_path(right_path); | |
3030 | ocfs2_free_path(left_path); | |
3031 | ||
3032 | return ret; | |
3033 | } | |
3034 | ||
70f18c08 JB |
3035 | static int ocfs2_remove_rightmost_path(handle_t *handle, |
3036 | struct ocfs2_extent_tree *et, | |
e7d4cb6b | 3037 | struct ocfs2_path *path, |
70f18c08 | 3038 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3039 | { |
3040 | int ret, subtree_index; | |
3041 | u32 cpos; | |
3042 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
3043 | struct ocfs2_extent_block *eb; |
3044 | struct ocfs2_extent_list *el; | |
3045 | ||
328d5752 | 3046 | |
6136ca5f | 3047 | ret = ocfs2_et_sanity_check(et); |
e7d4cb6b TM |
3048 | if (ret) |
3049 | goto out; | |
328d5752 MF |
3050 | /* |
3051 | * There's two ways we handle this depending on | |
3052 | * whether path is the only existing one. | |
3053 | */ | |
3054 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
3055 | handle->h_buffer_credits, | |
3056 | path); | |
3057 | if (ret) { | |
3058 | mlog_errno(ret); | |
3059 | goto out; | |
3060 | } | |
3061 | ||
d9a0a1f8 | 3062 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
328d5752 MF |
3063 | if (ret) { |
3064 | mlog_errno(ret); | |
3065 | goto out; | |
3066 | } | |
3067 | ||
3d03a305 JB |
3068 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3069 | path, &cpos); | |
328d5752 MF |
3070 | if (ret) { |
3071 | mlog_errno(ret); | |
3072 | goto out; | |
3073 | } | |
3074 | ||
3075 | if (cpos) { | |
3076 | /* | |
3077 | * We have a path to the left of this one - it needs | |
3078 | * an update too. | |
3079 | */ | |
ffdd7a54 | 3080 | left_path = ocfs2_new_path_from_path(path); |
328d5752 MF |
3081 | if (!left_path) { |
3082 | ret = -ENOMEM; | |
3083 | mlog_errno(ret); | |
3084 | goto out; | |
3085 | } | |
3086 | ||
facdb77f | 3087 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
3088 | if (ret) { |
3089 | mlog_errno(ret); | |
3090 | goto out; | |
3091 | } | |
3092 | ||
d9a0a1f8 | 3093 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
328d5752 MF |
3094 | if (ret) { |
3095 | mlog_errno(ret); | |
3096 | goto out; | |
3097 | } | |
3098 | ||
7dc02805 | 3099 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
328d5752 | 3100 | |
6641b0ce | 3101 | ocfs2_unlink_subtree(handle, et, left_path, path, |
328d5752 | 3102 | subtree_index, dealloc); |
09106bae | 3103 | ret = ocfs2_update_edge_lengths(handle, et, subtree_index, |
3c5e1068 TM |
3104 | left_path); |
3105 | if (ret) { | |
3106 | mlog_errno(ret); | |
3107 | goto out; | |
3108 | } | |
328d5752 MF |
3109 | |
3110 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 3111 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
3112 | } else { |
3113 | /* | |
3114 | * 'path' is also the leftmost path which | |
3115 | * means it must be the only one. This gets | |
3116 | * handled differently because we want to | |
70f18c08 | 3117 | * revert the root back to having extents |
328d5752 MF |
3118 | * in-line. |
3119 | */ | |
6641b0ce | 3120 | ocfs2_unlink_path(handle, et, dealloc, path, 1); |
328d5752 | 3121 | |
ce1d9ea6 | 3122 | el = et->et_root_el; |
328d5752 MF |
3123 | el->l_tree_depth = 0; |
3124 | el->l_next_free_rec = 0; | |
3125 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3126 | ||
35dc0aa3 | 3127 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
3128 | } |
3129 | ||
3130 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
3131 | ||
3132 | out: | |
3133 | ocfs2_free_path(left_path); | |
3134 | return ret; | |
3135 | } | |
3136 | ||
3137 | /* | |
3138 | * Left rotation of btree records. | |
3139 | * | |
3140 | * In many ways, this is (unsurprisingly) the opposite of right | |
3141 | * rotation. We start at some non-rightmost path containing an empty | |
3142 | * extent in the leaf block. The code works its way to the rightmost | |
3143 | * path by rotating records to the left in every subtree. | |
3144 | * | |
3145 | * This is used by any code which reduces the number of extent records | |
3146 | * in a leaf. After removal, an empty record should be placed in the | |
3147 | * leftmost list position. | |
3148 | * | |
3149 | * This won't handle a length update of the rightmost path records if | |
3150 | * the rightmost tree leaf record is removed so the caller is | |
3151 | * responsible for detecting and correcting that. | |
3152 | */ | |
70f18c08 JB |
3153 | static int ocfs2_rotate_tree_left(handle_t *handle, |
3154 | struct ocfs2_extent_tree *et, | |
328d5752 | 3155 | struct ocfs2_path *path, |
70f18c08 | 3156 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
328d5752 MF |
3157 | { |
3158 | int ret, orig_credits = handle->h_buffer_credits; | |
3159 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
3160 | struct ocfs2_extent_block *eb; | |
3161 | struct ocfs2_extent_list *el; | |
3162 | ||
3163 | el = path_leaf_el(path); | |
3164 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
3165 | return 0; | |
3166 | ||
3167 | if (path->p_tree_depth == 0) { | |
3168 | rightmost_no_delete: | |
3169 | /* | |
e7d4cb6b | 3170 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
3171 | * it up front. |
3172 | */ | |
70f18c08 | 3173 | ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path); |
328d5752 MF |
3174 | if (ret) |
3175 | mlog_errno(ret); | |
3176 | goto out; | |
3177 | } | |
3178 | ||
3179 | /* | |
3180 | * Handle rightmost branch now. There's several cases: | |
3181 | * 1) simple rotation leaving records in there. That's trivial. | |
3182 | * 2) rotation requiring a branch delete - there's no more | |
3183 | * records left. Two cases of this: | |
3184 | * a) There are branches to the left. | |
3185 | * b) This is also the leftmost (the only) branch. | |
3186 | * | |
3187 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
3188 | * 2a) we need the left branch so that we can update it with the unlink | |
70f18c08 | 3189 | * 2b) we need to bring the root back to inline extents. |
328d5752 MF |
3190 | */ |
3191 | ||
3192 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
3193 | el = &eb->h_list; | |
3194 | if (eb->h_next_leaf_blk == 0) { | |
3195 | /* | |
3196 | * This gets a bit tricky if we're going to delete the | |
3197 | * rightmost path. Get the other cases out of the way | |
3198 | * 1st. | |
3199 | */ | |
3200 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
3201 | goto rightmost_no_delete; | |
3202 | ||
3203 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
3204 | ret = -EIO; | |
70f18c08 JB |
3205 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3206 | "Owner %llu has empty extent block at %llu", | |
3207 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
328d5752 MF |
3208 | (unsigned long long)le64_to_cpu(eb->h_blkno)); |
3209 | goto out; | |
3210 | } | |
3211 | ||
3212 | /* | |
3213 | * XXX: The caller can not trust "path" any more after | |
3214 | * this as it will have been deleted. What do we do? | |
3215 | * | |
3216 | * In theory the rotate-for-merge code will never get | |
3217 | * here because it'll always ask for a rotate in a | |
3218 | * nonempty list. | |
3219 | */ | |
3220 | ||
70f18c08 JB |
3221 | ret = ocfs2_remove_rightmost_path(handle, et, path, |
3222 | dealloc); | |
328d5752 MF |
3223 | if (ret) |
3224 | mlog_errno(ret); | |
3225 | goto out; | |
3226 | } | |
3227 | ||
3228 | /* | |
3229 | * Now we can loop, remembering the path we get from -EAGAIN | |
3230 | * and restarting from there. | |
3231 | */ | |
3232 | try_rotate: | |
e46f74dc JB |
3233 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path, |
3234 | dealloc, &restart_path); | |
328d5752 MF |
3235 | if (ret && ret != -EAGAIN) { |
3236 | mlog_errno(ret); | |
3237 | goto out; | |
3238 | } | |
3239 | ||
3240 | while (ret == -EAGAIN) { | |
3241 | tmp_path = restart_path; | |
3242 | restart_path = NULL; | |
3243 | ||
e46f74dc | 3244 | ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, |
328d5752 | 3245 | tmp_path, dealloc, |
e46f74dc | 3246 | &restart_path); |
328d5752 MF |
3247 | if (ret && ret != -EAGAIN) { |
3248 | mlog_errno(ret); | |
3249 | goto out; | |
3250 | } | |
3251 | ||
3252 | ocfs2_free_path(tmp_path); | |
3253 | tmp_path = NULL; | |
3254 | ||
3255 | if (ret == 0) | |
3256 | goto try_rotate; | |
3257 | } | |
3258 | ||
3259 | out: | |
3260 | ocfs2_free_path(tmp_path); | |
3261 | ocfs2_free_path(restart_path); | |
3262 | return ret; | |
3263 | } | |
3264 | ||
3265 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
3266 | int index) | |
3267 | { | |
3268 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
3269 | unsigned int size; | |
3270 | ||
3271 | if (rec->e_leaf_clusters == 0) { | |
3272 | /* | |
3273 | * We consumed all of the merged-from record. An empty | |
3274 | * extent cannot exist anywhere but the 1st array | |
3275 | * position, so move things over if the merged-from | |
3276 | * record doesn't occupy that position. | |
3277 | * | |
3278 | * This creates a new empty extent so the caller | |
3279 | * should be smart enough to have removed any existing | |
3280 | * ones. | |
3281 | */ | |
3282 | if (index > 0) { | |
3283 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3284 | size = index * sizeof(struct ocfs2_extent_rec); | |
3285 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
3286 | } | |
3287 | ||
3288 | /* | |
3289 | * Always memset - the caller doesn't check whether it | |
3290 | * created an empty extent, so there could be junk in | |
3291 | * the other fields. | |
3292 | */ | |
3293 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
3294 | } | |
3295 | } | |
3296 | ||
4fe82c31 | 3297 | static int ocfs2_get_right_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3298 | struct ocfs2_path *left_path, |
3299 | struct ocfs2_path **ret_right_path) | |
3300 | { | |
3301 | int ret; | |
3302 | u32 right_cpos; | |
3303 | struct ocfs2_path *right_path = NULL; | |
3304 | struct ocfs2_extent_list *left_el; | |
3305 | ||
3306 | *ret_right_path = NULL; | |
3307 | ||
3308 | /* This function shouldn't be called for non-trees. */ | |
3309 | BUG_ON(left_path->p_tree_depth == 0); | |
3310 | ||
3311 | left_el = path_leaf_el(left_path); | |
3312 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
3313 | ||
4fe82c31 JB |
3314 | ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3315 | left_path, &right_cpos); | |
677b9752 TM |
3316 | if (ret) { |
3317 | mlog_errno(ret); | |
3318 | goto out; | |
3319 | } | |
3320 | ||
3321 | /* This function shouldn't be called for the rightmost leaf. */ | |
3322 | BUG_ON(right_cpos == 0); | |
3323 | ||
ffdd7a54 | 3324 | right_path = ocfs2_new_path_from_path(left_path); |
677b9752 TM |
3325 | if (!right_path) { |
3326 | ret = -ENOMEM; | |
3327 | mlog_errno(ret); | |
3328 | goto out; | |
3329 | } | |
3330 | ||
4fe82c31 | 3331 | ret = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
677b9752 TM |
3332 | if (ret) { |
3333 | mlog_errno(ret); | |
3334 | goto out; | |
3335 | } | |
3336 | ||
3337 | *ret_right_path = right_path; | |
3338 | out: | |
3339 | if (ret) | |
3340 | ocfs2_free_path(right_path); | |
3341 | return ret; | |
3342 | } | |
3343 | ||
328d5752 MF |
3344 | /* |
3345 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3346 | * onto the beginning of the record "next" to it. |
3347 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3348 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3349 | * next extent block. | |
328d5752 | 3350 | */ |
4fe82c31 | 3351 | static int ocfs2_merge_rec_right(struct ocfs2_path *left_path, |
677b9752 | 3352 | handle_t *handle, |
7dc02805 | 3353 | struct ocfs2_extent_tree *et, |
677b9752 TM |
3354 | struct ocfs2_extent_rec *split_rec, |
3355 | int index) | |
328d5752 | 3356 | { |
677b9752 | 3357 | int ret, next_free, i; |
328d5752 MF |
3358 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3359 | struct ocfs2_extent_rec *left_rec; | |
3360 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3361 | struct ocfs2_extent_list *right_el; |
3362 | struct ocfs2_path *right_path = NULL; | |
3363 | int subtree_index = 0; | |
3364 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3365 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3366 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3367 | |
3368 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3369 | left_rec = &el->l_recs[index]; |
677b9752 | 3370 | |
9d8df6aa | 3371 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3372 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3373 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3374 | ret = ocfs2_get_right_path(et, left_path, &right_path); |
677b9752 TM |
3375 | if (ret) { |
3376 | mlog_errno(ret); | |
3377 | goto out; | |
3378 | } | |
3379 | ||
3380 | right_el = path_leaf_el(right_path); | |
3381 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3382 | BUG_ON(next_free <= 0); | |
3383 | right_rec = &right_el->l_recs[0]; | |
3384 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3385 | BUG_ON(next_free <= 1); |
677b9752 TM |
3386 | right_rec = &right_el->l_recs[1]; |
3387 | } | |
3388 | ||
3389 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3390 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3391 | le32_to_cpu(right_rec->e_cpos)); | |
3392 | ||
7dc02805 JB |
3393 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3394 | right_path); | |
677b9752 TM |
3395 | |
3396 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3397 | handle->h_buffer_credits, | |
3398 | right_path); | |
3399 | if (ret) { | |
3400 | mlog_errno(ret); | |
3401 | goto out; | |
3402 | } | |
3403 | ||
3404 | root_bh = left_path->p_node[subtree_index].bh; | |
3405 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3406 | ||
7dc02805 | 3407 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3408 | subtree_index); |
677b9752 TM |
3409 | if (ret) { |
3410 | mlog_errno(ret); | |
3411 | goto out; | |
3412 | } | |
3413 | ||
3414 | for (i = subtree_index + 1; | |
3415 | i < path_num_items(right_path); i++) { | |
7dc02805 | 3416 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3417 | right_path, i); |
677b9752 TM |
3418 | if (ret) { |
3419 | mlog_errno(ret); | |
3420 | goto out; | |
3421 | } | |
3422 | ||
7dc02805 | 3423 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3424 | left_path, i); |
677b9752 TM |
3425 | if (ret) { |
3426 | mlog_errno(ret); | |
3427 | goto out; | |
3428 | } | |
3429 | } | |
3430 | ||
3431 | } else { | |
3432 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3433 | right_rec = &el->l_recs[index + 1]; | |
3434 | } | |
328d5752 | 3435 | |
7dc02805 | 3436 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path, |
13723d00 | 3437 | path_num_items(left_path) - 1); |
328d5752 MF |
3438 | if (ret) { |
3439 | mlog_errno(ret); | |
3440 | goto out; | |
3441 | } | |
3442 | ||
3443 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3444 | ||
3445 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3446 | le64_add_cpu(&right_rec->e_blkno, | |
7dc02805 JB |
3447 | -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3448 | split_clusters)); | |
328d5752 MF |
3449 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); |
3450 | ||
3451 | ocfs2_cleanup_merge(el, index); | |
3452 | ||
ec20cec7 | 3453 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3454 | if (right_path) { |
ec20cec7 | 3455 | ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); |
4619c73e JB |
3456 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
3457 | subtree_index); | |
677b9752 TM |
3458 | } |
3459 | out: | |
3460 | if (right_path) | |
3461 | ocfs2_free_path(right_path); | |
3462 | return ret; | |
3463 | } | |
3464 | ||
4fe82c31 | 3465 | static int ocfs2_get_left_path(struct ocfs2_extent_tree *et, |
677b9752 TM |
3466 | struct ocfs2_path *right_path, |
3467 | struct ocfs2_path **ret_left_path) | |
3468 | { | |
3469 | int ret; | |
3470 | u32 left_cpos; | |
3471 | struct ocfs2_path *left_path = NULL; | |
3472 | ||
3473 | *ret_left_path = NULL; | |
3474 | ||
3475 | /* This function shouldn't be called for non-trees. */ | |
3476 | BUG_ON(right_path->p_tree_depth == 0); | |
3477 | ||
4fe82c31 | 3478 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
677b9752 TM |
3479 | right_path, &left_cpos); |
3480 | if (ret) { | |
3481 | mlog_errno(ret); | |
3482 | goto out; | |
3483 | } | |
3484 | ||
3485 | /* This function shouldn't be called for the leftmost leaf. */ | |
3486 | BUG_ON(left_cpos == 0); | |
3487 | ||
ffdd7a54 | 3488 | left_path = ocfs2_new_path_from_path(right_path); |
677b9752 TM |
3489 | if (!left_path) { |
3490 | ret = -ENOMEM; | |
3491 | mlog_errno(ret); | |
3492 | goto out; | |
3493 | } | |
3494 | ||
4fe82c31 | 3495 | ret = ocfs2_find_path(et->et_ci, left_path, left_cpos); |
677b9752 TM |
3496 | if (ret) { |
3497 | mlog_errno(ret); | |
3498 | goto out; | |
3499 | } | |
3500 | ||
3501 | *ret_left_path = left_path; | |
328d5752 | 3502 | out: |
677b9752 TM |
3503 | if (ret) |
3504 | ocfs2_free_path(left_path); | |
328d5752 MF |
3505 | return ret; |
3506 | } | |
3507 | ||
3508 | /* | |
3509 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3510 | * onto the tail of the record "before" it. |
3511 | * For index > 0, the "before" means the extent rec at index - 1. | |
3512 | * | |
3513 | * For index == 0, the "before" means the last record of the previous | |
3514 | * extent block. And there is also a situation that we may need to | |
3515 | * remove the rightmost leaf extent block in the right_path and change | |
3516 | * the right path to indicate the new rightmost path. | |
328d5752 | 3517 | */ |
4fe82c31 | 3518 | static int ocfs2_merge_rec_left(struct ocfs2_path *right_path, |
328d5752 | 3519 | handle_t *handle, |
4fe82c31 | 3520 | struct ocfs2_extent_tree *et, |
328d5752 | 3521 | struct ocfs2_extent_rec *split_rec, |
677b9752 TM |
3522 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
3523 | int index) | |
328d5752 | 3524 | { |
677b9752 | 3525 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3526 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3527 | struct ocfs2_extent_rec *left_rec; | |
3528 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3529 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3530 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3531 | struct buffer_head *root_bh = NULL; | |
3532 | struct ocfs2_path *left_path = NULL; | |
3533 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3534 | |
677b9752 | 3535 | BUG_ON(index < 0); |
328d5752 | 3536 | |
328d5752 | 3537 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3538 | if (index == 0) { |
3539 | /* we meet with a cross extent block merge. */ | |
4fe82c31 | 3540 | ret = ocfs2_get_left_path(et, right_path, &left_path); |
677b9752 TM |
3541 | if (ret) { |
3542 | mlog_errno(ret); | |
3543 | goto out; | |
3544 | } | |
3545 | ||
3546 | left_el = path_leaf_el(left_path); | |
3547 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3548 | le16_to_cpu(left_el->l_count)); | |
3549 | ||
3550 | left_rec = &left_el->l_recs[ | |
3551 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3552 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3553 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3554 | le32_to_cpu(split_rec->e_cpos)); | |
3555 | ||
7dc02805 JB |
3556 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
3557 | right_path); | |
677b9752 TM |
3558 | |
3559 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3560 | handle->h_buffer_credits, | |
3561 | left_path); | |
3562 | if (ret) { | |
3563 | mlog_errno(ret); | |
3564 | goto out; | |
3565 | } | |
3566 | ||
3567 | root_bh = left_path->p_node[subtree_index].bh; | |
3568 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3569 | ||
4fe82c31 | 3570 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
13723d00 | 3571 | subtree_index); |
677b9752 TM |
3572 | if (ret) { |
3573 | mlog_errno(ret); | |
3574 | goto out; | |
3575 | } | |
3576 | ||
3577 | for (i = subtree_index + 1; | |
3578 | i < path_num_items(right_path); i++) { | |
4fe82c31 | 3579 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3580 | right_path, i); |
677b9752 TM |
3581 | if (ret) { |
3582 | mlog_errno(ret); | |
3583 | goto out; | |
3584 | } | |
3585 | ||
4fe82c31 | 3586 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, |
13723d00 | 3587 | left_path, i); |
677b9752 TM |
3588 | if (ret) { |
3589 | mlog_errno(ret); | |
3590 | goto out; | |
3591 | } | |
3592 | } | |
3593 | } else { | |
3594 | left_rec = &el->l_recs[index - 1]; | |
3595 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3596 | has_empty_extent = 1; | |
3597 | } | |
328d5752 | 3598 | |
4fe82c31 | 3599 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path, |
9047beab | 3600 | path_num_items(right_path) - 1); |
328d5752 MF |
3601 | if (ret) { |
3602 | mlog_errno(ret); | |
3603 | goto out; | |
3604 | } | |
3605 | ||
3606 | if (has_empty_extent && index == 1) { | |
3607 | /* | |
3608 | * The easy case - we can just plop the record right in. | |
3609 | */ | |
3610 | *left_rec = *split_rec; | |
3611 | ||
3612 | has_empty_extent = 0; | |
677b9752 | 3613 | } else |
328d5752 | 3614 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3615 | |
3616 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3617 | le64_add_cpu(&right_rec->e_blkno, | |
4fe82c31 JB |
3618 | ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci), |
3619 | split_clusters)); | |
328d5752 MF |
3620 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); |
3621 | ||
3622 | ocfs2_cleanup_merge(el, index); | |
3623 | ||
ec20cec7 | 3624 | ocfs2_journal_dirty(handle, bh); |
677b9752 | 3625 | if (left_path) { |
ec20cec7 | 3626 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); |
677b9752 TM |
3627 | |
3628 | /* | |
3629 | * In the situation that the right_rec is empty and the extent | |
3630 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3631 | * it and we need to delete the right extent block. | |
3632 | */ | |
3633 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3634 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3635 | ||
70f18c08 | 3636 | ret = ocfs2_remove_rightmost_path(handle, et, |
e7d4cb6b | 3637 | right_path, |
70f18c08 | 3638 | dealloc); |
677b9752 TM |
3639 | if (ret) { |
3640 | mlog_errno(ret); | |
3641 | goto out; | |
3642 | } | |
3643 | ||
3644 | /* Now the rightmost extent block has been deleted. | |
3645 | * So we use the new rightmost path. | |
3646 | */ | |
3647 | ocfs2_mv_path(right_path, left_path); | |
3648 | left_path = NULL; | |
3649 | } else | |
4619c73e | 3650 | ocfs2_complete_edge_insert(handle, left_path, |
677b9752 TM |
3651 | right_path, subtree_index); |
3652 | } | |
328d5752 | 3653 | out: |
677b9752 TM |
3654 | if (left_path) |
3655 | ocfs2_free_path(left_path); | |
328d5752 MF |
3656 | return ret; |
3657 | } | |
3658 | ||
c495dd24 JB |
3659 | static int ocfs2_try_to_merge_extent(handle_t *handle, |
3660 | struct ocfs2_extent_tree *et, | |
677b9752 | 3661 | struct ocfs2_path *path, |
328d5752 MF |
3662 | int split_index, |
3663 | struct ocfs2_extent_rec *split_rec, | |
3664 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
c495dd24 | 3665 | struct ocfs2_merge_ctxt *ctxt) |
328d5752 | 3666 | { |
518d7269 | 3667 | int ret = 0; |
677b9752 | 3668 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3669 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3670 | ||
3671 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3672 | ||
518d7269 TM |
3673 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3674 | /* | |
3675 | * The merge code will need to create an empty | |
3676 | * extent to take the place of the newly | |
3677 | * emptied slot. Remove any pre-existing empty | |
3678 | * extents - having more than one in a leaf is | |
3679 | * illegal. | |
3680 | */ | |
70f18c08 | 3681 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
518d7269 TM |
3682 | if (ret) { |
3683 | mlog_errno(ret); | |
3684 | goto out; | |
328d5752 | 3685 | } |
518d7269 TM |
3686 | split_index--; |
3687 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3688 | } |
3689 | ||
3690 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3691 | /* | |
3692 | * Left-right contig implies this. | |
3693 | */ | |
3694 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3695 | |
3696 | /* | |
3697 | * Since the leftright insert always covers the entire | |
3698 | * extent, this call will delete the insert record | |
3699 | * entirely, resulting in an empty extent record added to | |
3700 | * the extent block. | |
3701 | * | |
3702 | * Since the adding of an empty extent shifts | |
3703 | * everything back to the right, there's no need to | |
3704 | * update split_index here. | |
677b9752 TM |
3705 | * |
3706 | * When the split_index is zero, we need to merge it to the | |
3707 | * prevoius extent block. It is more efficient and easier | |
3708 | * if we do merge_right first and merge_left later. | |
328d5752 | 3709 | */ |
4fe82c31 | 3710 | ret = ocfs2_merge_rec_right(path, handle, et, split_rec, |
677b9752 | 3711 | split_index); |
328d5752 MF |
3712 | if (ret) { |
3713 | mlog_errno(ret); | |
3714 | goto out; | |
3715 | } | |
3716 | ||
3717 | /* | |
3718 | * We can only get this from logic error above. | |
3719 | */ | |
3720 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3721 | ||
677b9752 | 3722 | /* The merge left us with an empty extent, remove it. */ |
70f18c08 | 3723 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3724 | if (ret) { |
3725 | mlog_errno(ret); | |
3726 | goto out; | |
3727 | } | |
677b9752 | 3728 | |
328d5752 MF |
3729 | rec = &el->l_recs[split_index]; |
3730 | ||
3731 | /* | |
3732 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3733 | * we've merged it into the rec already. |
328d5752 | 3734 | */ |
4fe82c31 JB |
3735 | ret = ocfs2_merge_rec_left(path, handle, et, rec, |
3736 | dealloc, split_index); | |
677b9752 | 3737 | |
328d5752 MF |
3738 | if (ret) { |
3739 | mlog_errno(ret); | |
3740 | goto out; | |
3741 | } | |
3742 | ||
70f18c08 | 3743 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
328d5752 MF |
3744 | /* |
3745 | * Error from this last rotate is not critical, so | |
3746 | * print but don't bubble it up. | |
3747 | */ | |
3748 | if (ret) | |
3749 | mlog_errno(ret); | |
3750 | ret = 0; | |
3751 | } else { | |
3752 | /* | |
3753 | * Merge a record to the left or right. | |
3754 | * | |
3755 | * 'contig_type' is relative to the existing record, | |
3756 | * so for example, if we're "right contig", it's to | |
3757 | * the record on the left (hence the left merge). | |
3758 | */ | |
3759 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
4fe82c31 JB |
3760 | ret = ocfs2_merge_rec_left(path, handle, et, |
3761 | split_rec, dealloc, | |
328d5752 MF |
3762 | split_index); |
3763 | if (ret) { | |
3764 | mlog_errno(ret); | |
3765 | goto out; | |
3766 | } | |
3767 | } else { | |
4fe82c31 | 3768 | ret = ocfs2_merge_rec_right(path, handle, |
7dc02805 | 3769 | et, split_rec, |
328d5752 MF |
3770 | split_index); |
3771 | if (ret) { | |
3772 | mlog_errno(ret); | |
3773 | goto out; | |
3774 | } | |
3775 | } | |
3776 | ||
3777 | if (ctxt->c_split_covers_rec) { | |
3778 | /* | |
3779 | * The merge may have left an empty extent in | |
3780 | * our leaf. Try to rotate it away. | |
3781 | */ | |
70f18c08 JB |
3782 | ret = ocfs2_rotate_tree_left(handle, et, path, |
3783 | dealloc); | |
328d5752 MF |
3784 | if (ret) |
3785 | mlog_errno(ret); | |
3786 | ret = 0; | |
3787 | } | |
3788 | } | |
3789 | ||
3790 | out: | |
3791 | return ret; | |
3792 | } | |
3793 | ||
3794 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3795 | enum ocfs2_split_type split, | |
3796 | struct ocfs2_extent_rec *rec, | |
3797 | struct ocfs2_extent_rec *split_rec) | |
3798 | { | |
3799 | u64 len_blocks; | |
3800 | ||
3801 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3802 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3803 | ||
3804 | if (split == SPLIT_LEFT) { | |
3805 | /* | |
3806 | * Region is on the left edge of the existing | |
3807 | * record. | |
3808 | */ | |
3809 | le32_add_cpu(&rec->e_cpos, | |
3810 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3811 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3812 | le16_add_cpu(&rec->e_leaf_clusters, | |
3813 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3814 | } else { | |
3815 | /* | |
3816 | * Region is on the right edge of the existing | |
3817 | * record. | |
3818 | */ | |
3819 | le16_add_cpu(&rec->e_leaf_clusters, | |
3820 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3821 | } | |
3822 | } | |
3823 | ||
3824 | /* | |
3825 | * Do the final bits of extent record insertion at the target leaf | |
3826 | * list. If this leaf is part of an allocation tree, it is assumed | |
3827 | * that the tree above has been prepared. | |
3828 | */ | |
d5628623 JB |
3829 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et, |
3830 | struct ocfs2_extent_rec *insert_rec, | |
328d5752 | 3831 | struct ocfs2_extent_list *el, |
d5628623 | 3832 | struct ocfs2_insert_type *insert) |
328d5752 MF |
3833 | { |
3834 | int i = insert->ins_contig_index; | |
3835 | unsigned int range; | |
3836 | struct ocfs2_extent_rec *rec; | |
3837 | ||
3838 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3839 | ||
3840 | if (insert->ins_split != SPLIT_NONE) { | |
3841 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3842 | BUG_ON(i == -1); | |
3843 | rec = &el->l_recs[i]; | |
d5628623 JB |
3844 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
3845 | insert->ins_split, rec, | |
328d5752 MF |
3846 | insert_rec); |
3847 | goto rotate; | |
3848 | } | |
3849 | ||
3850 | /* | |
3851 | * Contiguous insert - either left or right. | |
3852 | */ | |
3853 | if (insert->ins_contig != CONTIG_NONE) { | |
3854 | rec = &el->l_recs[i]; | |
3855 | if (insert->ins_contig == CONTIG_LEFT) { | |
3856 | rec->e_blkno = insert_rec->e_blkno; | |
3857 | rec->e_cpos = insert_rec->e_cpos; | |
3858 | } | |
3859 | le16_add_cpu(&rec->e_leaf_clusters, | |
3860 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3861 | return; | |
3862 | } | |
3863 | ||
3864 | /* | |
3865 | * Handle insert into an empty leaf. | |
3866 | */ | |
3867 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3868 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3869 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3870 | el->l_recs[0] = *insert_rec; | |
3871 | el->l_next_free_rec = cpu_to_le16(1); | |
3872 | return; | |
3873 | } | |
3874 | ||
3875 | /* | |
3876 | * Appending insert. | |
3877 | */ | |
3878 | if (insert->ins_appending == APPEND_TAIL) { | |
3879 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3880 | rec = &el->l_recs[i]; | |
3881 | range = le32_to_cpu(rec->e_cpos) | |
3882 | + le16_to_cpu(rec->e_leaf_clusters); | |
3883 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3884 | ||
3885 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3886 | le16_to_cpu(el->l_count), | |
d5628623 | 3887 | "owner %llu, depth %u, count %u, next free %u, " |
328d5752 MF |
3888 | "rec.cpos %u, rec.clusters %u, " |
3889 | "insert.cpos %u, insert.clusters %u\n", | |
d5628623 | 3890 | ocfs2_metadata_cache_owner(et->et_ci), |
328d5752 MF |
3891 | le16_to_cpu(el->l_tree_depth), |
3892 | le16_to_cpu(el->l_count), | |
3893 | le16_to_cpu(el->l_next_free_rec), | |
3894 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3895 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3896 | le32_to_cpu(insert_rec->e_cpos), | |
3897 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3898 | i++; | |
3899 | el->l_recs[i] = *insert_rec; | |
3900 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3901 | return; | |
3902 | } | |
3903 | ||
3904 | rotate: | |
3905 | /* | |
3906 | * Ok, we have to rotate. | |
3907 | * | |
3908 | * At this point, it is safe to assume that inserting into an | |
3909 | * empty leaf and appending to a leaf have both been handled | |
3910 | * above. | |
3911 | * | |
3912 | * This leaf needs to have space, either by the empty 1st | |
3913 | * extent record, or by virtue of an l_next_rec < l_count. | |
3914 | */ | |
3915 | ocfs2_rotate_leaf(el, insert_rec); | |
3916 | } | |
3917 | ||
d401dc12 JB |
3918 | static void ocfs2_adjust_rightmost_records(handle_t *handle, |
3919 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3920 | struct ocfs2_path *path, |
3921 | struct ocfs2_extent_rec *insert_rec) | |
3922 | { | |
3923 | int ret, i, next_free; | |
3924 | struct buffer_head *bh; | |
3925 | struct ocfs2_extent_list *el; | |
3926 | struct ocfs2_extent_rec *rec; | |
3927 | ||
3928 | /* | |
3929 | * Update everything except the leaf block. | |
3930 | */ | |
3931 | for (i = 0; i < path->p_tree_depth; i++) { | |
3932 | bh = path->p_node[i].bh; | |
3933 | el = path->p_node[i].el; | |
3934 | ||
dcd0538f MF |
3935 | next_free = le16_to_cpu(el->l_next_free_rec); |
3936 | if (next_free == 0) { | |
d401dc12 JB |
3937 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
3938 | "Owner %llu has a bad extent list", | |
3939 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 3940 | ret = -EIO; |
328d5752 MF |
3941 | return; |
3942 | } | |
3943 | ||
3944 | rec = &el->l_recs[next_free - 1]; | |
3945 | ||
3946 | rec->e_int_clusters = insert_rec->e_cpos; | |
3947 | le32_add_cpu(&rec->e_int_clusters, | |
3948 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3949 | le32_add_cpu(&rec->e_int_clusters, | |
3950 | -le32_to_cpu(rec->e_cpos)); | |
3951 | ||
ec20cec7 | 3952 | ocfs2_journal_dirty(handle, bh); |
328d5752 MF |
3953 | } |
3954 | } | |
3955 | ||
d401dc12 JB |
3956 | static int ocfs2_append_rec_to_path(handle_t *handle, |
3957 | struct ocfs2_extent_tree *et, | |
328d5752 MF |
3958 | struct ocfs2_extent_rec *insert_rec, |
3959 | struct ocfs2_path *right_path, | |
3960 | struct ocfs2_path **ret_left_path) | |
3961 | { | |
3962 | int ret, next_free; | |
3963 | struct ocfs2_extent_list *el; | |
3964 | struct ocfs2_path *left_path = NULL; | |
3965 | ||
3966 | *ret_left_path = NULL; | |
3967 | ||
3968 | /* | |
3969 | * This shouldn't happen for non-trees. The extent rec cluster | |
3970 | * count manipulation below only works for interior nodes. | |
3971 | */ | |
3972 | BUG_ON(right_path->p_tree_depth == 0); | |
3973 | ||
3974 | /* | |
3975 | * If our appending insert is at the leftmost edge of a leaf, | |
3976 | * then we might need to update the rightmost records of the | |
3977 | * neighboring path. | |
3978 | */ | |
3979 | el = path_leaf_el(right_path); | |
3980 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3981 | if (next_free == 0 || | |
3982 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3983 | u32 left_cpos; | |
3984 | ||
d401dc12 JB |
3985 | ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci), |
3986 | right_path, &left_cpos); | |
328d5752 MF |
3987 | if (ret) { |
3988 | mlog_errno(ret); | |
dcd0538f MF |
3989 | goto out; |
3990 | } | |
3991 | ||
328d5752 MF |
3992 | mlog(0, "Append may need a left path update. cpos: %u, " |
3993 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3994 | left_cpos); | |
e48edee2 | 3995 | |
328d5752 MF |
3996 | /* |
3997 | * No need to worry if the append is already in the | |
3998 | * leftmost leaf. | |
3999 | */ | |
4000 | if (left_cpos) { | |
ffdd7a54 | 4001 | left_path = ocfs2_new_path_from_path(right_path); |
328d5752 MF |
4002 | if (!left_path) { |
4003 | ret = -ENOMEM; | |
4004 | mlog_errno(ret); | |
4005 | goto out; | |
4006 | } | |
dcd0538f | 4007 | |
d401dc12 | 4008 | ret = ocfs2_find_path(et->et_ci, left_path, |
facdb77f | 4009 | left_cpos); |
328d5752 MF |
4010 | if (ret) { |
4011 | mlog_errno(ret); | |
4012 | goto out; | |
4013 | } | |
dcd0538f | 4014 | |
328d5752 MF |
4015 | /* |
4016 | * ocfs2_insert_path() will pass the left_path to the | |
4017 | * journal for us. | |
4018 | */ | |
4019 | } | |
4020 | } | |
dcd0538f | 4021 | |
d401dc12 | 4022 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
328d5752 MF |
4023 | if (ret) { |
4024 | mlog_errno(ret); | |
4025 | goto out; | |
dcd0538f MF |
4026 | } |
4027 | ||
d401dc12 | 4028 | ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec); |
328d5752 | 4029 | |
dcd0538f MF |
4030 | *ret_left_path = left_path; |
4031 | ret = 0; | |
4032 | out: | |
4033 | if (ret != 0) | |
4034 | ocfs2_free_path(left_path); | |
4035 | ||
4036 | return ret; | |
4037 | } | |
4038 | ||
c38e52bb | 4039 | static void ocfs2_split_record(struct ocfs2_extent_tree *et, |
328d5752 MF |
4040 | struct ocfs2_path *left_path, |
4041 | struct ocfs2_path *right_path, | |
4042 | struct ocfs2_extent_rec *split_rec, | |
4043 | enum ocfs2_split_type split) | |
4044 | { | |
4045 | int index; | |
4046 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
4047 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
4048 | struct ocfs2_extent_rec *rec, *tmprec; | |
4049 | ||
c19a28e1 | 4050 | right_el = path_leaf_el(right_path); |
328d5752 MF |
4051 | if (left_path) |
4052 | left_el = path_leaf_el(left_path); | |
4053 | ||
4054 | el = right_el; | |
4055 | insert_el = right_el; | |
4056 | index = ocfs2_search_extent_list(el, cpos); | |
4057 | if (index != -1) { | |
4058 | if (index == 0 && left_path) { | |
4059 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
4060 | ||
4061 | /* | |
4062 | * This typically means that the record | |
4063 | * started in the left path but moved to the | |
4064 | * right as a result of rotation. We either | |
4065 | * move the existing record to the left, or we | |
4066 | * do the later insert there. | |
4067 | * | |
4068 | * In this case, the left path should always | |
4069 | * exist as the rotate code will have passed | |
4070 | * it back for a post-insert update. | |
4071 | */ | |
4072 | ||
4073 | if (split == SPLIT_LEFT) { | |
4074 | /* | |
4075 | * It's a left split. Since we know | |
4076 | * that the rotate code gave us an | |
4077 | * empty extent in the left path, we | |
4078 | * can just do the insert there. | |
4079 | */ | |
4080 | insert_el = left_el; | |
4081 | } else { | |
4082 | /* | |
4083 | * Right split - we have to move the | |
4084 | * existing record over to the left | |
4085 | * leaf. The insert will be into the | |
4086 | * newly created empty extent in the | |
4087 | * right leaf. | |
4088 | */ | |
4089 | tmprec = &right_el->l_recs[index]; | |
4090 | ocfs2_rotate_leaf(left_el, tmprec); | |
4091 | el = left_el; | |
4092 | ||
4093 | memset(tmprec, 0, sizeof(*tmprec)); | |
4094 | index = ocfs2_search_extent_list(left_el, cpos); | |
4095 | BUG_ON(index == -1); | |
4096 | } | |
4097 | } | |
4098 | } else { | |
4099 | BUG_ON(!left_path); | |
4100 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
4101 | /* | |
4102 | * Left path is easy - we can just allow the insert to | |
4103 | * happen. | |
4104 | */ | |
4105 | el = left_el; | |
4106 | insert_el = left_el; | |
4107 | index = ocfs2_search_extent_list(el, cpos); | |
4108 | BUG_ON(index == -1); | |
4109 | } | |
4110 | ||
4111 | rec = &el->l_recs[index]; | |
c38e52bb JB |
4112 | ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4113 | split, rec, split_rec); | |
328d5752 MF |
4114 | ocfs2_rotate_leaf(insert_el, split_rec); |
4115 | } | |
4116 | ||
dcd0538f | 4117 | /* |
e7d4cb6b TM |
4118 | * This function only does inserts on an allocation b-tree. For tree |
4119 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
4120 | * |
4121 | * right_path is the path we want to do the actual insert | |
4122 | * in. left_path should only be passed in if we need to update that | |
4123 | * portion of the tree after an edge insert. | |
4124 | */ | |
3505bec0 | 4125 | static int ocfs2_insert_path(handle_t *handle, |
7dc02805 | 4126 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4127 | struct ocfs2_path *left_path, |
4128 | struct ocfs2_path *right_path, | |
4129 | struct ocfs2_extent_rec *insert_rec, | |
4130 | struct ocfs2_insert_type *insert) | |
4131 | { | |
4132 | int ret, subtree_index; | |
4133 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 4134 | |
dcd0538f | 4135 | if (left_path) { |
dcd0538f MF |
4136 | /* |
4137 | * There's a chance that left_path got passed back to | |
4138 | * us without being accounted for in the | |
4139 | * journal. Extend our transaction here to be sure we | |
4140 | * can change those blocks. | |
4141 | */ | |
c901fb00 | 4142 | ret = ocfs2_extend_trans(handle, left_path->p_tree_depth); |
dcd0538f MF |
4143 | if (ret < 0) { |
4144 | mlog_errno(ret); | |
4145 | goto out; | |
4146 | } | |
4147 | ||
7dc02805 | 4148 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
dcd0538f MF |
4149 | if (ret < 0) { |
4150 | mlog_errno(ret); | |
4151 | goto out; | |
4152 | } | |
4153 | } | |
4154 | ||
e8aed345 MF |
4155 | /* |
4156 | * Pass both paths to the journal. The majority of inserts | |
4157 | * will be touching all components anyway. | |
4158 | */ | |
7dc02805 | 4159 | ret = ocfs2_journal_access_path(et->et_ci, handle, right_path); |
e8aed345 MF |
4160 | if (ret < 0) { |
4161 | mlog_errno(ret); | |
4162 | goto out; | |
4163 | } | |
4164 | ||
328d5752 MF |
4165 | if (insert->ins_split != SPLIT_NONE) { |
4166 | /* | |
4167 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 4168 | * of splits, but it's easier to just let one separate |
328d5752 MF |
4169 | * function sort it all out. |
4170 | */ | |
c38e52bb | 4171 | ocfs2_split_record(et, left_path, right_path, |
328d5752 | 4172 | insert_rec, insert->ins_split); |
e8aed345 MF |
4173 | |
4174 | /* | |
4175 | * Split might have modified either leaf and we don't | |
4176 | * have a guarantee that the later edge insert will | |
4177 | * dirty this for us. | |
4178 | */ | |
4179 | if (left_path) | |
ec20cec7 JB |
4180 | ocfs2_journal_dirty(handle, |
4181 | path_leaf_bh(left_path)); | |
328d5752 | 4182 | } else |
d5628623 JB |
4183 | ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path), |
4184 | insert); | |
dcd0538f | 4185 | |
ec20cec7 | 4186 | ocfs2_journal_dirty(handle, leaf_bh); |
dcd0538f MF |
4187 | |
4188 | if (left_path) { | |
4189 | /* | |
4190 | * The rotate code has indicated that we need to fix | |
4191 | * up portions of the tree after the insert. | |
4192 | * | |
4193 | * XXX: Should we extend the transaction here? | |
4194 | */ | |
7dc02805 | 4195 | subtree_index = ocfs2_find_subtree_root(et, left_path, |
dcd0538f | 4196 | right_path); |
4619c73e JB |
4197 | ocfs2_complete_edge_insert(handle, left_path, right_path, |
4198 | subtree_index); | |
dcd0538f MF |
4199 | } |
4200 | ||
4201 | ret = 0; | |
4202 | out: | |
4203 | return ret; | |
4204 | } | |
4205 | ||
3505bec0 | 4206 | static int ocfs2_do_insert_extent(handle_t *handle, |
e7d4cb6b | 4207 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4208 | struct ocfs2_extent_rec *insert_rec, |
4209 | struct ocfs2_insert_type *type) | |
4210 | { | |
4211 | int ret, rotate = 0; | |
4212 | u32 cpos; | |
4213 | struct ocfs2_path *right_path = NULL; | |
4214 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
4215 | struct ocfs2_extent_list *el; |
4216 | ||
ce1d9ea6 | 4217 | el = et->et_root_el; |
dcd0538f | 4218 | |
d9a0a1f8 | 4219 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4220 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
4221 | if (ret) { |
4222 | mlog_errno(ret); | |
4223 | goto out; | |
4224 | } | |
4225 | ||
4226 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
d5628623 | 4227 | ocfs2_insert_at_leaf(et, insert_rec, el, type); |
dcd0538f MF |
4228 | goto out_update_clusters; |
4229 | } | |
4230 | ||
ffdd7a54 | 4231 | right_path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4232 | if (!right_path) { |
4233 | ret = -ENOMEM; | |
4234 | mlog_errno(ret); | |
4235 | goto out; | |
4236 | } | |
4237 | ||
4238 | /* | |
4239 | * Determine the path to start with. Rotations need the | |
4240 | * rightmost path, everything else can go directly to the | |
4241 | * target leaf. | |
4242 | */ | |
4243 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
4244 | if (type->ins_appending == APPEND_NONE && | |
4245 | type->ins_contig == CONTIG_NONE) { | |
4246 | rotate = 1; | |
4247 | cpos = UINT_MAX; | |
4248 | } | |
4249 | ||
facdb77f | 4250 | ret = ocfs2_find_path(et->et_ci, right_path, cpos); |
dcd0538f MF |
4251 | if (ret) { |
4252 | mlog_errno(ret); | |
4253 | goto out; | |
4254 | } | |
4255 | ||
4256 | /* | |
4257 | * Rotations and appends need special treatment - they modify | |
4258 | * parts of the tree's above them. | |
4259 | * | |
4260 | * Both might pass back a path immediate to the left of the | |
4261 | * one being inserted to. This will be cause | |
4262 | * ocfs2_insert_path() to modify the rightmost records of | |
4263 | * left_path to account for an edge insert. | |
4264 | * | |
4265 | * XXX: When modifying this code, keep in mind that an insert | |
4266 | * can wind up skipping both of these two special cases... | |
4267 | */ | |
4268 | if (rotate) { | |
1bbf0b8d | 4269 | ret = ocfs2_rotate_tree_right(handle, et, type->ins_split, |
dcd0538f MF |
4270 | le32_to_cpu(insert_rec->e_cpos), |
4271 | right_path, &left_path); | |
4272 | if (ret) { | |
4273 | mlog_errno(ret); | |
4274 | goto out; | |
4275 | } | |
e8aed345 MF |
4276 | |
4277 | /* | |
4278 | * ocfs2_rotate_tree_right() might have extended the | |
4279 | * transaction without re-journaling our tree root. | |
4280 | */ | |
d9a0a1f8 | 4281 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4282 | OCFS2_JOURNAL_ACCESS_WRITE); |
e8aed345 MF |
4283 | if (ret) { |
4284 | mlog_errno(ret); | |
4285 | goto out; | |
4286 | } | |
dcd0538f MF |
4287 | } else if (type->ins_appending == APPEND_TAIL |
4288 | && type->ins_contig != CONTIG_LEFT) { | |
d401dc12 | 4289 | ret = ocfs2_append_rec_to_path(handle, et, insert_rec, |
dcd0538f MF |
4290 | right_path, &left_path); |
4291 | if (ret) { | |
4292 | mlog_errno(ret); | |
4293 | goto out; | |
4294 | } | |
4295 | } | |
4296 | ||
3505bec0 | 4297 | ret = ocfs2_insert_path(handle, et, left_path, right_path, |
dcd0538f MF |
4298 | insert_rec, type); |
4299 | if (ret) { | |
4300 | mlog_errno(ret); | |
4301 | goto out; | |
4302 | } | |
4303 | ||
4304 | out_update_clusters: | |
328d5752 | 4305 | if (type->ins_split == SPLIT_NONE) |
6136ca5f | 4306 | ocfs2_et_update_clusters(et, |
35dc0aa3 | 4307 | le16_to_cpu(insert_rec->e_leaf_clusters)); |
dcd0538f | 4308 | |
ec20cec7 | 4309 | ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
4310 | |
4311 | out: | |
4312 | ocfs2_free_path(left_path); | |
4313 | ocfs2_free_path(right_path); | |
4314 | ||
4315 | return ret; | |
4316 | } | |
4317 | ||
328d5752 | 4318 | static enum ocfs2_contig_type |
a2970291 JB |
4319 | ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et, |
4320 | struct ocfs2_path *path, | |
328d5752 MF |
4321 | struct ocfs2_extent_list *el, int index, |
4322 | struct ocfs2_extent_rec *split_rec) | |
4323 | { | |
ad5a4d70 | 4324 | int status; |
328d5752 | 4325 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4326 | u32 left_cpos, right_cpos; |
4327 | struct ocfs2_extent_rec *rec = NULL; | |
4328 | struct ocfs2_extent_list *new_el; | |
4329 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4330 | struct buffer_head *bh; | |
4331 | struct ocfs2_extent_block *eb; | |
a2970291 | 4332 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
ad5a4d70 TM |
4333 | |
4334 | if (index > 0) { | |
4335 | rec = &el->l_recs[index - 1]; | |
4336 | } else if (path->p_tree_depth > 0) { | |
a2970291 | 4337 | status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
ad5a4d70 TM |
4338 | if (status) |
4339 | goto out; | |
4340 | ||
4341 | if (left_cpos != 0) { | |
ffdd7a54 | 4342 | left_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4343 | if (!left_path) |
4344 | goto out; | |
4345 | ||
a2970291 JB |
4346 | status = ocfs2_find_path(et->et_ci, left_path, |
4347 | left_cpos); | |
ad5a4d70 TM |
4348 | if (status) |
4349 | goto out; | |
4350 | ||
4351 | new_el = path_leaf_el(left_path); | |
4352 | ||
4353 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4354 | le16_to_cpu(new_el->l_count)) { | |
4355 | bh = path_leaf_bh(left_path); | |
4356 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4357 | ocfs2_error(sb, |
5e96581a JB |
4358 | "Extent block #%llu has an " |
4359 | "invalid l_next_free_rec of " | |
4360 | "%d. It should have " | |
4361 | "matched the l_count of %d", | |
4362 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4363 | le16_to_cpu(new_el->l_next_free_rec), | |
4364 | le16_to_cpu(new_el->l_count)); | |
4365 | status = -EINVAL; | |
ad5a4d70 TM |
4366 | goto out; |
4367 | } | |
4368 | rec = &new_el->l_recs[ | |
4369 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4370 | } | |
4371 | } | |
328d5752 MF |
4372 | |
4373 | /* | |
4374 | * We're careful to check for an empty extent record here - | |
4375 | * the merge code will know what to do if it sees one. | |
4376 | */ | |
ad5a4d70 | 4377 | if (rec) { |
328d5752 MF |
4378 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4379 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4380 | ret = CONTIG_RIGHT; | |
4381 | } else { | |
853a3a14 | 4382 | ret = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4383 | } |
4384 | } | |
4385 | ||
ad5a4d70 TM |
4386 | rec = NULL; |
4387 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4388 | rec = &el->l_recs[index + 1]; | |
4389 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4390 | path->p_tree_depth > 0) { | |
a2970291 | 4391 | status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos); |
ad5a4d70 TM |
4392 | if (status) |
4393 | goto out; | |
4394 | ||
4395 | if (right_cpos == 0) | |
4396 | goto out; | |
4397 | ||
ffdd7a54 | 4398 | right_path = ocfs2_new_path_from_path(path); |
ad5a4d70 TM |
4399 | if (!right_path) |
4400 | goto out; | |
4401 | ||
a2970291 | 4402 | status = ocfs2_find_path(et->et_ci, right_path, right_cpos); |
ad5a4d70 TM |
4403 | if (status) |
4404 | goto out; | |
4405 | ||
4406 | new_el = path_leaf_el(right_path); | |
4407 | rec = &new_el->l_recs[0]; | |
4408 | if (ocfs2_is_empty_extent(rec)) { | |
4409 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4410 | bh = path_leaf_bh(right_path); | |
4411 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
a2970291 | 4412 | ocfs2_error(sb, |
5e96581a JB |
4413 | "Extent block #%llu has an " |
4414 | "invalid l_next_free_rec of %d", | |
4415 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
4416 | le16_to_cpu(new_el->l_next_free_rec)); | |
4417 | status = -EINVAL; | |
ad5a4d70 TM |
4418 | goto out; |
4419 | } | |
4420 | rec = &new_el->l_recs[1]; | |
4421 | } | |
4422 | } | |
4423 | ||
4424 | if (rec) { | |
328d5752 MF |
4425 | enum ocfs2_contig_type contig_type; |
4426 | ||
853a3a14 | 4427 | contig_type = ocfs2_et_extent_contig(et, rec, split_rec); |
328d5752 MF |
4428 | |
4429 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4430 | ret = CONTIG_LEFTRIGHT; | |
4431 | else if (ret == CONTIG_NONE) | |
4432 | ret = contig_type; | |
4433 | } | |
4434 | ||
ad5a4d70 TM |
4435 | out: |
4436 | if (left_path) | |
4437 | ocfs2_free_path(left_path); | |
4438 | if (right_path) | |
4439 | ocfs2_free_path(right_path); | |
4440 | ||
328d5752 MF |
4441 | return ret; |
4442 | } | |
4443 | ||
1ef61b33 | 4444 | static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4445 | struct ocfs2_insert_type *insert, |
4446 | struct ocfs2_extent_list *el, | |
1ef61b33 | 4447 | struct ocfs2_extent_rec *insert_rec) |
dcd0538f MF |
4448 | { |
4449 | int i; | |
4450 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4451 | ||
e48edee2 MF |
4452 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4453 | ||
dcd0538f | 4454 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
853a3a14 TM |
4455 | contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i], |
4456 | insert_rec); | |
dcd0538f MF |
4457 | if (contig_type != CONTIG_NONE) { |
4458 | insert->ins_contig_index = i; | |
4459 | break; | |
4460 | } | |
4461 | } | |
4462 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4463 | |
4464 | if (insert->ins_contig != CONTIG_NONE) { | |
4465 | struct ocfs2_extent_rec *rec = | |
4466 | &el->l_recs[insert->ins_contig_index]; | |
4467 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4468 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4469 | ||
4470 | /* | |
4471 | * Caller might want us to limit the size of extents, don't | |
4472 | * calculate contiguousness if we might exceed that limit. | |
4473 | */ | |
ce1d9ea6 JB |
4474 | if (et->et_max_leaf_clusters && |
4475 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4476 | insert->ins_contig = CONTIG_NONE; |
4477 | } | |
dcd0538f MF |
4478 | } |
4479 | ||
4480 | /* | |
4481 | * This should only be called against the righmost leaf extent list. | |
4482 | * | |
4483 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4484 | * insert at the tail of the rightmost leaf. | |
4485 | * | |
e7d4cb6b TM |
4486 | * This should also work against the root extent list for tree's with 0 |
4487 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4488 | * then the logic here makes sense. |
4489 | */ | |
4490 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4491 | struct ocfs2_extent_list *el, | |
4492 | struct ocfs2_extent_rec *insert_rec) | |
4493 | { | |
4494 | int i; | |
4495 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4496 | struct ocfs2_extent_rec *rec; | |
4497 | ||
4498 | insert->ins_appending = APPEND_NONE; | |
4499 | ||
e48edee2 | 4500 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4501 | |
4502 | if (!el->l_next_free_rec) | |
4503 | goto set_tail_append; | |
4504 | ||
4505 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4506 | /* Were all records empty? */ | |
4507 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4508 | goto set_tail_append; | |
4509 | } | |
4510 | ||
4511 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4512 | rec = &el->l_recs[i]; | |
4513 | ||
e48edee2 MF |
4514 | if (cpos >= |
4515 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4516 | goto set_tail_append; |
4517 | ||
4518 | return; | |
4519 | ||
4520 | set_tail_append: | |
4521 | insert->ins_appending = APPEND_TAIL; | |
4522 | } | |
4523 | ||
4524 | /* | |
4525 | * Helper function called at the begining of an insert. | |
4526 | * | |
4527 | * This computes a few things that are commonly used in the process of | |
4528 | * inserting into the btree: | |
4529 | * - Whether the new extent is contiguous with an existing one. | |
4530 | * - The current tree depth. | |
4531 | * - Whether the insert is an appending one. | |
4532 | * - The total # of free records in the tree. | |
4533 | * | |
4534 | * All of the information is stored on the ocfs2_insert_type | |
4535 | * structure. | |
4536 | */ | |
627961b7 | 4537 | static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et, |
dcd0538f MF |
4538 | struct buffer_head **last_eb_bh, |
4539 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4540 | int *free_records, |
dcd0538f MF |
4541 | struct ocfs2_insert_type *insert) |
4542 | { | |
4543 | int ret; | |
dcd0538f MF |
4544 | struct ocfs2_extent_block *eb; |
4545 | struct ocfs2_extent_list *el; | |
4546 | struct ocfs2_path *path = NULL; | |
4547 | struct buffer_head *bh = NULL; | |
4548 | ||
328d5752 MF |
4549 | insert->ins_split = SPLIT_NONE; |
4550 | ||
ce1d9ea6 | 4551 | el = et->et_root_el; |
dcd0538f MF |
4552 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4553 | ||
4554 | if (el->l_tree_depth) { | |
4555 | /* | |
4556 | * If we have tree depth, we read in the | |
4557 | * rightmost extent block ahead of time as | |
4558 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4559 | * may want it later. | |
4560 | */ | |
3d03a305 | 4561 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
4562 | ocfs2_et_get_last_eb_blk(et), |
4563 | &bh); | |
dcd0538f MF |
4564 | if (ret) { |
4565 | mlog_exit(ret); | |
4566 | goto out; | |
4567 | } | |
ccd979bd | 4568 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4569 | el = &eb->h_list; |
dcd0538f | 4570 | } |
ccd979bd | 4571 | |
dcd0538f MF |
4572 | /* |
4573 | * Unless we have a contiguous insert, we'll need to know if | |
4574 | * there is room left in our allocation tree for another | |
4575 | * extent record. | |
4576 | * | |
4577 | * XXX: This test is simplistic, we can search for empty | |
4578 | * extent records too. | |
4579 | */ | |
c77534f6 | 4580 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4581 | le16_to_cpu(el->l_next_free_rec); |
4582 | ||
4583 | if (!insert->ins_tree_depth) { | |
1ef61b33 | 4584 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4585 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4586 | return 0; | |
ccd979bd MF |
4587 | } |
4588 | ||
ffdd7a54 | 4589 | path = ocfs2_new_path_from_et(et); |
dcd0538f MF |
4590 | if (!path) { |
4591 | ret = -ENOMEM; | |
4592 | mlog_errno(ret); | |
4593 | goto out; | |
4594 | } | |
ccd979bd | 4595 | |
dcd0538f MF |
4596 | /* |
4597 | * In the case that we're inserting past what the tree | |
4598 | * currently accounts for, ocfs2_find_path() will return for | |
4599 | * us the rightmost tree path. This is accounted for below in | |
4600 | * the appending code. | |
4601 | */ | |
facdb77f | 4602 | ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos)); |
dcd0538f MF |
4603 | if (ret) { |
4604 | mlog_errno(ret); | |
4605 | goto out; | |
4606 | } | |
ccd979bd | 4607 | |
dcd0538f MF |
4608 | el = path_leaf_el(path); |
4609 | ||
4610 | /* | |
4611 | * Now that we have the path, there's two things we want to determine: | |
4612 | * 1) Contiguousness (also set contig_index if this is so) | |
4613 | * | |
4614 | * 2) Are we doing an append? We can trivially break this up | |
4615 | * into two types of appends: simple record append, or a | |
4616 | * rotate inside the tail leaf. | |
4617 | */ | |
1ef61b33 | 4618 | ocfs2_figure_contig_type(et, insert, el, insert_rec); |
dcd0538f MF |
4619 | |
4620 | /* | |
4621 | * The insert code isn't quite ready to deal with all cases of | |
4622 | * left contiguousness. Specifically, if it's an insert into | |
4623 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4624 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4625 | * left. For now, just catch that case and fool the layers |
4626 | * above us. This works just fine for tree_depth == 0, which | |
4627 | * is why we allow that above. | |
4628 | */ | |
4629 | if (insert->ins_contig == CONTIG_LEFT && | |
4630 | insert->ins_contig_index == 0) | |
4631 | insert->ins_contig = CONTIG_NONE; | |
4632 | ||
4633 | /* | |
4634 | * Ok, so we can simply compare against last_eb to figure out | |
4635 | * whether the path doesn't exist. This will only happen in | |
4636 | * the case that we're doing a tail append, so maybe we can | |
4637 | * take advantage of that information somehow. | |
4638 | */ | |
35dc0aa3 | 4639 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4640 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4641 | /* |
4642 | * Ok, ocfs2_find_path() returned us the rightmost | |
4643 | * tree path. This might be an appending insert. There are | |
4644 | * two cases: | |
4645 | * 1) We're doing a true append at the tail: | |
4646 | * -This might even be off the end of the leaf | |
4647 | * 2) We're "appending" by rotating in the tail | |
4648 | */ | |
4649 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4650 | } | |
4651 | ||
4652 | out: | |
4653 | ocfs2_free_path(path); | |
4654 | ||
4655 | if (ret == 0) | |
4656 | *last_eb_bh = bh; | |
4657 | else | |
4658 | brelse(bh); | |
4659 | return ret; | |
ccd979bd MF |
4660 | } |
4661 | ||
dcd0538f | 4662 | /* |
cc79d8c1 | 4663 | * Insert an extent into a btree. |
dcd0538f | 4664 | * |
cc79d8c1 | 4665 | * The caller needs to update the owning btree's cluster count. |
dcd0538f | 4666 | */ |
cc79d8c1 | 4667 | int ocfs2_insert_extent(handle_t *handle, |
f99b9b7c JB |
4668 | struct ocfs2_extent_tree *et, |
4669 | u32 cpos, | |
4670 | u64 start_blk, | |
4671 | u32 new_clusters, | |
4672 | u8 flags, | |
4673 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 4674 | { |
c3afcbb3 | 4675 | int status; |
c77534f6 | 4676 | int uninitialized_var(free_records); |
ccd979bd | 4677 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4678 | struct ocfs2_insert_type insert = {0, }; |
4679 | struct ocfs2_extent_rec rec; | |
4680 | ||
cc79d8c1 JB |
4681 | mlog(0, "add %u clusters at position %u to owner %llu\n", |
4682 | new_clusters, cpos, | |
4683 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
dcd0538f | 4684 | |
e48edee2 | 4685 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4686 | rec.e_cpos = cpu_to_le32(cpos); |
4687 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4688 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4689 | rec.e_flags = flags; |
6136ca5f | 4690 | status = ocfs2_et_insert_check(et, &rec); |
1e61ee79 JB |
4691 | if (status) { |
4692 | mlog_errno(status); | |
4693 | goto bail; | |
4694 | } | |
dcd0538f | 4695 | |
627961b7 | 4696 | status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec, |
c77534f6 | 4697 | &free_records, &insert); |
dcd0538f MF |
4698 | if (status < 0) { |
4699 | mlog_errno(status); | |
4700 | goto bail; | |
ccd979bd MF |
4701 | } |
4702 | ||
dcd0538f MF |
4703 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4704 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4705 | "Insert.tree_depth: %d\n", | |
4706 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4707 | free_records, insert.ins_tree_depth); |
ccd979bd | 4708 | |
c77534f6 | 4709 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
d401dc12 | 4710 | status = ocfs2_grow_tree(handle, et, |
328d5752 | 4711 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4712 | meta_ac); |
4713 | if (status) { | |
ccd979bd MF |
4714 | mlog_errno(status); |
4715 | goto bail; | |
4716 | } | |
ccd979bd MF |
4717 | } |
4718 | ||
dcd0538f | 4719 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
3505bec0 | 4720 | status = ocfs2_do_insert_extent(handle, et, &rec, &insert); |
ccd979bd MF |
4721 | if (status < 0) |
4722 | mlog_errno(status); | |
92ba470c JB |
4723 | else |
4724 | ocfs2_et_extent_map_insert(et, &rec); | |
ccd979bd MF |
4725 | |
4726 | bail: | |
a81cb88b | 4727 | brelse(last_eb_bh); |
ccd979bd | 4728 | |
f56654c4 TM |
4729 | mlog_exit(status); |
4730 | return status; | |
4731 | } | |
4732 | ||
0eb8d47e TM |
4733 | /* |
4734 | * Allcate and add clusters into the extent b-tree. | |
4735 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
f99b9b7c | 4736 | * The extent b-tree's root is specified by et, and |
0eb8d47e TM |
4737 | * it is not limited to the file storage. Any extent tree can use this |
4738 | * function if it implements the proper ocfs2_extent_tree. | |
4739 | */ | |
cbee7e1a JB |
4740 | int ocfs2_add_clusters_in_btree(handle_t *handle, |
4741 | struct ocfs2_extent_tree *et, | |
0eb8d47e TM |
4742 | u32 *logical_offset, |
4743 | u32 clusters_to_add, | |
4744 | int mark_unwritten, | |
0eb8d47e TM |
4745 | struct ocfs2_alloc_context *data_ac, |
4746 | struct ocfs2_alloc_context *meta_ac, | |
f99b9b7c | 4747 | enum ocfs2_alloc_restarted *reason_ret) |
0eb8d47e TM |
4748 | { |
4749 | int status = 0; | |
4750 | int free_extents; | |
4751 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4752 | u32 bit_off, num_bits; | |
4753 | u64 block; | |
4754 | u8 flags = 0; | |
cbee7e1a JB |
4755 | struct ocfs2_super *osb = |
4756 | OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci)); | |
0eb8d47e TM |
4757 | |
4758 | BUG_ON(!clusters_to_add); | |
4759 | ||
4760 | if (mark_unwritten) | |
4761 | flags = OCFS2_EXT_UNWRITTEN; | |
4762 | ||
3d03a305 | 4763 | free_extents = ocfs2_num_free_extents(osb, et); |
0eb8d47e TM |
4764 | if (free_extents < 0) { |
4765 | status = free_extents; | |
4766 | mlog_errno(status); | |
4767 | goto leave; | |
4768 | } | |
4769 | ||
4770 | /* there are two cases which could cause us to EAGAIN in the | |
4771 | * we-need-more-metadata case: | |
4772 | * 1) we haven't reserved *any* | |
4773 | * 2) we are so fragmented, we've needed to add metadata too | |
4774 | * many times. */ | |
4775 | if (!free_extents && !meta_ac) { | |
4776 | mlog(0, "we haven't reserved any metadata!\n"); | |
4777 | status = -EAGAIN; | |
4778 | reason = RESTART_META; | |
4779 | goto leave; | |
4780 | } else if ((!free_extents) | |
4781 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
f99b9b7c | 4782 | < ocfs2_extend_meta_needed(et->et_root_el))) { |
0eb8d47e TM |
4783 | mlog(0, "filesystem is really fragmented...\n"); |
4784 | status = -EAGAIN; | |
4785 | reason = RESTART_META; | |
4786 | goto leave; | |
4787 | } | |
4788 | ||
4789 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4790 | clusters_to_add, &bit_off, &num_bits); | |
4791 | if (status < 0) { | |
4792 | if (status != -ENOSPC) | |
4793 | mlog_errno(status); | |
4794 | goto leave; | |
4795 | } | |
4796 | ||
4797 | BUG_ON(num_bits > clusters_to_add); | |
4798 | ||
13723d00 | 4799 | /* reserve our write early -- insert_extent may update the tree root */ |
d9a0a1f8 | 4800 | status = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 4801 | OCFS2_JOURNAL_ACCESS_WRITE); |
0eb8d47e TM |
4802 | if (status < 0) { |
4803 | mlog_errno(status); | |
4804 | goto leave; | |
4805 | } | |
4806 | ||
4807 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
cbee7e1a JB |
4808 | mlog(0, "Allocating %u clusters at block %u for owner %llu\n", |
4809 | num_bits, bit_off, | |
4810 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci)); | |
cc79d8c1 | 4811 | status = ocfs2_insert_extent(handle, et, *logical_offset, block, |
f99b9b7c | 4812 | num_bits, flags, meta_ac); |
0eb8d47e TM |
4813 | if (status < 0) { |
4814 | mlog_errno(status); | |
4815 | goto leave; | |
4816 | } | |
4817 | ||
ec20cec7 | 4818 | ocfs2_journal_dirty(handle, et->et_root_bh); |
0eb8d47e TM |
4819 | |
4820 | clusters_to_add -= num_bits; | |
4821 | *logical_offset += num_bits; | |
4822 | ||
4823 | if (clusters_to_add) { | |
4824 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4825 | clusters_to_add); | |
4826 | status = -EAGAIN; | |
4827 | reason = RESTART_TRANS; | |
4828 | } | |
4829 | ||
4830 | leave: | |
4831 | mlog_exit(status); | |
4832 | if (reason_ret) | |
4833 | *reason_ret = reason; | |
4834 | return status; | |
4835 | } | |
4836 | ||
328d5752 MF |
4837 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4838 | struct ocfs2_extent_rec *split_rec, | |
4839 | u32 cpos, | |
4840 | struct ocfs2_extent_rec *rec) | |
4841 | { | |
4842 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4843 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4844 | ||
4845 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4846 | ||
4847 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4848 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4849 | ||
4850 | split_rec->e_blkno = rec->e_blkno; | |
4851 | le64_add_cpu(&split_rec->e_blkno, | |
4852 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4853 | ||
4854 | split_rec->e_flags = rec->e_flags; | |
4855 | } | |
4856 | ||
d231129f | 4857 | static int ocfs2_split_and_insert(handle_t *handle, |
e7d4cb6b | 4858 | struct ocfs2_extent_tree *et, |
d231129f | 4859 | struct ocfs2_path *path, |
328d5752 MF |
4860 | struct buffer_head **last_eb_bh, |
4861 | int split_index, | |
4862 | struct ocfs2_extent_rec *orig_split_rec, | |
4863 | struct ocfs2_alloc_context *meta_ac) | |
4864 | { | |
4865 | int ret = 0, depth; | |
4866 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4867 | struct ocfs2_extent_rec tmprec; | |
4868 | struct ocfs2_extent_list *rightmost_el; | |
4869 | struct ocfs2_extent_rec rec; | |
4870 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4871 | struct ocfs2_insert_type insert; | |
4872 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4873 | |
4874 | leftright: | |
4875 | /* | |
4876 | * Store a copy of the record on the stack - it might move | |
4877 | * around as the tree is manipulated below. | |
4878 | */ | |
4879 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4880 | ||
ce1d9ea6 | 4881 | rightmost_el = et->et_root_el; |
328d5752 MF |
4882 | |
4883 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4884 | if (depth) { | |
4885 | BUG_ON(!(*last_eb_bh)); | |
4886 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4887 | rightmost_el = &eb->h_list; | |
4888 | } | |
4889 | ||
4890 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4891 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 4892 | ret = ocfs2_grow_tree(handle, et, |
e7d4cb6b | 4893 | &depth, last_eb_bh, meta_ac); |
328d5752 MF |
4894 | if (ret) { |
4895 | mlog_errno(ret); | |
4896 | goto out; | |
4897 | } | |
328d5752 MF |
4898 | } |
4899 | ||
4900 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4901 | insert.ins_appending = APPEND_NONE; | |
4902 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4903 | insert.ins_tree_depth = depth; |
4904 | ||
4905 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4906 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4907 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4908 | le16_to_cpu(rec.e_leaf_clusters); | |
4909 | ||
4910 | if (split_rec.e_cpos == rec.e_cpos) { | |
4911 | insert.ins_split = SPLIT_LEFT; | |
4912 | } else if (insert_range == rec_range) { | |
4913 | insert.ins_split = SPLIT_RIGHT; | |
4914 | } else { | |
4915 | /* | |
4916 | * Left/right split. We fake this as a right split | |
4917 | * first and then make a second pass as a left split. | |
4918 | */ | |
4919 | insert.ins_split = SPLIT_RIGHT; | |
4920 | ||
d231129f JB |
4921 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
4922 | &tmprec, insert_range, &rec); | |
328d5752 MF |
4923 | |
4924 | split_rec = tmprec; | |
4925 | ||
4926 | BUG_ON(do_leftright); | |
4927 | do_leftright = 1; | |
4928 | } | |
4929 | ||
3505bec0 | 4930 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
328d5752 MF |
4931 | if (ret) { |
4932 | mlog_errno(ret); | |
4933 | goto out; | |
4934 | } | |
4935 | ||
4936 | if (do_leftright == 1) { | |
4937 | u32 cpos; | |
4938 | struct ocfs2_extent_list *el; | |
4939 | ||
4940 | do_leftright++; | |
4941 | split_rec = *orig_split_rec; | |
4942 | ||
4943 | ocfs2_reinit_path(path, 1); | |
4944 | ||
4945 | cpos = le32_to_cpu(split_rec.e_cpos); | |
facdb77f | 4946 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
328d5752 MF |
4947 | if (ret) { |
4948 | mlog_errno(ret); | |
4949 | goto out; | |
4950 | } | |
4951 | ||
4952 | el = path_leaf_el(path); | |
4953 | split_index = ocfs2_search_extent_list(el, cpos); | |
4954 | goto leftright; | |
4955 | } | |
4956 | out: | |
4957 | ||
4958 | return ret; | |
4959 | } | |
4960 | ||
f3868d0f JB |
4961 | static int ocfs2_replace_extent_rec(handle_t *handle, |
4962 | struct ocfs2_extent_tree *et, | |
47be12e4 TM |
4963 | struct ocfs2_path *path, |
4964 | struct ocfs2_extent_list *el, | |
4965 | int split_index, | |
4966 | struct ocfs2_extent_rec *split_rec) | |
4967 | { | |
4968 | int ret; | |
4969 | ||
f3868d0f | 4970 | ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path, |
47be12e4 TM |
4971 | path_num_items(path) - 1); |
4972 | if (ret) { | |
4973 | mlog_errno(ret); | |
4974 | goto out; | |
4975 | } | |
4976 | ||
4977 | el->l_recs[split_index] = *split_rec; | |
4978 | ||
4979 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
4980 | out: | |
4981 | return ret; | |
4982 | } | |
4983 | ||
328d5752 | 4984 | /* |
555936bf TM |
4985 | * Split part or all of the extent record at split_index in the leaf |
4986 | * pointed to by path. Merge with the contiguous extent record if needed. | |
328d5752 MF |
4987 | * |
4988 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4989 | * | |
4990 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4991 | * of the tree is required. All other cases will degrade into a less | |
4992 | * optimal tree layout. | |
4993 | * | |
e7d4cb6b TM |
4994 | * last_eb_bh should be the rightmost leaf block for any extent |
4995 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4996 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4997 | * |
4998 | * This code is optimized for readability - several passes might be | |
4999 | * made over certain portions of the tree. All of those blocks will | |
5000 | * have been brought into cache (and pinned via the journal), so the | |
5001 | * extra overhead is not expressed in terms of disk reads. | |
5002 | */ | |
e2e9f608 TM |
5003 | int ocfs2_split_extent(handle_t *handle, |
5004 | struct ocfs2_extent_tree *et, | |
5005 | struct ocfs2_path *path, | |
5006 | int split_index, | |
5007 | struct ocfs2_extent_rec *split_rec, | |
5008 | struct ocfs2_alloc_context *meta_ac, | |
5009 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
328d5752 MF |
5010 | { |
5011 | int ret = 0; | |
5012 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 5013 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
5014 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
5015 | struct ocfs2_merge_ctxt ctxt; | |
5016 | struct ocfs2_extent_list *rightmost_el; | |
5017 | ||
328d5752 MF |
5018 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || |
5019 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
5020 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
5021 | ret = -EIO; | |
5022 | mlog_errno(ret); | |
5023 | goto out; | |
5024 | } | |
5025 | ||
a2970291 | 5026 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el, |
328d5752 MF |
5027 | split_index, |
5028 | split_rec); | |
5029 | ||
5030 | /* | |
5031 | * The core merge / split code wants to know how much room is | |
a1cf076b | 5032 | * left in this allocation tree, so we pass the |
328d5752 MF |
5033 | * rightmost extent list. |
5034 | */ | |
5035 | if (path->p_tree_depth) { | |
5036 | struct ocfs2_extent_block *eb; | |
328d5752 | 5037 | |
3d03a305 | 5038 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5039 | ocfs2_et_get_last_eb_blk(et), |
5040 | &last_eb_bh); | |
328d5752 MF |
5041 | if (ret) { |
5042 | mlog_exit(ret); | |
5043 | goto out; | |
5044 | } | |
5045 | ||
5046 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
328d5752 MF |
5047 | rightmost_el = &eb->h_list; |
5048 | } else | |
5049 | rightmost_el = path_root_el(path); | |
5050 | ||
328d5752 MF |
5051 | if (rec->e_cpos == split_rec->e_cpos && |
5052 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
5053 | ctxt.c_split_covers_rec = 1; | |
5054 | else | |
5055 | ctxt.c_split_covers_rec = 0; | |
5056 | ||
5057 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
5058 | ||
015452b1 MF |
5059 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
5060 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
5061 | ctxt.c_split_covers_rec); | |
328d5752 MF |
5062 | |
5063 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
5064 | if (ctxt.c_split_covers_rec) | |
f3868d0f | 5065 | ret = ocfs2_replace_extent_rec(handle, et, path, el, |
47be12e4 | 5066 | split_index, split_rec); |
328d5752 | 5067 | else |
d231129f | 5068 | ret = ocfs2_split_and_insert(handle, et, path, |
328d5752 MF |
5069 | &last_eb_bh, split_index, |
5070 | split_rec, meta_ac); | |
5071 | if (ret) | |
5072 | mlog_errno(ret); | |
5073 | } else { | |
c495dd24 | 5074 | ret = ocfs2_try_to_merge_extent(handle, et, path, |
328d5752 | 5075 | split_index, split_rec, |
c495dd24 | 5076 | dealloc, &ctxt); |
328d5752 MF |
5077 | if (ret) |
5078 | mlog_errno(ret); | |
5079 | } | |
5080 | ||
328d5752 MF |
5081 | out: |
5082 | brelse(last_eb_bh); | |
5083 | return ret; | |
5084 | } | |
5085 | ||
5086 | /* | |
555936bf TM |
5087 | * Change the flags of the already-existing extent at cpos for len clusters. |
5088 | * | |
5089 | * new_flags: the flags we want to set. | |
5090 | * clear_flags: the flags we want to clear. | |
5091 | * phys: the new physical offset we want this new extent starts from. | |
328d5752 MF |
5092 | * |
5093 | * If the existing extent is larger than the request, initiate a | |
5094 | * split. An attempt will be made at merging with adjacent extents. | |
5095 | * | |
5096 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5097 | */ | |
1aa75fea TM |
5098 | int ocfs2_change_extent_flag(handle_t *handle, |
5099 | struct ocfs2_extent_tree *et, | |
5100 | u32 cpos, u32 len, u32 phys, | |
5101 | struct ocfs2_alloc_context *meta_ac, | |
5102 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
5103 | int new_flags, int clear_flags) | |
328d5752 MF |
5104 | { |
5105 | int ret, index; | |
555936bf TM |
5106 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
5107 | u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys); | |
328d5752 MF |
5108 | struct ocfs2_extent_rec split_rec; |
5109 | struct ocfs2_path *left_path = NULL; | |
5110 | struct ocfs2_extent_list *el; | |
555936bf | 5111 | struct ocfs2_extent_rec *rec; |
328d5752 | 5112 | |
ffdd7a54 | 5113 | left_path = ocfs2_new_path_from_et(et); |
328d5752 MF |
5114 | if (!left_path) { |
5115 | ret = -ENOMEM; | |
5116 | mlog_errno(ret); | |
5117 | goto out; | |
5118 | } | |
5119 | ||
facdb77f | 5120 | ret = ocfs2_find_path(et->et_ci, left_path, cpos); |
328d5752 MF |
5121 | if (ret) { |
5122 | mlog_errno(ret); | |
5123 | goto out; | |
5124 | } | |
5125 | el = path_leaf_el(left_path); | |
5126 | ||
5127 | index = ocfs2_search_extent_list(el, cpos); | |
5128 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
555936bf TM |
5129 | ocfs2_error(sb, |
5130 | "Owner %llu has an extent at cpos %u which can no " | |
328d5752 | 5131 | "longer be found.\n", |
555936bf TM |
5132 | (unsigned long long) |
5133 | ocfs2_metadata_cache_owner(et->et_ci), cpos); | |
328d5752 MF |
5134 | ret = -EROFS; |
5135 | goto out; | |
5136 | } | |
5137 | ||
555936bf TM |
5138 | ret = -EIO; |
5139 | rec = &el->l_recs[index]; | |
5140 | if (new_flags && (rec->e_flags & new_flags)) { | |
5141 | mlog(ML_ERROR, "Owner %llu tried to set %d flags on an " | |
5142 | "extent that already had them", | |
5143 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5144 | new_flags); | |
5145 | goto out; | |
5146 | } | |
5147 | ||
5148 | if (clear_flags && !(rec->e_flags & clear_flags)) { | |
5149 | mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an " | |
5150 | "extent that didn't have them", | |
5151 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
5152 | clear_flags); | |
5153 | goto out; | |
5154 | } | |
5155 | ||
328d5752 MF |
5156 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); |
5157 | split_rec.e_cpos = cpu_to_le32(cpos); | |
5158 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
5159 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
555936bf TM |
5160 | split_rec.e_flags = rec->e_flags; |
5161 | if (new_flags) | |
5162 | split_rec.e_flags |= new_flags; | |
5163 | if (clear_flags) | |
5164 | split_rec.e_flags &= ~clear_flags; | |
5165 | ||
e2e9f608 TM |
5166 | ret = ocfs2_split_extent(handle, et, left_path, |
5167 | index, &split_rec, meta_ac, | |
5168 | dealloc); | |
328d5752 MF |
5169 | if (ret) |
5170 | mlog_errno(ret); | |
5171 | ||
5172 | out: | |
5173 | ocfs2_free_path(left_path); | |
5174 | return ret; | |
555936bf TM |
5175 | |
5176 | } | |
5177 | ||
5178 | /* | |
5179 | * Mark the already-existing extent at cpos as written for len clusters. | |
5180 | * This removes the unwritten extent flag. | |
5181 | * | |
5182 | * If the existing extent is larger than the request, initiate a | |
5183 | * split. An attempt will be made at merging with adjacent extents. | |
5184 | * | |
5185 | * The caller is responsible for passing down meta_ac if we'll need it. | |
5186 | */ | |
5187 | int ocfs2_mark_extent_written(struct inode *inode, | |
5188 | struct ocfs2_extent_tree *et, | |
5189 | handle_t *handle, u32 cpos, u32 len, u32 phys, | |
5190 | struct ocfs2_alloc_context *meta_ac, | |
5191 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5192 | { | |
5193 | int ret; | |
5194 | ||
5195 | mlog(0, "Inode %lu cpos %u, len %u, phys clusters %u\n", | |
5196 | inode->i_ino, cpos, len, phys); | |
5197 | ||
5198 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { | |
5199 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
5200 | "that are being written to, but the feature bit " | |
5201 | "is not set in the super block.", | |
5202 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
5203 | ret = -EROFS; | |
5204 | goto out; | |
5205 | } | |
5206 | ||
5207 | /* | |
5208 | * XXX: This should be fixed up so that we just re-insert the | |
5209 | * next extent records. | |
5210 | */ | |
5211 | ocfs2_et_extent_map_truncate(et, 0); | |
5212 | ||
5213 | ret = ocfs2_change_extent_flag(handle, et, cpos, | |
5214 | len, phys, meta_ac, dealloc, | |
5215 | 0, OCFS2_EXT_UNWRITTEN); | |
5216 | if (ret) | |
5217 | mlog_errno(ret); | |
5218 | ||
5219 | out: | |
5220 | return ret; | |
328d5752 MF |
5221 | } |
5222 | ||
dbdcf6a4 JB |
5223 | static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et, |
5224 | struct ocfs2_path *path, | |
d0c7d708 MF |
5225 | int index, u32 new_range, |
5226 | struct ocfs2_alloc_context *meta_ac) | |
5227 | { | |
c901fb00 | 5228 | int ret, depth, credits; |
d0c7d708 MF |
5229 | struct buffer_head *last_eb_bh = NULL; |
5230 | struct ocfs2_extent_block *eb; | |
5231 | struct ocfs2_extent_list *rightmost_el, *el; | |
5232 | struct ocfs2_extent_rec split_rec; | |
5233 | struct ocfs2_extent_rec *rec; | |
5234 | struct ocfs2_insert_type insert; | |
5235 | ||
5236 | /* | |
5237 | * Setup the record to split before we grow the tree. | |
5238 | */ | |
5239 | el = path_leaf_el(path); | |
5240 | rec = &el->l_recs[index]; | |
dbdcf6a4 JB |
5241 | ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci), |
5242 | &split_rec, new_range, rec); | |
d0c7d708 MF |
5243 | |
5244 | depth = path->p_tree_depth; | |
5245 | if (depth > 0) { | |
3d03a305 | 5246 | ret = ocfs2_read_extent_block(et->et_ci, |
5e96581a JB |
5247 | ocfs2_et_get_last_eb_blk(et), |
5248 | &last_eb_bh); | |
d0c7d708 MF |
5249 | if (ret < 0) { |
5250 | mlog_errno(ret); | |
5251 | goto out; | |
5252 | } | |
5253 | ||
5254 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
5255 | rightmost_el = &eb->h_list; | |
5256 | } else | |
5257 | rightmost_el = path_leaf_el(path); | |
5258 | ||
c901fb00 TM |
5259 | credits = path->p_tree_depth + |
5260 | ocfs2_extend_meta_needed(et->et_root_el); | |
d0c7d708 MF |
5261 | ret = ocfs2_extend_trans(handle, credits); |
5262 | if (ret) { | |
5263 | mlog_errno(ret); | |
5264 | goto out; | |
5265 | } | |
5266 | ||
5267 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
5268 | le16_to_cpu(rightmost_el->l_count)) { | |
d401dc12 | 5269 | ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
5270 | meta_ac); |
5271 | if (ret) { | |
5272 | mlog_errno(ret); | |
5273 | goto out; | |
5274 | } | |
d0c7d708 MF |
5275 | } |
5276 | ||
5277 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
5278 | insert.ins_appending = APPEND_NONE; | |
5279 | insert.ins_contig = CONTIG_NONE; | |
5280 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5281 | insert.ins_tree_depth = depth; |
5282 | ||
3505bec0 | 5283 | ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert); |
d0c7d708 MF |
5284 | if (ret) |
5285 | mlog_errno(ret); | |
5286 | ||
5287 | out: | |
5288 | brelse(last_eb_bh); | |
5289 | return ret; | |
5290 | } | |
5291 | ||
043beebb JB |
5292 | static int ocfs2_truncate_rec(handle_t *handle, |
5293 | struct ocfs2_extent_tree *et, | |
d0c7d708 MF |
5294 | struct ocfs2_path *path, int index, |
5295 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
043beebb | 5296 | u32 cpos, u32 len) |
d0c7d708 MF |
5297 | { |
5298 | int ret; | |
5299 | u32 left_cpos, rec_range, trunc_range; | |
5300 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
043beebb | 5301 | struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci); |
d0c7d708 MF |
5302 | struct ocfs2_path *left_path = NULL; |
5303 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5304 | struct ocfs2_extent_rec *rec; | |
5305 | struct ocfs2_extent_block *eb; | |
5306 | ||
5307 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
70f18c08 | 5308 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5309 | if (ret) { |
5310 | mlog_errno(ret); | |
5311 | goto out; | |
5312 | } | |
5313 | ||
5314 | index--; | |
5315 | } | |
5316 | ||
5317 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5318 | path->p_tree_depth) { | |
5319 | /* | |
5320 | * Check whether this is the rightmost tree record. If | |
5321 | * we remove all of this record or part of its right | |
5322 | * edge then an update of the record lengths above it | |
5323 | * will be required. | |
5324 | */ | |
5325 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5326 | if (eb->h_next_leaf_blk == 0) | |
5327 | is_rightmost_tree_rec = 1; | |
5328 | } | |
5329 | ||
5330 | rec = &el->l_recs[index]; | |
5331 | if (index == 0 && path->p_tree_depth && | |
5332 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5333 | /* | |
5334 | * Changing the leftmost offset (via partial or whole | |
5335 | * record truncate) of an interior (or rightmost) path | |
5336 | * means we have to update the subtree that is formed | |
5337 | * by this leaf and the one to it's left. | |
5338 | * | |
5339 | * There are two cases we can skip: | |
043beebb | 5340 | * 1) Path is the leftmost one in our btree. |
d0c7d708 MF |
5341 | * 2) The leaf is rightmost and will be empty after |
5342 | * we remove the extent record - the rotate code | |
5343 | * knows how to update the newly formed edge. | |
5344 | */ | |
5345 | ||
043beebb | 5346 | ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos); |
d0c7d708 MF |
5347 | if (ret) { |
5348 | mlog_errno(ret); | |
5349 | goto out; | |
5350 | } | |
5351 | ||
5352 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
ffdd7a54 | 5353 | left_path = ocfs2_new_path_from_path(path); |
d0c7d708 MF |
5354 | if (!left_path) { |
5355 | ret = -ENOMEM; | |
5356 | mlog_errno(ret); | |
5357 | goto out; | |
5358 | } | |
5359 | ||
facdb77f JB |
5360 | ret = ocfs2_find_path(et->et_ci, left_path, |
5361 | left_cpos); | |
d0c7d708 MF |
5362 | if (ret) { |
5363 | mlog_errno(ret); | |
5364 | goto out; | |
5365 | } | |
5366 | } | |
5367 | } | |
5368 | ||
5369 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5370 | handle->h_buffer_credits, | |
5371 | path); | |
5372 | if (ret) { | |
5373 | mlog_errno(ret); | |
5374 | goto out; | |
5375 | } | |
5376 | ||
d9a0a1f8 | 5377 | ret = ocfs2_journal_access_path(et->et_ci, handle, path); |
d0c7d708 MF |
5378 | if (ret) { |
5379 | mlog_errno(ret); | |
5380 | goto out; | |
5381 | } | |
5382 | ||
d9a0a1f8 | 5383 | ret = ocfs2_journal_access_path(et->et_ci, handle, left_path); |
d0c7d708 MF |
5384 | if (ret) { |
5385 | mlog_errno(ret); | |
5386 | goto out; | |
5387 | } | |
5388 | ||
5389 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5390 | trunc_range = cpos + len; | |
5391 | ||
5392 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5393 | int next_free; | |
5394 | ||
5395 | memset(rec, 0, sizeof(*rec)); | |
5396 | ocfs2_cleanup_merge(el, index); | |
5397 | wants_rotate = 1; | |
5398 | ||
5399 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5400 | if (is_rightmost_tree_rec && next_free > 1) { | |
5401 | /* | |
5402 | * We skip the edge update if this path will | |
5403 | * be deleted by the rotate code. | |
5404 | */ | |
5405 | rec = &el->l_recs[next_free - 1]; | |
d401dc12 | 5406 | ocfs2_adjust_rightmost_records(handle, et, path, |
d0c7d708 MF |
5407 | rec); |
5408 | } | |
5409 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5410 | /* Remove leftmost portion of the record. */ | |
5411 | le32_add_cpu(&rec->e_cpos, len); | |
5412 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5413 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5414 | } else if (rec_range == trunc_range) { | |
5415 | /* Remove rightmost portion of the record */ | |
5416 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5417 | if (is_rightmost_tree_rec) | |
d401dc12 | 5418 | ocfs2_adjust_rightmost_records(handle, et, path, rec); |
d0c7d708 MF |
5419 | } else { |
5420 | /* Caller should have trapped this. */ | |
043beebb JB |
5421 | mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) " |
5422 | "(%u, %u)\n", | |
5423 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5424 | le32_to_cpu(rec->e_cpos), |
5425 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5426 | BUG(); | |
5427 | } | |
5428 | ||
5429 | if (left_path) { | |
5430 | int subtree_index; | |
5431 | ||
7dc02805 | 5432 | subtree_index = ocfs2_find_subtree_root(et, left_path, path); |
4619c73e | 5433 | ocfs2_complete_edge_insert(handle, left_path, path, |
d0c7d708 MF |
5434 | subtree_index); |
5435 | } | |
5436 | ||
5437 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5438 | ||
70f18c08 | 5439 | ret = ocfs2_rotate_tree_left(handle, et, path, dealloc); |
d0c7d708 MF |
5440 | if (ret) { |
5441 | mlog_errno(ret); | |
5442 | goto out; | |
5443 | } | |
5444 | ||
5445 | out: | |
5446 | ocfs2_free_path(left_path); | |
5447 | return ret; | |
5448 | } | |
5449 | ||
dbdcf6a4 | 5450 | int ocfs2_remove_extent(handle_t *handle, |
f99b9b7c | 5451 | struct ocfs2_extent_tree *et, |
dbdcf6a4 | 5452 | u32 cpos, u32 len, |
063c4561 | 5453 | struct ocfs2_alloc_context *meta_ac, |
f99b9b7c | 5454 | struct ocfs2_cached_dealloc_ctxt *dealloc) |
d0c7d708 MF |
5455 | { |
5456 | int ret, index; | |
5457 | u32 rec_range, trunc_range; | |
5458 | struct ocfs2_extent_rec *rec; | |
5459 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5460 | struct ocfs2_path *path = NULL; |
d0c7d708 | 5461 | |
4c911eef JB |
5462 | /* |
5463 | * XXX: Why are we truncating to 0 instead of wherever this | |
5464 | * affects us? | |
5465 | */ | |
5466 | ocfs2_et_extent_map_truncate(et, 0); | |
d0c7d708 | 5467 | |
ffdd7a54 | 5468 | path = ocfs2_new_path_from_et(et); |
d0c7d708 MF |
5469 | if (!path) { |
5470 | ret = -ENOMEM; | |
5471 | mlog_errno(ret); | |
5472 | goto out; | |
5473 | } | |
5474 | ||
facdb77f | 5475 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5476 | if (ret) { |
5477 | mlog_errno(ret); | |
5478 | goto out; | |
5479 | } | |
5480 | ||
5481 | el = path_leaf_el(path); | |
5482 | index = ocfs2_search_extent_list(el, cpos); | |
5483 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5484 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5485 | "Owner %llu has an extent at cpos %u which can no " | |
d0c7d708 | 5486 | "longer be found.\n", |
dbdcf6a4 JB |
5487 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5488 | cpos); | |
d0c7d708 MF |
5489 | ret = -EROFS; |
5490 | goto out; | |
5491 | } | |
5492 | ||
5493 | /* | |
5494 | * We have 3 cases of extent removal: | |
5495 | * 1) Range covers the entire extent rec | |
5496 | * 2) Range begins or ends on one edge of the extent rec | |
5497 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5498 | * | |
5499 | * For case 1 we remove the extent rec and left rotate to | |
5500 | * fill the hole. | |
5501 | * | |
5502 | * For case 2 we just shrink the existing extent rec, with a | |
5503 | * tree update if the shrinking edge is also the edge of an | |
5504 | * extent block. | |
5505 | * | |
5506 | * For case 3 we do a right split to turn the extent rec into | |
5507 | * something case 2 can handle. | |
5508 | */ | |
5509 | rec = &el->l_recs[index]; | |
5510 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5511 | trunc_range = cpos + len; | |
5512 | ||
5513 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5514 | ||
dbdcf6a4 | 5515 | mlog(0, "Owner %llu, remove (cpos %u, len %u). Existing index %d " |
d0c7d708 | 5516 | "(cpos %u, len %u)\n", |
dbdcf6a4 JB |
5517 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
5518 | cpos, len, index, | |
d0c7d708 MF |
5519 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); |
5520 | ||
5521 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
043beebb JB |
5522 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5523 | cpos, len); | |
d0c7d708 MF |
5524 | if (ret) { |
5525 | mlog_errno(ret); | |
5526 | goto out; | |
5527 | } | |
5528 | } else { | |
dbdcf6a4 | 5529 | ret = ocfs2_split_tree(handle, et, path, index, |
d0c7d708 MF |
5530 | trunc_range, meta_ac); |
5531 | if (ret) { | |
5532 | mlog_errno(ret); | |
5533 | goto out; | |
5534 | } | |
5535 | ||
5536 | /* | |
5537 | * The split could have manipulated the tree enough to | |
5538 | * move the record location, so we have to look for it again. | |
5539 | */ | |
5540 | ocfs2_reinit_path(path, 1); | |
5541 | ||
facdb77f | 5542 | ret = ocfs2_find_path(et->et_ci, path, cpos); |
d0c7d708 MF |
5543 | if (ret) { |
5544 | mlog_errno(ret); | |
5545 | goto out; | |
5546 | } | |
5547 | ||
5548 | el = path_leaf_el(path); | |
5549 | index = ocfs2_search_extent_list(el, cpos); | |
5550 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
dbdcf6a4 JB |
5551 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5552 | "Owner %llu: split at cpos %u lost record.", | |
5553 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), | |
d0c7d708 MF |
5554 | cpos); |
5555 | ret = -EROFS; | |
5556 | goto out; | |
5557 | } | |
5558 | ||
5559 | /* | |
5560 | * Double check our values here. If anything is fishy, | |
5561 | * it's easier to catch it at the top level. | |
5562 | */ | |
5563 | rec = &el->l_recs[index]; | |
5564 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5565 | ocfs2_rec_clusters(el, rec); | |
5566 | if (rec_range != trunc_range) { | |
dbdcf6a4 JB |
5567 | ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci), |
5568 | "Owner %llu: error after split at cpos %u" | |
d0c7d708 | 5569 | "trunc len %u, existing record is (%u,%u)", |
dbdcf6a4 | 5570 | (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), |
d0c7d708 MF |
5571 | cpos, len, le32_to_cpu(rec->e_cpos), |
5572 | ocfs2_rec_clusters(el, rec)); | |
5573 | ret = -EROFS; | |
5574 | goto out; | |
5575 | } | |
5576 | ||
043beebb JB |
5577 | ret = ocfs2_truncate_rec(handle, et, path, index, dealloc, |
5578 | cpos, len); | |
d0c7d708 MF |
5579 | if (ret) { |
5580 | mlog_errno(ret); | |
5581 | goto out; | |
5582 | } | |
5583 | } | |
5584 | ||
5585 | out: | |
5586 | ocfs2_free_path(path); | |
5587 | return ret; | |
5588 | } | |
5589 | ||
fecc0112 MF |
5590 | int ocfs2_remove_btree_range(struct inode *inode, |
5591 | struct ocfs2_extent_tree *et, | |
5592 | u32 cpos, u32 phys_cpos, u32 len, | |
5593 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
5594 | { | |
5595 | int ret; | |
5596 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
5597 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
5598 | struct inode *tl_inode = osb->osb_tl_inode; | |
5599 | handle_t *handle; | |
5600 | struct ocfs2_alloc_context *meta_ac = NULL; | |
5601 | ||
5602 | ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac); | |
5603 | if (ret) { | |
5604 | mlog_errno(ret); | |
5605 | return ret; | |
5606 | } | |
5607 | ||
5608 | mutex_lock(&tl_inode->i_mutex); | |
5609 | ||
5610 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5611 | ret = __ocfs2_flush_truncate_log(osb); | |
5612 | if (ret < 0) { | |
5613 | mlog_errno(ret); | |
5614 | goto out; | |
5615 | } | |
5616 | } | |
5617 | ||
a90714c1 | 5618 | handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb)); |
fecc0112 MF |
5619 | if (IS_ERR(handle)) { |
5620 | ret = PTR_ERR(handle); | |
5621 | mlog_errno(ret); | |
5622 | goto out; | |
5623 | } | |
5624 | ||
d9a0a1f8 | 5625 | ret = ocfs2_et_root_journal_access(handle, et, |
13723d00 | 5626 | OCFS2_JOURNAL_ACCESS_WRITE); |
fecc0112 MF |
5627 | if (ret) { |
5628 | mlog_errno(ret); | |
5629 | goto out; | |
5630 | } | |
5631 | ||
5dd4056d | 5632 | dquot_free_space_nodirty(inode, |
fd4ef231 MF |
5633 | ocfs2_clusters_to_bytes(inode->i_sb, len)); |
5634 | ||
dbdcf6a4 | 5635 | ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc); |
fecc0112 MF |
5636 | if (ret) { |
5637 | mlog_errno(ret); | |
5638 | goto out_commit; | |
5639 | } | |
5640 | ||
6136ca5f | 5641 | ocfs2_et_update_clusters(et, -len); |
fecc0112 | 5642 | |
ec20cec7 | 5643 | ocfs2_journal_dirty(handle, et->et_root_bh); |
fecc0112 MF |
5644 | |
5645 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
5646 | if (ret) | |
5647 | mlog_errno(ret); | |
5648 | ||
5649 | out_commit: | |
5650 | ocfs2_commit_trans(osb, handle); | |
5651 | out: | |
5652 | mutex_unlock(&tl_inode->i_mutex); | |
5653 | ||
5654 | if (meta_ac) | |
5655 | ocfs2_free_alloc_context(meta_ac); | |
5656 | ||
5657 | return ret; | |
5658 | } | |
5659 | ||
063c4561 | 5660 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5661 | { |
5662 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5663 | struct ocfs2_dinode *di; | |
5664 | struct ocfs2_truncate_log *tl; | |
5665 | ||
5666 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5667 | tl = &di->id2.i_dealloc; | |
5668 | ||
5669 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5670 | "slot %d, invalid truncate log parameters: used = " | |
5671 | "%u, count = %u\n", osb->slot_num, | |
5672 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5673 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5674 | } | |
5675 | ||
5676 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5677 | unsigned int new_start) | |
5678 | { | |
5679 | unsigned int tail_index; | |
5680 | unsigned int current_tail; | |
5681 | ||
5682 | /* No records, nothing to coalesce */ | |
5683 | if (!le16_to_cpu(tl->tl_used)) | |
5684 | return 0; | |
5685 | ||
5686 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5687 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5688 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5689 | ||
5690 | return current_tail == new_start; | |
5691 | } | |
5692 | ||
063c4561 MF |
5693 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5694 | handle_t *handle, | |
5695 | u64 start_blk, | |
5696 | unsigned int num_clusters) | |
ccd979bd MF |
5697 | { |
5698 | int status, index; | |
5699 | unsigned int start_cluster, tl_count; | |
5700 | struct inode *tl_inode = osb->osb_tl_inode; | |
5701 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5702 | struct ocfs2_dinode *di; | |
5703 | struct ocfs2_truncate_log *tl; | |
5704 | ||
b0697053 MF |
5705 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5706 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5707 | |
1b1dcc1b | 5708 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5709 | |
5710 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5711 | ||
5712 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5713 | |
10995aa2 JB |
5714 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5715 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5716 | * corruption is a code bug */ | |
5717 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5718 | ||
5719 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
5720 | tl_count = le16_to_cpu(tl->tl_count); |
5721 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5722 | tl_count == 0, | |
b0697053 MF |
5723 | "Truncate record count on #%llu invalid " |
5724 | "wanted %u, actual %u\n", | |
5725 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5726 | ocfs2_truncate_recs_per_inode(osb->sb), |
5727 | le16_to_cpu(tl->tl_count)); | |
5728 | ||
5729 | /* Caller should have known to flush before calling us. */ | |
5730 | index = le16_to_cpu(tl->tl_used); | |
5731 | if (index >= tl_count) { | |
5732 | status = -ENOSPC; | |
5733 | mlog_errno(status); | |
5734 | goto bail; | |
5735 | } | |
5736 | ||
0cf2f763 | 5737 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5738 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5739 | if (status < 0) { |
5740 | mlog_errno(status); | |
5741 | goto bail; | |
5742 | } | |
5743 | ||
5744 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5745 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5746 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5747 | |
5748 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5749 | /* | |
5750 | * Move index back to the record we are coalescing with. | |
5751 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5752 | */ | |
5753 | index--; | |
5754 | ||
5755 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5756 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5757 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5758 | num_clusters); | |
5759 | } else { | |
5760 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5761 | tl->tl_used = cpu_to_le16(index + 1); | |
5762 | } | |
5763 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5764 | ||
ec20cec7 | 5765 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd MF |
5766 | |
5767 | bail: | |
5768 | mlog_exit(status); | |
5769 | return status; | |
5770 | } | |
5771 | ||
5772 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5773 | handle_t *handle, |
ccd979bd MF |
5774 | struct inode *data_alloc_inode, |
5775 | struct buffer_head *data_alloc_bh) | |
5776 | { | |
5777 | int status = 0; | |
5778 | int i; | |
5779 | unsigned int num_clusters; | |
5780 | u64 start_blk; | |
5781 | struct ocfs2_truncate_rec rec; | |
5782 | struct ocfs2_dinode *di; | |
5783 | struct ocfs2_truncate_log *tl; | |
5784 | struct inode *tl_inode = osb->osb_tl_inode; | |
5785 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5786 | ||
5787 | mlog_entry_void(); | |
5788 | ||
5789 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5790 | tl = &di->id2.i_dealloc; | |
5791 | i = le16_to_cpu(tl->tl_used) - 1; | |
5792 | while (i >= 0) { | |
5793 | /* Caller has given us at least enough credits to | |
5794 | * update the truncate log dinode */ | |
0cf2f763 | 5795 | status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh, |
13723d00 | 5796 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
5797 | if (status < 0) { |
5798 | mlog_errno(status); | |
5799 | goto bail; | |
5800 | } | |
5801 | ||
5802 | tl->tl_used = cpu_to_le16(i); | |
5803 | ||
ec20cec7 | 5804 | ocfs2_journal_dirty(handle, tl_bh); |
ccd979bd MF |
5805 | |
5806 | /* TODO: Perhaps we can calculate the bulk of the | |
5807 | * credits up front rather than extending like | |
5808 | * this. */ | |
5809 | status = ocfs2_extend_trans(handle, | |
5810 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5811 | if (status < 0) { | |
5812 | mlog_errno(status); | |
5813 | goto bail; | |
5814 | } | |
5815 | ||
5816 | rec = tl->tl_recs[i]; | |
5817 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5818 | le32_to_cpu(rec.t_start)); | |
5819 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5820 | ||
5821 | /* if start_blk is not set, we ignore the record as | |
5822 | * invalid. */ | |
5823 | if (start_blk) { | |
5824 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5825 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5826 | ||
5827 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5828 | data_alloc_bh, start_blk, | |
5829 | num_clusters); | |
5830 | if (status < 0) { | |
5831 | mlog_errno(status); | |
5832 | goto bail; | |
5833 | } | |
5834 | } | |
5835 | i--; | |
5836 | } | |
5837 | ||
5838 | bail: | |
5839 | mlog_exit(status); | |
5840 | return status; | |
5841 | } | |
5842 | ||
1b1dcc1b | 5843 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5844 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5845 | { |
5846 | int status; | |
5847 | unsigned int num_to_flush; | |
1fabe148 | 5848 | handle_t *handle; |
ccd979bd MF |
5849 | struct inode *tl_inode = osb->osb_tl_inode; |
5850 | struct inode *data_alloc_inode = NULL; | |
5851 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5852 | struct buffer_head *data_alloc_bh = NULL; | |
5853 | struct ocfs2_dinode *di; | |
5854 | struct ocfs2_truncate_log *tl; | |
5855 | ||
5856 | mlog_entry_void(); | |
5857 | ||
1b1dcc1b | 5858 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5859 | |
5860 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 5861 | |
10995aa2 JB |
5862 | /* tl_bh is loaded from ocfs2_truncate_log_init(). It's validated |
5863 | * by the underlying call to ocfs2_read_inode_block(), so any | |
5864 | * corruption is a code bug */ | |
5865 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
5866 | ||
5867 | tl = &di->id2.i_dealloc; | |
ccd979bd | 5868 | num_to_flush = le16_to_cpu(tl->tl_used); |
b0697053 MF |
5869 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5870 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5871 | if (!num_to_flush) { |
5872 | status = 0; | |
e08dc8b9 | 5873 | goto out; |
ccd979bd MF |
5874 | } |
5875 | ||
5876 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5877 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5878 | OCFS2_INVALID_SLOT); | |
5879 | if (!data_alloc_inode) { | |
5880 | status = -EINVAL; | |
5881 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5882 | goto out; |
ccd979bd MF |
5883 | } |
5884 | ||
e08dc8b9 MF |
5885 | mutex_lock(&data_alloc_inode->i_mutex); |
5886 | ||
e63aecb6 | 5887 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5888 | if (status < 0) { |
5889 | mlog_errno(status); | |
e08dc8b9 | 5890 | goto out_mutex; |
ccd979bd MF |
5891 | } |
5892 | ||
65eff9cc | 5893 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5894 | if (IS_ERR(handle)) { |
5895 | status = PTR_ERR(handle); | |
ccd979bd | 5896 | mlog_errno(status); |
e08dc8b9 | 5897 | goto out_unlock; |
ccd979bd MF |
5898 | } |
5899 | ||
5900 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5901 | data_alloc_bh); | |
e08dc8b9 | 5902 | if (status < 0) |
ccd979bd | 5903 | mlog_errno(status); |
ccd979bd | 5904 | |
02dc1af4 | 5905 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5906 | |
e08dc8b9 MF |
5907 | out_unlock: |
5908 | brelse(data_alloc_bh); | |
e63aecb6 | 5909 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5910 | |
e08dc8b9 MF |
5911 | out_mutex: |
5912 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5913 | iput(data_alloc_inode); | |
ccd979bd | 5914 | |
e08dc8b9 | 5915 | out: |
ccd979bd MF |
5916 | mlog_exit(status); |
5917 | return status; | |
5918 | } | |
5919 | ||
5920 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5921 | { | |
5922 | int status; | |
5923 | struct inode *tl_inode = osb->osb_tl_inode; | |
5924 | ||
1b1dcc1b | 5925 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5926 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5927 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5928 | |
5929 | return status; | |
5930 | } | |
5931 | ||
c4028958 | 5932 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5933 | { |
5934 | int status; | |
c4028958 DH |
5935 | struct ocfs2_super *osb = |
5936 | container_of(work, struct ocfs2_super, | |
5937 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5938 | |
5939 | mlog_entry_void(); | |
5940 | ||
5941 | status = ocfs2_flush_truncate_log(osb); | |
5942 | if (status < 0) | |
5943 | mlog_errno(status); | |
4d0ddb2c | 5944 | else |
b89c5428 | 5945 | ocfs2_init_steal_slots(osb); |
ccd979bd MF |
5946 | |
5947 | mlog_exit(status); | |
5948 | } | |
5949 | ||
5950 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5951 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5952 | int cancel) | |
5953 | { | |
5954 | if (osb->osb_tl_inode) { | |
5955 | /* We want to push off log flushes while truncates are | |
5956 | * still running. */ | |
5957 | if (cancel) | |
5958 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5959 | ||
5960 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5961 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5962 | } | |
5963 | } | |
5964 | ||
5965 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5966 | int slot_num, | |
5967 | struct inode **tl_inode, | |
5968 | struct buffer_head **tl_bh) | |
5969 | { | |
5970 | int status; | |
5971 | struct inode *inode = NULL; | |
5972 | struct buffer_head *bh = NULL; | |
5973 | ||
5974 | inode = ocfs2_get_system_file_inode(osb, | |
5975 | TRUNCATE_LOG_SYSTEM_INODE, | |
5976 | slot_num); | |
5977 | if (!inode) { | |
5978 | status = -EINVAL; | |
5979 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5980 | goto bail; | |
5981 | } | |
5982 | ||
b657c95c | 5983 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
5984 | if (status < 0) { |
5985 | iput(inode); | |
5986 | mlog_errno(status); | |
5987 | goto bail; | |
5988 | } | |
5989 | ||
5990 | *tl_inode = inode; | |
5991 | *tl_bh = bh; | |
5992 | bail: | |
5993 | mlog_exit(status); | |
5994 | return status; | |
5995 | } | |
5996 | ||
5997 | /* called during the 1st stage of node recovery. we stamp a clean | |
5998 | * truncate log and pass back a copy for processing later. if the | |
5999 | * truncate log does not require processing, a *tl_copy is set to | |
6000 | * NULL. */ | |
6001 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
6002 | int slot_num, | |
6003 | struct ocfs2_dinode **tl_copy) | |
6004 | { | |
6005 | int status; | |
6006 | struct inode *tl_inode = NULL; | |
6007 | struct buffer_head *tl_bh = NULL; | |
6008 | struct ocfs2_dinode *di; | |
6009 | struct ocfs2_truncate_log *tl; | |
6010 | ||
6011 | *tl_copy = NULL; | |
6012 | ||
6013 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
6014 | ||
6015 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
6016 | if (status < 0) { | |
6017 | mlog_errno(status); | |
6018 | goto bail; | |
6019 | } | |
6020 | ||
6021 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
ccd979bd | 6022 | |
10995aa2 JB |
6023 | /* tl_bh is loaded from ocfs2_get_truncate_log_info(). It's |
6024 | * validated by the underlying call to ocfs2_read_inode_block(), | |
6025 | * so any corruption is a code bug */ | |
6026 | BUG_ON(!OCFS2_IS_VALID_DINODE(di)); | |
6027 | ||
6028 | tl = &di->id2.i_dealloc; | |
ccd979bd MF |
6029 | if (le16_to_cpu(tl->tl_used)) { |
6030 | mlog(0, "We'll have %u logs to recover\n", | |
6031 | le16_to_cpu(tl->tl_used)); | |
6032 | ||
6033 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
6034 | if (!(*tl_copy)) { | |
6035 | status = -ENOMEM; | |
6036 | mlog_errno(status); | |
6037 | goto bail; | |
6038 | } | |
6039 | ||
6040 | /* Assuming the write-out below goes well, this copy | |
6041 | * will be passed back to recovery for processing. */ | |
6042 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
6043 | ||
6044 | /* All we need to do to clear the truncate log is set | |
6045 | * tl_used. */ | |
6046 | tl->tl_used = 0; | |
6047 | ||
13723d00 | 6048 | ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check); |
8cb471e8 | 6049 | status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode)); |
ccd979bd MF |
6050 | if (status < 0) { |
6051 | mlog_errno(status); | |
6052 | goto bail; | |
6053 | } | |
6054 | } | |
6055 | ||
6056 | bail: | |
6057 | if (tl_inode) | |
6058 | iput(tl_inode); | |
a81cb88b | 6059 | brelse(tl_bh); |
ccd979bd MF |
6060 | |
6061 | if (status < 0 && (*tl_copy)) { | |
6062 | kfree(*tl_copy); | |
6063 | *tl_copy = NULL; | |
6064 | } | |
6065 | ||
6066 | mlog_exit(status); | |
6067 | return status; | |
6068 | } | |
6069 | ||
6070 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
6071 | struct ocfs2_dinode *tl_copy) | |
6072 | { | |
6073 | int status = 0; | |
6074 | int i; | |
6075 | unsigned int clusters, num_recs, start_cluster; | |
6076 | u64 start_blk; | |
1fabe148 | 6077 | handle_t *handle; |
ccd979bd MF |
6078 | struct inode *tl_inode = osb->osb_tl_inode; |
6079 | struct ocfs2_truncate_log *tl; | |
6080 | ||
6081 | mlog_entry_void(); | |
6082 | ||
6083 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
6084 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
6085 | return -EINVAL; | |
6086 | } | |
6087 | ||
6088 | tl = &tl_copy->id2.i_dealloc; | |
6089 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 6090 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 6091 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 6092 | |
1b1dcc1b | 6093 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6094 | for(i = 0; i < num_recs; i++) { |
6095 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6096 | status = __ocfs2_flush_truncate_log(osb); | |
6097 | if (status < 0) { | |
6098 | mlog_errno(status); | |
6099 | goto bail_up; | |
6100 | } | |
6101 | } | |
6102 | ||
65eff9cc | 6103 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
6104 | if (IS_ERR(handle)) { |
6105 | status = PTR_ERR(handle); | |
6106 | mlog_errno(status); | |
6107 | goto bail_up; | |
6108 | } | |
6109 | ||
6110 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
6111 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
6112 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
6113 | ||
6114 | status = ocfs2_truncate_log_append(osb, handle, | |
6115 | start_blk, clusters); | |
02dc1af4 | 6116 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6117 | if (status < 0) { |
6118 | mlog_errno(status); | |
6119 | goto bail_up; | |
6120 | } | |
6121 | } | |
6122 | ||
6123 | bail_up: | |
1b1dcc1b | 6124 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6125 | |
6126 | mlog_exit(status); | |
6127 | return status; | |
6128 | } | |
6129 | ||
6130 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
6131 | { | |
6132 | int status; | |
6133 | struct inode *tl_inode = osb->osb_tl_inode; | |
6134 | ||
6135 | mlog_entry_void(); | |
6136 | ||
6137 | if (tl_inode) { | |
6138 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
6139 | flush_workqueue(ocfs2_wq); | |
6140 | ||
6141 | status = ocfs2_flush_truncate_log(osb); | |
6142 | if (status < 0) | |
6143 | mlog_errno(status); | |
6144 | ||
6145 | brelse(osb->osb_tl_bh); | |
6146 | iput(osb->osb_tl_inode); | |
6147 | } | |
6148 | ||
6149 | mlog_exit_void(); | |
6150 | } | |
6151 | ||
6152 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
6153 | { | |
6154 | int status; | |
6155 | struct inode *tl_inode = NULL; | |
6156 | struct buffer_head *tl_bh = NULL; | |
6157 | ||
6158 | mlog_entry_void(); | |
6159 | ||
6160 | status = ocfs2_get_truncate_log_info(osb, | |
6161 | osb->slot_num, | |
6162 | &tl_inode, | |
6163 | &tl_bh); | |
6164 | if (status < 0) | |
6165 | mlog_errno(status); | |
6166 | ||
6167 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
6168 | * osb->osb_tl_inode so we don't set any of the osb variables | |
6169 | * until we're sure all is well. */ | |
c4028958 DH |
6170 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
6171 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
6172 | osb->osb_tl_bh = tl_bh; |
6173 | osb->osb_tl_inode = tl_inode; | |
6174 | ||
6175 | mlog_exit(status); | |
6176 | return status; | |
6177 | } | |
6178 | ||
2b604351 MF |
6179 | /* |
6180 | * Delayed de-allocation of suballocator blocks. | |
6181 | * | |
6182 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
6183 | * | |
6184 | * The locking for this can get extremely complicated, especially when | |
6185 | * the suballocator inodes to delete from aren't known until deep | |
6186 | * within an unrelated codepath. | |
6187 | * | |
6188 | * ocfs2_extent_block structures are a good example of this - an inode | |
6189 | * btree could have been grown by any number of nodes each allocating | |
6190 | * out of their own suballoc inode. | |
6191 | * | |
6192 | * These structures allow the delay of block de-allocation until a | |
6193 | * later time, when locking of multiple cluster inodes won't cause | |
6194 | * deadlock. | |
6195 | */ | |
6196 | ||
6197 | /* | |
2891d290 TM |
6198 | * Describe a single bit freed from a suballocator. For the block |
6199 | * suballocators, it represents one block. For the global cluster | |
6200 | * allocator, it represents some clusters and free_bit indicates | |
6201 | * clusters number. | |
2b604351 MF |
6202 | */ |
6203 | struct ocfs2_cached_block_free { | |
6204 | struct ocfs2_cached_block_free *free_next; | |
6205 | u64 free_blk; | |
6206 | unsigned int free_bit; | |
6207 | }; | |
6208 | ||
6209 | struct ocfs2_per_slot_free_list { | |
6210 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
6211 | int f_inode_type; | |
6212 | int f_slot; | |
6213 | struct ocfs2_cached_block_free *f_first; | |
6214 | }; | |
6215 | ||
2891d290 TM |
6216 | static int ocfs2_free_cached_blocks(struct ocfs2_super *osb, |
6217 | int sysfile_type, | |
6218 | int slot, | |
6219 | struct ocfs2_cached_block_free *head) | |
2b604351 MF |
6220 | { |
6221 | int ret; | |
6222 | u64 bg_blkno; | |
6223 | handle_t *handle; | |
6224 | struct inode *inode; | |
6225 | struct buffer_head *di_bh = NULL; | |
6226 | struct ocfs2_cached_block_free *tmp; | |
6227 | ||
6228 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
6229 | if (!inode) { | |
6230 | ret = -EINVAL; | |
6231 | mlog_errno(ret); | |
6232 | goto out; | |
6233 | } | |
6234 | ||
6235 | mutex_lock(&inode->i_mutex); | |
6236 | ||
e63aecb6 | 6237 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
6238 | if (ret) { |
6239 | mlog_errno(ret); | |
6240 | goto out_mutex; | |
6241 | } | |
6242 | ||
6243 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
6244 | if (IS_ERR(handle)) { | |
6245 | ret = PTR_ERR(handle); | |
6246 | mlog_errno(ret); | |
6247 | goto out_unlock; | |
6248 | } | |
6249 | ||
6250 | while (head) { | |
6251 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
6252 | head->free_bit); | |
6253 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
6254 | head->free_bit, (unsigned long long)head->free_blk); | |
6255 | ||
6256 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
6257 | head->free_bit, bg_blkno, 1); | |
6258 | if (ret) { | |
6259 | mlog_errno(ret); | |
6260 | goto out_journal; | |
6261 | } | |
6262 | ||
6263 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
6264 | if (ret) { | |
6265 | mlog_errno(ret); | |
6266 | goto out_journal; | |
6267 | } | |
6268 | ||
6269 | tmp = head; | |
6270 | head = head->free_next; | |
6271 | kfree(tmp); | |
6272 | } | |
6273 | ||
6274 | out_journal: | |
6275 | ocfs2_commit_trans(osb, handle); | |
6276 | ||
6277 | out_unlock: | |
e63aecb6 | 6278 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
6279 | brelse(di_bh); |
6280 | out_mutex: | |
6281 | mutex_unlock(&inode->i_mutex); | |
6282 | iput(inode); | |
6283 | out: | |
6284 | while(head) { | |
6285 | /* Premature exit may have left some dangling items. */ | |
6286 | tmp = head; | |
6287 | head = head->free_next; | |
6288 | kfree(tmp); | |
6289 | } | |
6290 | ||
6291 | return ret; | |
6292 | } | |
6293 | ||
2891d290 TM |
6294 | int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6295 | u64 blkno, unsigned int bit) | |
6296 | { | |
6297 | int ret = 0; | |
6298 | struct ocfs2_cached_block_free *item; | |
6299 | ||
6300 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6301 | if (item == NULL) { | |
6302 | ret = -ENOMEM; | |
6303 | mlog_errno(ret); | |
6304 | return ret; | |
6305 | } | |
6306 | ||
6307 | mlog(0, "Insert clusters: (bit %u, blk %llu)\n", | |
6308 | bit, (unsigned long long)blkno); | |
6309 | ||
6310 | item->free_blk = blkno; | |
6311 | item->free_bit = bit; | |
6312 | item->free_next = ctxt->c_global_allocator; | |
6313 | ||
6314 | ctxt->c_global_allocator = item; | |
6315 | return ret; | |
6316 | } | |
6317 | ||
6318 | static int ocfs2_free_cached_clusters(struct ocfs2_super *osb, | |
6319 | struct ocfs2_cached_block_free *head) | |
6320 | { | |
6321 | struct ocfs2_cached_block_free *tmp; | |
6322 | struct inode *tl_inode = osb->osb_tl_inode; | |
6323 | handle_t *handle; | |
6324 | int ret = 0; | |
6325 | ||
6326 | mutex_lock(&tl_inode->i_mutex); | |
6327 | ||
6328 | while (head) { | |
6329 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6330 | ret = __ocfs2_flush_truncate_log(osb); | |
6331 | if (ret < 0) { | |
6332 | mlog_errno(ret); | |
6333 | break; | |
6334 | } | |
6335 | } | |
6336 | ||
6337 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); | |
6338 | if (IS_ERR(handle)) { | |
6339 | ret = PTR_ERR(handle); | |
6340 | mlog_errno(ret); | |
6341 | break; | |
6342 | } | |
6343 | ||
6344 | ret = ocfs2_truncate_log_append(osb, handle, head->free_blk, | |
6345 | head->free_bit); | |
6346 | ||
6347 | ocfs2_commit_trans(osb, handle); | |
6348 | tmp = head; | |
6349 | head = head->free_next; | |
6350 | kfree(tmp); | |
6351 | ||
6352 | if (ret < 0) { | |
6353 | mlog_errno(ret); | |
6354 | break; | |
6355 | } | |
6356 | } | |
6357 | ||
6358 | mutex_unlock(&tl_inode->i_mutex); | |
6359 | ||
6360 | while (head) { | |
6361 | /* Premature exit may have left some dangling items. */ | |
6362 | tmp = head; | |
6363 | head = head->free_next; | |
6364 | kfree(tmp); | |
6365 | } | |
6366 | ||
6367 | return ret; | |
6368 | } | |
6369 | ||
2b604351 MF |
6370 | int ocfs2_run_deallocs(struct ocfs2_super *osb, |
6371 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6372 | { | |
6373 | int ret = 0, ret2; | |
6374 | struct ocfs2_per_slot_free_list *fl; | |
6375 | ||
6376 | if (!ctxt) | |
6377 | return 0; | |
6378 | ||
6379 | while (ctxt->c_first_suballocator) { | |
6380 | fl = ctxt->c_first_suballocator; | |
6381 | ||
6382 | if (fl->f_first) { | |
6383 | mlog(0, "Free items: (type %u, slot %d)\n", | |
6384 | fl->f_inode_type, fl->f_slot); | |
2891d290 TM |
6385 | ret2 = ocfs2_free_cached_blocks(osb, |
6386 | fl->f_inode_type, | |
6387 | fl->f_slot, | |
6388 | fl->f_first); | |
2b604351 MF |
6389 | if (ret2) |
6390 | mlog_errno(ret2); | |
6391 | if (!ret) | |
6392 | ret = ret2; | |
6393 | } | |
6394 | ||
6395 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
6396 | kfree(fl); | |
6397 | } | |
6398 | ||
2891d290 TM |
6399 | if (ctxt->c_global_allocator) { |
6400 | ret2 = ocfs2_free_cached_clusters(osb, | |
6401 | ctxt->c_global_allocator); | |
6402 | if (ret2) | |
6403 | mlog_errno(ret2); | |
6404 | if (!ret) | |
6405 | ret = ret2; | |
6406 | ||
6407 | ctxt->c_global_allocator = NULL; | |
6408 | } | |
6409 | ||
2b604351 MF |
6410 | return ret; |
6411 | } | |
6412 | ||
6413 | static struct ocfs2_per_slot_free_list * | |
6414 | ocfs2_find_per_slot_free_list(int type, | |
6415 | int slot, | |
6416 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
6417 | { | |
6418 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
6419 | ||
6420 | while (fl) { | |
6421 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
6422 | return fl; | |
6423 | ||
6424 | fl = fl->f_next_suballocator; | |
6425 | } | |
6426 | ||
6427 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
6428 | if (fl) { | |
6429 | fl->f_inode_type = type; | |
6430 | fl->f_slot = slot; | |
6431 | fl->f_first = NULL; | |
6432 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
6433 | ||
6434 | ctxt->c_first_suballocator = fl; | |
6435 | } | |
6436 | return fl; | |
6437 | } | |
6438 | ||
1823cb0b TM |
6439 | int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6440 | int type, int slot, u64 blkno, | |
6441 | unsigned int bit) | |
2b604351 MF |
6442 | { |
6443 | int ret; | |
6444 | struct ocfs2_per_slot_free_list *fl; | |
6445 | struct ocfs2_cached_block_free *item; | |
6446 | ||
6447 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6448 | if (fl == NULL) { | |
6449 | ret = -ENOMEM; | |
6450 | mlog_errno(ret); | |
6451 | goto out; | |
6452 | } | |
6453 | ||
6454 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6455 | if (item == NULL) { | |
6456 | ret = -ENOMEM; | |
6457 | mlog_errno(ret); | |
6458 | goto out; | |
6459 | } | |
6460 | ||
6461 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6462 | type, slot, bit, (unsigned long long)blkno); | |
6463 | ||
6464 | item->free_blk = blkno; | |
6465 | item->free_bit = bit; | |
6466 | item->free_next = fl->f_first; | |
6467 | ||
6468 | fl->f_first = item; | |
6469 | ||
6470 | ret = 0; | |
6471 | out: | |
6472 | return ret; | |
6473 | } | |
6474 | ||
59a5e416 MF |
6475 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6476 | struct ocfs2_extent_block *eb) | |
6477 | { | |
6478 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6479 | le16_to_cpu(eb->h_suballoc_slot), | |
6480 | le64_to_cpu(eb->h_blkno), | |
6481 | le16_to_cpu(eb->h_suballoc_bit)); | |
6482 | } | |
6483 | ||
ccd979bd MF |
6484 | /* This function will figure out whether the currently last extent |
6485 | * block will be deleted, and if it will, what the new last extent | |
6486 | * block will be so we can update his h_next_leaf_blk field, as well | |
6487 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6488 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6489 | unsigned int clusters_to_del, |
dcd0538f | 6490 | struct ocfs2_path *path, |
ccd979bd MF |
6491 | struct buffer_head **new_last_eb) |
6492 | { | |
3a0782d0 | 6493 | int next_free, ret = 0; |
dcd0538f | 6494 | u32 cpos; |
3a0782d0 | 6495 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6496 | struct ocfs2_extent_block *eb; |
6497 | struct ocfs2_extent_list *el; | |
6498 | struct buffer_head *bh = NULL; | |
6499 | ||
6500 | *new_last_eb = NULL; | |
6501 | ||
ccd979bd | 6502 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6503 | if (!path->p_tree_depth) |
6504 | goto out; | |
ccd979bd MF |
6505 | |
6506 | /* trunc to zero special case - this makes tree_depth = 0 | |
6507 | * regardless of what it is. */ | |
3a0782d0 | 6508 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6509 | goto out; |
ccd979bd | 6510 | |
dcd0538f | 6511 | el = path_leaf_el(path); |
ccd979bd MF |
6512 | BUG_ON(!el->l_next_free_rec); |
6513 | ||
3a0782d0 MF |
6514 | /* |
6515 | * Make sure that this extent list will actually be empty | |
6516 | * after we clear away the data. We can shortcut out if | |
6517 | * there's more than one non-empty extent in the | |
6518 | * list. Otherwise, a check of the remaining extent is | |
6519 | * necessary. | |
6520 | */ | |
6521 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6522 | rec = NULL; | |
dcd0538f | 6523 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6524 | if (next_free > 2) |
dcd0538f | 6525 | goto out; |
3a0782d0 MF |
6526 | |
6527 | /* We may have a valid extent in index 1, check it. */ | |
6528 | if (next_free == 2) | |
6529 | rec = &el->l_recs[1]; | |
6530 | ||
6531 | /* | |
6532 | * Fall through - no more nonempty extents, so we want | |
6533 | * to delete this leaf. | |
6534 | */ | |
6535 | } else { | |
6536 | if (next_free > 1) | |
6537 | goto out; | |
6538 | ||
6539 | rec = &el->l_recs[0]; | |
6540 | } | |
6541 | ||
6542 | if (rec) { | |
6543 | /* | |
6544 | * Check it we'll only be trimming off the end of this | |
6545 | * cluster. | |
6546 | */ | |
e48edee2 | 6547 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6548 | goto out; |
6549 | } | |
ccd979bd | 6550 | |
dcd0538f MF |
6551 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6552 | if (ret) { | |
6553 | mlog_errno(ret); | |
6554 | goto out; | |
6555 | } | |
ccd979bd | 6556 | |
facdb77f | 6557 | ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh); |
dcd0538f MF |
6558 | if (ret) { |
6559 | mlog_errno(ret); | |
6560 | goto out; | |
6561 | } | |
ccd979bd | 6562 | |
dcd0538f MF |
6563 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6564 | el = &eb->h_list; | |
5e96581a JB |
6565 | |
6566 | /* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block(). | |
6567 | * Any corruption is a code bug. */ | |
6568 | BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb)); | |
ccd979bd MF |
6569 | |
6570 | *new_last_eb = bh; | |
6571 | get_bh(*new_last_eb); | |
dcd0538f MF |
6572 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6573 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6574 | out: | |
6575 | brelse(bh); | |
ccd979bd | 6576 | |
dcd0538f | 6577 | return ret; |
ccd979bd MF |
6578 | } |
6579 | ||
3a0782d0 MF |
6580 | /* |
6581 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6582 | * during truncate. | |
6583 | * | |
6584 | * The caller needs to: | |
6585 | * - start journaling of each path component. | |
6586 | * - compute and fully set up any new last ext block | |
6587 | */ | |
6588 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6589 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
bcbbb24a | 6590 | u32 clusters_to_del, u64 *delete_start, u8 *flags) |
3a0782d0 MF |
6591 | { |
6592 | int ret, i, index = path->p_tree_depth; | |
6593 | u32 new_edge = 0; | |
6594 | u64 deleted_eb = 0; | |
6595 | struct buffer_head *bh; | |
6596 | struct ocfs2_extent_list *el; | |
6597 | struct ocfs2_extent_rec *rec; | |
6598 | ||
6599 | *delete_start = 0; | |
bcbbb24a | 6600 | *flags = 0; |
3a0782d0 MF |
6601 | |
6602 | while (index >= 0) { | |
6603 | bh = path->p_node[index].bh; | |
6604 | el = path->p_node[index].el; | |
6605 | ||
6606 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6607 | index, (unsigned long long)bh->b_blocknr); | |
6608 | ||
6609 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6610 | ||
6611 | if (index != | |
6612 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6613 | ocfs2_error(inode->i_sb, | |
6614 | "Inode %lu has invalid ext. block %llu", | |
6615 | inode->i_ino, | |
6616 | (unsigned long long)bh->b_blocknr); | |
6617 | ret = -EROFS; | |
6618 | goto out; | |
6619 | } | |
6620 | ||
6621 | find_tail_record: | |
6622 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6623 | rec = &el->l_recs[i]; | |
6624 | ||
6625 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6626 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6627 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6628 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6629 | le16_to_cpu(el->l_next_free_rec)); | |
6630 | ||
e48edee2 | 6631 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6632 | |
6633 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6634 | /* | |
6635 | * If the leaf block contains a single empty | |
6636 | * extent and no records, we can just remove | |
6637 | * the block. | |
6638 | */ | |
6639 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6640 | memset(rec, 0, | |
6641 | sizeof(struct ocfs2_extent_rec)); | |
6642 | el->l_next_free_rec = cpu_to_le16(0); | |
6643 | ||
6644 | goto delete; | |
6645 | } | |
6646 | ||
6647 | /* | |
6648 | * Remove any empty extents by shifting things | |
6649 | * left. That should make life much easier on | |
6650 | * the code below. This condition is rare | |
6651 | * enough that we shouldn't see a performance | |
6652 | * hit. | |
6653 | */ | |
6654 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6655 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6656 | ||
6657 | for(i = 0; | |
6658 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6659 | el->l_recs[i] = el->l_recs[i + 1]; | |
6660 | ||
6661 | memset(&el->l_recs[i], 0, | |
6662 | sizeof(struct ocfs2_extent_rec)); | |
6663 | ||
6664 | /* | |
6665 | * We've modified our extent list. The | |
6666 | * simplest way to handle this change | |
6667 | * is to being the search from the | |
6668 | * start again. | |
6669 | */ | |
6670 | goto find_tail_record; | |
6671 | } | |
6672 | ||
e48edee2 | 6673 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6674 | |
6675 | /* | |
6676 | * We'll use "new_edge" on our way back up the | |
6677 | * tree to know what our rightmost cpos is. | |
6678 | */ | |
e48edee2 | 6679 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6680 | new_edge += le32_to_cpu(rec->e_cpos); |
6681 | ||
6682 | /* | |
6683 | * The caller will use this to delete data blocks. | |
6684 | */ | |
6685 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6686 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6687 | le16_to_cpu(rec->e_leaf_clusters)); |
bcbbb24a | 6688 | *flags = rec->e_flags; |
3a0782d0 MF |
6689 | |
6690 | /* | |
6691 | * If it's now empty, remove this record. | |
6692 | */ | |
e48edee2 | 6693 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6694 | memset(rec, 0, |
6695 | sizeof(struct ocfs2_extent_rec)); | |
6696 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6697 | } | |
6698 | } else { | |
6699 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6700 | memset(rec, 0, | |
6701 | sizeof(struct ocfs2_extent_rec)); | |
6702 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6703 | ||
6704 | goto delete; | |
6705 | } | |
6706 | ||
6707 | /* Can this actually happen? */ | |
6708 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6709 | goto delete; | |
6710 | ||
6711 | /* | |
6712 | * We never actually deleted any clusters | |
6713 | * because our leaf was empty. There's no | |
6714 | * reason to adjust the rightmost edge then. | |
6715 | */ | |
6716 | if (new_edge == 0) | |
6717 | goto delete; | |
6718 | ||
e48edee2 MF |
6719 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6720 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6721 | -le32_to_cpu(rec->e_cpos)); |
6722 | ||
6723 | /* | |
6724 | * A deleted child record should have been | |
6725 | * caught above. | |
6726 | */ | |
e48edee2 | 6727 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6728 | } |
6729 | ||
6730 | delete: | |
ec20cec7 | 6731 | ocfs2_journal_dirty(handle, bh); |
3a0782d0 MF |
6732 | |
6733 | mlog(0, "extent list container %llu, after: record %d: " | |
6734 | "(%u, %u, %llu), next = %u.\n", | |
6735 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6736 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6737 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6738 | le16_to_cpu(el->l_next_free_rec)); | |
6739 | ||
6740 | /* | |
6741 | * We must be careful to only attempt delete of an | |
6742 | * extent block (and not the root inode block). | |
6743 | */ | |
6744 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6745 | struct ocfs2_extent_block *eb = | |
6746 | (struct ocfs2_extent_block *)bh->b_data; | |
6747 | ||
6748 | /* | |
6749 | * Save this for use when processing the | |
6750 | * parent block. | |
6751 | */ | |
6752 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6753 | ||
6754 | mlog(0, "deleting this extent block.\n"); | |
6755 | ||
8cb471e8 | 6756 | ocfs2_remove_from_cache(INODE_CACHE(inode), bh); |
3a0782d0 | 6757 | |
e48edee2 | 6758 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6759 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6760 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6761 | ||
59a5e416 MF |
6762 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6763 | /* An error here is not fatal. */ | |
6764 | if (ret < 0) | |
6765 | mlog_errno(ret); | |
3a0782d0 MF |
6766 | } else { |
6767 | deleted_eb = 0; | |
6768 | } | |
6769 | ||
6770 | index--; | |
6771 | } | |
6772 | ||
6773 | ret = 0; | |
6774 | out: | |
6775 | return ret; | |
6776 | } | |
6777 | ||
ccd979bd MF |
6778 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6779 | unsigned int clusters_to_del, | |
6780 | struct inode *inode, | |
6781 | struct buffer_head *fe_bh, | |
1fabe148 | 6782 | handle_t *handle, |
dcd0538f | 6783 | struct ocfs2_truncate_context *tc, |
bcbbb24a TM |
6784 | struct ocfs2_path *path, |
6785 | struct ocfs2_alloc_context *meta_ac) | |
ccd979bd | 6786 | { |
3a0782d0 | 6787 | int status; |
ccd979bd | 6788 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6789 | struct ocfs2_extent_block *last_eb = NULL; |
6790 | struct ocfs2_extent_list *el; | |
ccd979bd | 6791 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd | 6792 | u64 delete_blk = 0; |
bcbbb24a | 6793 | u8 rec_flags; |
ccd979bd MF |
6794 | |
6795 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6796 | ||
3a0782d0 | 6797 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6798 | path, &last_eb_bh); |
ccd979bd MF |
6799 | if (status < 0) { |
6800 | mlog_errno(status); | |
6801 | goto bail; | |
6802 | } | |
dcd0538f MF |
6803 | |
6804 | /* | |
6805 | * Each component will be touched, so we might as well journal | |
6806 | * here to avoid having to handle errors later. | |
6807 | */ | |
0cf2f763 | 6808 | status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path); |
3a0782d0 MF |
6809 | if (status < 0) { |
6810 | mlog_errno(status); | |
6811 | goto bail; | |
dcd0538f MF |
6812 | } |
6813 | ||
6814 | if (last_eb_bh) { | |
0cf2f763 | 6815 | status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh, |
13723d00 | 6816 | OCFS2_JOURNAL_ACCESS_WRITE); |
dcd0538f MF |
6817 | if (status < 0) { |
6818 | mlog_errno(status); | |
6819 | goto bail; | |
6820 | } | |
6821 | ||
ccd979bd | 6822 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6823 | } |
ccd979bd | 6824 | |
dcd0538f MF |
6825 | el = &(fe->id2.i_list); |
6826 | ||
6827 | /* | |
6828 | * Lower levels depend on this never happening, but it's best | |
6829 | * to check it up here before changing the tree. | |
6830 | */ | |
e48edee2 | 6831 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6832 | ocfs2_error(inode->i_sb, |
6833 | "Inode %lu has an empty extent record, depth %u\n", | |
6834 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6835 | status = -EROFS; |
ccd979bd MF |
6836 | goto bail; |
6837 | } | |
ccd979bd | 6838 | |
5dd4056d | 6839 | dquot_free_space_nodirty(inode, |
a90714c1 | 6840 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_del)); |
ccd979bd MF |
6841 | spin_lock(&OCFS2_I(inode)->ip_lock); |
6842 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6843 | clusters_to_del; | |
6844 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6845 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6846 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6847 | |
3a0782d0 | 6848 | status = ocfs2_trim_tree(inode, path, handle, tc, |
bcbbb24a | 6849 | clusters_to_del, &delete_blk, &rec_flags); |
3a0782d0 MF |
6850 | if (status) { |
6851 | mlog_errno(status); | |
6852 | goto bail; | |
ccd979bd MF |
6853 | } |
6854 | ||
dcd0538f | 6855 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6856 | /* trunc to zero is a special case. */ |
6857 | el->l_tree_depth = 0; | |
6858 | fe->i_last_eb_blk = 0; | |
6859 | } else if (last_eb) | |
6860 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6861 | ||
ec20cec7 | 6862 | ocfs2_journal_dirty(handle, fe_bh); |
ccd979bd MF |
6863 | |
6864 | if (last_eb) { | |
6865 | /* If there will be a new last extent block, then by | |
6866 | * definition, there cannot be any leaves to the right of | |
6867 | * him. */ | |
ccd979bd | 6868 | last_eb->h_next_leaf_blk = 0; |
ec20cec7 | 6869 | ocfs2_journal_dirty(handle, last_eb_bh); |
ccd979bd MF |
6870 | } |
6871 | ||
3a0782d0 | 6872 | if (delete_blk) { |
bcbbb24a TM |
6873 | if (rec_flags & OCFS2_EXT_REFCOUNTED) |
6874 | status = ocfs2_decrease_refcount(inode, handle, | |
6875 | ocfs2_blocks_to_clusters(osb->sb, | |
6876 | delete_blk), | |
6877 | clusters_to_del, meta_ac, | |
6ae23c55 | 6878 | &tc->tc_dealloc, 1); |
bcbbb24a TM |
6879 | else |
6880 | status = ocfs2_truncate_log_append(osb, handle, | |
6881 | delete_blk, | |
6882 | clusters_to_del); | |
ccd979bd MF |
6883 | if (status < 0) { |
6884 | mlog_errno(status); | |
6885 | goto bail; | |
6886 | } | |
ccd979bd MF |
6887 | } |
6888 | status = 0; | |
6889 | bail: | |
60e2ec48 | 6890 | brelse(last_eb_bh); |
ccd979bd MF |
6891 | mlog_exit(status); |
6892 | return status; | |
6893 | } | |
6894 | ||
2b4e30fb | 6895 | static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh) |
60b11392 MF |
6896 | { |
6897 | set_buffer_uptodate(bh); | |
6898 | mark_buffer_dirty(bh); | |
6899 | return 0; | |
6900 | } | |
6901 | ||
6f70fa51 TM |
6902 | void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6903 | unsigned int from, unsigned int to, | |
6904 | struct page *page, int zero, u64 *phys) | |
1d410a6e MF |
6905 | { |
6906 | int ret, partial = 0; | |
6907 | ||
6908 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6909 | if (ret) | |
6910 | mlog_errno(ret); | |
6911 | ||
6912 | if (zero) | |
eebd2aa3 | 6913 | zero_user_segment(page, from, to); |
1d410a6e MF |
6914 | |
6915 | /* | |
6916 | * Need to set the buffers we zero'd into uptodate | |
6917 | * here if they aren't - ocfs2_map_page_blocks() | |
6918 | * might've skipped some | |
6919 | */ | |
2b4e30fb JB |
6920 | ret = walk_page_buffers(handle, page_buffers(page), |
6921 | from, to, &partial, | |
6922 | ocfs2_zero_func); | |
6923 | if (ret < 0) | |
6924 | mlog_errno(ret); | |
6925 | else if (ocfs2_should_order_data(inode)) { | |
6926 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
1d410a6e MF |
6927 | if (ret < 0) |
6928 | mlog_errno(ret); | |
6929 | } | |
6930 | ||
6931 | if (!partial) | |
6932 | SetPageUptodate(page); | |
6933 | ||
6934 | flush_dcache_page(page); | |
6935 | } | |
6936 | ||
35edec1d MF |
6937 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6938 | loff_t end, struct page **pages, | |
6939 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6940 | { |
1d410a6e | 6941 | int i; |
60b11392 MF |
6942 | struct page *page; |
6943 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6944 | struct super_block *sb = inode->i_sb; | |
6945 | ||
6946 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6947 | ||
6948 | if (numpages == 0) | |
6949 | goto out; | |
6950 | ||
35edec1d | 6951 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6952 | for(i = 0; i < numpages; i++) { |
6953 | page = pages[i]; | |
6954 | ||
35edec1d MF |
6955 | from = start & (PAGE_CACHE_SIZE - 1); |
6956 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6957 | to = end & (PAGE_CACHE_SIZE - 1); | |
6958 | ||
60b11392 MF |
6959 | BUG_ON(from > PAGE_CACHE_SIZE); |
6960 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6961 | ||
1d410a6e MF |
6962 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6963 | &phys); | |
60b11392 | 6964 | |
35edec1d | 6965 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6966 | } |
6967 | out: | |
1d410a6e MF |
6968 | if (pages) |
6969 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6970 | } |
6971 | ||
6f70fa51 TM |
6972 | int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end, |
6973 | struct page **pages, int *num) | |
60b11392 | 6974 | { |
1d410a6e | 6975 | int numpages, ret = 0; |
60b11392 MF |
6976 | struct address_space *mapping = inode->i_mapping; |
6977 | unsigned long index; | |
35edec1d | 6978 | loff_t last_page_bytes; |
60b11392 | 6979 | |
35edec1d | 6980 | BUG_ON(start > end); |
60b11392 | 6981 | |
1d410a6e | 6982 | numpages = 0; |
35edec1d MF |
6983 | last_page_bytes = PAGE_ALIGN(end); |
6984 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6985 | do { |
6986 | pages[numpages] = grab_cache_page(mapping, index); | |
6987 | if (!pages[numpages]) { | |
6988 | ret = -ENOMEM; | |
6989 | mlog_errno(ret); | |
6990 | goto out; | |
6991 | } | |
6992 | ||
6993 | numpages++; | |
6994 | index++; | |
35edec1d | 6995 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6996 | |
6997 | out: | |
6998 | if (ret != 0) { | |
1d410a6e MF |
6999 | if (pages) |
7000 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
7001 | numpages = 0; |
7002 | } | |
7003 | ||
7004 | *num = numpages; | |
7005 | ||
7006 | return ret; | |
7007 | } | |
7008 | ||
6f70fa51 TM |
7009 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
7010 | struct page **pages, int *num) | |
7011 | { | |
7012 | struct super_block *sb = inode->i_sb; | |
7013 | ||
7014 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != | |
7015 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
7016 | ||
7017 | return ocfs2_grab_pages(inode, start, end, pages, num); | |
7018 | } | |
7019 | ||
60b11392 MF |
7020 | /* |
7021 | * Zero the area past i_size but still within an allocated | |
7022 | * cluster. This avoids exposing nonzero data on subsequent file | |
7023 | * extends. | |
7024 | * | |
7025 | * We need to call this before i_size is updated on the inode because | |
7026 | * otherwise block_write_full_page() will skip writeout of pages past | |
7027 | * i_size. The new_i_size parameter is passed for this reason. | |
7028 | */ | |
35edec1d MF |
7029 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
7030 | u64 range_start, u64 range_end) | |
60b11392 | 7031 | { |
1d410a6e | 7032 | int ret = 0, numpages; |
60b11392 MF |
7033 | struct page **pages = NULL; |
7034 | u64 phys; | |
1d410a6e MF |
7035 | unsigned int ext_flags; |
7036 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
7037 | |
7038 | /* | |
7039 | * File systems which don't support sparse files zero on every | |
7040 | * extend. | |
7041 | */ | |
1d410a6e | 7042 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
7043 | return 0; |
7044 | ||
1d410a6e | 7045 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
7046 | sizeof(struct page *), GFP_NOFS); |
7047 | if (pages == NULL) { | |
7048 | ret = -ENOMEM; | |
7049 | mlog_errno(ret); | |
7050 | goto out; | |
7051 | } | |
7052 | ||
1d410a6e MF |
7053 | if (range_start == range_end) |
7054 | goto out; | |
7055 | ||
7056 | ret = ocfs2_extent_map_get_blocks(inode, | |
7057 | range_start >> sb->s_blocksize_bits, | |
7058 | &phys, NULL, &ext_flags); | |
60b11392 MF |
7059 | if (ret) { |
7060 | mlog_errno(ret); | |
7061 | goto out; | |
7062 | } | |
7063 | ||
1d410a6e MF |
7064 | /* |
7065 | * Tail is a hole, or is marked unwritten. In either case, we | |
7066 | * can count on read and write to return/push zero's. | |
7067 | */ | |
7068 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
7069 | goto out; |
7070 | ||
1d410a6e MF |
7071 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
7072 | &numpages); | |
7073 | if (ret) { | |
7074 | mlog_errno(ret); | |
7075 | goto out; | |
7076 | } | |
7077 | ||
35edec1d MF |
7078 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
7079 | numpages, phys, handle); | |
60b11392 MF |
7080 | |
7081 | /* | |
7082 | * Initiate writeout of the pages we zero'd here. We don't | |
7083 | * wait on them - the truncate_inode_pages() call later will | |
7084 | * do that for us. | |
7085 | */ | |
2cfd30ad CH |
7086 | ret = filemap_fdatawrite_range(inode->i_mapping, range_start, |
7087 | range_end - 1); | |
60b11392 MF |
7088 | if (ret) |
7089 | mlog_errno(ret); | |
7090 | ||
7091 | out: | |
7092 | if (pages) | |
7093 | kfree(pages); | |
7094 | ||
7095 | return ret; | |
7096 | } | |
7097 | ||
fdd77704 TY |
7098 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
7099 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
7100 | { |
7101 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 7102 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 7103 | |
fdd77704 TY |
7104 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
7105 | memset(&di->id2, 0, blocksize - | |
7106 | offsetof(struct ocfs2_dinode, id2) - | |
7107 | xattrsize); | |
7108 | else | |
7109 | memset(&di->id2, 0, blocksize - | |
7110 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
7111 | } |
7112 | ||
5b6a3a2b MF |
7113 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
7114 | struct ocfs2_dinode *di) | |
7115 | { | |
fdd77704 | 7116 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
7117 | di->id2.i_list.l_tree_depth = 0; |
7118 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
7119 | di->id2.i_list.l_count = cpu_to_le16( |
7120 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
7121 | } |
7122 | ||
1afc32b9 MF |
7123 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
7124 | { | |
7125 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7126 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7127 | ||
7128 | spin_lock(&oi->ip_lock); | |
7129 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
7130 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7131 | spin_unlock(&oi->ip_lock); | |
7132 | ||
7133 | /* | |
7134 | * We clear the entire i_data structure here so that all | |
7135 | * fields can be properly initialized. | |
7136 | */ | |
fdd77704 | 7137 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 7138 | |
fdd77704 TY |
7139 | idata->id_count = cpu_to_le16( |
7140 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
7141 | } |
7142 | ||
7143 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
7144 | struct buffer_head *di_bh) | |
7145 | { | |
7146 | int ret, i, has_data, num_pages = 0; | |
7147 | handle_t *handle; | |
7148 | u64 uninitialized_var(block); | |
7149 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
7150 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7151 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
7152 | struct ocfs2_alloc_context *data_ac = NULL; |
7153 | struct page **pages = NULL; | |
7154 | loff_t end = osb->s_clustersize; | |
f99b9b7c | 7155 | struct ocfs2_extent_tree et; |
a90714c1 | 7156 | int did_quota = 0; |
1afc32b9 MF |
7157 | |
7158 | has_data = i_size_read(inode) ? 1 : 0; | |
7159 | ||
7160 | if (has_data) { | |
7161 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
7162 | sizeof(struct page *), GFP_NOFS); | |
7163 | if (pages == NULL) { | |
7164 | ret = -ENOMEM; | |
7165 | mlog_errno(ret); | |
7166 | goto out; | |
7167 | } | |
7168 | ||
7169 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
7170 | if (ret) { | |
7171 | mlog_errno(ret); | |
7172 | goto out; | |
7173 | } | |
7174 | } | |
7175 | ||
a90714c1 JK |
7176 | handle = ocfs2_start_trans(osb, |
7177 | ocfs2_inline_to_extents_credits(osb->sb)); | |
1afc32b9 MF |
7178 | if (IS_ERR(handle)) { |
7179 | ret = PTR_ERR(handle); | |
7180 | mlog_errno(ret); | |
7181 | goto out_unlock; | |
7182 | } | |
7183 | ||
0cf2f763 | 7184 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7185 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7186 | if (ret) { |
7187 | mlog_errno(ret); | |
7188 | goto out_commit; | |
7189 | } | |
7190 | ||
7191 | if (has_data) { | |
7192 | u32 bit_off, num; | |
7193 | unsigned int page_end; | |
7194 | u64 phys; | |
7195 | ||
5dd4056d CH |
7196 | ret = dquot_alloc_space_nodirty(inode, |
7197 | ocfs2_clusters_to_bytes(osb->sb, 1)); | |
7198 | if (ret) | |
a90714c1 | 7199 | goto out_commit; |
a90714c1 JK |
7200 | did_quota = 1; |
7201 | ||
4fe370af MF |
7202 | data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv; |
7203 | ||
1afc32b9 MF |
7204 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, |
7205 | &num); | |
7206 | if (ret) { | |
7207 | mlog_errno(ret); | |
7208 | goto out_commit; | |
7209 | } | |
7210 | ||
7211 | /* | |
7212 | * Save two copies, one for insert, and one that can | |
7213 | * be changed by ocfs2_map_and_dirty_page() below. | |
7214 | */ | |
7215 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
7216 | ||
7217 | /* | |
7218 | * Non sparse file systems zero on extend, so no need | |
7219 | * to do that now. | |
7220 | */ | |
7221 | if (!ocfs2_sparse_alloc(osb) && | |
7222 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
7223 | end = PAGE_CACHE_SIZE; | |
7224 | ||
7225 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
7226 | if (ret) { | |
7227 | mlog_errno(ret); | |
7228 | goto out_commit; | |
7229 | } | |
7230 | ||
7231 | /* | |
7232 | * This should populate the 1st page for us and mark | |
7233 | * it up to date. | |
7234 | */ | |
7235 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
7236 | if (ret) { | |
7237 | mlog_errno(ret); | |
7238 | goto out_commit; | |
7239 | } | |
7240 | ||
7241 | page_end = PAGE_CACHE_SIZE; | |
7242 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
7243 | page_end = osb->s_clustersize; | |
7244 | ||
7245 | for (i = 0; i < num_pages; i++) | |
7246 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
7247 | pages[i], i > 0, &phys); | |
7248 | } | |
7249 | ||
7250 | spin_lock(&oi->ip_lock); | |
7251 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
7252 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
7253 | spin_unlock(&oi->ip_lock); | |
7254 | ||
5b6a3a2b | 7255 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
7256 | |
7257 | ocfs2_journal_dirty(handle, di_bh); | |
7258 | ||
7259 | if (has_data) { | |
7260 | /* | |
7261 | * An error at this point should be extremely rare. If | |
7262 | * this proves to be false, we could always re-build | |
7263 | * the in-inode data from our pages. | |
7264 | */ | |
5e404e9e | 7265 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
cc79d8c1 | 7266 | ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL); |
1afc32b9 MF |
7267 | if (ret) { |
7268 | mlog_errno(ret); | |
7269 | goto out_commit; | |
7270 | } | |
7271 | ||
7272 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7273 | } | |
7274 | ||
7275 | out_commit: | |
a90714c1 | 7276 | if (ret < 0 && did_quota) |
5dd4056d | 7277 | dquot_free_space_nodirty(inode, |
a90714c1 JK |
7278 | ocfs2_clusters_to_bytes(osb->sb, 1)); |
7279 | ||
1afc32b9 MF |
7280 | ocfs2_commit_trans(osb, handle); |
7281 | ||
7282 | out_unlock: | |
7283 | if (data_ac) | |
7284 | ocfs2_free_alloc_context(data_ac); | |
7285 | ||
7286 | out: | |
7287 | if (pages) { | |
7288 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
7289 | kfree(pages); | |
7290 | } | |
7291 | ||
7292 | return ret; | |
7293 | } | |
7294 | ||
ccd979bd MF |
7295 | /* |
7296 | * It is expected, that by the time you call this function, | |
7297 | * inode->i_size and fe->i_size have been adjusted. | |
7298 | * | |
7299 | * WARNING: This will kfree the truncate context | |
7300 | */ | |
7301 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
7302 | struct inode *inode, | |
7303 | struct buffer_head *fe_bh, | |
7304 | struct ocfs2_truncate_context *tc) | |
7305 | { | |
7306 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 7307 | u32 clusters_to_del, new_highest_cpos, range; |
bcbbb24a | 7308 | u64 blkno = 0; |
ccd979bd | 7309 | struct ocfs2_extent_list *el; |
1fabe148 | 7310 | handle_t *handle = NULL; |
ccd979bd | 7311 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 7312 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 7313 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
bcbbb24a TM |
7314 | struct ocfs2_alloc_context *meta_ac = NULL; |
7315 | struct ocfs2_refcount_tree *ref_tree = NULL; | |
ccd979bd MF |
7316 | |
7317 | mlog_entry_void(); | |
7318 | ||
dcd0538f | 7319 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
7320 | i_size_read(inode)); |
7321 | ||
13723d00 JB |
7322 | path = ocfs2_new_path(fe_bh, &di->id2.i_list, |
7323 | ocfs2_journal_access_di); | |
dcd0538f MF |
7324 | if (!path) { |
7325 | status = -ENOMEM; | |
7326 | mlog_errno(status); | |
7327 | goto bail; | |
7328 | } | |
83418978 MF |
7329 | |
7330 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
7331 | ||
ccd979bd | 7332 | start: |
3a0782d0 MF |
7333 | /* |
7334 | * Check that we still have allocation to delete. | |
7335 | */ | |
7336 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
7337 | status = 0; | |
7338 | goto bail; | |
7339 | } | |
7340 | ||
bcbbb24a TM |
7341 | credits = 0; |
7342 | ||
dcd0538f MF |
7343 | /* |
7344 | * Truncate always works against the rightmost tree branch. | |
7345 | */ | |
facdb77f | 7346 | status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX); |
dcd0538f MF |
7347 | if (status) { |
7348 | mlog_errno(status); | |
7349 | goto bail; | |
ccd979bd MF |
7350 | } |
7351 | ||
dcd0538f MF |
7352 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
7353 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
7354 | ||
7355 | /* | |
7356 | * By now, el will point to the extent list on the bottom most | |
7357 | * portion of this tree. Only the tail record is considered in | |
7358 | * each pass. | |
7359 | * | |
7360 | * We handle the following cases, in order: | |
7361 | * - empty extent: delete the remaining branch | |
7362 | * - remove the entire record | |
7363 | * - remove a partial record | |
7364 | * - no record needs to be removed (truncate has completed) | |
7365 | */ | |
7366 | el = path_leaf_el(path); | |
3a0782d0 MF |
7367 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
7368 | ocfs2_error(inode->i_sb, | |
7369 | "Inode %llu has empty extent block at %llu\n", | |
7370 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7371 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7372 | status = -EROFS; | |
7373 | goto bail; | |
7374 | } | |
7375 | ||
ccd979bd | 7376 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 7377 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 7378 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
7379 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
7380 | clusters_to_del = 0; | |
7381 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 7382 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
bcbbb24a | 7383 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); |
dcd0538f | 7384 | } else if (range > new_highest_cpos) { |
e48edee2 | 7385 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 7386 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f | 7387 | new_highest_cpos; |
bcbbb24a TM |
7388 | blkno = le64_to_cpu(el->l_recs[i].e_blkno) + |
7389 | ocfs2_clusters_to_blocks(inode->i_sb, | |
7390 | ocfs2_rec_clusters(el, &el->l_recs[i]) - | |
7391 | clusters_to_del); | |
dcd0538f MF |
7392 | } else { |
7393 | status = 0; | |
7394 | goto bail; | |
7395 | } | |
ccd979bd | 7396 | |
dcd0538f MF |
7397 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
7398 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
7399 | ||
bcbbb24a TM |
7400 | if (el->l_recs[i].e_flags & OCFS2_EXT_REFCOUNTED && clusters_to_del) { |
7401 | BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & | |
7402 | OCFS2_HAS_REFCOUNT_FL)); | |
7403 | ||
7404 | status = ocfs2_lock_refcount_tree(osb, | |
7405 | le64_to_cpu(di->i_refcount_loc), | |
7406 | 1, &ref_tree, NULL); | |
7407 | if (status) { | |
7408 | mlog_errno(status); | |
7409 | goto bail; | |
7410 | } | |
7411 | ||
7412 | status = ocfs2_prepare_refcount_change_for_del(inode, fe_bh, | |
7413 | blkno, | |
7414 | clusters_to_del, | |
7415 | &credits, | |
7416 | &meta_ac); | |
7417 | if (status < 0) { | |
7418 | mlog_errno(status); | |
7419 | goto bail; | |
7420 | } | |
7421 | } | |
7422 | ||
1b1dcc1b | 7423 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
7424 | tl_sem = 1; |
7425 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
7426 | * record is free for use. If there isn't any, we flush to get | |
7427 | * an empty truncate log. */ | |
7428 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
7429 | status = __ocfs2_flush_truncate_log(osb); | |
7430 | if (status < 0) { | |
7431 | mlog_errno(status); | |
7432 | goto bail; | |
7433 | } | |
7434 | } | |
7435 | ||
bcbbb24a | 7436 | credits += ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, |
dcd0538f MF |
7437 | (struct ocfs2_dinode *)fe_bh->b_data, |
7438 | el); | |
65eff9cc | 7439 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
7440 | if (IS_ERR(handle)) { |
7441 | status = PTR_ERR(handle); | |
7442 | handle = NULL; | |
7443 | mlog_errno(status); | |
7444 | goto bail; | |
7445 | } | |
7446 | ||
dcd0538f | 7447 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
bcbbb24a | 7448 | tc, path, meta_ac); |
ccd979bd MF |
7449 | if (status < 0) { |
7450 | mlog_errno(status); | |
7451 | goto bail; | |
7452 | } | |
7453 | ||
1b1dcc1b | 7454 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7455 | tl_sem = 0; |
7456 | ||
02dc1af4 | 7457 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
7458 | handle = NULL; |
7459 | ||
dcd0538f MF |
7460 | ocfs2_reinit_path(path, 1); |
7461 | ||
bcbbb24a TM |
7462 | if (meta_ac) { |
7463 | ocfs2_free_alloc_context(meta_ac); | |
7464 | meta_ac = NULL; | |
7465 | } | |
7466 | ||
7467 | if (ref_tree) { | |
7468 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
7469 | ref_tree = NULL; | |
7470 | } | |
7471 | ||
dcd0538f | 7472 | /* |
3a0782d0 MF |
7473 | * The check above will catch the case where we've truncated |
7474 | * away all allocation. | |
dcd0538f | 7475 | */ |
3a0782d0 MF |
7476 | goto start; |
7477 | ||
ccd979bd | 7478 | bail: |
ccd979bd MF |
7479 | |
7480 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
7481 | ||
7482 | if (tl_sem) | |
1b1dcc1b | 7483 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
7484 | |
7485 | if (handle) | |
02dc1af4 | 7486 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 7487 | |
bcbbb24a TM |
7488 | if (meta_ac) |
7489 | ocfs2_free_alloc_context(meta_ac); | |
7490 | ||
7491 | if (ref_tree) | |
7492 | ocfs2_unlock_refcount_tree(osb, ref_tree, 1); | |
7493 | ||
59a5e416 MF |
7494 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
7495 | ||
dcd0538f | 7496 | ocfs2_free_path(path); |
ccd979bd MF |
7497 | |
7498 | /* This will drop the ext_alloc cluster lock for us */ | |
7499 | ocfs2_free_truncate_context(tc); | |
7500 | ||
7501 | mlog_exit(status); | |
7502 | return status; | |
7503 | } | |
7504 | ||
ccd979bd | 7505 | /* |
59a5e416 | 7506 | * Expects the inode to already be locked. |
ccd979bd MF |
7507 | */ |
7508 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7509 | struct inode *inode, | |
7510 | struct buffer_head *fe_bh, | |
7511 | struct ocfs2_truncate_context **tc) | |
7512 | { | |
59a5e416 | 7513 | int status; |
ccd979bd MF |
7514 | unsigned int new_i_clusters; |
7515 | struct ocfs2_dinode *fe; | |
7516 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7517 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7518 | |
7519 | mlog_entry_void(); | |
7520 | ||
7521 | *tc = NULL; | |
7522 | ||
7523 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7524 | i_size_read(inode)); | |
7525 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7526 | ||
7527 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7528 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7529 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7530 | |
cd861280 | 7531 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7532 | if (!(*tc)) { |
7533 | status = -ENOMEM; | |
7534 | mlog_errno(status); | |
7535 | goto bail; | |
7536 | } | |
59a5e416 | 7537 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7538 | |
ccd979bd | 7539 | if (fe->id2.i_list.l_tree_depth) { |
3d03a305 | 7540 | status = ocfs2_read_extent_block(INODE_CACHE(inode), |
5e96581a JB |
7541 | le64_to_cpu(fe->i_last_eb_blk), |
7542 | &last_eb_bh); | |
ccd979bd MF |
7543 | if (status < 0) { |
7544 | mlog_errno(status); | |
7545 | goto bail; | |
7546 | } | |
7547 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
ccd979bd MF |
7548 | } |
7549 | ||
7550 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7551 | ||
ccd979bd MF |
7552 | status = 0; |
7553 | bail: | |
7554 | if (status < 0) { | |
7555 | if (*tc) | |
7556 | ocfs2_free_truncate_context(*tc); | |
7557 | *tc = NULL; | |
7558 | } | |
7559 | mlog_exit_void(); | |
7560 | return status; | |
7561 | } | |
7562 | ||
1afc32b9 MF |
7563 | /* |
7564 | * 'start' is inclusive, 'end' is not. | |
7565 | */ | |
7566 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7567 | unsigned int start, unsigned int end, int trunc) | |
7568 | { | |
7569 | int ret; | |
7570 | unsigned int numbytes; | |
7571 | handle_t *handle; | |
7572 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7573 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7574 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7575 | ||
7576 | if (end > i_size_read(inode)) | |
7577 | end = i_size_read(inode); | |
7578 | ||
7579 | BUG_ON(start >= end); | |
7580 | ||
7581 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7582 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7583 | !ocfs2_supports_inline_data(osb)) { | |
7584 | ocfs2_error(inode->i_sb, | |
7585 | "Inline data flags for inode %llu don't agree! " | |
7586 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7587 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7588 | le16_to_cpu(di->i_dyn_features), | |
7589 | OCFS2_I(inode)->ip_dyn_features, | |
7590 | osb->s_feature_incompat); | |
7591 | ret = -EROFS; | |
7592 | goto out; | |
7593 | } | |
7594 | ||
7595 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7596 | if (IS_ERR(handle)) { | |
7597 | ret = PTR_ERR(handle); | |
7598 | mlog_errno(ret); | |
7599 | goto out; | |
7600 | } | |
7601 | ||
0cf2f763 | 7602 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 7603 | OCFS2_JOURNAL_ACCESS_WRITE); |
1afc32b9 MF |
7604 | if (ret) { |
7605 | mlog_errno(ret); | |
7606 | goto out_commit; | |
7607 | } | |
7608 | ||
7609 | numbytes = end - start; | |
7610 | memset(idata->id_data + start, 0, numbytes); | |
7611 | ||
7612 | /* | |
7613 | * No need to worry about the data page here - it's been | |
7614 | * truncated already and inline data doesn't need it for | |
7615 | * pushing zero's to disk, so we'll let readpage pick it up | |
7616 | * later. | |
7617 | */ | |
7618 | if (trunc) { | |
7619 | i_size_write(inode, start); | |
7620 | di->i_size = cpu_to_le64(start); | |
7621 | } | |
7622 | ||
7623 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7624 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7625 | ||
7626 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7627 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7628 | ||
7629 | ocfs2_journal_dirty(handle, di_bh); | |
7630 | ||
7631 | out_commit: | |
7632 | ocfs2_commit_trans(osb, handle); | |
7633 | ||
7634 | out: | |
7635 | return ret; | |
7636 | } | |
7637 | ||
ccd979bd MF |
7638 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7639 | { | |
59a5e416 MF |
7640 | /* |
7641 | * The caller is responsible for completing deallocation | |
7642 | * before freeing the context. | |
7643 | */ | |
7644 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7645 | mlog(ML_NOTICE, | |
7646 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd | 7647 | |
a81cb88b | 7648 | brelse(tc->tc_last_eb_bh); |
ccd979bd MF |
7649 | |
7650 | kfree(tc); | |
7651 | } |