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> |
ccd979bd MF |
31 | |
32 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
60b11392 | 38 | #include "aops.h" |
ccd979bd MF |
39 | #include "dlmglue.h" |
40 | #include "extent_map.h" | |
41 | #include "inode.h" | |
42 | #include "journal.h" | |
43 | #include "localalloc.h" | |
44 | #include "suballoc.h" | |
45 | #include "sysfile.h" | |
46 | #include "file.h" | |
47 | #include "super.h" | |
48 | #include "uptodate.h" | |
49 | ||
50 | #include "buffer_head_io.h" | |
51 | ||
e7d4cb6b TM |
52 | /* |
53 | * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract | |
54 | * the b-tree operations in ocfs2. Now all the b-tree operations are not | |
55 | * limited to ocfs2_dinode only. Any data which need to allocate clusters | |
56 | * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree | |
57 | * and operation. | |
58 | * | |
59 | * ocfs2_extent_tree contains info for the root of the b-tree, it must have a | |
60 | * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree | |
61 | * functions. | |
62 | * ocfs2_extent_tree_operations abstract the normal operations we do for | |
63 | * the root of extent b-tree. | |
64 | */ | |
65 | struct ocfs2_extent_tree; | |
66 | ||
67 | struct ocfs2_extent_tree_operations { | |
35dc0aa3 JB |
68 | void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et, |
69 | u64 blkno); | |
70 | u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et); | |
71 | void (*eo_update_clusters)(struct inode *inode, | |
72 | struct ocfs2_extent_tree *et, | |
73 | u32 new_clusters); | |
74 | int (*eo_sanity_check)(struct inode *inode, struct ocfs2_extent_tree *et); | |
0ce1010f JB |
75 | |
76 | /* These are internal to ocfs2_extent_tree and don't have | |
77 | * accessor functions */ | |
78 | void (*eo_fill_root_el)(struct ocfs2_extent_tree *et); | |
943cced3 JB |
79 | void (*eo_fill_max_leaf_clusters)(struct inode *inode, |
80 | struct ocfs2_extent_tree *et); | |
e7d4cb6b TM |
81 | }; |
82 | ||
83 | struct ocfs2_extent_tree { | |
ce1d9ea6 JB |
84 | enum ocfs2_extent_tree_type et_type; |
85 | struct ocfs2_extent_tree_operations *et_ops; | |
86 | struct buffer_head *et_root_bh; | |
87 | struct ocfs2_extent_list *et_root_el; | |
ea5efa15 | 88 | void *et_object; |
ce1d9ea6 | 89 | unsigned int et_max_leaf_clusters; |
e7d4cb6b TM |
90 | }; |
91 | ||
0ce1010f JB |
92 | static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et) |
93 | { | |
94 | struct ocfs2_dinode *di = et->et_object; | |
95 | ||
96 | et->et_root_el = &di->id2.i_list; | |
97 | } | |
98 | ||
e7d4cb6b TM |
99 | static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et, |
100 | u64 blkno) | |
101 | { | |
ea5efa15 | 102 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 103 | |
ce1d9ea6 | 104 | BUG_ON(et->et_type != OCFS2_DINODE_EXTENT); |
e7d4cb6b TM |
105 | di->i_last_eb_blk = cpu_to_le64(blkno); |
106 | } | |
107 | ||
108 | static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
109 | { | |
ea5efa15 | 110 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b | 111 | |
ce1d9ea6 | 112 | BUG_ON(et->et_type != OCFS2_DINODE_EXTENT); |
e7d4cb6b TM |
113 | return le64_to_cpu(di->i_last_eb_blk); |
114 | } | |
115 | ||
116 | static void ocfs2_dinode_update_clusters(struct inode *inode, | |
117 | struct ocfs2_extent_tree *et, | |
118 | u32 clusters) | |
119 | { | |
ea5efa15 | 120 | struct ocfs2_dinode *di = et->et_object; |
e7d4cb6b TM |
121 | |
122 | le32_add_cpu(&di->i_clusters, clusters); | |
123 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
124 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters); | |
125 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
126 | } | |
127 | ||
128 | static int ocfs2_dinode_sanity_check(struct inode *inode, | |
129 | struct ocfs2_extent_tree *et) | |
130 | { | |
131 | int ret = 0; | |
132 | struct ocfs2_dinode *di; | |
133 | ||
ce1d9ea6 | 134 | BUG_ON(et->et_type != OCFS2_DINODE_EXTENT); |
e7d4cb6b | 135 | |
ea5efa15 | 136 | di = et->et_object; |
e7d4cb6b TM |
137 | if (!OCFS2_IS_VALID_DINODE(di)) { |
138 | ret = -EIO; | |
139 | ocfs2_error(inode->i_sb, | |
140 | "Inode %llu has invalid path root", | |
141 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
142 | } | |
143 | ||
144 | return ret; | |
145 | } | |
146 | ||
147 | static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = { | |
35dc0aa3 JB |
148 | .eo_set_last_eb_blk = ocfs2_dinode_set_last_eb_blk, |
149 | .eo_get_last_eb_blk = ocfs2_dinode_get_last_eb_blk, | |
150 | .eo_update_clusters = ocfs2_dinode_update_clusters, | |
151 | .eo_sanity_check = ocfs2_dinode_sanity_check, | |
0ce1010f | 152 | .eo_fill_root_el = ocfs2_dinode_fill_root_el, |
e7d4cb6b TM |
153 | }; |
154 | ||
0ce1010f JB |
155 | static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et) |
156 | { | |
157 | struct ocfs2_xattr_value_root *xv = et->et_object; | |
158 | ||
159 | et->et_root_el = &xv->xr_list; | |
160 | } | |
161 | ||
f56654c4 TM |
162 | static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et, |
163 | u64 blkno) | |
164 | { | |
165 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 166 | (struct ocfs2_xattr_value_root *)et->et_object; |
f56654c4 TM |
167 | |
168 | xv->xr_last_eb_blk = cpu_to_le64(blkno); | |
169 | } | |
170 | ||
171 | static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
172 | { | |
173 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 174 | (struct ocfs2_xattr_value_root *) et->et_object; |
f56654c4 TM |
175 | |
176 | return le64_to_cpu(xv->xr_last_eb_blk); | |
177 | } | |
178 | ||
179 | static void ocfs2_xattr_value_update_clusters(struct inode *inode, | |
180 | struct ocfs2_extent_tree *et, | |
181 | u32 clusters) | |
182 | { | |
183 | struct ocfs2_xattr_value_root *xv = | |
ea5efa15 | 184 | (struct ocfs2_xattr_value_root *)et->et_object; |
f56654c4 TM |
185 | |
186 | le32_add_cpu(&xv->xr_clusters, clusters); | |
187 | } | |
188 | ||
189 | static int ocfs2_xattr_value_sanity_check(struct inode *inode, | |
190 | struct ocfs2_extent_tree *et) | |
191 | { | |
192 | return 0; | |
193 | } | |
194 | ||
195 | static struct ocfs2_extent_tree_operations ocfs2_xattr_et_ops = { | |
35dc0aa3 JB |
196 | .eo_set_last_eb_blk = ocfs2_xattr_value_set_last_eb_blk, |
197 | .eo_get_last_eb_blk = ocfs2_xattr_value_get_last_eb_blk, | |
198 | .eo_update_clusters = ocfs2_xattr_value_update_clusters, | |
199 | .eo_sanity_check = ocfs2_xattr_value_sanity_check, | |
0ce1010f | 200 | .eo_fill_root_el = ocfs2_xattr_value_fill_root_el, |
f56654c4 TM |
201 | }; |
202 | ||
0ce1010f JB |
203 | static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et) |
204 | { | |
205 | struct ocfs2_xattr_block *xb = et->et_object; | |
206 | ||
207 | et->et_root_el = &xb->xb_attrs.xb_root.xt_list; | |
208 | } | |
209 | ||
943cced3 JB |
210 | static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct inode *inode, |
211 | struct ocfs2_extent_tree *et) | |
212 | { | |
213 | et->et_max_leaf_clusters = | |
214 | ocfs2_clusters_for_bytes(inode->i_sb, | |
215 | OCFS2_MAX_XATTR_TREE_LEAF_SIZE); | |
216 | } | |
217 | ||
ba492615 TM |
218 | static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et, |
219 | u64 blkno) | |
220 | { | |
ea5efa15 | 221 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
222 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
223 | ||
224 | xt->xt_last_eb_blk = cpu_to_le64(blkno); | |
225 | } | |
226 | ||
227 | static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et) | |
228 | { | |
ea5efa15 | 229 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
230 | struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root; |
231 | ||
232 | return le64_to_cpu(xt->xt_last_eb_blk); | |
233 | } | |
234 | ||
235 | static void ocfs2_xattr_tree_update_clusters(struct inode *inode, | |
236 | struct ocfs2_extent_tree *et, | |
237 | u32 clusters) | |
238 | { | |
ea5efa15 | 239 | struct ocfs2_xattr_block *xb = et->et_object; |
ba492615 TM |
240 | |
241 | le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters); | |
242 | } | |
243 | ||
244 | static int ocfs2_xattr_tree_sanity_check(struct inode *inode, | |
245 | struct ocfs2_extent_tree *et) | |
246 | { | |
247 | return 0; | |
248 | } | |
249 | ||
250 | static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = { | |
35dc0aa3 JB |
251 | .eo_set_last_eb_blk = ocfs2_xattr_tree_set_last_eb_blk, |
252 | .eo_get_last_eb_blk = ocfs2_xattr_tree_get_last_eb_blk, | |
253 | .eo_update_clusters = ocfs2_xattr_tree_update_clusters, | |
254 | .eo_sanity_check = ocfs2_xattr_tree_sanity_check, | |
0ce1010f | 255 | .eo_fill_root_el = ocfs2_xattr_tree_fill_root_el, |
943cced3 | 256 | .eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters, |
ba492615 TM |
257 | }; |
258 | ||
dc0ce61a JB |
259 | static void ocfs2_get_extent_tree(struct ocfs2_extent_tree *et, |
260 | struct inode *inode, | |
261 | struct buffer_head *bh, | |
262 | enum ocfs2_extent_tree_type et_type, | |
ea5efa15 | 263 | void *obj) |
e7d4cb6b | 264 | { |
ce1d9ea6 | 265 | et->et_type = et_type; |
e7d4cb6b | 266 | get_bh(bh); |
ce1d9ea6 | 267 | et->et_root_bh = bh; |
ea5efa15 JB |
268 | if (!obj) |
269 | obj = (void *)bh->b_data; | |
270 | et->et_object = obj; | |
e7d4cb6b | 271 | |
e7d4cb6b | 272 | if (et_type == OCFS2_DINODE_EXTENT) { |
ce1d9ea6 | 273 | et->et_ops = &ocfs2_dinode_et_ops; |
f56654c4 | 274 | } else if (et_type == OCFS2_XATTR_VALUE_EXTENT) { |
ce1d9ea6 | 275 | et->et_ops = &ocfs2_xattr_et_ops; |
ba492615 | 276 | } else if (et_type == OCFS2_XATTR_TREE_EXTENT) { |
ce1d9ea6 | 277 | et->et_ops = &ocfs2_xattr_tree_et_ops; |
e7d4cb6b | 278 | } |
0ce1010f JB |
279 | |
280 | et->et_ops->eo_fill_root_el(et); | |
943cced3 JB |
281 | if (!et->et_ops->eo_fill_max_leaf_clusters) |
282 | et->et_max_leaf_clusters = 0; | |
283 | else | |
284 | et->et_ops->eo_fill_max_leaf_clusters(inode, et); | |
e7d4cb6b TM |
285 | } |
286 | ||
dc0ce61a | 287 | static void ocfs2_put_extent_tree(struct ocfs2_extent_tree *et) |
e7d4cb6b | 288 | { |
dc0ce61a | 289 | brelse(et->et_root_bh); |
e7d4cb6b TM |
290 | } |
291 | ||
35dc0aa3 JB |
292 | static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et, |
293 | u64 new_last_eb_blk) | |
e7d4cb6b | 294 | { |
ce1d9ea6 | 295 | et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk); |
e7d4cb6b TM |
296 | } |
297 | ||
35dc0aa3 | 298 | static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et) |
e7d4cb6b | 299 | { |
ce1d9ea6 | 300 | return et->et_ops->eo_get_last_eb_blk(et); |
e7d4cb6b TM |
301 | } |
302 | ||
35dc0aa3 JB |
303 | static inline void ocfs2_et_update_clusters(struct inode *inode, |
304 | struct ocfs2_extent_tree *et, | |
305 | u32 clusters) | |
306 | { | |
ce1d9ea6 | 307 | et->et_ops->eo_update_clusters(inode, et, clusters); |
35dc0aa3 JB |
308 | } |
309 | ||
310 | static inline int ocfs2_et_sanity_check(struct inode *inode, | |
311 | struct ocfs2_extent_tree *et) | |
e7d4cb6b | 312 | { |
ce1d9ea6 | 313 | return et->et_ops->eo_sanity_check(inode, et); |
e7d4cb6b TM |
314 | } |
315 | ||
dcd0538f | 316 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc); |
59a5e416 MF |
317 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
318 | struct ocfs2_extent_block *eb); | |
ccd979bd | 319 | |
dcd0538f MF |
320 | /* |
321 | * Structures which describe a path through a btree, and functions to | |
322 | * manipulate them. | |
323 | * | |
324 | * The idea here is to be as generic as possible with the tree | |
325 | * manipulation code. | |
326 | */ | |
327 | struct ocfs2_path_item { | |
328 | struct buffer_head *bh; | |
329 | struct ocfs2_extent_list *el; | |
330 | }; | |
ccd979bd | 331 | |
dcd0538f | 332 | #define OCFS2_MAX_PATH_DEPTH 5 |
ccd979bd | 333 | |
dcd0538f MF |
334 | struct ocfs2_path { |
335 | int p_tree_depth; | |
336 | struct ocfs2_path_item p_node[OCFS2_MAX_PATH_DEPTH]; | |
337 | }; | |
ccd979bd | 338 | |
dcd0538f MF |
339 | #define path_root_bh(_path) ((_path)->p_node[0].bh) |
340 | #define path_root_el(_path) ((_path)->p_node[0].el) | |
341 | #define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh) | |
342 | #define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el) | |
343 | #define path_num_items(_path) ((_path)->p_tree_depth + 1) | |
ccd979bd | 344 | |
dcd0538f MF |
345 | /* |
346 | * Reset the actual path elements so that we can re-use the structure | |
347 | * to build another path. Generally, this involves freeing the buffer | |
348 | * heads. | |
349 | */ | |
350 | static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root) | |
351 | { | |
352 | int i, start = 0, depth = 0; | |
353 | struct ocfs2_path_item *node; | |
ccd979bd | 354 | |
dcd0538f MF |
355 | if (keep_root) |
356 | start = 1; | |
ccd979bd | 357 | |
dcd0538f MF |
358 | for(i = start; i < path_num_items(path); i++) { |
359 | node = &path->p_node[i]; | |
360 | ||
361 | brelse(node->bh); | |
362 | node->bh = NULL; | |
363 | node->el = NULL; | |
364 | } | |
365 | ||
366 | /* | |
367 | * Tree depth may change during truncate, or insert. If we're | |
368 | * keeping the root extent list, then make sure that our path | |
369 | * structure reflects the proper depth. | |
370 | */ | |
371 | if (keep_root) | |
372 | depth = le16_to_cpu(path_root_el(path)->l_tree_depth); | |
373 | ||
374 | path->p_tree_depth = depth; | |
375 | } | |
376 | ||
377 | static void ocfs2_free_path(struct ocfs2_path *path) | |
378 | { | |
379 | if (path) { | |
380 | ocfs2_reinit_path(path, 0); | |
381 | kfree(path); | |
382 | } | |
383 | } | |
384 | ||
328d5752 MF |
385 | /* |
386 | * All the elements of src into dest. After this call, src could be freed | |
387 | * without affecting dest. | |
388 | * | |
389 | * Both paths should have the same root. Any non-root elements of dest | |
390 | * will be freed. | |
391 | */ | |
392 | static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
393 | { | |
394 | int i; | |
395 | ||
396 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
397 | BUG_ON(path_root_el(dest) != path_root_el(src)); | |
398 | ||
399 | ocfs2_reinit_path(dest, 1); | |
400 | ||
401 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
402 | dest->p_node[i].bh = src->p_node[i].bh; | |
403 | dest->p_node[i].el = src->p_node[i].el; | |
404 | ||
405 | if (dest->p_node[i].bh) | |
406 | get_bh(dest->p_node[i].bh); | |
407 | } | |
408 | } | |
409 | ||
dcd0538f MF |
410 | /* |
411 | * Make the *dest path the same as src and re-initialize src path to | |
412 | * have a root only. | |
413 | */ | |
414 | static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src) | |
415 | { | |
416 | int i; | |
417 | ||
418 | BUG_ON(path_root_bh(dest) != path_root_bh(src)); | |
419 | ||
420 | for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) { | |
421 | brelse(dest->p_node[i].bh); | |
422 | ||
423 | dest->p_node[i].bh = src->p_node[i].bh; | |
424 | dest->p_node[i].el = src->p_node[i].el; | |
425 | ||
426 | src->p_node[i].bh = NULL; | |
427 | src->p_node[i].el = NULL; | |
428 | } | |
429 | } | |
430 | ||
431 | /* | |
432 | * Insert an extent block at given index. | |
433 | * | |
434 | * This will not take an additional reference on eb_bh. | |
435 | */ | |
436 | static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index, | |
437 | struct buffer_head *eb_bh) | |
438 | { | |
439 | struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data; | |
440 | ||
441 | /* | |
442 | * Right now, no root bh is an extent block, so this helps | |
443 | * catch code errors with dinode trees. The assertion can be | |
444 | * safely removed if we ever need to insert extent block | |
445 | * structures at the root. | |
446 | */ | |
447 | BUG_ON(index == 0); | |
448 | ||
449 | path->p_node[index].bh = eb_bh; | |
450 | path->p_node[index].el = &eb->h_list; | |
451 | } | |
452 | ||
453 | static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh, | |
454 | struct ocfs2_extent_list *root_el) | |
455 | { | |
456 | struct ocfs2_path *path; | |
ccd979bd | 457 | |
dcd0538f MF |
458 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH); |
459 | ||
460 | path = kzalloc(sizeof(*path), GFP_NOFS); | |
461 | if (path) { | |
462 | path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth); | |
463 | get_bh(root_bh); | |
464 | path_root_bh(path) = root_bh; | |
465 | path_root_el(path) = root_el; | |
466 | } | |
467 | ||
468 | return path; | |
469 | } | |
470 | ||
dcd0538f MF |
471 | /* |
472 | * Convenience function to journal all components in a path. | |
473 | */ | |
474 | static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle, | |
475 | struct ocfs2_path *path) | |
476 | { | |
477 | int i, ret = 0; | |
478 | ||
479 | if (!path) | |
480 | goto out; | |
481 | ||
482 | for(i = 0; i < path_num_items(path); i++) { | |
483 | ret = ocfs2_journal_access(handle, inode, path->p_node[i].bh, | |
484 | OCFS2_JOURNAL_ACCESS_WRITE); | |
485 | if (ret < 0) { | |
486 | mlog_errno(ret); | |
487 | goto out; | |
488 | } | |
489 | } | |
490 | ||
491 | out: | |
492 | return ret; | |
493 | } | |
494 | ||
328d5752 MF |
495 | /* |
496 | * Return the index of the extent record which contains cluster #v_cluster. | |
497 | * -1 is returned if it was not found. | |
498 | * | |
499 | * Should work fine on interior and exterior nodes. | |
500 | */ | |
501 | int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster) | |
502 | { | |
503 | int ret = -1; | |
504 | int i; | |
505 | struct ocfs2_extent_rec *rec; | |
506 | u32 rec_end, rec_start, clusters; | |
507 | ||
508 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
509 | rec = &el->l_recs[i]; | |
510 | ||
511 | rec_start = le32_to_cpu(rec->e_cpos); | |
512 | clusters = ocfs2_rec_clusters(el, rec); | |
513 | ||
514 | rec_end = rec_start + clusters; | |
515 | ||
516 | if (v_cluster >= rec_start && v_cluster < rec_end) { | |
517 | ret = i; | |
518 | break; | |
519 | } | |
520 | } | |
521 | ||
522 | return ret; | |
523 | } | |
524 | ||
dcd0538f MF |
525 | enum ocfs2_contig_type { |
526 | CONTIG_NONE = 0, | |
527 | CONTIG_LEFT, | |
328d5752 MF |
528 | CONTIG_RIGHT, |
529 | CONTIG_LEFTRIGHT, | |
dcd0538f MF |
530 | }; |
531 | ||
e48edee2 MF |
532 | |
533 | /* | |
534 | * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and | |
535 | * ocfs2_extent_contig only work properly against leaf nodes! | |
536 | */ | |
dcd0538f MF |
537 | static int ocfs2_block_extent_contig(struct super_block *sb, |
538 | struct ocfs2_extent_rec *ext, | |
539 | u64 blkno) | |
ccd979bd | 540 | { |
e48edee2 MF |
541 | u64 blk_end = le64_to_cpu(ext->e_blkno); |
542 | ||
543 | blk_end += ocfs2_clusters_to_blocks(sb, | |
544 | le16_to_cpu(ext->e_leaf_clusters)); | |
545 | ||
546 | return blkno == blk_end; | |
ccd979bd MF |
547 | } |
548 | ||
dcd0538f MF |
549 | static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left, |
550 | struct ocfs2_extent_rec *right) | |
551 | { | |
e48edee2 MF |
552 | u32 left_range; |
553 | ||
554 | left_range = le32_to_cpu(left->e_cpos) + | |
555 | le16_to_cpu(left->e_leaf_clusters); | |
556 | ||
557 | return (left_range == le32_to_cpu(right->e_cpos)); | |
dcd0538f MF |
558 | } |
559 | ||
560 | static enum ocfs2_contig_type | |
561 | ocfs2_extent_contig(struct inode *inode, | |
562 | struct ocfs2_extent_rec *ext, | |
563 | struct ocfs2_extent_rec *insert_rec) | |
564 | { | |
565 | u64 blkno = le64_to_cpu(insert_rec->e_blkno); | |
566 | ||
328d5752 MF |
567 | /* |
568 | * Refuse to coalesce extent records with different flag | |
569 | * fields - we don't want to mix unwritten extents with user | |
570 | * data. | |
571 | */ | |
572 | if (ext->e_flags != insert_rec->e_flags) | |
573 | return CONTIG_NONE; | |
574 | ||
dcd0538f MF |
575 | if (ocfs2_extents_adjacent(ext, insert_rec) && |
576 | ocfs2_block_extent_contig(inode->i_sb, ext, blkno)) | |
577 | return CONTIG_RIGHT; | |
578 | ||
579 | blkno = le64_to_cpu(ext->e_blkno); | |
580 | if (ocfs2_extents_adjacent(insert_rec, ext) && | |
581 | ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno)) | |
582 | return CONTIG_LEFT; | |
583 | ||
584 | return CONTIG_NONE; | |
585 | } | |
586 | ||
587 | /* | |
588 | * NOTE: We can have pretty much any combination of contiguousness and | |
589 | * appending. | |
590 | * | |
591 | * The usefulness of APPEND_TAIL is more in that it lets us know that | |
592 | * we'll have to update the path to that leaf. | |
593 | */ | |
594 | enum ocfs2_append_type { | |
595 | APPEND_NONE = 0, | |
596 | APPEND_TAIL, | |
597 | }; | |
598 | ||
328d5752 MF |
599 | enum ocfs2_split_type { |
600 | SPLIT_NONE = 0, | |
601 | SPLIT_LEFT, | |
602 | SPLIT_RIGHT, | |
603 | }; | |
604 | ||
dcd0538f | 605 | struct ocfs2_insert_type { |
328d5752 | 606 | enum ocfs2_split_type ins_split; |
dcd0538f MF |
607 | enum ocfs2_append_type ins_appending; |
608 | enum ocfs2_contig_type ins_contig; | |
609 | int ins_contig_index; | |
dcd0538f MF |
610 | int ins_tree_depth; |
611 | }; | |
612 | ||
328d5752 MF |
613 | struct ocfs2_merge_ctxt { |
614 | enum ocfs2_contig_type c_contig_type; | |
615 | int c_has_empty_extent; | |
616 | int c_split_covers_rec; | |
328d5752 MF |
617 | }; |
618 | ||
ccd979bd MF |
619 | /* |
620 | * How many free extents have we got before we need more meta data? | |
621 | */ | |
622 | int ocfs2_num_free_extents(struct ocfs2_super *osb, | |
623 | struct inode *inode, | |
e7d4cb6b | 624 | struct buffer_head *root_bh, |
f56654c4 | 625 | enum ocfs2_extent_tree_type type, |
ea5efa15 | 626 | void *obj) |
ccd979bd MF |
627 | { |
628 | int retval; | |
e7d4cb6b | 629 | struct ocfs2_extent_list *el = NULL; |
ccd979bd MF |
630 | struct ocfs2_extent_block *eb; |
631 | struct buffer_head *eb_bh = NULL; | |
e7d4cb6b | 632 | u64 last_eb_blk = 0; |
1c25d93a | 633 | struct ocfs2_extent_tree et; |
ccd979bd MF |
634 | |
635 | mlog_entry_void(); | |
636 | ||
1c25d93a JB |
637 | ocfs2_get_extent_tree(&et, inode, root_bh, type, obj); |
638 | el = et.et_root_el; | |
639 | last_eb_blk = ocfs2_et_get_last_eb_blk(&et); | |
ccd979bd | 640 | |
e7d4cb6b TM |
641 | if (last_eb_blk) { |
642 | retval = ocfs2_read_block(osb, last_eb_blk, | |
ccd979bd MF |
643 | &eb_bh, OCFS2_BH_CACHED, inode); |
644 | if (retval < 0) { | |
645 | mlog_errno(retval); | |
646 | goto bail; | |
647 | } | |
648 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
649 | el = &eb->h_list; | |
e7d4cb6b | 650 | } |
ccd979bd MF |
651 | |
652 | BUG_ON(el->l_tree_depth != 0); | |
653 | ||
654 | retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec); | |
655 | bail: | |
656 | if (eb_bh) | |
657 | brelse(eb_bh); | |
658 | ||
1c25d93a | 659 | ocfs2_put_extent_tree(&et); |
ccd979bd MF |
660 | mlog_exit(retval); |
661 | return retval; | |
662 | } | |
663 | ||
664 | /* expects array to already be allocated | |
665 | * | |
666 | * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and | |
667 | * l_count for you | |
668 | */ | |
669 | static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb, | |
1fabe148 | 670 | handle_t *handle, |
ccd979bd MF |
671 | struct inode *inode, |
672 | int wanted, | |
673 | struct ocfs2_alloc_context *meta_ac, | |
674 | struct buffer_head *bhs[]) | |
675 | { | |
676 | int count, status, i; | |
677 | u16 suballoc_bit_start; | |
678 | u32 num_got; | |
679 | u64 first_blkno; | |
680 | struct ocfs2_extent_block *eb; | |
681 | ||
682 | mlog_entry_void(); | |
683 | ||
684 | count = 0; | |
685 | while (count < wanted) { | |
686 | status = ocfs2_claim_metadata(osb, | |
687 | handle, | |
688 | meta_ac, | |
689 | wanted - count, | |
690 | &suballoc_bit_start, | |
691 | &num_got, | |
692 | &first_blkno); | |
693 | if (status < 0) { | |
694 | mlog_errno(status); | |
695 | goto bail; | |
696 | } | |
697 | ||
698 | for(i = count; i < (num_got + count); i++) { | |
699 | bhs[i] = sb_getblk(osb->sb, first_blkno); | |
700 | if (bhs[i] == NULL) { | |
701 | status = -EIO; | |
702 | mlog_errno(status); | |
703 | goto bail; | |
704 | } | |
705 | ocfs2_set_new_buffer_uptodate(inode, bhs[i]); | |
706 | ||
707 | status = ocfs2_journal_access(handle, inode, bhs[i], | |
708 | OCFS2_JOURNAL_ACCESS_CREATE); | |
709 | if (status < 0) { | |
710 | mlog_errno(status); | |
711 | goto bail; | |
712 | } | |
713 | ||
714 | memset(bhs[i]->b_data, 0, osb->sb->s_blocksize); | |
715 | eb = (struct ocfs2_extent_block *) bhs[i]->b_data; | |
716 | /* Ok, setup the minimal stuff here. */ | |
717 | strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE); | |
718 | eb->h_blkno = cpu_to_le64(first_blkno); | |
719 | eb->h_fs_generation = cpu_to_le32(osb->fs_generation); | |
ccd979bd | 720 | eb->h_suballoc_slot = cpu_to_le16(osb->slot_num); |
ccd979bd MF |
721 | eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start); |
722 | eb->h_list.l_count = | |
723 | cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb)); | |
724 | ||
725 | suballoc_bit_start++; | |
726 | first_blkno++; | |
727 | ||
728 | /* We'll also be dirtied by the caller, so | |
729 | * this isn't absolutely necessary. */ | |
730 | status = ocfs2_journal_dirty(handle, bhs[i]); | |
731 | if (status < 0) { | |
732 | mlog_errno(status); | |
733 | goto bail; | |
734 | } | |
735 | } | |
736 | ||
737 | count += num_got; | |
738 | } | |
739 | ||
740 | status = 0; | |
741 | bail: | |
742 | if (status < 0) { | |
743 | for(i = 0; i < wanted; i++) { | |
744 | if (bhs[i]) | |
745 | brelse(bhs[i]); | |
746 | bhs[i] = NULL; | |
747 | } | |
748 | } | |
749 | mlog_exit(status); | |
750 | return status; | |
751 | } | |
752 | ||
dcd0538f MF |
753 | /* |
754 | * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth(). | |
755 | * | |
756 | * Returns the sum of the rightmost extent rec logical offset and | |
757 | * cluster count. | |
758 | * | |
759 | * ocfs2_add_branch() uses this to determine what logical cluster | |
760 | * value should be populated into the leftmost new branch records. | |
761 | * | |
762 | * ocfs2_shift_tree_depth() uses this to determine the # clusters | |
763 | * value for the new topmost tree record. | |
764 | */ | |
765 | static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list *el) | |
766 | { | |
767 | int i; | |
768 | ||
769 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
770 | ||
771 | return le32_to_cpu(el->l_recs[i].e_cpos) + | |
e48edee2 | 772 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
773 | } |
774 | ||
ccd979bd MF |
775 | /* |
776 | * Add an entire tree branch to our inode. eb_bh is the extent block | |
777 | * to start at, if we don't want to start the branch at the dinode | |
778 | * structure. | |
779 | * | |
780 | * last_eb_bh is required as we have to update it's next_leaf pointer | |
781 | * for the new last extent block. | |
782 | * | |
783 | * the new branch will be 'empty' in the sense that every block will | |
e48edee2 | 784 | * contain a single record with cluster count == 0. |
ccd979bd MF |
785 | */ |
786 | static int ocfs2_add_branch(struct ocfs2_super *osb, | |
1fabe148 | 787 | handle_t *handle, |
ccd979bd | 788 | struct inode *inode, |
e7d4cb6b | 789 | struct ocfs2_extent_tree *et, |
ccd979bd | 790 | struct buffer_head *eb_bh, |
328d5752 | 791 | struct buffer_head **last_eb_bh, |
ccd979bd MF |
792 | struct ocfs2_alloc_context *meta_ac) |
793 | { | |
794 | int status, new_blocks, i; | |
795 | u64 next_blkno, new_last_eb_blk; | |
796 | struct buffer_head *bh; | |
797 | struct buffer_head **new_eb_bhs = NULL; | |
ccd979bd MF |
798 | struct ocfs2_extent_block *eb; |
799 | struct ocfs2_extent_list *eb_el; | |
800 | struct ocfs2_extent_list *el; | |
dcd0538f | 801 | u32 new_cpos; |
ccd979bd MF |
802 | |
803 | mlog_entry_void(); | |
804 | ||
328d5752 | 805 | BUG_ON(!last_eb_bh || !*last_eb_bh); |
ccd979bd | 806 | |
ccd979bd MF |
807 | if (eb_bh) { |
808 | eb = (struct ocfs2_extent_block *) eb_bh->b_data; | |
809 | el = &eb->h_list; | |
810 | } else | |
ce1d9ea6 | 811 | el = et->et_root_el; |
ccd979bd MF |
812 | |
813 | /* we never add a branch to a leaf. */ | |
814 | BUG_ON(!el->l_tree_depth); | |
815 | ||
816 | new_blocks = le16_to_cpu(el->l_tree_depth); | |
817 | ||
818 | /* allocate the number of new eb blocks we need */ | |
819 | new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *), | |
820 | GFP_KERNEL); | |
821 | if (!new_eb_bhs) { | |
822 | status = -ENOMEM; | |
823 | mlog_errno(status); | |
824 | goto bail; | |
825 | } | |
826 | ||
827 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks, | |
828 | meta_ac, new_eb_bhs); | |
829 | if (status < 0) { | |
830 | mlog_errno(status); | |
831 | goto bail; | |
832 | } | |
833 | ||
328d5752 | 834 | eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data; |
dcd0538f MF |
835 | new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list); |
836 | ||
ccd979bd MF |
837 | /* Note: new_eb_bhs[new_blocks - 1] is the guy which will be |
838 | * linked with the rest of the tree. | |
839 | * conversly, new_eb_bhs[0] is the new bottommost leaf. | |
840 | * | |
841 | * when we leave the loop, new_last_eb_blk will point to the | |
842 | * newest leaf, and next_blkno will point to the topmost extent | |
843 | * block. */ | |
844 | next_blkno = new_last_eb_blk = 0; | |
845 | for(i = 0; i < new_blocks; i++) { | |
846 | bh = new_eb_bhs[i]; | |
847 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
848 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
849 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
850 | status = -EIO; | |
851 | goto bail; | |
852 | } | |
853 | eb_el = &eb->h_list; | |
854 | ||
855 | status = ocfs2_journal_access(handle, inode, bh, | |
856 | OCFS2_JOURNAL_ACCESS_CREATE); | |
857 | if (status < 0) { | |
858 | mlog_errno(status); | |
859 | goto bail; | |
860 | } | |
861 | ||
862 | eb->h_next_leaf_blk = 0; | |
863 | eb_el->l_tree_depth = cpu_to_le16(i); | |
864 | eb_el->l_next_free_rec = cpu_to_le16(1); | |
dcd0538f MF |
865 | /* |
866 | * This actually counts as an empty extent as | |
867 | * c_clusters == 0 | |
868 | */ | |
869 | eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos); | |
ccd979bd | 870 | eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno); |
e48edee2 MF |
871 | /* |
872 | * eb_el isn't always an interior node, but even leaf | |
873 | * nodes want a zero'd flags and reserved field so | |
874 | * this gets the whole 32 bits regardless of use. | |
875 | */ | |
876 | eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0); | |
ccd979bd MF |
877 | if (!eb_el->l_tree_depth) |
878 | new_last_eb_blk = le64_to_cpu(eb->h_blkno); | |
879 | ||
880 | status = ocfs2_journal_dirty(handle, bh); | |
881 | if (status < 0) { | |
882 | mlog_errno(status); | |
883 | goto bail; | |
884 | } | |
885 | ||
886 | next_blkno = le64_to_cpu(eb->h_blkno); | |
887 | } | |
888 | ||
889 | /* This is a bit hairy. We want to update up to three blocks | |
890 | * here without leaving any of them in an inconsistent state | |
891 | * in case of error. We don't have to worry about | |
892 | * journal_dirty erroring as it won't unless we've aborted the | |
893 | * handle (in which case we would never be here) so reserving | |
894 | * the write with journal_access is all we need to do. */ | |
328d5752 | 895 | status = ocfs2_journal_access(handle, inode, *last_eb_bh, |
ccd979bd MF |
896 | OCFS2_JOURNAL_ACCESS_WRITE); |
897 | if (status < 0) { | |
898 | mlog_errno(status); | |
899 | goto bail; | |
900 | } | |
ce1d9ea6 | 901 | status = ocfs2_journal_access(handle, inode, et->et_root_bh, |
ccd979bd MF |
902 | OCFS2_JOURNAL_ACCESS_WRITE); |
903 | if (status < 0) { | |
904 | mlog_errno(status); | |
905 | goto bail; | |
906 | } | |
907 | if (eb_bh) { | |
908 | status = ocfs2_journal_access(handle, inode, eb_bh, | |
909 | OCFS2_JOURNAL_ACCESS_WRITE); | |
910 | if (status < 0) { | |
911 | mlog_errno(status); | |
912 | goto bail; | |
913 | } | |
914 | } | |
915 | ||
916 | /* Link the new branch into the rest of the tree (el will | |
e7d4cb6b | 917 | * either be on the root_bh, or the extent block passed in. */ |
ccd979bd MF |
918 | i = le16_to_cpu(el->l_next_free_rec); |
919 | el->l_recs[i].e_blkno = cpu_to_le64(next_blkno); | |
dcd0538f | 920 | el->l_recs[i].e_cpos = cpu_to_le32(new_cpos); |
e48edee2 | 921 | el->l_recs[i].e_int_clusters = 0; |
ccd979bd MF |
922 | le16_add_cpu(&el->l_next_free_rec, 1); |
923 | ||
924 | /* fe needs a new last extent block pointer, as does the | |
925 | * next_leaf on the previously last-extent-block. */ | |
35dc0aa3 | 926 | ocfs2_et_set_last_eb_blk(et, new_last_eb_blk); |
ccd979bd | 927 | |
328d5752 | 928 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; |
ccd979bd MF |
929 | eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk); |
930 | ||
328d5752 | 931 | status = ocfs2_journal_dirty(handle, *last_eb_bh); |
ccd979bd MF |
932 | if (status < 0) |
933 | mlog_errno(status); | |
ce1d9ea6 | 934 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
935 | if (status < 0) |
936 | mlog_errno(status); | |
937 | if (eb_bh) { | |
938 | status = ocfs2_journal_dirty(handle, eb_bh); | |
939 | if (status < 0) | |
940 | mlog_errno(status); | |
941 | } | |
942 | ||
328d5752 MF |
943 | /* |
944 | * Some callers want to track the rightmost leaf so pass it | |
945 | * back here. | |
946 | */ | |
947 | brelse(*last_eb_bh); | |
948 | get_bh(new_eb_bhs[0]); | |
949 | *last_eb_bh = new_eb_bhs[0]; | |
950 | ||
ccd979bd MF |
951 | status = 0; |
952 | bail: | |
953 | if (new_eb_bhs) { | |
954 | for (i = 0; i < new_blocks; i++) | |
955 | if (new_eb_bhs[i]) | |
956 | brelse(new_eb_bhs[i]); | |
957 | kfree(new_eb_bhs); | |
958 | } | |
959 | ||
960 | mlog_exit(status); | |
961 | return status; | |
962 | } | |
963 | ||
964 | /* | |
965 | * adds another level to the allocation tree. | |
966 | * returns back the new extent block so you can add a branch to it | |
967 | * after this call. | |
968 | */ | |
969 | static int ocfs2_shift_tree_depth(struct ocfs2_super *osb, | |
1fabe148 | 970 | handle_t *handle, |
ccd979bd | 971 | struct inode *inode, |
e7d4cb6b | 972 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
973 | struct ocfs2_alloc_context *meta_ac, |
974 | struct buffer_head **ret_new_eb_bh) | |
975 | { | |
976 | int status, i; | |
dcd0538f | 977 | u32 new_clusters; |
ccd979bd | 978 | struct buffer_head *new_eb_bh = NULL; |
ccd979bd | 979 | struct ocfs2_extent_block *eb; |
e7d4cb6b | 980 | struct ocfs2_extent_list *root_el; |
ccd979bd MF |
981 | struct ocfs2_extent_list *eb_el; |
982 | ||
983 | mlog_entry_void(); | |
984 | ||
985 | status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac, | |
986 | &new_eb_bh); | |
987 | if (status < 0) { | |
988 | mlog_errno(status); | |
989 | goto bail; | |
990 | } | |
991 | ||
992 | eb = (struct ocfs2_extent_block *) new_eb_bh->b_data; | |
993 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
994 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
995 | status = -EIO; | |
996 | goto bail; | |
997 | } | |
998 | ||
999 | eb_el = &eb->h_list; | |
ce1d9ea6 | 1000 | root_el = et->et_root_el; |
ccd979bd MF |
1001 | |
1002 | status = ocfs2_journal_access(handle, inode, new_eb_bh, | |
1003 | OCFS2_JOURNAL_ACCESS_CREATE); | |
1004 | if (status < 0) { | |
1005 | mlog_errno(status); | |
1006 | goto bail; | |
1007 | } | |
1008 | ||
e7d4cb6b TM |
1009 | /* copy the root extent list data into the new extent block */ |
1010 | eb_el->l_tree_depth = root_el->l_tree_depth; | |
1011 | eb_el->l_next_free_rec = root_el->l_next_free_rec; | |
1012 | for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1013 | eb_el->l_recs[i] = root_el->l_recs[i]; | |
ccd979bd MF |
1014 | |
1015 | status = ocfs2_journal_dirty(handle, new_eb_bh); | |
1016 | if (status < 0) { | |
1017 | mlog_errno(status); | |
1018 | goto bail; | |
1019 | } | |
1020 | ||
ce1d9ea6 | 1021 | status = ocfs2_journal_access(handle, inode, et->et_root_bh, |
ccd979bd MF |
1022 | OCFS2_JOURNAL_ACCESS_WRITE); |
1023 | if (status < 0) { | |
1024 | mlog_errno(status); | |
1025 | goto bail; | |
1026 | } | |
1027 | ||
dcd0538f MF |
1028 | new_clusters = ocfs2_sum_rightmost_rec(eb_el); |
1029 | ||
e7d4cb6b TM |
1030 | /* update root_bh now */ |
1031 | le16_add_cpu(&root_el->l_tree_depth, 1); | |
1032 | root_el->l_recs[0].e_cpos = 0; | |
1033 | root_el->l_recs[0].e_blkno = eb->h_blkno; | |
1034 | root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters); | |
1035 | for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) | |
1036 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); | |
1037 | root_el->l_next_free_rec = cpu_to_le16(1); | |
ccd979bd MF |
1038 | |
1039 | /* If this is our 1st tree depth shift, then last_eb_blk | |
1040 | * becomes the allocated extent block */ | |
e7d4cb6b | 1041 | if (root_el->l_tree_depth == cpu_to_le16(1)) |
35dc0aa3 | 1042 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
ccd979bd | 1043 | |
ce1d9ea6 | 1044 | status = ocfs2_journal_dirty(handle, et->et_root_bh); |
ccd979bd MF |
1045 | if (status < 0) { |
1046 | mlog_errno(status); | |
1047 | goto bail; | |
1048 | } | |
1049 | ||
1050 | *ret_new_eb_bh = new_eb_bh; | |
1051 | new_eb_bh = NULL; | |
1052 | status = 0; | |
1053 | bail: | |
1054 | if (new_eb_bh) | |
1055 | brelse(new_eb_bh); | |
1056 | ||
1057 | mlog_exit(status); | |
1058 | return status; | |
1059 | } | |
1060 | ||
ccd979bd MF |
1061 | /* |
1062 | * Should only be called when there is no space left in any of the | |
1063 | * leaf nodes. What we want to do is find the lowest tree depth | |
1064 | * non-leaf extent block with room for new records. There are three | |
1065 | * valid results of this search: | |
1066 | * | |
1067 | * 1) a lowest extent block is found, then we pass it back in | |
1068 | * *lowest_eb_bh and return '0' | |
1069 | * | |
e7d4cb6b | 1070 | * 2) the search fails to find anything, but the root_el has room. We |
ccd979bd MF |
1071 | * pass NULL back in *lowest_eb_bh, but still return '0' |
1072 | * | |
e7d4cb6b | 1073 | * 3) the search fails to find anything AND the root_el is full, in |
ccd979bd MF |
1074 | * which case we return > 0 |
1075 | * | |
1076 | * return status < 0 indicates an error. | |
1077 | */ | |
1078 | static int ocfs2_find_branch_target(struct ocfs2_super *osb, | |
1079 | struct inode *inode, | |
e7d4cb6b | 1080 | struct ocfs2_extent_tree *et, |
ccd979bd MF |
1081 | struct buffer_head **target_bh) |
1082 | { | |
1083 | int status = 0, i; | |
1084 | u64 blkno; | |
ccd979bd MF |
1085 | struct ocfs2_extent_block *eb; |
1086 | struct ocfs2_extent_list *el; | |
1087 | struct buffer_head *bh = NULL; | |
1088 | struct buffer_head *lowest_bh = NULL; | |
1089 | ||
1090 | mlog_entry_void(); | |
1091 | ||
1092 | *target_bh = NULL; | |
1093 | ||
ce1d9ea6 | 1094 | el = et->et_root_el; |
ccd979bd MF |
1095 | |
1096 | while(le16_to_cpu(el->l_tree_depth) > 1) { | |
1097 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
b0697053 | 1098 | ocfs2_error(inode->i_sb, "Dinode %llu has empty " |
ccd979bd | 1099 | "extent list (next_free_rec == 0)", |
b0697053 | 1100 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
ccd979bd MF |
1101 | status = -EIO; |
1102 | goto bail; | |
1103 | } | |
1104 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
1105 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1106 | if (!blkno) { | |
b0697053 | 1107 | ocfs2_error(inode->i_sb, "Dinode %llu has extent " |
ccd979bd MF |
1108 | "list where extent # %d has no physical " |
1109 | "block start", | |
b0697053 | 1110 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i); |
ccd979bd MF |
1111 | status = -EIO; |
1112 | goto bail; | |
1113 | } | |
1114 | ||
1115 | if (bh) { | |
1116 | brelse(bh); | |
1117 | bh = NULL; | |
1118 | } | |
1119 | ||
1120 | status = ocfs2_read_block(osb, blkno, &bh, OCFS2_BH_CACHED, | |
1121 | inode); | |
1122 | if (status < 0) { | |
1123 | mlog_errno(status); | |
1124 | goto bail; | |
1125 | } | |
dcd0538f MF |
1126 | |
1127 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1128 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1129 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1130 | status = -EIO; | |
1131 | goto bail; | |
1132 | } | |
1133 | el = &eb->h_list; | |
1134 | ||
1135 | if (le16_to_cpu(el->l_next_free_rec) < | |
1136 | le16_to_cpu(el->l_count)) { | |
1137 | if (lowest_bh) | |
1138 | brelse(lowest_bh); | |
1139 | lowest_bh = bh; | |
1140 | get_bh(lowest_bh); | |
1141 | } | |
1142 | } | |
1143 | ||
1144 | /* If we didn't find one and the fe doesn't have any room, | |
1145 | * then return '1' */ | |
ce1d9ea6 | 1146 | el = et->et_root_el; |
e7d4cb6b | 1147 | if (!lowest_bh && (el->l_next_free_rec == el->l_count)) |
dcd0538f MF |
1148 | status = 1; |
1149 | ||
1150 | *target_bh = lowest_bh; | |
1151 | bail: | |
1152 | if (bh) | |
1153 | brelse(bh); | |
1154 | ||
1155 | mlog_exit(status); | |
1156 | return status; | |
1157 | } | |
1158 | ||
c3afcbb3 MF |
1159 | /* |
1160 | * Grow a b-tree so that it has more records. | |
1161 | * | |
1162 | * We might shift the tree depth in which case existing paths should | |
1163 | * be considered invalid. | |
1164 | * | |
1165 | * Tree depth after the grow is returned via *final_depth. | |
328d5752 MF |
1166 | * |
1167 | * *last_eb_bh will be updated by ocfs2_add_branch(). | |
c3afcbb3 MF |
1168 | */ |
1169 | static int ocfs2_grow_tree(struct inode *inode, handle_t *handle, | |
e7d4cb6b | 1170 | struct ocfs2_extent_tree *et, int *final_depth, |
328d5752 | 1171 | struct buffer_head **last_eb_bh, |
c3afcbb3 MF |
1172 | struct ocfs2_alloc_context *meta_ac) |
1173 | { | |
1174 | int ret, shift; | |
ce1d9ea6 | 1175 | struct ocfs2_extent_list *el = et->et_root_el; |
e7d4cb6b | 1176 | int depth = le16_to_cpu(el->l_tree_depth); |
c3afcbb3 MF |
1177 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1178 | struct buffer_head *bh = NULL; | |
1179 | ||
1180 | BUG_ON(meta_ac == NULL); | |
1181 | ||
e7d4cb6b | 1182 | shift = ocfs2_find_branch_target(osb, inode, et, &bh); |
c3afcbb3 MF |
1183 | if (shift < 0) { |
1184 | ret = shift; | |
1185 | mlog_errno(ret); | |
1186 | goto out; | |
1187 | } | |
1188 | ||
1189 | /* We traveled all the way to the bottom of the allocation tree | |
1190 | * and didn't find room for any more extents - we need to add | |
1191 | * another tree level */ | |
1192 | if (shift) { | |
1193 | BUG_ON(bh); | |
1194 | mlog(0, "need to shift tree depth (current = %d)\n", depth); | |
1195 | ||
1196 | /* ocfs2_shift_tree_depth will return us a buffer with | |
1197 | * the new extent block (so we can pass that to | |
1198 | * ocfs2_add_branch). */ | |
e7d4cb6b | 1199 | ret = ocfs2_shift_tree_depth(osb, handle, inode, et, |
c3afcbb3 MF |
1200 | meta_ac, &bh); |
1201 | if (ret < 0) { | |
1202 | mlog_errno(ret); | |
1203 | goto out; | |
1204 | } | |
1205 | depth++; | |
328d5752 MF |
1206 | if (depth == 1) { |
1207 | /* | |
1208 | * Special case: we have room now if we shifted from | |
1209 | * tree_depth 0, so no more work needs to be done. | |
1210 | * | |
1211 | * We won't be calling add_branch, so pass | |
1212 | * back *last_eb_bh as the new leaf. At depth | |
1213 | * zero, it should always be null so there's | |
1214 | * no reason to brelse. | |
1215 | */ | |
1216 | BUG_ON(*last_eb_bh); | |
1217 | get_bh(bh); | |
1218 | *last_eb_bh = bh; | |
c3afcbb3 | 1219 | goto out; |
328d5752 | 1220 | } |
c3afcbb3 MF |
1221 | } |
1222 | ||
1223 | /* call ocfs2_add_branch to add the final part of the tree with | |
1224 | * the new data. */ | |
1225 | mlog(0, "add branch. bh = %p\n", bh); | |
e7d4cb6b | 1226 | ret = ocfs2_add_branch(osb, handle, inode, et, bh, last_eb_bh, |
c3afcbb3 MF |
1227 | meta_ac); |
1228 | if (ret < 0) { | |
1229 | mlog_errno(ret); | |
1230 | goto out; | |
1231 | } | |
1232 | ||
1233 | out: | |
1234 | if (final_depth) | |
1235 | *final_depth = depth; | |
1236 | brelse(bh); | |
1237 | return ret; | |
1238 | } | |
1239 | ||
dcd0538f MF |
1240 | /* |
1241 | * This function will discard the rightmost extent record. | |
1242 | */ | |
1243 | static void ocfs2_shift_records_right(struct ocfs2_extent_list *el) | |
1244 | { | |
1245 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1246 | int count = le16_to_cpu(el->l_count); | |
1247 | unsigned int num_bytes; | |
1248 | ||
1249 | BUG_ON(!next_free); | |
1250 | /* This will cause us to go off the end of our extent list. */ | |
1251 | BUG_ON(next_free >= count); | |
1252 | ||
1253 | num_bytes = sizeof(struct ocfs2_extent_rec) * next_free; | |
1254 | ||
1255 | memmove(&el->l_recs[1], &el->l_recs[0], num_bytes); | |
1256 | } | |
1257 | ||
1258 | static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el, | |
1259 | struct ocfs2_extent_rec *insert_rec) | |
1260 | { | |
1261 | int i, insert_index, next_free, has_empty, num_bytes; | |
1262 | u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos); | |
1263 | struct ocfs2_extent_rec *rec; | |
1264 | ||
1265 | next_free = le16_to_cpu(el->l_next_free_rec); | |
1266 | has_empty = ocfs2_is_empty_extent(&el->l_recs[0]); | |
1267 | ||
1268 | BUG_ON(!next_free); | |
1269 | ||
1270 | /* The tree code before us didn't allow enough room in the leaf. */ | |
b1f3550f | 1271 | BUG_ON(el->l_next_free_rec == el->l_count && !has_empty); |
dcd0538f MF |
1272 | |
1273 | /* | |
1274 | * The easiest way to approach this is to just remove the | |
1275 | * empty extent and temporarily decrement next_free. | |
1276 | */ | |
1277 | if (has_empty) { | |
1278 | /* | |
1279 | * If next_free was 1 (only an empty extent), this | |
1280 | * loop won't execute, which is fine. We still want | |
1281 | * the decrement above to happen. | |
1282 | */ | |
1283 | for(i = 0; i < (next_free - 1); i++) | |
1284 | el->l_recs[i] = el->l_recs[i+1]; | |
1285 | ||
1286 | next_free--; | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Figure out what the new record index should be. | |
1291 | */ | |
1292 | for(i = 0; i < next_free; i++) { | |
1293 | rec = &el->l_recs[i]; | |
1294 | ||
1295 | if (insert_cpos < le32_to_cpu(rec->e_cpos)) | |
1296 | break; | |
1297 | } | |
1298 | insert_index = i; | |
1299 | ||
1300 | mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n", | |
1301 | insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count)); | |
1302 | ||
1303 | BUG_ON(insert_index < 0); | |
1304 | BUG_ON(insert_index >= le16_to_cpu(el->l_count)); | |
1305 | BUG_ON(insert_index > next_free); | |
1306 | ||
1307 | /* | |
1308 | * No need to memmove if we're just adding to the tail. | |
1309 | */ | |
1310 | if (insert_index != next_free) { | |
1311 | BUG_ON(next_free >= le16_to_cpu(el->l_count)); | |
1312 | ||
1313 | num_bytes = next_free - insert_index; | |
1314 | num_bytes *= sizeof(struct ocfs2_extent_rec); | |
1315 | memmove(&el->l_recs[insert_index + 1], | |
1316 | &el->l_recs[insert_index], | |
1317 | num_bytes); | |
1318 | } | |
1319 | ||
1320 | /* | |
1321 | * Either we had an empty extent, and need to re-increment or | |
1322 | * there was no empty extent on a non full rightmost leaf node, | |
1323 | * in which case we still need to increment. | |
1324 | */ | |
1325 | next_free++; | |
1326 | el->l_next_free_rec = cpu_to_le16(next_free); | |
1327 | /* | |
1328 | * Make sure none of the math above just messed up our tree. | |
1329 | */ | |
1330 | BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count)); | |
1331 | ||
1332 | el->l_recs[insert_index] = *insert_rec; | |
1333 | ||
1334 | } | |
1335 | ||
328d5752 MF |
1336 | static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el) |
1337 | { | |
1338 | int size, num_recs = le16_to_cpu(el->l_next_free_rec); | |
1339 | ||
1340 | BUG_ON(num_recs == 0); | |
1341 | ||
1342 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
1343 | num_recs--; | |
1344 | size = num_recs * sizeof(struct ocfs2_extent_rec); | |
1345 | memmove(&el->l_recs[0], &el->l_recs[1], size); | |
1346 | memset(&el->l_recs[num_recs], 0, | |
1347 | sizeof(struct ocfs2_extent_rec)); | |
1348 | el->l_next_free_rec = cpu_to_le16(num_recs); | |
1349 | } | |
1350 | } | |
1351 | ||
dcd0538f MF |
1352 | /* |
1353 | * Create an empty extent record . | |
1354 | * | |
1355 | * l_next_free_rec may be updated. | |
1356 | * | |
1357 | * If an empty extent already exists do nothing. | |
1358 | */ | |
1359 | static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el) | |
1360 | { | |
1361 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
1362 | ||
e48edee2 MF |
1363 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
1364 | ||
dcd0538f MF |
1365 | if (next_free == 0) |
1366 | goto set_and_inc; | |
1367 | ||
1368 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
1369 | return; | |
1370 | ||
1371 | mlog_bug_on_msg(el->l_count == el->l_next_free_rec, | |
1372 | "Asked to create an empty extent in a full list:\n" | |
1373 | "count = %u, tree depth = %u", | |
1374 | le16_to_cpu(el->l_count), | |
1375 | le16_to_cpu(el->l_tree_depth)); | |
1376 | ||
1377 | ocfs2_shift_records_right(el); | |
1378 | ||
1379 | set_and_inc: | |
1380 | le16_add_cpu(&el->l_next_free_rec, 1); | |
1381 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1382 | } | |
1383 | ||
1384 | /* | |
1385 | * For a rotation which involves two leaf nodes, the "root node" is | |
1386 | * the lowest level tree node which contains a path to both leafs. This | |
1387 | * resulting set of information can be used to form a complete "subtree" | |
1388 | * | |
1389 | * This function is passed two full paths from the dinode down to a | |
1390 | * pair of adjacent leaves. It's task is to figure out which path | |
1391 | * index contains the subtree root - this can be the root index itself | |
1392 | * in a worst-case rotation. | |
1393 | * | |
1394 | * The array index of the subtree root is passed back. | |
1395 | */ | |
1396 | static int ocfs2_find_subtree_root(struct inode *inode, | |
1397 | struct ocfs2_path *left, | |
1398 | struct ocfs2_path *right) | |
1399 | { | |
1400 | int i = 0; | |
1401 | ||
1402 | /* | |
1403 | * Check that the caller passed in two paths from the same tree. | |
1404 | */ | |
1405 | BUG_ON(path_root_bh(left) != path_root_bh(right)); | |
1406 | ||
1407 | do { | |
1408 | i++; | |
1409 | ||
1410 | /* | |
1411 | * The caller didn't pass two adjacent paths. | |
1412 | */ | |
1413 | mlog_bug_on_msg(i > left->p_tree_depth, | |
1414 | "Inode %lu, left depth %u, right depth %u\n" | |
1415 | "left leaf blk %llu, right leaf blk %llu\n", | |
1416 | inode->i_ino, left->p_tree_depth, | |
1417 | right->p_tree_depth, | |
1418 | (unsigned long long)path_leaf_bh(left)->b_blocknr, | |
1419 | (unsigned long long)path_leaf_bh(right)->b_blocknr); | |
1420 | } while (left->p_node[i].bh->b_blocknr == | |
1421 | right->p_node[i].bh->b_blocknr); | |
1422 | ||
1423 | return i - 1; | |
1424 | } | |
1425 | ||
1426 | typedef void (path_insert_t)(void *, struct buffer_head *); | |
1427 | ||
1428 | /* | |
1429 | * Traverse a btree path in search of cpos, starting at root_el. | |
1430 | * | |
1431 | * This code can be called with a cpos larger than the tree, in which | |
1432 | * case it will return the rightmost path. | |
1433 | */ | |
1434 | static int __ocfs2_find_path(struct inode *inode, | |
1435 | struct ocfs2_extent_list *root_el, u32 cpos, | |
1436 | path_insert_t *func, void *data) | |
1437 | { | |
1438 | int i, ret = 0; | |
1439 | u32 range; | |
1440 | u64 blkno; | |
1441 | struct buffer_head *bh = NULL; | |
1442 | struct ocfs2_extent_block *eb; | |
1443 | struct ocfs2_extent_list *el; | |
1444 | struct ocfs2_extent_rec *rec; | |
1445 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1446 | ||
1447 | el = root_el; | |
1448 | while (el->l_tree_depth) { | |
1449 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
1450 | ocfs2_error(inode->i_sb, | |
1451 | "Inode %llu has empty extent list at " | |
1452 | "depth %u\n", | |
1453 | (unsigned long long)oi->ip_blkno, | |
1454 | le16_to_cpu(el->l_tree_depth)); | |
1455 | ret = -EROFS; | |
1456 | goto out; | |
1457 | ||
1458 | } | |
1459 | ||
1460 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) { | |
1461 | rec = &el->l_recs[i]; | |
1462 | ||
1463 | /* | |
1464 | * In the case that cpos is off the allocation | |
1465 | * tree, this should just wind up returning the | |
1466 | * rightmost record. | |
1467 | */ | |
1468 | range = le32_to_cpu(rec->e_cpos) + | |
e48edee2 | 1469 | ocfs2_rec_clusters(el, rec); |
dcd0538f MF |
1470 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) |
1471 | break; | |
1472 | } | |
1473 | ||
1474 | blkno = le64_to_cpu(el->l_recs[i].e_blkno); | |
1475 | if (blkno == 0) { | |
1476 | ocfs2_error(inode->i_sb, | |
1477 | "Inode %llu has bad blkno in extent list " | |
1478 | "at depth %u (index %d)\n", | |
1479 | (unsigned long long)oi->ip_blkno, | |
1480 | le16_to_cpu(el->l_tree_depth), i); | |
1481 | ret = -EROFS; | |
1482 | goto out; | |
1483 | } | |
1484 | ||
1485 | brelse(bh); | |
1486 | bh = NULL; | |
1487 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno, | |
1488 | &bh, OCFS2_BH_CACHED, inode); | |
1489 | if (ret) { | |
1490 | mlog_errno(ret); | |
1491 | goto out; | |
1492 | } | |
1493 | ||
1494 | eb = (struct ocfs2_extent_block *) bh->b_data; | |
1495 | el = &eb->h_list; | |
1496 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
1497 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
1498 | ret = -EIO; | |
1499 | goto out; | |
1500 | } | |
1501 | ||
1502 | if (le16_to_cpu(el->l_next_free_rec) > | |
1503 | le16_to_cpu(el->l_count)) { | |
1504 | ocfs2_error(inode->i_sb, | |
1505 | "Inode %llu has bad count in extent list " | |
1506 | "at block %llu (next free=%u, count=%u)\n", | |
1507 | (unsigned long long)oi->ip_blkno, | |
1508 | (unsigned long long)bh->b_blocknr, | |
1509 | le16_to_cpu(el->l_next_free_rec), | |
1510 | le16_to_cpu(el->l_count)); | |
1511 | ret = -EROFS; | |
1512 | goto out; | |
1513 | } | |
1514 | ||
1515 | if (func) | |
1516 | func(data, bh); | |
1517 | } | |
1518 | ||
1519 | out: | |
1520 | /* | |
1521 | * Catch any trailing bh that the loop didn't handle. | |
1522 | */ | |
1523 | brelse(bh); | |
1524 | ||
1525 | return ret; | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * Given an initialized path (that is, it has a valid root extent | |
1530 | * list), this function will traverse the btree in search of the path | |
1531 | * which would contain cpos. | |
1532 | * | |
1533 | * The path traveled is recorded in the path structure. | |
1534 | * | |
1535 | * Note that this will not do any comparisons on leaf node extent | |
1536 | * records, so it will work fine in the case that we just added a tree | |
1537 | * branch. | |
1538 | */ | |
1539 | struct find_path_data { | |
1540 | int index; | |
1541 | struct ocfs2_path *path; | |
1542 | }; | |
1543 | static void find_path_ins(void *data, struct buffer_head *bh) | |
1544 | { | |
1545 | struct find_path_data *fp = data; | |
1546 | ||
1547 | get_bh(bh); | |
1548 | ocfs2_path_insert_eb(fp->path, fp->index, bh); | |
1549 | fp->index++; | |
1550 | } | |
1551 | static int ocfs2_find_path(struct inode *inode, struct ocfs2_path *path, | |
1552 | u32 cpos) | |
1553 | { | |
1554 | struct find_path_data data; | |
1555 | ||
1556 | data.index = 1; | |
1557 | data.path = path; | |
1558 | return __ocfs2_find_path(inode, path_root_el(path), cpos, | |
1559 | find_path_ins, &data); | |
1560 | } | |
1561 | ||
1562 | static void find_leaf_ins(void *data, struct buffer_head *bh) | |
1563 | { | |
1564 | struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data; | |
1565 | struct ocfs2_extent_list *el = &eb->h_list; | |
1566 | struct buffer_head **ret = data; | |
1567 | ||
1568 | /* We want to retain only the leaf block. */ | |
1569 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
1570 | get_bh(bh); | |
1571 | *ret = bh; | |
1572 | } | |
1573 | } | |
1574 | /* | |
1575 | * Find the leaf block in the tree which would contain cpos. No | |
1576 | * checking of the actual leaf is done. | |
1577 | * | |
1578 | * Some paths want to call this instead of allocating a path structure | |
1579 | * and calling ocfs2_find_path(). | |
1580 | * | |
1581 | * This function doesn't handle non btree extent lists. | |
1582 | */ | |
363041a5 MF |
1583 | int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el, |
1584 | u32 cpos, struct buffer_head **leaf_bh) | |
dcd0538f MF |
1585 | { |
1586 | int ret; | |
1587 | struct buffer_head *bh = NULL; | |
1588 | ||
1589 | ret = __ocfs2_find_path(inode, root_el, cpos, find_leaf_ins, &bh); | |
1590 | if (ret) { | |
1591 | mlog_errno(ret); | |
1592 | goto out; | |
1593 | } | |
1594 | ||
1595 | *leaf_bh = bh; | |
1596 | out: | |
1597 | return ret; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Adjust the adjacent records (left_rec, right_rec) involved in a rotation. | |
1602 | * | |
1603 | * Basically, we've moved stuff around at the bottom of the tree and | |
1604 | * we need to fix up the extent records above the changes to reflect | |
1605 | * the new changes. | |
1606 | * | |
1607 | * left_rec: the record on the left. | |
1608 | * left_child_el: is the child list pointed to by left_rec | |
1609 | * right_rec: the record to the right of left_rec | |
1610 | * right_child_el: is the child list pointed to by right_rec | |
1611 | * | |
1612 | * By definition, this only works on interior nodes. | |
1613 | */ | |
1614 | static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec, | |
1615 | struct ocfs2_extent_list *left_child_el, | |
1616 | struct ocfs2_extent_rec *right_rec, | |
1617 | struct ocfs2_extent_list *right_child_el) | |
1618 | { | |
1619 | u32 left_clusters, right_end; | |
1620 | ||
1621 | /* | |
1622 | * Interior nodes never have holes. Their cpos is the cpos of | |
1623 | * the leftmost record in their child list. Their cluster | |
1624 | * count covers the full theoretical range of their child list | |
1625 | * - the range between their cpos and the cpos of the record | |
1626 | * immediately to their right. | |
1627 | */ | |
1628 | left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos); | |
328d5752 MF |
1629 | if (ocfs2_is_empty_extent(&right_child_el->l_recs[0])) { |
1630 | BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1); | |
1631 | left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos); | |
1632 | } | |
dcd0538f | 1633 | left_clusters -= le32_to_cpu(left_rec->e_cpos); |
e48edee2 | 1634 | left_rec->e_int_clusters = cpu_to_le32(left_clusters); |
dcd0538f MF |
1635 | |
1636 | /* | |
1637 | * Calculate the rightmost cluster count boundary before | |
e48edee2 | 1638 | * moving cpos - we will need to adjust clusters after |
dcd0538f MF |
1639 | * updating e_cpos to keep the same highest cluster count. |
1640 | */ | |
1641 | right_end = le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1642 | right_end += le32_to_cpu(right_rec->e_int_clusters); |
dcd0538f MF |
1643 | |
1644 | right_rec->e_cpos = left_rec->e_cpos; | |
1645 | le32_add_cpu(&right_rec->e_cpos, left_clusters); | |
1646 | ||
1647 | right_end -= le32_to_cpu(right_rec->e_cpos); | |
e48edee2 | 1648 | right_rec->e_int_clusters = cpu_to_le32(right_end); |
dcd0538f MF |
1649 | } |
1650 | ||
1651 | /* | |
1652 | * Adjust the adjacent root node records involved in a | |
1653 | * rotation. left_el_blkno is passed in as a key so that we can easily | |
1654 | * find it's index in the root list. | |
1655 | */ | |
1656 | static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el, | |
1657 | struct ocfs2_extent_list *left_el, | |
1658 | struct ocfs2_extent_list *right_el, | |
1659 | u64 left_el_blkno) | |
1660 | { | |
1661 | int i; | |
1662 | ||
1663 | BUG_ON(le16_to_cpu(root_el->l_tree_depth) <= | |
1664 | le16_to_cpu(left_el->l_tree_depth)); | |
1665 | ||
1666 | for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) { | |
1667 | if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno) | |
1668 | break; | |
1669 | } | |
1670 | ||
1671 | /* | |
1672 | * The path walking code should have never returned a root and | |
1673 | * two paths which are not adjacent. | |
1674 | */ | |
1675 | BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1)); | |
1676 | ||
1677 | ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el, | |
1678 | &root_el->l_recs[i + 1], right_el); | |
1679 | } | |
1680 | ||
1681 | /* | |
1682 | * We've changed a leaf block (in right_path) and need to reflect that | |
1683 | * change back up the subtree. | |
1684 | * | |
1685 | * This happens in multiple places: | |
1686 | * - When we've moved an extent record from the left path leaf to the right | |
1687 | * path leaf to make room for an empty extent in the left path leaf. | |
1688 | * - When our insert into the right path leaf is at the leftmost edge | |
1689 | * and requires an update of the path immediately to it's left. This | |
1690 | * can occur at the end of some types of rotation and appending inserts. | |
677b9752 TM |
1691 | * - When we've adjusted the last extent record in the left path leaf and the |
1692 | * 1st extent record in the right path leaf during cross extent block merge. | |
dcd0538f MF |
1693 | */ |
1694 | static void ocfs2_complete_edge_insert(struct inode *inode, handle_t *handle, | |
1695 | struct ocfs2_path *left_path, | |
1696 | struct ocfs2_path *right_path, | |
1697 | int subtree_index) | |
1698 | { | |
1699 | int ret, i, idx; | |
1700 | struct ocfs2_extent_list *el, *left_el, *right_el; | |
1701 | struct ocfs2_extent_rec *left_rec, *right_rec; | |
1702 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
1703 | ||
1704 | /* | |
1705 | * Update the counts and position values within all the | |
1706 | * interior nodes to reflect the leaf rotation we just did. | |
1707 | * | |
1708 | * The root node is handled below the loop. | |
1709 | * | |
1710 | * We begin the loop with right_el and left_el pointing to the | |
1711 | * leaf lists and work our way up. | |
1712 | * | |
1713 | * NOTE: within this loop, left_el and right_el always refer | |
1714 | * to the *child* lists. | |
1715 | */ | |
1716 | left_el = path_leaf_el(left_path); | |
1717 | right_el = path_leaf_el(right_path); | |
1718 | for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) { | |
1719 | mlog(0, "Adjust records at index %u\n", i); | |
1720 | ||
1721 | /* | |
1722 | * One nice property of knowing that all of these | |
1723 | * nodes are below the root is that we only deal with | |
1724 | * the leftmost right node record and the rightmost | |
1725 | * left node record. | |
1726 | */ | |
1727 | el = left_path->p_node[i].el; | |
1728 | idx = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1729 | left_rec = &el->l_recs[idx]; | |
1730 | ||
1731 | el = right_path->p_node[i].el; | |
1732 | right_rec = &el->l_recs[0]; | |
1733 | ||
1734 | ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec, | |
1735 | right_el); | |
1736 | ||
1737 | ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh); | |
1738 | if (ret) | |
1739 | mlog_errno(ret); | |
1740 | ||
1741 | ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh); | |
1742 | if (ret) | |
1743 | mlog_errno(ret); | |
1744 | ||
1745 | /* | |
1746 | * Setup our list pointers now so that the current | |
1747 | * parents become children in the next iteration. | |
1748 | */ | |
1749 | left_el = left_path->p_node[i].el; | |
1750 | right_el = right_path->p_node[i].el; | |
1751 | } | |
1752 | ||
1753 | /* | |
1754 | * At the root node, adjust the two adjacent records which | |
1755 | * begin our path to the leaves. | |
1756 | */ | |
1757 | ||
1758 | el = left_path->p_node[subtree_index].el; | |
1759 | left_el = left_path->p_node[subtree_index + 1].el; | |
1760 | right_el = right_path->p_node[subtree_index + 1].el; | |
1761 | ||
1762 | ocfs2_adjust_root_records(el, left_el, right_el, | |
1763 | left_path->p_node[subtree_index + 1].bh->b_blocknr); | |
1764 | ||
1765 | root_bh = left_path->p_node[subtree_index].bh; | |
1766 | ||
1767 | ret = ocfs2_journal_dirty(handle, root_bh); | |
1768 | if (ret) | |
1769 | mlog_errno(ret); | |
1770 | } | |
1771 | ||
1772 | static int ocfs2_rotate_subtree_right(struct inode *inode, | |
1773 | handle_t *handle, | |
1774 | struct ocfs2_path *left_path, | |
1775 | struct ocfs2_path *right_path, | |
1776 | int subtree_index) | |
1777 | { | |
1778 | int ret, i; | |
1779 | struct buffer_head *right_leaf_bh; | |
1780 | struct buffer_head *left_leaf_bh = NULL; | |
1781 | struct buffer_head *root_bh; | |
1782 | struct ocfs2_extent_list *right_el, *left_el; | |
1783 | struct ocfs2_extent_rec move_rec; | |
1784 | ||
1785 | left_leaf_bh = path_leaf_bh(left_path); | |
1786 | left_el = path_leaf_el(left_path); | |
1787 | ||
1788 | if (left_el->l_next_free_rec != left_el->l_count) { | |
1789 | ocfs2_error(inode->i_sb, | |
1790 | "Inode %llu has non-full interior leaf node %llu" | |
1791 | "(next free = %u)", | |
1792 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1793 | (unsigned long long)left_leaf_bh->b_blocknr, | |
1794 | le16_to_cpu(left_el->l_next_free_rec)); | |
1795 | return -EROFS; | |
1796 | } | |
1797 | ||
1798 | /* | |
1799 | * This extent block may already have an empty record, so we | |
1800 | * return early if so. | |
1801 | */ | |
1802 | if (ocfs2_is_empty_extent(&left_el->l_recs[0])) | |
1803 | return 0; | |
1804 | ||
1805 | root_bh = left_path->p_node[subtree_index].bh; | |
1806 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
1807 | ||
1808 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
1809 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1810 | if (ret) { | |
1811 | mlog_errno(ret); | |
1812 | goto out; | |
1813 | } | |
1814 | ||
1815 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
1816 | ret = ocfs2_journal_access(handle, inode, | |
1817 | right_path->p_node[i].bh, | |
1818 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1819 | if (ret) { | |
1820 | mlog_errno(ret); | |
1821 | goto out; | |
1822 | } | |
1823 | ||
1824 | ret = ocfs2_journal_access(handle, inode, | |
1825 | left_path->p_node[i].bh, | |
1826 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1827 | if (ret) { | |
1828 | mlog_errno(ret); | |
1829 | goto out; | |
1830 | } | |
1831 | } | |
1832 | ||
1833 | right_leaf_bh = path_leaf_bh(right_path); | |
1834 | right_el = path_leaf_el(right_path); | |
1835 | ||
1836 | /* This is a code error, not a disk corruption. */ | |
1837 | mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails " | |
1838 | "because rightmost leaf block %llu is empty\n", | |
1839 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1840 | (unsigned long long)right_leaf_bh->b_blocknr); | |
1841 | ||
1842 | ocfs2_create_empty_extent(right_el); | |
1843 | ||
1844 | ret = ocfs2_journal_dirty(handle, right_leaf_bh); | |
1845 | if (ret) { | |
1846 | mlog_errno(ret); | |
1847 | goto out; | |
1848 | } | |
1849 | ||
1850 | /* Do the copy now. */ | |
1851 | i = le16_to_cpu(left_el->l_next_free_rec) - 1; | |
1852 | move_rec = left_el->l_recs[i]; | |
1853 | right_el->l_recs[0] = move_rec; | |
1854 | ||
1855 | /* | |
1856 | * Clear out the record we just copied and shift everything | |
1857 | * over, leaving an empty extent in the left leaf. | |
1858 | * | |
1859 | * We temporarily subtract from next_free_rec so that the | |
1860 | * shift will lose the tail record (which is now defunct). | |
1861 | */ | |
1862 | le16_add_cpu(&left_el->l_next_free_rec, -1); | |
1863 | ocfs2_shift_records_right(left_el); | |
1864 | memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
1865 | le16_add_cpu(&left_el->l_next_free_rec, 1); | |
1866 | ||
1867 | ret = ocfs2_journal_dirty(handle, left_leaf_bh); | |
1868 | if (ret) { | |
1869 | mlog_errno(ret); | |
1870 | goto out; | |
1871 | } | |
1872 | ||
1873 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
1874 | subtree_index); | |
1875 | ||
1876 | out: | |
1877 | return ret; | |
1878 | } | |
1879 | ||
1880 | /* | |
1881 | * Given a full path, determine what cpos value would return us a path | |
1882 | * containing the leaf immediately to the left of the current one. | |
1883 | * | |
1884 | * Will return zero if the path passed in is already the leftmost path. | |
1885 | */ | |
1886 | static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb, | |
1887 | struct ocfs2_path *path, u32 *cpos) | |
1888 | { | |
1889 | int i, j, ret = 0; | |
1890 | u64 blkno; | |
1891 | struct ocfs2_extent_list *el; | |
1892 | ||
e48edee2 MF |
1893 | BUG_ON(path->p_tree_depth == 0); |
1894 | ||
dcd0538f MF |
1895 | *cpos = 0; |
1896 | ||
1897 | blkno = path_leaf_bh(path)->b_blocknr; | |
1898 | ||
1899 | /* Start at the tree node just above the leaf and work our way up. */ | |
1900 | i = path->p_tree_depth - 1; | |
1901 | while (i >= 0) { | |
1902 | el = path->p_node[i].el; | |
1903 | ||
1904 | /* | |
1905 | * Find the extent record just before the one in our | |
1906 | * path. | |
1907 | */ | |
1908 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
1909 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
1910 | if (j == 0) { | |
1911 | if (i == 0) { | |
1912 | /* | |
1913 | * We've determined that the | |
1914 | * path specified is already | |
1915 | * the leftmost one - return a | |
1916 | * cpos of zero. | |
1917 | */ | |
1918 | goto out; | |
1919 | } | |
1920 | /* | |
1921 | * The leftmost record points to our | |
1922 | * leaf - we need to travel up the | |
1923 | * tree one level. | |
1924 | */ | |
1925 | goto next_node; | |
1926 | } | |
1927 | ||
1928 | *cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos); | |
e48edee2 MF |
1929 | *cpos = *cpos + ocfs2_rec_clusters(el, |
1930 | &el->l_recs[j - 1]); | |
1931 | *cpos = *cpos - 1; | |
dcd0538f MF |
1932 | goto out; |
1933 | } | |
1934 | } | |
1935 | ||
1936 | /* | |
1937 | * If we got here, we never found a valid node where | |
1938 | * the tree indicated one should be. | |
1939 | */ | |
1940 | ocfs2_error(sb, | |
1941 | "Invalid extent tree at extent block %llu\n", | |
1942 | (unsigned long long)blkno); | |
1943 | ret = -EROFS; | |
1944 | goto out; | |
1945 | ||
1946 | next_node: | |
1947 | blkno = path->p_node[i].bh->b_blocknr; | |
1948 | i--; | |
1949 | } | |
1950 | ||
1951 | out: | |
1952 | return ret; | |
1953 | } | |
1954 | ||
328d5752 MF |
1955 | /* |
1956 | * Extend the transaction by enough credits to complete the rotation, | |
1957 | * and still leave at least the original number of credits allocated | |
1958 | * to this transaction. | |
1959 | */ | |
dcd0538f | 1960 | static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth, |
328d5752 | 1961 | int op_credits, |
dcd0538f MF |
1962 | struct ocfs2_path *path) |
1963 | { | |
328d5752 | 1964 | int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits; |
dcd0538f MF |
1965 | |
1966 | if (handle->h_buffer_credits < credits) | |
1967 | return ocfs2_extend_trans(handle, credits); | |
1968 | ||
1969 | return 0; | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * Trap the case where we're inserting into the theoretical range past | |
1974 | * the _actual_ left leaf range. Otherwise, we'll rotate a record | |
1975 | * whose cpos is less than ours into the right leaf. | |
1976 | * | |
1977 | * It's only necessary to look at the rightmost record of the left | |
1978 | * leaf because the logic that calls us should ensure that the | |
1979 | * theoretical ranges in the path components above the leaves are | |
1980 | * correct. | |
1981 | */ | |
1982 | static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path, | |
1983 | u32 insert_cpos) | |
1984 | { | |
1985 | struct ocfs2_extent_list *left_el; | |
1986 | struct ocfs2_extent_rec *rec; | |
1987 | int next_free; | |
1988 | ||
1989 | left_el = path_leaf_el(left_path); | |
1990 | next_free = le16_to_cpu(left_el->l_next_free_rec); | |
1991 | rec = &left_el->l_recs[next_free - 1]; | |
1992 | ||
1993 | if (insert_cpos > le32_to_cpu(rec->e_cpos)) | |
1994 | return 1; | |
1995 | return 0; | |
1996 | } | |
1997 | ||
328d5752 MF |
1998 | static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos) |
1999 | { | |
2000 | int next_free = le16_to_cpu(el->l_next_free_rec); | |
2001 | unsigned int range; | |
2002 | struct ocfs2_extent_rec *rec; | |
2003 | ||
2004 | if (next_free == 0) | |
2005 | return 0; | |
2006 | ||
2007 | rec = &el->l_recs[0]; | |
2008 | if (ocfs2_is_empty_extent(rec)) { | |
2009 | /* Empty list. */ | |
2010 | if (next_free == 1) | |
2011 | return 0; | |
2012 | rec = &el->l_recs[1]; | |
2013 | } | |
2014 | ||
2015 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
2016 | if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range) | |
2017 | return 1; | |
2018 | return 0; | |
2019 | } | |
2020 | ||
dcd0538f MF |
2021 | /* |
2022 | * Rotate all the records in a btree right one record, starting at insert_cpos. | |
2023 | * | |
2024 | * The path to the rightmost leaf should be passed in. | |
2025 | * | |
2026 | * The array is assumed to be large enough to hold an entire path (tree depth). | |
2027 | * | |
2028 | * Upon succesful return from this function: | |
2029 | * | |
2030 | * - The 'right_path' array will contain a path to the leaf block | |
2031 | * whose range contains e_cpos. | |
2032 | * - That leaf block will have a single empty extent in list index 0. | |
2033 | * - In the case that the rotation requires a post-insert update, | |
2034 | * *ret_left_path will contain a valid path which can be passed to | |
2035 | * ocfs2_insert_path(). | |
2036 | */ | |
2037 | static int ocfs2_rotate_tree_right(struct inode *inode, | |
2038 | handle_t *handle, | |
328d5752 | 2039 | enum ocfs2_split_type split, |
dcd0538f MF |
2040 | u32 insert_cpos, |
2041 | struct ocfs2_path *right_path, | |
2042 | struct ocfs2_path **ret_left_path) | |
2043 | { | |
328d5752 | 2044 | int ret, start, orig_credits = handle->h_buffer_credits; |
dcd0538f MF |
2045 | u32 cpos; |
2046 | struct ocfs2_path *left_path = NULL; | |
2047 | ||
2048 | *ret_left_path = NULL; | |
2049 | ||
2050 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
2051 | path_root_el(right_path)); | |
2052 | if (!left_path) { | |
2053 | ret = -ENOMEM; | |
2054 | mlog_errno(ret); | |
2055 | goto out; | |
2056 | } | |
2057 | ||
2058 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, &cpos); | |
2059 | if (ret) { | |
2060 | mlog_errno(ret); | |
2061 | goto out; | |
2062 | } | |
2063 | ||
2064 | mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos); | |
2065 | ||
2066 | /* | |
2067 | * What we want to do here is: | |
2068 | * | |
2069 | * 1) Start with the rightmost path. | |
2070 | * | |
2071 | * 2) Determine a path to the leaf block directly to the left | |
2072 | * of that leaf. | |
2073 | * | |
2074 | * 3) Determine the 'subtree root' - the lowest level tree node | |
2075 | * which contains a path to both leaves. | |
2076 | * | |
2077 | * 4) Rotate the subtree. | |
2078 | * | |
2079 | * 5) Find the next subtree by considering the left path to be | |
2080 | * the new right path. | |
2081 | * | |
2082 | * The check at the top of this while loop also accepts | |
2083 | * insert_cpos == cpos because cpos is only a _theoretical_ | |
2084 | * value to get us the left path - insert_cpos might very well | |
2085 | * be filling that hole. | |
2086 | * | |
2087 | * Stop at a cpos of '0' because we either started at the | |
2088 | * leftmost branch (i.e., a tree with one branch and a | |
2089 | * rotation inside of it), or we've gone as far as we can in | |
2090 | * rotating subtrees. | |
2091 | */ | |
2092 | while (cpos && insert_cpos <= cpos) { | |
2093 | mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n", | |
2094 | insert_cpos, cpos); | |
2095 | ||
2096 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2097 | if (ret) { | |
2098 | mlog_errno(ret); | |
2099 | goto out; | |
2100 | } | |
2101 | ||
2102 | mlog_bug_on_msg(path_leaf_bh(left_path) == | |
2103 | path_leaf_bh(right_path), | |
2104 | "Inode %lu: error during insert of %u " | |
2105 | "(left path cpos %u) results in two identical " | |
2106 | "paths ending at %llu\n", | |
2107 | inode->i_ino, insert_cpos, cpos, | |
2108 | (unsigned long long) | |
2109 | path_leaf_bh(left_path)->b_blocknr); | |
2110 | ||
328d5752 MF |
2111 | if (split == SPLIT_NONE && |
2112 | ocfs2_rotate_requires_path_adjustment(left_path, | |
dcd0538f | 2113 | insert_cpos)) { |
dcd0538f MF |
2114 | |
2115 | /* | |
2116 | * We've rotated the tree as much as we | |
2117 | * should. The rest is up to | |
2118 | * ocfs2_insert_path() to complete, after the | |
2119 | * record insertion. We indicate this | |
2120 | * situation by returning the left path. | |
2121 | * | |
2122 | * The reason we don't adjust the records here | |
2123 | * before the record insert is that an error | |
2124 | * later might break the rule where a parent | |
2125 | * record e_cpos will reflect the actual | |
2126 | * e_cpos of the 1st nonempty record of the | |
2127 | * child list. | |
2128 | */ | |
2129 | *ret_left_path = left_path; | |
2130 | goto out_ret_path; | |
2131 | } | |
2132 | ||
2133 | start = ocfs2_find_subtree_root(inode, left_path, right_path); | |
2134 | ||
2135 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2136 | start, | |
2137 | (unsigned long long) right_path->p_node[start].bh->b_blocknr, | |
2138 | right_path->p_tree_depth); | |
2139 | ||
2140 | ret = ocfs2_extend_rotate_transaction(handle, start, | |
328d5752 | 2141 | orig_credits, right_path); |
dcd0538f MF |
2142 | if (ret) { |
2143 | mlog_errno(ret); | |
2144 | goto out; | |
2145 | } | |
2146 | ||
2147 | ret = ocfs2_rotate_subtree_right(inode, handle, left_path, | |
2148 | right_path, start); | |
2149 | if (ret) { | |
2150 | mlog_errno(ret); | |
2151 | goto out; | |
2152 | } | |
2153 | ||
328d5752 MF |
2154 | if (split != SPLIT_NONE && |
2155 | ocfs2_leftmost_rec_contains(path_leaf_el(right_path), | |
2156 | insert_cpos)) { | |
2157 | /* | |
2158 | * A rotate moves the rightmost left leaf | |
2159 | * record over to the leftmost right leaf | |
2160 | * slot. If we're doing an extent split | |
2161 | * instead of a real insert, then we have to | |
2162 | * check that the extent to be split wasn't | |
2163 | * just moved over. If it was, then we can | |
2164 | * exit here, passing left_path back - | |
2165 | * ocfs2_split_extent() is smart enough to | |
2166 | * search both leaves. | |
2167 | */ | |
2168 | *ret_left_path = left_path; | |
2169 | goto out_ret_path; | |
2170 | } | |
2171 | ||
dcd0538f MF |
2172 | /* |
2173 | * There is no need to re-read the next right path | |
2174 | * as we know that it'll be our current left | |
2175 | * path. Optimize by copying values instead. | |
2176 | */ | |
2177 | ocfs2_mv_path(right_path, left_path); | |
2178 | ||
2179 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
2180 | &cpos); | |
2181 | if (ret) { | |
2182 | mlog_errno(ret); | |
2183 | goto out; | |
2184 | } | |
2185 | } | |
2186 | ||
2187 | out: | |
2188 | ocfs2_free_path(left_path); | |
2189 | ||
2190 | out_ret_path: | |
2191 | return ret; | |
2192 | } | |
2193 | ||
328d5752 MF |
2194 | static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle, |
2195 | struct ocfs2_path *path) | |
dcd0538f | 2196 | { |
328d5752 | 2197 | int i, idx; |
dcd0538f | 2198 | struct ocfs2_extent_rec *rec; |
328d5752 MF |
2199 | struct ocfs2_extent_list *el; |
2200 | struct ocfs2_extent_block *eb; | |
2201 | u32 range; | |
dcd0538f | 2202 | |
328d5752 MF |
2203 | /* Path should always be rightmost. */ |
2204 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2205 | BUG_ON(eb->h_next_leaf_blk != 0ULL); | |
dcd0538f | 2206 | |
328d5752 MF |
2207 | el = &eb->h_list; |
2208 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
2209 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2210 | rec = &el->l_recs[idx]; | |
2211 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
dcd0538f | 2212 | |
328d5752 MF |
2213 | for (i = 0; i < path->p_tree_depth; i++) { |
2214 | el = path->p_node[i].el; | |
2215 | idx = le16_to_cpu(el->l_next_free_rec) - 1; | |
2216 | rec = &el->l_recs[idx]; | |
dcd0538f | 2217 | |
328d5752 MF |
2218 | rec->e_int_clusters = cpu_to_le32(range); |
2219 | le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos)); | |
dcd0538f | 2220 | |
328d5752 | 2221 | ocfs2_journal_dirty(handle, path->p_node[i].bh); |
dcd0538f | 2222 | } |
dcd0538f MF |
2223 | } |
2224 | ||
328d5752 MF |
2225 | static void ocfs2_unlink_path(struct inode *inode, handle_t *handle, |
2226 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2227 | struct ocfs2_path *path, int unlink_start) | |
dcd0538f | 2228 | { |
328d5752 MF |
2229 | int ret, i; |
2230 | struct ocfs2_extent_block *eb; | |
2231 | struct ocfs2_extent_list *el; | |
2232 | struct buffer_head *bh; | |
2233 | ||
2234 | for(i = unlink_start; i < path_num_items(path); i++) { | |
2235 | bh = path->p_node[i].bh; | |
2236 | ||
2237 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
2238 | /* | |
2239 | * Not all nodes might have had their final count | |
2240 | * decremented by the caller - handle this here. | |
2241 | */ | |
2242 | el = &eb->h_list; | |
2243 | if (le16_to_cpu(el->l_next_free_rec) > 1) { | |
2244 | mlog(ML_ERROR, | |
2245 | "Inode %llu, attempted to remove extent block " | |
2246 | "%llu with %u records\n", | |
2247 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2248 | (unsigned long long)le64_to_cpu(eb->h_blkno), | |
2249 | le16_to_cpu(el->l_next_free_rec)); | |
2250 | ||
2251 | ocfs2_journal_dirty(handle, bh); | |
2252 | ocfs2_remove_from_cache(inode, bh); | |
2253 | continue; | |
2254 | } | |
2255 | ||
2256 | el->l_next_free_rec = 0; | |
2257 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2258 | ||
2259 | ocfs2_journal_dirty(handle, bh); | |
2260 | ||
2261 | ret = ocfs2_cache_extent_block_free(dealloc, eb); | |
2262 | if (ret) | |
2263 | mlog_errno(ret); | |
2264 | ||
2265 | ocfs2_remove_from_cache(inode, bh); | |
2266 | } | |
dcd0538f MF |
2267 | } |
2268 | ||
328d5752 MF |
2269 | static void ocfs2_unlink_subtree(struct inode *inode, handle_t *handle, |
2270 | struct ocfs2_path *left_path, | |
2271 | struct ocfs2_path *right_path, | |
2272 | int subtree_index, | |
2273 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
dcd0538f | 2274 | { |
328d5752 MF |
2275 | int i; |
2276 | struct buffer_head *root_bh = left_path->p_node[subtree_index].bh; | |
2277 | struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el; | |
dcd0538f | 2278 | struct ocfs2_extent_list *el; |
328d5752 | 2279 | struct ocfs2_extent_block *eb; |
dcd0538f | 2280 | |
328d5752 | 2281 | el = path_leaf_el(left_path); |
dcd0538f | 2282 | |
328d5752 | 2283 | eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data; |
e48edee2 | 2284 | |
328d5752 MF |
2285 | for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++) |
2286 | if (root_el->l_recs[i].e_blkno == eb->h_blkno) | |
2287 | break; | |
dcd0538f | 2288 | |
328d5752 | 2289 | BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec)); |
dcd0538f | 2290 | |
328d5752 MF |
2291 | memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec)); |
2292 | le16_add_cpu(&root_el->l_next_free_rec, -1); | |
2293 | ||
2294 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
2295 | eb->h_next_leaf_blk = 0; | |
2296 | ||
2297 | ocfs2_journal_dirty(handle, root_bh); | |
2298 | ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2299 | ||
2300 | ocfs2_unlink_path(inode, handle, dealloc, right_path, | |
2301 | subtree_index + 1); | |
2302 | } | |
2303 | ||
2304 | static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle, | |
2305 | struct ocfs2_path *left_path, | |
2306 | struct ocfs2_path *right_path, | |
2307 | int subtree_index, | |
2308 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2309 | int *deleted, |
2310 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2311 | { |
2312 | int ret, i, del_right_subtree = 0, right_has_empty = 0; | |
e7d4cb6b | 2313 | struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path); |
328d5752 MF |
2314 | struct ocfs2_extent_list *right_leaf_el, *left_leaf_el; |
2315 | struct ocfs2_extent_block *eb; | |
2316 | ||
2317 | *deleted = 0; | |
2318 | ||
2319 | right_leaf_el = path_leaf_el(right_path); | |
2320 | left_leaf_el = path_leaf_el(left_path); | |
2321 | root_bh = left_path->p_node[subtree_index].bh; | |
2322 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
2323 | ||
2324 | if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0])) | |
2325 | return 0; | |
dcd0538f | 2326 | |
328d5752 MF |
2327 | eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data; |
2328 | if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) { | |
dcd0538f | 2329 | /* |
328d5752 MF |
2330 | * It's legal for us to proceed if the right leaf is |
2331 | * the rightmost one and it has an empty extent. There | |
2332 | * are two cases to handle - whether the leaf will be | |
2333 | * empty after removal or not. If the leaf isn't empty | |
2334 | * then just remove the empty extent up front. The | |
2335 | * next block will handle empty leaves by flagging | |
2336 | * them for unlink. | |
2337 | * | |
2338 | * Non rightmost leaves will throw -EAGAIN and the | |
2339 | * caller can manually move the subtree and retry. | |
dcd0538f | 2340 | */ |
dcd0538f | 2341 | |
328d5752 MF |
2342 | if (eb->h_next_leaf_blk != 0ULL) |
2343 | return -EAGAIN; | |
2344 | ||
2345 | if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) { | |
2346 | ret = ocfs2_journal_access(handle, inode, | |
2347 | path_leaf_bh(right_path), | |
2348 | OCFS2_JOURNAL_ACCESS_WRITE); | |
dcd0538f MF |
2349 | if (ret) { |
2350 | mlog_errno(ret); | |
2351 | goto out; | |
2352 | } | |
2353 | ||
328d5752 MF |
2354 | ocfs2_remove_empty_extent(right_leaf_el); |
2355 | } else | |
2356 | right_has_empty = 1; | |
dcd0538f MF |
2357 | } |
2358 | ||
328d5752 MF |
2359 | if (eb->h_next_leaf_blk == 0ULL && |
2360 | le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) { | |
2361 | /* | |
2362 | * We have to update i_last_eb_blk during the meta | |
2363 | * data delete. | |
2364 | */ | |
e7d4cb6b | 2365 | ret = ocfs2_journal_access(handle, inode, et_root_bh, |
328d5752 MF |
2366 | OCFS2_JOURNAL_ACCESS_WRITE); |
2367 | if (ret) { | |
2368 | mlog_errno(ret); | |
2369 | goto out; | |
2370 | } | |
2371 | ||
2372 | del_right_subtree = 1; | |
2373 | } | |
2374 | ||
2375 | /* | |
2376 | * Getting here with an empty extent in the right path implies | |
2377 | * that it's the rightmost path and will be deleted. | |
2378 | */ | |
2379 | BUG_ON(right_has_empty && !del_right_subtree); | |
2380 | ||
2381 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
2382 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2383 | if (ret) { | |
2384 | mlog_errno(ret); | |
2385 | goto out; | |
2386 | } | |
2387 | ||
2388 | for(i = subtree_index + 1; i < path_num_items(right_path); i++) { | |
2389 | ret = ocfs2_journal_access(handle, inode, | |
2390 | right_path->p_node[i].bh, | |
2391 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2392 | if (ret) { | |
2393 | mlog_errno(ret); | |
2394 | goto out; | |
2395 | } | |
2396 | ||
2397 | ret = ocfs2_journal_access(handle, inode, | |
2398 | left_path->p_node[i].bh, | |
2399 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2400 | if (ret) { | |
2401 | mlog_errno(ret); | |
2402 | goto out; | |
2403 | } | |
2404 | } | |
2405 | ||
2406 | if (!right_has_empty) { | |
2407 | /* | |
2408 | * Only do this if we're moving a real | |
2409 | * record. Otherwise, the action is delayed until | |
2410 | * after removal of the right path in which case we | |
2411 | * can do a simple shift to remove the empty extent. | |
2412 | */ | |
2413 | ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]); | |
2414 | memset(&right_leaf_el->l_recs[0], 0, | |
2415 | sizeof(struct ocfs2_extent_rec)); | |
2416 | } | |
2417 | if (eb->h_next_leaf_blk == 0ULL) { | |
2418 | /* | |
2419 | * Move recs over to get rid of empty extent, decrease | |
2420 | * next_free. This is allowed to remove the last | |
2421 | * extent in our leaf (setting l_next_free_rec to | |
2422 | * zero) - the delete code below won't care. | |
2423 | */ | |
2424 | ocfs2_remove_empty_extent(right_leaf_el); | |
2425 | } | |
2426 | ||
2427 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
2428 | if (ret) | |
2429 | mlog_errno(ret); | |
2430 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
2431 | if (ret) | |
2432 | mlog_errno(ret); | |
2433 | ||
2434 | if (del_right_subtree) { | |
2435 | ocfs2_unlink_subtree(inode, handle, left_path, right_path, | |
2436 | subtree_index, dealloc); | |
2437 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2438 | ||
2439 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2440 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2441 | |
2442 | /* | |
2443 | * Removal of the extent in the left leaf was skipped | |
2444 | * above so we could delete the right path | |
2445 | * 1st. | |
2446 | */ | |
2447 | if (right_has_empty) | |
2448 | ocfs2_remove_empty_extent(left_leaf_el); | |
2449 | ||
e7d4cb6b | 2450 | ret = ocfs2_journal_dirty(handle, et_root_bh); |
328d5752 MF |
2451 | if (ret) |
2452 | mlog_errno(ret); | |
2453 | ||
2454 | *deleted = 1; | |
2455 | } else | |
2456 | ocfs2_complete_edge_insert(inode, handle, left_path, right_path, | |
2457 | subtree_index); | |
2458 | ||
2459 | out: | |
2460 | return ret; | |
2461 | } | |
2462 | ||
2463 | /* | |
2464 | * Given a full path, determine what cpos value would return us a path | |
2465 | * containing the leaf immediately to the right of the current one. | |
2466 | * | |
2467 | * Will return zero if the path passed in is already the rightmost path. | |
2468 | * | |
2469 | * This looks similar, but is subtly different to | |
2470 | * ocfs2_find_cpos_for_left_leaf(). | |
2471 | */ | |
2472 | static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb, | |
2473 | struct ocfs2_path *path, u32 *cpos) | |
2474 | { | |
2475 | int i, j, ret = 0; | |
2476 | u64 blkno; | |
2477 | struct ocfs2_extent_list *el; | |
2478 | ||
2479 | *cpos = 0; | |
2480 | ||
2481 | if (path->p_tree_depth == 0) | |
2482 | return 0; | |
2483 | ||
2484 | blkno = path_leaf_bh(path)->b_blocknr; | |
2485 | ||
2486 | /* Start at the tree node just above the leaf and work our way up. */ | |
2487 | i = path->p_tree_depth - 1; | |
2488 | while (i >= 0) { | |
2489 | int next_free; | |
2490 | ||
2491 | el = path->p_node[i].el; | |
2492 | ||
2493 | /* | |
2494 | * Find the extent record just after the one in our | |
2495 | * path. | |
2496 | */ | |
2497 | next_free = le16_to_cpu(el->l_next_free_rec); | |
2498 | for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) { | |
2499 | if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) { | |
2500 | if (j == (next_free - 1)) { | |
2501 | if (i == 0) { | |
2502 | /* | |
2503 | * We've determined that the | |
2504 | * path specified is already | |
2505 | * the rightmost one - return a | |
2506 | * cpos of zero. | |
2507 | */ | |
2508 | goto out; | |
2509 | } | |
2510 | /* | |
2511 | * The rightmost record points to our | |
2512 | * leaf - we need to travel up the | |
2513 | * tree one level. | |
2514 | */ | |
2515 | goto next_node; | |
2516 | } | |
2517 | ||
2518 | *cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos); | |
2519 | goto out; | |
2520 | } | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * If we got here, we never found a valid node where | |
2525 | * the tree indicated one should be. | |
2526 | */ | |
2527 | ocfs2_error(sb, | |
2528 | "Invalid extent tree at extent block %llu\n", | |
2529 | (unsigned long long)blkno); | |
2530 | ret = -EROFS; | |
2531 | goto out; | |
2532 | ||
2533 | next_node: | |
2534 | blkno = path->p_node[i].bh->b_blocknr; | |
2535 | i--; | |
2536 | } | |
2537 | ||
2538 | out: | |
2539 | return ret; | |
2540 | } | |
2541 | ||
2542 | static int ocfs2_rotate_rightmost_leaf_left(struct inode *inode, | |
2543 | handle_t *handle, | |
2544 | struct buffer_head *bh, | |
2545 | struct ocfs2_extent_list *el) | |
2546 | { | |
2547 | int ret; | |
2548 | ||
2549 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2550 | return 0; | |
2551 | ||
2552 | ret = ocfs2_journal_access(handle, inode, bh, | |
2553 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2554 | if (ret) { | |
2555 | mlog_errno(ret); | |
2556 | goto out; | |
2557 | } | |
2558 | ||
2559 | ocfs2_remove_empty_extent(el); | |
2560 | ||
2561 | ret = ocfs2_journal_dirty(handle, bh); | |
2562 | if (ret) | |
2563 | mlog_errno(ret); | |
2564 | ||
2565 | out: | |
2566 | return ret; | |
2567 | } | |
2568 | ||
2569 | static int __ocfs2_rotate_tree_left(struct inode *inode, | |
2570 | handle_t *handle, int orig_credits, | |
2571 | struct ocfs2_path *path, | |
2572 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
2573 | struct ocfs2_path **empty_extent_path, |
2574 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2575 | { |
2576 | int ret, subtree_root, deleted; | |
2577 | u32 right_cpos; | |
2578 | struct ocfs2_path *left_path = NULL; | |
2579 | struct ocfs2_path *right_path = NULL; | |
2580 | ||
2581 | BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0]))); | |
2582 | ||
2583 | *empty_extent_path = NULL; | |
2584 | ||
2585 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, path, | |
2586 | &right_cpos); | |
2587 | if (ret) { | |
2588 | mlog_errno(ret); | |
2589 | goto out; | |
2590 | } | |
2591 | ||
2592 | left_path = ocfs2_new_path(path_root_bh(path), | |
2593 | path_root_el(path)); | |
2594 | if (!left_path) { | |
2595 | ret = -ENOMEM; | |
2596 | mlog_errno(ret); | |
2597 | goto out; | |
2598 | } | |
2599 | ||
2600 | ocfs2_cp_path(left_path, path); | |
2601 | ||
2602 | right_path = ocfs2_new_path(path_root_bh(path), | |
2603 | path_root_el(path)); | |
2604 | if (!right_path) { | |
2605 | ret = -ENOMEM; | |
2606 | mlog_errno(ret); | |
2607 | goto out; | |
2608 | } | |
2609 | ||
2610 | while (right_cpos) { | |
2611 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
2612 | if (ret) { | |
2613 | mlog_errno(ret); | |
2614 | goto out; | |
2615 | } | |
2616 | ||
2617 | subtree_root = ocfs2_find_subtree_root(inode, left_path, | |
2618 | right_path); | |
2619 | ||
2620 | mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n", | |
2621 | subtree_root, | |
2622 | (unsigned long long) | |
2623 | right_path->p_node[subtree_root].bh->b_blocknr, | |
2624 | right_path->p_tree_depth); | |
2625 | ||
2626 | ret = ocfs2_extend_rotate_transaction(handle, subtree_root, | |
2627 | orig_credits, left_path); | |
2628 | if (ret) { | |
2629 | mlog_errno(ret); | |
2630 | goto out; | |
2631 | } | |
2632 | ||
e8aed345 MF |
2633 | /* |
2634 | * Caller might still want to make changes to the | |
2635 | * tree root, so re-add it to the journal here. | |
2636 | */ | |
2637 | ret = ocfs2_journal_access(handle, inode, | |
2638 | path_root_bh(left_path), | |
2639 | OCFS2_JOURNAL_ACCESS_WRITE); | |
2640 | if (ret) { | |
2641 | mlog_errno(ret); | |
2642 | goto out; | |
2643 | } | |
2644 | ||
328d5752 MF |
2645 | ret = ocfs2_rotate_subtree_left(inode, handle, left_path, |
2646 | right_path, subtree_root, | |
e7d4cb6b | 2647 | dealloc, &deleted, et); |
328d5752 MF |
2648 | if (ret == -EAGAIN) { |
2649 | /* | |
2650 | * The rotation has to temporarily stop due to | |
2651 | * the right subtree having an empty | |
2652 | * extent. Pass it back to the caller for a | |
2653 | * fixup. | |
2654 | */ | |
2655 | *empty_extent_path = right_path; | |
2656 | right_path = NULL; | |
2657 | goto out; | |
2658 | } | |
2659 | if (ret) { | |
2660 | mlog_errno(ret); | |
2661 | goto out; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * The subtree rotate might have removed records on | |
2666 | * the rightmost edge. If so, then rotation is | |
2667 | * complete. | |
2668 | */ | |
2669 | if (deleted) | |
2670 | break; | |
2671 | ||
2672 | ocfs2_mv_path(left_path, right_path); | |
2673 | ||
2674 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
2675 | &right_cpos); | |
2676 | if (ret) { | |
2677 | mlog_errno(ret); | |
2678 | goto out; | |
2679 | } | |
2680 | } | |
2681 | ||
2682 | out: | |
2683 | ocfs2_free_path(right_path); | |
2684 | ocfs2_free_path(left_path); | |
2685 | ||
2686 | return ret; | |
2687 | } | |
2688 | ||
2689 | static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle, | |
e7d4cb6b TM |
2690 | struct ocfs2_path *path, |
2691 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
2692 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2693 | { |
2694 | int ret, subtree_index; | |
2695 | u32 cpos; | |
2696 | struct ocfs2_path *left_path = NULL; | |
328d5752 MF |
2697 | struct ocfs2_extent_block *eb; |
2698 | struct ocfs2_extent_list *el; | |
2699 | ||
328d5752 | 2700 | |
35dc0aa3 | 2701 | ret = ocfs2_et_sanity_check(inode, et); |
e7d4cb6b TM |
2702 | if (ret) |
2703 | goto out; | |
328d5752 MF |
2704 | /* |
2705 | * There's two ways we handle this depending on | |
2706 | * whether path is the only existing one. | |
2707 | */ | |
2708 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
2709 | handle->h_buffer_credits, | |
2710 | path); | |
2711 | if (ret) { | |
2712 | mlog_errno(ret); | |
2713 | goto out; | |
2714 | } | |
2715 | ||
2716 | ret = ocfs2_journal_access_path(inode, handle, path); | |
2717 | if (ret) { | |
2718 | mlog_errno(ret); | |
2719 | goto out; | |
2720 | } | |
2721 | ||
2722 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); | |
2723 | if (ret) { | |
2724 | mlog_errno(ret); | |
2725 | goto out; | |
2726 | } | |
2727 | ||
2728 | if (cpos) { | |
2729 | /* | |
2730 | * We have a path to the left of this one - it needs | |
2731 | * an update too. | |
2732 | */ | |
2733 | left_path = ocfs2_new_path(path_root_bh(path), | |
2734 | path_root_el(path)); | |
2735 | if (!left_path) { | |
2736 | ret = -ENOMEM; | |
2737 | mlog_errno(ret); | |
2738 | goto out; | |
2739 | } | |
2740 | ||
2741 | ret = ocfs2_find_path(inode, left_path, cpos); | |
2742 | if (ret) { | |
2743 | mlog_errno(ret); | |
2744 | goto out; | |
2745 | } | |
2746 | ||
2747 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
2748 | if (ret) { | |
2749 | mlog_errno(ret); | |
2750 | goto out; | |
2751 | } | |
2752 | ||
2753 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
2754 | ||
2755 | ocfs2_unlink_subtree(inode, handle, left_path, path, | |
2756 | subtree_index, dealloc); | |
2757 | ocfs2_update_edge_lengths(inode, handle, left_path); | |
2758 | ||
2759 | eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data; | |
35dc0aa3 | 2760 | ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno)); |
328d5752 MF |
2761 | } else { |
2762 | /* | |
2763 | * 'path' is also the leftmost path which | |
2764 | * means it must be the only one. This gets | |
2765 | * handled differently because we want to | |
2766 | * revert the inode back to having extents | |
2767 | * in-line. | |
2768 | */ | |
2769 | ocfs2_unlink_path(inode, handle, dealloc, path, 1); | |
2770 | ||
ce1d9ea6 | 2771 | el = et->et_root_el; |
328d5752 MF |
2772 | el->l_tree_depth = 0; |
2773 | el->l_next_free_rec = 0; | |
2774 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2775 | ||
35dc0aa3 | 2776 | ocfs2_et_set_last_eb_blk(et, 0); |
328d5752 MF |
2777 | } |
2778 | ||
2779 | ocfs2_journal_dirty(handle, path_root_bh(path)); | |
2780 | ||
2781 | out: | |
2782 | ocfs2_free_path(left_path); | |
2783 | return ret; | |
2784 | } | |
2785 | ||
2786 | /* | |
2787 | * Left rotation of btree records. | |
2788 | * | |
2789 | * In many ways, this is (unsurprisingly) the opposite of right | |
2790 | * rotation. We start at some non-rightmost path containing an empty | |
2791 | * extent in the leaf block. The code works its way to the rightmost | |
2792 | * path by rotating records to the left in every subtree. | |
2793 | * | |
2794 | * This is used by any code which reduces the number of extent records | |
2795 | * in a leaf. After removal, an empty record should be placed in the | |
2796 | * leftmost list position. | |
2797 | * | |
2798 | * This won't handle a length update of the rightmost path records if | |
2799 | * the rightmost tree leaf record is removed so the caller is | |
2800 | * responsible for detecting and correcting that. | |
2801 | */ | |
2802 | static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle, | |
2803 | struct ocfs2_path *path, | |
e7d4cb6b TM |
2804 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
2805 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
2806 | { |
2807 | int ret, orig_credits = handle->h_buffer_credits; | |
2808 | struct ocfs2_path *tmp_path = NULL, *restart_path = NULL; | |
2809 | struct ocfs2_extent_block *eb; | |
2810 | struct ocfs2_extent_list *el; | |
2811 | ||
2812 | el = path_leaf_el(path); | |
2813 | if (!ocfs2_is_empty_extent(&el->l_recs[0])) | |
2814 | return 0; | |
2815 | ||
2816 | if (path->p_tree_depth == 0) { | |
2817 | rightmost_no_delete: | |
2818 | /* | |
e7d4cb6b | 2819 | * Inline extents. This is trivially handled, so do |
328d5752 MF |
2820 | * it up front. |
2821 | */ | |
2822 | ret = ocfs2_rotate_rightmost_leaf_left(inode, handle, | |
2823 | path_leaf_bh(path), | |
2824 | path_leaf_el(path)); | |
2825 | if (ret) | |
2826 | mlog_errno(ret); | |
2827 | goto out; | |
2828 | } | |
2829 | ||
2830 | /* | |
2831 | * Handle rightmost branch now. There's several cases: | |
2832 | * 1) simple rotation leaving records in there. That's trivial. | |
2833 | * 2) rotation requiring a branch delete - there's no more | |
2834 | * records left. Two cases of this: | |
2835 | * a) There are branches to the left. | |
2836 | * b) This is also the leftmost (the only) branch. | |
2837 | * | |
2838 | * 1) is handled via ocfs2_rotate_rightmost_leaf_left() | |
2839 | * 2a) we need the left branch so that we can update it with the unlink | |
2840 | * 2b) we need to bring the inode back to inline extents. | |
2841 | */ | |
2842 | ||
2843 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
2844 | el = &eb->h_list; | |
2845 | if (eb->h_next_leaf_blk == 0) { | |
2846 | /* | |
2847 | * This gets a bit tricky if we're going to delete the | |
2848 | * rightmost path. Get the other cases out of the way | |
2849 | * 1st. | |
2850 | */ | |
2851 | if (le16_to_cpu(el->l_next_free_rec) > 1) | |
2852 | goto rightmost_no_delete; | |
2853 | ||
2854 | if (le16_to_cpu(el->l_next_free_rec) == 0) { | |
2855 | ret = -EIO; | |
2856 | ocfs2_error(inode->i_sb, | |
2857 | "Inode %llu has empty extent block at %llu", | |
2858 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2859 | (unsigned long long)le64_to_cpu(eb->h_blkno)); | |
2860 | goto out; | |
2861 | } | |
2862 | ||
2863 | /* | |
2864 | * XXX: The caller can not trust "path" any more after | |
2865 | * this as it will have been deleted. What do we do? | |
2866 | * | |
2867 | * In theory the rotate-for-merge code will never get | |
2868 | * here because it'll always ask for a rotate in a | |
2869 | * nonempty list. | |
2870 | */ | |
2871 | ||
2872 | ret = ocfs2_remove_rightmost_path(inode, handle, path, | |
e7d4cb6b | 2873 | dealloc, et); |
328d5752 MF |
2874 | if (ret) |
2875 | mlog_errno(ret); | |
2876 | goto out; | |
2877 | } | |
2878 | ||
2879 | /* | |
2880 | * Now we can loop, remembering the path we get from -EAGAIN | |
2881 | * and restarting from there. | |
2882 | */ | |
2883 | try_rotate: | |
2884 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path, | |
e7d4cb6b | 2885 | dealloc, &restart_path, et); |
328d5752 MF |
2886 | if (ret && ret != -EAGAIN) { |
2887 | mlog_errno(ret); | |
2888 | goto out; | |
2889 | } | |
2890 | ||
2891 | while (ret == -EAGAIN) { | |
2892 | tmp_path = restart_path; | |
2893 | restart_path = NULL; | |
2894 | ||
2895 | ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, | |
2896 | tmp_path, dealloc, | |
e7d4cb6b | 2897 | &restart_path, et); |
328d5752 MF |
2898 | if (ret && ret != -EAGAIN) { |
2899 | mlog_errno(ret); | |
2900 | goto out; | |
2901 | } | |
2902 | ||
2903 | ocfs2_free_path(tmp_path); | |
2904 | tmp_path = NULL; | |
2905 | ||
2906 | if (ret == 0) | |
2907 | goto try_rotate; | |
2908 | } | |
2909 | ||
2910 | out: | |
2911 | ocfs2_free_path(tmp_path); | |
2912 | ocfs2_free_path(restart_path); | |
2913 | return ret; | |
2914 | } | |
2915 | ||
2916 | static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el, | |
2917 | int index) | |
2918 | { | |
2919 | struct ocfs2_extent_rec *rec = &el->l_recs[index]; | |
2920 | unsigned int size; | |
2921 | ||
2922 | if (rec->e_leaf_clusters == 0) { | |
2923 | /* | |
2924 | * We consumed all of the merged-from record. An empty | |
2925 | * extent cannot exist anywhere but the 1st array | |
2926 | * position, so move things over if the merged-from | |
2927 | * record doesn't occupy that position. | |
2928 | * | |
2929 | * This creates a new empty extent so the caller | |
2930 | * should be smart enough to have removed any existing | |
2931 | * ones. | |
2932 | */ | |
2933 | if (index > 0) { | |
2934 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
2935 | size = index * sizeof(struct ocfs2_extent_rec); | |
2936 | memmove(&el->l_recs[1], &el->l_recs[0], size); | |
2937 | } | |
2938 | ||
2939 | /* | |
2940 | * Always memset - the caller doesn't check whether it | |
2941 | * created an empty extent, so there could be junk in | |
2942 | * the other fields. | |
2943 | */ | |
2944 | memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec)); | |
2945 | } | |
2946 | } | |
2947 | ||
677b9752 TM |
2948 | static int ocfs2_get_right_path(struct inode *inode, |
2949 | struct ocfs2_path *left_path, | |
2950 | struct ocfs2_path **ret_right_path) | |
2951 | { | |
2952 | int ret; | |
2953 | u32 right_cpos; | |
2954 | struct ocfs2_path *right_path = NULL; | |
2955 | struct ocfs2_extent_list *left_el; | |
2956 | ||
2957 | *ret_right_path = NULL; | |
2958 | ||
2959 | /* This function shouldn't be called for non-trees. */ | |
2960 | BUG_ON(left_path->p_tree_depth == 0); | |
2961 | ||
2962 | left_el = path_leaf_el(left_path); | |
2963 | BUG_ON(left_el->l_next_free_rec != left_el->l_count); | |
2964 | ||
2965 | ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path, | |
2966 | &right_cpos); | |
2967 | if (ret) { | |
2968 | mlog_errno(ret); | |
2969 | goto out; | |
2970 | } | |
2971 | ||
2972 | /* This function shouldn't be called for the rightmost leaf. */ | |
2973 | BUG_ON(right_cpos == 0); | |
2974 | ||
2975 | right_path = ocfs2_new_path(path_root_bh(left_path), | |
2976 | path_root_el(left_path)); | |
2977 | if (!right_path) { | |
2978 | ret = -ENOMEM; | |
2979 | mlog_errno(ret); | |
2980 | goto out; | |
2981 | } | |
2982 | ||
2983 | ret = ocfs2_find_path(inode, right_path, right_cpos); | |
2984 | if (ret) { | |
2985 | mlog_errno(ret); | |
2986 | goto out; | |
2987 | } | |
2988 | ||
2989 | *ret_right_path = right_path; | |
2990 | out: | |
2991 | if (ret) | |
2992 | ocfs2_free_path(right_path); | |
2993 | return ret; | |
2994 | } | |
2995 | ||
328d5752 MF |
2996 | /* |
2997 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
2998 | * onto the beginning of the record "next" to it. |
2999 | * For index < l_count - 1, the next means the extent rec at index + 1. | |
3000 | * For index == l_count - 1, the "next" means the 1st extent rec of the | |
3001 | * next extent block. | |
328d5752 | 3002 | */ |
677b9752 TM |
3003 | static int ocfs2_merge_rec_right(struct inode *inode, |
3004 | struct ocfs2_path *left_path, | |
3005 | handle_t *handle, | |
3006 | struct ocfs2_extent_rec *split_rec, | |
3007 | int index) | |
328d5752 | 3008 | { |
677b9752 | 3009 | int ret, next_free, i; |
328d5752 MF |
3010 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3011 | struct ocfs2_extent_rec *left_rec; | |
3012 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3013 | struct ocfs2_extent_list *right_el; |
3014 | struct ocfs2_path *right_path = NULL; | |
3015 | int subtree_index = 0; | |
3016 | struct ocfs2_extent_list *el = path_leaf_el(left_path); | |
3017 | struct buffer_head *bh = path_leaf_bh(left_path); | |
3018 | struct buffer_head *root_bh = NULL; | |
328d5752 MF |
3019 | |
3020 | BUG_ON(index >= le16_to_cpu(el->l_next_free_rec)); | |
328d5752 | 3021 | left_rec = &el->l_recs[index]; |
677b9752 | 3022 | |
9d8df6aa | 3023 | if (index == le16_to_cpu(el->l_next_free_rec) - 1 && |
677b9752 TM |
3024 | le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) { |
3025 | /* we meet with a cross extent block merge. */ | |
3026 | ret = ocfs2_get_right_path(inode, left_path, &right_path); | |
3027 | if (ret) { | |
3028 | mlog_errno(ret); | |
3029 | goto out; | |
3030 | } | |
3031 | ||
3032 | right_el = path_leaf_el(right_path); | |
3033 | next_free = le16_to_cpu(right_el->l_next_free_rec); | |
3034 | BUG_ON(next_free <= 0); | |
3035 | right_rec = &right_el->l_recs[0]; | |
3036 | if (ocfs2_is_empty_extent(right_rec)) { | |
9d8df6aa | 3037 | BUG_ON(next_free <= 1); |
677b9752 TM |
3038 | right_rec = &right_el->l_recs[1]; |
3039 | } | |
3040 | ||
3041 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3042 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3043 | le32_to_cpu(right_rec->e_cpos)); | |
3044 | ||
3045 | subtree_index = ocfs2_find_subtree_root(inode, | |
3046 | left_path, right_path); | |
3047 | ||
3048 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3049 | handle->h_buffer_credits, | |
3050 | right_path); | |
3051 | if (ret) { | |
3052 | mlog_errno(ret); | |
3053 | goto out; | |
3054 | } | |
3055 | ||
3056 | root_bh = left_path->p_node[subtree_index].bh; | |
3057 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3058 | ||
3059 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3060 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3061 | if (ret) { | |
3062 | mlog_errno(ret); | |
3063 | goto out; | |
3064 | } | |
3065 | ||
3066 | for (i = subtree_index + 1; | |
3067 | i < path_num_items(right_path); i++) { | |
3068 | ret = ocfs2_journal_access(handle, inode, | |
3069 | right_path->p_node[i].bh, | |
3070 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3071 | if (ret) { | |
3072 | mlog_errno(ret); | |
3073 | goto out; | |
3074 | } | |
3075 | ||
3076 | ret = ocfs2_journal_access(handle, inode, | |
3077 | left_path->p_node[i].bh, | |
3078 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3079 | if (ret) { | |
3080 | mlog_errno(ret); | |
3081 | goto out; | |
3082 | } | |
3083 | } | |
3084 | ||
3085 | } else { | |
3086 | BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1); | |
3087 | right_rec = &el->l_recs[index + 1]; | |
3088 | } | |
328d5752 MF |
3089 | |
3090 | ret = ocfs2_journal_access(handle, inode, bh, | |
3091 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3092 | if (ret) { | |
3093 | mlog_errno(ret); | |
3094 | goto out; | |
3095 | } | |
3096 | ||
3097 | le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters); | |
3098 | ||
3099 | le32_add_cpu(&right_rec->e_cpos, -split_clusters); | |
3100 | le64_add_cpu(&right_rec->e_blkno, | |
3101 | -ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3102 | le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters); | |
3103 | ||
3104 | ocfs2_cleanup_merge(el, index); | |
3105 | ||
3106 | ret = ocfs2_journal_dirty(handle, bh); | |
3107 | if (ret) | |
3108 | mlog_errno(ret); | |
3109 | ||
677b9752 TM |
3110 | if (right_path) { |
3111 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path)); | |
3112 | if (ret) | |
3113 | mlog_errno(ret); | |
3114 | ||
3115 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3116 | right_path, subtree_index); | |
3117 | } | |
3118 | out: | |
3119 | if (right_path) | |
3120 | ocfs2_free_path(right_path); | |
3121 | return ret; | |
3122 | } | |
3123 | ||
3124 | static int ocfs2_get_left_path(struct inode *inode, | |
3125 | struct ocfs2_path *right_path, | |
3126 | struct ocfs2_path **ret_left_path) | |
3127 | { | |
3128 | int ret; | |
3129 | u32 left_cpos; | |
3130 | struct ocfs2_path *left_path = NULL; | |
3131 | ||
3132 | *ret_left_path = NULL; | |
3133 | ||
3134 | /* This function shouldn't be called for non-trees. */ | |
3135 | BUG_ON(right_path->p_tree_depth == 0); | |
3136 | ||
3137 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
3138 | right_path, &left_cpos); | |
3139 | if (ret) { | |
3140 | mlog_errno(ret); | |
3141 | goto out; | |
3142 | } | |
3143 | ||
3144 | /* This function shouldn't be called for the leftmost leaf. */ | |
3145 | BUG_ON(left_cpos == 0); | |
3146 | ||
3147 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3148 | path_root_el(right_path)); | |
3149 | if (!left_path) { | |
3150 | ret = -ENOMEM; | |
3151 | mlog_errno(ret); | |
3152 | goto out; | |
3153 | } | |
3154 | ||
3155 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
3156 | if (ret) { | |
3157 | mlog_errno(ret); | |
3158 | goto out; | |
3159 | } | |
3160 | ||
3161 | *ret_left_path = left_path; | |
328d5752 | 3162 | out: |
677b9752 TM |
3163 | if (ret) |
3164 | ocfs2_free_path(left_path); | |
328d5752 MF |
3165 | return ret; |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * Remove split_rec clusters from the record at index and merge them | |
677b9752 TM |
3170 | * onto the tail of the record "before" it. |
3171 | * For index > 0, the "before" means the extent rec at index - 1. | |
3172 | * | |
3173 | * For index == 0, the "before" means the last record of the previous | |
3174 | * extent block. And there is also a situation that we may need to | |
3175 | * remove the rightmost leaf extent block in the right_path and change | |
3176 | * the right path to indicate the new rightmost path. | |
328d5752 | 3177 | */ |
677b9752 TM |
3178 | static int ocfs2_merge_rec_left(struct inode *inode, |
3179 | struct ocfs2_path *right_path, | |
328d5752 MF |
3180 | handle_t *handle, |
3181 | struct ocfs2_extent_rec *split_rec, | |
677b9752 | 3182 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
e7d4cb6b | 3183 | struct ocfs2_extent_tree *et, |
677b9752 | 3184 | int index) |
328d5752 | 3185 | { |
677b9752 | 3186 | int ret, i, subtree_index = 0, has_empty_extent = 0; |
328d5752 MF |
3187 | unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters); |
3188 | struct ocfs2_extent_rec *left_rec; | |
3189 | struct ocfs2_extent_rec *right_rec; | |
677b9752 TM |
3190 | struct ocfs2_extent_list *el = path_leaf_el(right_path); |
3191 | struct buffer_head *bh = path_leaf_bh(right_path); | |
3192 | struct buffer_head *root_bh = NULL; | |
3193 | struct ocfs2_path *left_path = NULL; | |
3194 | struct ocfs2_extent_list *left_el; | |
328d5752 | 3195 | |
677b9752 | 3196 | BUG_ON(index < 0); |
328d5752 | 3197 | |
328d5752 | 3198 | right_rec = &el->l_recs[index]; |
677b9752 TM |
3199 | if (index == 0) { |
3200 | /* we meet with a cross extent block merge. */ | |
3201 | ret = ocfs2_get_left_path(inode, right_path, &left_path); | |
3202 | if (ret) { | |
3203 | mlog_errno(ret); | |
3204 | goto out; | |
3205 | } | |
3206 | ||
3207 | left_el = path_leaf_el(left_path); | |
3208 | BUG_ON(le16_to_cpu(left_el->l_next_free_rec) != | |
3209 | le16_to_cpu(left_el->l_count)); | |
3210 | ||
3211 | left_rec = &left_el->l_recs[ | |
3212 | le16_to_cpu(left_el->l_next_free_rec) - 1]; | |
3213 | BUG_ON(le32_to_cpu(left_rec->e_cpos) + | |
3214 | le16_to_cpu(left_rec->e_leaf_clusters) != | |
3215 | le32_to_cpu(split_rec->e_cpos)); | |
3216 | ||
3217 | subtree_index = ocfs2_find_subtree_root(inode, | |
3218 | left_path, right_path); | |
3219 | ||
3220 | ret = ocfs2_extend_rotate_transaction(handle, subtree_index, | |
3221 | handle->h_buffer_credits, | |
3222 | left_path); | |
3223 | if (ret) { | |
3224 | mlog_errno(ret); | |
3225 | goto out; | |
3226 | } | |
3227 | ||
3228 | root_bh = left_path->p_node[subtree_index].bh; | |
3229 | BUG_ON(root_bh != right_path->p_node[subtree_index].bh); | |
3230 | ||
3231 | ret = ocfs2_journal_access(handle, inode, root_bh, | |
3232 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3233 | if (ret) { | |
3234 | mlog_errno(ret); | |
3235 | goto out; | |
3236 | } | |
3237 | ||
3238 | for (i = subtree_index + 1; | |
3239 | i < path_num_items(right_path); i++) { | |
3240 | ret = ocfs2_journal_access(handle, inode, | |
3241 | right_path->p_node[i].bh, | |
3242 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3243 | if (ret) { | |
3244 | mlog_errno(ret); | |
3245 | goto out; | |
3246 | } | |
3247 | ||
3248 | ret = ocfs2_journal_access(handle, inode, | |
3249 | left_path->p_node[i].bh, | |
3250 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3251 | if (ret) { | |
3252 | mlog_errno(ret); | |
3253 | goto out; | |
3254 | } | |
3255 | } | |
3256 | } else { | |
3257 | left_rec = &el->l_recs[index - 1]; | |
3258 | if (ocfs2_is_empty_extent(&el->l_recs[0])) | |
3259 | has_empty_extent = 1; | |
3260 | } | |
328d5752 MF |
3261 | |
3262 | ret = ocfs2_journal_access(handle, inode, bh, | |
3263 | OCFS2_JOURNAL_ACCESS_WRITE); | |
3264 | if (ret) { | |
3265 | mlog_errno(ret); | |
3266 | goto out; | |
3267 | } | |
3268 | ||
3269 | if (has_empty_extent && index == 1) { | |
3270 | /* | |
3271 | * The easy case - we can just plop the record right in. | |
3272 | */ | |
3273 | *left_rec = *split_rec; | |
3274 | ||
3275 | has_empty_extent = 0; | |
677b9752 | 3276 | } else |
328d5752 | 3277 | le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters); |
328d5752 MF |
3278 | |
3279 | le32_add_cpu(&right_rec->e_cpos, split_clusters); | |
3280 | le64_add_cpu(&right_rec->e_blkno, | |
3281 | ocfs2_clusters_to_blocks(inode->i_sb, split_clusters)); | |
3282 | le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters); | |
3283 | ||
3284 | ocfs2_cleanup_merge(el, index); | |
3285 | ||
3286 | ret = ocfs2_journal_dirty(handle, bh); | |
3287 | if (ret) | |
3288 | mlog_errno(ret); | |
3289 | ||
677b9752 TM |
3290 | if (left_path) { |
3291 | ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path)); | |
3292 | if (ret) | |
3293 | mlog_errno(ret); | |
3294 | ||
3295 | /* | |
3296 | * In the situation that the right_rec is empty and the extent | |
3297 | * block is empty also, ocfs2_complete_edge_insert can't handle | |
3298 | * it and we need to delete the right extent block. | |
3299 | */ | |
3300 | if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 && | |
3301 | le16_to_cpu(el->l_next_free_rec) == 1) { | |
3302 | ||
3303 | ret = ocfs2_remove_rightmost_path(inode, handle, | |
e7d4cb6b TM |
3304 | right_path, |
3305 | dealloc, et); | |
677b9752 TM |
3306 | if (ret) { |
3307 | mlog_errno(ret); | |
3308 | goto out; | |
3309 | } | |
3310 | ||
3311 | /* Now the rightmost extent block has been deleted. | |
3312 | * So we use the new rightmost path. | |
3313 | */ | |
3314 | ocfs2_mv_path(right_path, left_path); | |
3315 | left_path = NULL; | |
3316 | } else | |
3317 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3318 | right_path, subtree_index); | |
3319 | } | |
328d5752 | 3320 | out: |
677b9752 TM |
3321 | if (left_path) |
3322 | ocfs2_free_path(left_path); | |
328d5752 MF |
3323 | return ret; |
3324 | } | |
3325 | ||
3326 | static int ocfs2_try_to_merge_extent(struct inode *inode, | |
3327 | handle_t *handle, | |
677b9752 | 3328 | struct ocfs2_path *path, |
328d5752 MF |
3329 | int split_index, |
3330 | struct ocfs2_extent_rec *split_rec, | |
3331 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
3332 | struct ocfs2_merge_ctxt *ctxt, |
3333 | struct ocfs2_extent_tree *et) | |
328d5752 MF |
3334 | |
3335 | { | |
518d7269 | 3336 | int ret = 0; |
677b9752 | 3337 | struct ocfs2_extent_list *el = path_leaf_el(path); |
328d5752 MF |
3338 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
3339 | ||
3340 | BUG_ON(ctxt->c_contig_type == CONTIG_NONE); | |
3341 | ||
518d7269 TM |
3342 | if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) { |
3343 | /* | |
3344 | * The merge code will need to create an empty | |
3345 | * extent to take the place of the newly | |
3346 | * emptied slot. Remove any pre-existing empty | |
3347 | * extents - having more than one in a leaf is | |
3348 | * illegal. | |
3349 | */ | |
677b9752 | 3350 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3351 | dealloc, et); |
518d7269 TM |
3352 | if (ret) { |
3353 | mlog_errno(ret); | |
3354 | goto out; | |
328d5752 | 3355 | } |
518d7269 TM |
3356 | split_index--; |
3357 | rec = &el->l_recs[split_index]; | |
328d5752 MF |
3358 | } |
3359 | ||
3360 | if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) { | |
3361 | /* | |
3362 | * Left-right contig implies this. | |
3363 | */ | |
3364 | BUG_ON(!ctxt->c_split_covers_rec); | |
328d5752 MF |
3365 | |
3366 | /* | |
3367 | * Since the leftright insert always covers the entire | |
3368 | * extent, this call will delete the insert record | |
3369 | * entirely, resulting in an empty extent record added to | |
3370 | * the extent block. | |
3371 | * | |
3372 | * Since the adding of an empty extent shifts | |
3373 | * everything back to the right, there's no need to | |
3374 | * update split_index here. | |
677b9752 TM |
3375 | * |
3376 | * When the split_index is zero, we need to merge it to the | |
3377 | * prevoius extent block. It is more efficient and easier | |
3378 | * if we do merge_right first and merge_left later. | |
328d5752 | 3379 | */ |
677b9752 TM |
3380 | ret = ocfs2_merge_rec_right(inode, path, |
3381 | handle, split_rec, | |
3382 | split_index); | |
328d5752 MF |
3383 | if (ret) { |
3384 | mlog_errno(ret); | |
3385 | goto out; | |
3386 | } | |
3387 | ||
3388 | /* | |
3389 | * We can only get this from logic error above. | |
3390 | */ | |
3391 | BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0])); | |
3392 | ||
677b9752 | 3393 | /* The merge left us with an empty extent, remove it. */ |
e7d4cb6b TM |
3394 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
3395 | dealloc, et); | |
328d5752 MF |
3396 | if (ret) { |
3397 | mlog_errno(ret); | |
3398 | goto out; | |
3399 | } | |
677b9752 | 3400 | |
328d5752 MF |
3401 | rec = &el->l_recs[split_index]; |
3402 | ||
3403 | /* | |
3404 | * Note that we don't pass split_rec here on purpose - | |
677b9752 | 3405 | * we've merged it into the rec already. |
328d5752 | 3406 | */ |
677b9752 TM |
3407 | ret = ocfs2_merge_rec_left(inode, path, |
3408 | handle, rec, | |
e7d4cb6b | 3409 | dealloc, et, |
677b9752 TM |
3410 | split_index); |
3411 | ||
328d5752 MF |
3412 | if (ret) { |
3413 | mlog_errno(ret); | |
3414 | goto out; | |
3415 | } | |
3416 | ||
677b9752 | 3417 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3418 | dealloc, et); |
328d5752 MF |
3419 | /* |
3420 | * Error from this last rotate is not critical, so | |
3421 | * print but don't bubble it up. | |
3422 | */ | |
3423 | if (ret) | |
3424 | mlog_errno(ret); | |
3425 | ret = 0; | |
3426 | } else { | |
3427 | /* | |
3428 | * Merge a record to the left or right. | |
3429 | * | |
3430 | * 'contig_type' is relative to the existing record, | |
3431 | * so for example, if we're "right contig", it's to | |
3432 | * the record on the left (hence the left merge). | |
3433 | */ | |
3434 | if (ctxt->c_contig_type == CONTIG_RIGHT) { | |
3435 | ret = ocfs2_merge_rec_left(inode, | |
677b9752 TM |
3436 | path, |
3437 | handle, split_rec, | |
e7d4cb6b | 3438 | dealloc, et, |
328d5752 MF |
3439 | split_index); |
3440 | if (ret) { | |
3441 | mlog_errno(ret); | |
3442 | goto out; | |
3443 | } | |
3444 | } else { | |
3445 | ret = ocfs2_merge_rec_right(inode, | |
677b9752 TM |
3446 | path, |
3447 | handle, split_rec, | |
328d5752 MF |
3448 | split_index); |
3449 | if (ret) { | |
3450 | mlog_errno(ret); | |
3451 | goto out; | |
3452 | } | |
3453 | } | |
3454 | ||
3455 | if (ctxt->c_split_covers_rec) { | |
3456 | /* | |
3457 | * The merge may have left an empty extent in | |
3458 | * our leaf. Try to rotate it away. | |
3459 | */ | |
677b9752 | 3460 | ret = ocfs2_rotate_tree_left(inode, handle, path, |
e7d4cb6b | 3461 | dealloc, et); |
328d5752 MF |
3462 | if (ret) |
3463 | mlog_errno(ret); | |
3464 | ret = 0; | |
3465 | } | |
3466 | } | |
3467 | ||
3468 | out: | |
3469 | return ret; | |
3470 | } | |
3471 | ||
3472 | static void ocfs2_subtract_from_rec(struct super_block *sb, | |
3473 | enum ocfs2_split_type split, | |
3474 | struct ocfs2_extent_rec *rec, | |
3475 | struct ocfs2_extent_rec *split_rec) | |
3476 | { | |
3477 | u64 len_blocks; | |
3478 | ||
3479 | len_blocks = ocfs2_clusters_to_blocks(sb, | |
3480 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3481 | ||
3482 | if (split == SPLIT_LEFT) { | |
3483 | /* | |
3484 | * Region is on the left edge of the existing | |
3485 | * record. | |
3486 | */ | |
3487 | le32_add_cpu(&rec->e_cpos, | |
3488 | le16_to_cpu(split_rec->e_leaf_clusters)); | |
3489 | le64_add_cpu(&rec->e_blkno, len_blocks); | |
3490 | le16_add_cpu(&rec->e_leaf_clusters, | |
3491 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3492 | } else { | |
3493 | /* | |
3494 | * Region is on the right edge of the existing | |
3495 | * record. | |
3496 | */ | |
3497 | le16_add_cpu(&rec->e_leaf_clusters, | |
3498 | -le16_to_cpu(split_rec->e_leaf_clusters)); | |
3499 | } | |
3500 | } | |
3501 | ||
3502 | /* | |
3503 | * Do the final bits of extent record insertion at the target leaf | |
3504 | * list. If this leaf is part of an allocation tree, it is assumed | |
3505 | * that the tree above has been prepared. | |
3506 | */ | |
3507 | static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec, | |
3508 | struct ocfs2_extent_list *el, | |
3509 | struct ocfs2_insert_type *insert, | |
3510 | struct inode *inode) | |
3511 | { | |
3512 | int i = insert->ins_contig_index; | |
3513 | unsigned int range; | |
3514 | struct ocfs2_extent_rec *rec; | |
3515 | ||
3516 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); | |
3517 | ||
3518 | if (insert->ins_split != SPLIT_NONE) { | |
3519 | i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos)); | |
3520 | BUG_ON(i == -1); | |
3521 | rec = &el->l_recs[i]; | |
3522 | ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec, | |
3523 | insert_rec); | |
3524 | goto rotate; | |
3525 | } | |
3526 | ||
3527 | /* | |
3528 | * Contiguous insert - either left or right. | |
3529 | */ | |
3530 | if (insert->ins_contig != CONTIG_NONE) { | |
3531 | rec = &el->l_recs[i]; | |
3532 | if (insert->ins_contig == CONTIG_LEFT) { | |
3533 | rec->e_blkno = insert_rec->e_blkno; | |
3534 | rec->e_cpos = insert_rec->e_cpos; | |
3535 | } | |
3536 | le16_add_cpu(&rec->e_leaf_clusters, | |
3537 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3538 | return; | |
3539 | } | |
3540 | ||
3541 | /* | |
3542 | * Handle insert into an empty leaf. | |
3543 | */ | |
3544 | if (le16_to_cpu(el->l_next_free_rec) == 0 || | |
3545 | ((le16_to_cpu(el->l_next_free_rec) == 1) && | |
3546 | ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3547 | el->l_recs[0] = *insert_rec; | |
3548 | el->l_next_free_rec = cpu_to_le16(1); | |
3549 | return; | |
3550 | } | |
3551 | ||
3552 | /* | |
3553 | * Appending insert. | |
3554 | */ | |
3555 | if (insert->ins_appending == APPEND_TAIL) { | |
3556 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
3557 | rec = &el->l_recs[i]; | |
3558 | range = le32_to_cpu(rec->e_cpos) | |
3559 | + le16_to_cpu(rec->e_leaf_clusters); | |
3560 | BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range); | |
3561 | ||
3562 | mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >= | |
3563 | le16_to_cpu(el->l_count), | |
3564 | "inode %lu, depth %u, count %u, next free %u, " | |
3565 | "rec.cpos %u, rec.clusters %u, " | |
3566 | "insert.cpos %u, insert.clusters %u\n", | |
3567 | inode->i_ino, | |
3568 | le16_to_cpu(el->l_tree_depth), | |
3569 | le16_to_cpu(el->l_count), | |
3570 | le16_to_cpu(el->l_next_free_rec), | |
3571 | le32_to_cpu(el->l_recs[i].e_cpos), | |
3572 | le16_to_cpu(el->l_recs[i].e_leaf_clusters), | |
3573 | le32_to_cpu(insert_rec->e_cpos), | |
3574 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3575 | i++; | |
3576 | el->l_recs[i] = *insert_rec; | |
3577 | le16_add_cpu(&el->l_next_free_rec, 1); | |
3578 | return; | |
3579 | } | |
3580 | ||
3581 | rotate: | |
3582 | /* | |
3583 | * Ok, we have to rotate. | |
3584 | * | |
3585 | * At this point, it is safe to assume that inserting into an | |
3586 | * empty leaf and appending to a leaf have both been handled | |
3587 | * above. | |
3588 | * | |
3589 | * This leaf needs to have space, either by the empty 1st | |
3590 | * extent record, or by virtue of an l_next_rec < l_count. | |
3591 | */ | |
3592 | ocfs2_rotate_leaf(el, insert_rec); | |
3593 | } | |
3594 | ||
328d5752 MF |
3595 | static void ocfs2_adjust_rightmost_records(struct inode *inode, |
3596 | handle_t *handle, | |
3597 | struct ocfs2_path *path, | |
3598 | struct ocfs2_extent_rec *insert_rec) | |
3599 | { | |
3600 | int ret, i, next_free; | |
3601 | struct buffer_head *bh; | |
3602 | struct ocfs2_extent_list *el; | |
3603 | struct ocfs2_extent_rec *rec; | |
3604 | ||
3605 | /* | |
3606 | * Update everything except the leaf block. | |
3607 | */ | |
3608 | for (i = 0; i < path->p_tree_depth; i++) { | |
3609 | bh = path->p_node[i].bh; | |
3610 | el = path->p_node[i].el; | |
3611 | ||
dcd0538f MF |
3612 | next_free = le16_to_cpu(el->l_next_free_rec); |
3613 | if (next_free == 0) { | |
3614 | ocfs2_error(inode->i_sb, | |
3615 | "Dinode %llu has a bad extent list", | |
3616 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
3617 | ret = -EIO; | |
328d5752 MF |
3618 | return; |
3619 | } | |
3620 | ||
3621 | rec = &el->l_recs[next_free - 1]; | |
3622 | ||
3623 | rec->e_int_clusters = insert_rec->e_cpos; | |
3624 | le32_add_cpu(&rec->e_int_clusters, | |
3625 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
3626 | le32_add_cpu(&rec->e_int_clusters, | |
3627 | -le32_to_cpu(rec->e_cpos)); | |
3628 | ||
3629 | ret = ocfs2_journal_dirty(handle, bh); | |
3630 | if (ret) | |
3631 | mlog_errno(ret); | |
3632 | ||
3633 | } | |
3634 | } | |
3635 | ||
3636 | static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle, | |
3637 | struct ocfs2_extent_rec *insert_rec, | |
3638 | struct ocfs2_path *right_path, | |
3639 | struct ocfs2_path **ret_left_path) | |
3640 | { | |
3641 | int ret, next_free; | |
3642 | struct ocfs2_extent_list *el; | |
3643 | struct ocfs2_path *left_path = NULL; | |
3644 | ||
3645 | *ret_left_path = NULL; | |
3646 | ||
3647 | /* | |
3648 | * This shouldn't happen for non-trees. The extent rec cluster | |
3649 | * count manipulation below only works for interior nodes. | |
3650 | */ | |
3651 | BUG_ON(right_path->p_tree_depth == 0); | |
3652 | ||
3653 | /* | |
3654 | * If our appending insert is at the leftmost edge of a leaf, | |
3655 | * then we might need to update the rightmost records of the | |
3656 | * neighboring path. | |
3657 | */ | |
3658 | el = path_leaf_el(right_path); | |
3659 | next_free = le16_to_cpu(el->l_next_free_rec); | |
3660 | if (next_free == 0 || | |
3661 | (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) { | |
3662 | u32 left_cpos; | |
3663 | ||
3664 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, | |
3665 | &left_cpos); | |
3666 | if (ret) { | |
3667 | mlog_errno(ret); | |
dcd0538f MF |
3668 | goto out; |
3669 | } | |
3670 | ||
328d5752 MF |
3671 | mlog(0, "Append may need a left path update. cpos: %u, " |
3672 | "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos), | |
3673 | left_cpos); | |
e48edee2 | 3674 | |
328d5752 MF |
3675 | /* |
3676 | * No need to worry if the append is already in the | |
3677 | * leftmost leaf. | |
3678 | */ | |
3679 | if (left_cpos) { | |
3680 | left_path = ocfs2_new_path(path_root_bh(right_path), | |
3681 | path_root_el(right_path)); | |
3682 | if (!left_path) { | |
3683 | ret = -ENOMEM; | |
3684 | mlog_errno(ret); | |
3685 | goto out; | |
3686 | } | |
dcd0538f | 3687 | |
328d5752 MF |
3688 | ret = ocfs2_find_path(inode, left_path, left_cpos); |
3689 | if (ret) { | |
3690 | mlog_errno(ret); | |
3691 | goto out; | |
3692 | } | |
dcd0538f | 3693 | |
328d5752 MF |
3694 | /* |
3695 | * ocfs2_insert_path() will pass the left_path to the | |
3696 | * journal for us. | |
3697 | */ | |
3698 | } | |
3699 | } | |
dcd0538f | 3700 | |
328d5752 MF |
3701 | ret = ocfs2_journal_access_path(inode, handle, right_path); |
3702 | if (ret) { | |
3703 | mlog_errno(ret); | |
3704 | goto out; | |
dcd0538f MF |
3705 | } |
3706 | ||
328d5752 MF |
3707 | ocfs2_adjust_rightmost_records(inode, handle, right_path, insert_rec); |
3708 | ||
dcd0538f MF |
3709 | *ret_left_path = left_path; |
3710 | ret = 0; | |
3711 | out: | |
3712 | if (ret != 0) | |
3713 | ocfs2_free_path(left_path); | |
3714 | ||
3715 | return ret; | |
3716 | } | |
3717 | ||
328d5752 MF |
3718 | static void ocfs2_split_record(struct inode *inode, |
3719 | struct ocfs2_path *left_path, | |
3720 | struct ocfs2_path *right_path, | |
3721 | struct ocfs2_extent_rec *split_rec, | |
3722 | enum ocfs2_split_type split) | |
3723 | { | |
3724 | int index; | |
3725 | u32 cpos = le32_to_cpu(split_rec->e_cpos); | |
3726 | struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el; | |
3727 | struct ocfs2_extent_rec *rec, *tmprec; | |
3728 | ||
3729 | right_el = path_leaf_el(right_path);; | |
3730 | if (left_path) | |
3731 | left_el = path_leaf_el(left_path); | |
3732 | ||
3733 | el = right_el; | |
3734 | insert_el = right_el; | |
3735 | index = ocfs2_search_extent_list(el, cpos); | |
3736 | if (index != -1) { | |
3737 | if (index == 0 && left_path) { | |
3738 | BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0])); | |
3739 | ||
3740 | /* | |
3741 | * This typically means that the record | |
3742 | * started in the left path but moved to the | |
3743 | * right as a result of rotation. We either | |
3744 | * move the existing record to the left, or we | |
3745 | * do the later insert there. | |
3746 | * | |
3747 | * In this case, the left path should always | |
3748 | * exist as the rotate code will have passed | |
3749 | * it back for a post-insert update. | |
3750 | */ | |
3751 | ||
3752 | if (split == SPLIT_LEFT) { | |
3753 | /* | |
3754 | * It's a left split. Since we know | |
3755 | * that the rotate code gave us an | |
3756 | * empty extent in the left path, we | |
3757 | * can just do the insert there. | |
3758 | */ | |
3759 | insert_el = left_el; | |
3760 | } else { | |
3761 | /* | |
3762 | * Right split - we have to move the | |
3763 | * existing record over to the left | |
3764 | * leaf. The insert will be into the | |
3765 | * newly created empty extent in the | |
3766 | * right leaf. | |
3767 | */ | |
3768 | tmprec = &right_el->l_recs[index]; | |
3769 | ocfs2_rotate_leaf(left_el, tmprec); | |
3770 | el = left_el; | |
3771 | ||
3772 | memset(tmprec, 0, sizeof(*tmprec)); | |
3773 | index = ocfs2_search_extent_list(left_el, cpos); | |
3774 | BUG_ON(index == -1); | |
3775 | } | |
3776 | } | |
3777 | } else { | |
3778 | BUG_ON(!left_path); | |
3779 | BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0])); | |
3780 | /* | |
3781 | * Left path is easy - we can just allow the insert to | |
3782 | * happen. | |
3783 | */ | |
3784 | el = left_el; | |
3785 | insert_el = left_el; | |
3786 | index = ocfs2_search_extent_list(el, cpos); | |
3787 | BUG_ON(index == -1); | |
3788 | } | |
3789 | ||
3790 | rec = &el->l_recs[index]; | |
3791 | ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec); | |
3792 | ocfs2_rotate_leaf(insert_el, split_rec); | |
3793 | } | |
3794 | ||
dcd0538f | 3795 | /* |
e7d4cb6b TM |
3796 | * This function only does inserts on an allocation b-tree. For tree |
3797 | * depth = 0, ocfs2_insert_at_leaf() is called directly. | |
dcd0538f MF |
3798 | * |
3799 | * right_path is the path we want to do the actual insert | |
3800 | * in. left_path should only be passed in if we need to update that | |
3801 | * portion of the tree after an edge insert. | |
3802 | */ | |
3803 | static int ocfs2_insert_path(struct inode *inode, | |
3804 | handle_t *handle, | |
3805 | struct ocfs2_path *left_path, | |
3806 | struct ocfs2_path *right_path, | |
3807 | struct ocfs2_extent_rec *insert_rec, | |
3808 | struct ocfs2_insert_type *insert) | |
3809 | { | |
3810 | int ret, subtree_index; | |
3811 | struct buffer_head *leaf_bh = path_leaf_bh(right_path); | |
dcd0538f | 3812 | |
dcd0538f MF |
3813 | if (left_path) { |
3814 | int credits = handle->h_buffer_credits; | |
3815 | ||
3816 | /* | |
3817 | * There's a chance that left_path got passed back to | |
3818 | * us without being accounted for in the | |
3819 | * journal. Extend our transaction here to be sure we | |
3820 | * can change those blocks. | |
3821 | */ | |
3822 | credits += left_path->p_tree_depth; | |
3823 | ||
3824 | ret = ocfs2_extend_trans(handle, credits); | |
3825 | if (ret < 0) { | |
3826 | mlog_errno(ret); | |
3827 | goto out; | |
3828 | } | |
3829 | ||
3830 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
3831 | if (ret < 0) { | |
3832 | mlog_errno(ret); | |
3833 | goto out; | |
3834 | } | |
3835 | } | |
3836 | ||
e8aed345 MF |
3837 | /* |
3838 | * Pass both paths to the journal. The majority of inserts | |
3839 | * will be touching all components anyway. | |
3840 | */ | |
3841 | ret = ocfs2_journal_access_path(inode, handle, right_path); | |
3842 | if (ret < 0) { | |
3843 | mlog_errno(ret); | |
3844 | goto out; | |
3845 | } | |
3846 | ||
328d5752 MF |
3847 | if (insert->ins_split != SPLIT_NONE) { |
3848 | /* | |
3849 | * We could call ocfs2_insert_at_leaf() for some types | |
c78bad11 | 3850 | * of splits, but it's easier to just let one separate |
328d5752 MF |
3851 | * function sort it all out. |
3852 | */ | |
3853 | ocfs2_split_record(inode, left_path, right_path, | |
3854 | insert_rec, insert->ins_split); | |
e8aed345 MF |
3855 | |
3856 | /* | |
3857 | * Split might have modified either leaf and we don't | |
3858 | * have a guarantee that the later edge insert will | |
3859 | * dirty this for us. | |
3860 | */ | |
3861 | if (left_path) | |
3862 | ret = ocfs2_journal_dirty(handle, | |
3863 | path_leaf_bh(left_path)); | |
3864 | if (ret) | |
3865 | mlog_errno(ret); | |
328d5752 MF |
3866 | } else |
3867 | ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path), | |
3868 | insert, inode); | |
dcd0538f | 3869 | |
dcd0538f MF |
3870 | ret = ocfs2_journal_dirty(handle, leaf_bh); |
3871 | if (ret) | |
3872 | mlog_errno(ret); | |
3873 | ||
3874 | if (left_path) { | |
3875 | /* | |
3876 | * The rotate code has indicated that we need to fix | |
3877 | * up portions of the tree after the insert. | |
3878 | * | |
3879 | * XXX: Should we extend the transaction here? | |
3880 | */ | |
3881 | subtree_index = ocfs2_find_subtree_root(inode, left_path, | |
3882 | right_path); | |
3883 | ocfs2_complete_edge_insert(inode, handle, left_path, | |
3884 | right_path, subtree_index); | |
3885 | } | |
3886 | ||
3887 | ret = 0; | |
3888 | out: | |
3889 | return ret; | |
3890 | } | |
3891 | ||
3892 | static int ocfs2_do_insert_extent(struct inode *inode, | |
3893 | handle_t *handle, | |
e7d4cb6b | 3894 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
3895 | struct ocfs2_extent_rec *insert_rec, |
3896 | struct ocfs2_insert_type *type) | |
3897 | { | |
3898 | int ret, rotate = 0; | |
3899 | u32 cpos; | |
3900 | struct ocfs2_path *right_path = NULL; | |
3901 | struct ocfs2_path *left_path = NULL; | |
dcd0538f MF |
3902 | struct ocfs2_extent_list *el; |
3903 | ||
ce1d9ea6 | 3904 | el = et->et_root_el; |
dcd0538f | 3905 | |
ce1d9ea6 | 3906 | ret = ocfs2_journal_access(handle, inode, et->et_root_bh, |
dcd0538f MF |
3907 | OCFS2_JOURNAL_ACCESS_WRITE); |
3908 | if (ret) { | |
3909 | mlog_errno(ret); | |
3910 | goto out; | |
3911 | } | |
3912 | ||
3913 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
3914 | ocfs2_insert_at_leaf(insert_rec, el, type, inode); | |
3915 | goto out_update_clusters; | |
3916 | } | |
3917 | ||
ce1d9ea6 | 3918 | right_path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
dcd0538f MF |
3919 | if (!right_path) { |
3920 | ret = -ENOMEM; | |
3921 | mlog_errno(ret); | |
3922 | goto out; | |
3923 | } | |
3924 | ||
3925 | /* | |
3926 | * Determine the path to start with. Rotations need the | |
3927 | * rightmost path, everything else can go directly to the | |
3928 | * target leaf. | |
3929 | */ | |
3930 | cpos = le32_to_cpu(insert_rec->e_cpos); | |
3931 | if (type->ins_appending == APPEND_NONE && | |
3932 | type->ins_contig == CONTIG_NONE) { | |
3933 | rotate = 1; | |
3934 | cpos = UINT_MAX; | |
3935 | } | |
3936 | ||
3937 | ret = ocfs2_find_path(inode, right_path, cpos); | |
3938 | if (ret) { | |
3939 | mlog_errno(ret); | |
3940 | goto out; | |
3941 | } | |
3942 | ||
3943 | /* | |
3944 | * Rotations and appends need special treatment - they modify | |
3945 | * parts of the tree's above them. | |
3946 | * | |
3947 | * Both might pass back a path immediate to the left of the | |
3948 | * one being inserted to. This will be cause | |
3949 | * ocfs2_insert_path() to modify the rightmost records of | |
3950 | * left_path to account for an edge insert. | |
3951 | * | |
3952 | * XXX: When modifying this code, keep in mind that an insert | |
3953 | * can wind up skipping both of these two special cases... | |
3954 | */ | |
3955 | if (rotate) { | |
328d5752 | 3956 | ret = ocfs2_rotate_tree_right(inode, handle, type->ins_split, |
dcd0538f MF |
3957 | le32_to_cpu(insert_rec->e_cpos), |
3958 | right_path, &left_path); | |
3959 | if (ret) { | |
3960 | mlog_errno(ret); | |
3961 | goto out; | |
3962 | } | |
e8aed345 MF |
3963 | |
3964 | /* | |
3965 | * ocfs2_rotate_tree_right() might have extended the | |
3966 | * transaction without re-journaling our tree root. | |
3967 | */ | |
ce1d9ea6 | 3968 | ret = ocfs2_journal_access(handle, inode, et->et_root_bh, |
e8aed345 MF |
3969 | OCFS2_JOURNAL_ACCESS_WRITE); |
3970 | if (ret) { | |
3971 | mlog_errno(ret); | |
3972 | goto out; | |
3973 | } | |
dcd0538f MF |
3974 | } else if (type->ins_appending == APPEND_TAIL |
3975 | && type->ins_contig != CONTIG_LEFT) { | |
3976 | ret = ocfs2_append_rec_to_path(inode, handle, insert_rec, | |
3977 | right_path, &left_path); | |
3978 | if (ret) { | |
3979 | mlog_errno(ret); | |
3980 | goto out; | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | ret = ocfs2_insert_path(inode, handle, left_path, right_path, | |
3985 | insert_rec, type); | |
3986 | if (ret) { | |
3987 | mlog_errno(ret); | |
3988 | goto out; | |
3989 | } | |
3990 | ||
3991 | out_update_clusters: | |
328d5752 | 3992 | if (type->ins_split == SPLIT_NONE) |
35dc0aa3 JB |
3993 | ocfs2_et_update_clusters(inode, et, |
3994 | le16_to_cpu(insert_rec->e_leaf_clusters)); | |
dcd0538f | 3995 | |
ce1d9ea6 | 3996 | ret = ocfs2_journal_dirty(handle, et->et_root_bh); |
dcd0538f MF |
3997 | if (ret) |
3998 | mlog_errno(ret); | |
3999 | ||
4000 | out: | |
4001 | ocfs2_free_path(left_path); | |
4002 | ocfs2_free_path(right_path); | |
4003 | ||
4004 | return ret; | |
4005 | } | |
4006 | ||
328d5752 | 4007 | static enum ocfs2_contig_type |
ad5a4d70 | 4008 | ocfs2_figure_merge_contig_type(struct inode *inode, struct ocfs2_path *path, |
328d5752 MF |
4009 | struct ocfs2_extent_list *el, int index, |
4010 | struct ocfs2_extent_rec *split_rec) | |
4011 | { | |
ad5a4d70 | 4012 | int status; |
328d5752 | 4013 | enum ocfs2_contig_type ret = CONTIG_NONE; |
ad5a4d70 TM |
4014 | u32 left_cpos, right_cpos; |
4015 | struct ocfs2_extent_rec *rec = NULL; | |
4016 | struct ocfs2_extent_list *new_el; | |
4017 | struct ocfs2_path *left_path = NULL, *right_path = NULL; | |
4018 | struct buffer_head *bh; | |
4019 | struct ocfs2_extent_block *eb; | |
4020 | ||
4021 | if (index > 0) { | |
4022 | rec = &el->l_recs[index - 1]; | |
4023 | } else if (path->p_tree_depth > 0) { | |
4024 | status = ocfs2_find_cpos_for_left_leaf(inode->i_sb, | |
4025 | path, &left_cpos); | |
4026 | if (status) | |
4027 | goto out; | |
4028 | ||
4029 | if (left_cpos != 0) { | |
4030 | left_path = ocfs2_new_path(path_root_bh(path), | |
4031 | path_root_el(path)); | |
4032 | if (!left_path) | |
4033 | goto out; | |
4034 | ||
4035 | status = ocfs2_find_path(inode, left_path, left_cpos); | |
4036 | if (status) | |
4037 | goto out; | |
4038 | ||
4039 | new_el = path_leaf_el(left_path); | |
4040 | ||
4041 | if (le16_to_cpu(new_el->l_next_free_rec) != | |
4042 | le16_to_cpu(new_el->l_count)) { | |
4043 | bh = path_leaf_bh(left_path); | |
4044 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
4045 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, | |
4046 | eb); | |
4047 | goto out; | |
4048 | } | |
4049 | rec = &new_el->l_recs[ | |
4050 | le16_to_cpu(new_el->l_next_free_rec) - 1]; | |
4051 | } | |
4052 | } | |
328d5752 MF |
4053 | |
4054 | /* | |
4055 | * We're careful to check for an empty extent record here - | |
4056 | * the merge code will know what to do if it sees one. | |
4057 | */ | |
ad5a4d70 | 4058 | if (rec) { |
328d5752 MF |
4059 | if (index == 1 && ocfs2_is_empty_extent(rec)) { |
4060 | if (split_rec->e_cpos == el->l_recs[index].e_cpos) | |
4061 | ret = CONTIG_RIGHT; | |
4062 | } else { | |
4063 | ret = ocfs2_extent_contig(inode, rec, split_rec); | |
4064 | } | |
4065 | } | |
4066 | ||
ad5a4d70 TM |
4067 | rec = NULL; |
4068 | if (index < (le16_to_cpu(el->l_next_free_rec) - 1)) | |
4069 | rec = &el->l_recs[index + 1]; | |
4070 | else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) && | |
4071 | path->p_tree_depth > 0) { | |
4072 | status = ocfs2_find_cpos_for_right_leaf(inode->i_sb, | |
4073 | path, &right_cpos); | |
4074 | if (status) | |
4075 | goto out; | |
4076 | ||
4077 | if (right_cpos == 0) | |
4078 | goto out; | |
4079 | ||
4080 | right_path = ocfs2_new_path(path_root_bh(path), | |
4081 | path_root_el(path)); | |
4082 | if (!right_path) | |
4083 | goto out; | |
4084 | ||
4085 | status = ocfs2_find_path(inode, right_path, right_cpos); | |
4086 | if (status) | |
4087 | goto out; | |
4088 | ||
4089 | new_el = path_leaf_el(right_path); | |
4090 | rec = &new_el->l_recs[0]; | |
4091 | if (ocfs2_is_empty_extent(rec)) { | |
4092 | if (le16_to_cpu(new_el->l_next_free_rec) <= 1) { | |
4093 | bh = path_leaf_bh(right_path); | |
4094 | eb = (struct ocfs2_extent_block *)bh->b_data; | |
4095 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, | |
4096 | eb); | |
4097 | goto out; | |
4098 | } | |
4099 | rec = &new_el->l_recs[1]; | |
4100 | } | |
4101 | } | |
4102 | ||
4103 | if (rec) { | |
328d5752 MF |
4104 | enum ocfs2_contig_type contig_type; |
4105 | ||
328d5752 MF |
4106 | contig_type = ocfs2_extent_contig(inode, rec, split_rec); |
4107 | ||
4108 | if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT) | |
4109 | ret = CONTIG_LEFTRIGHT; | |
4110 | else if (ret == CONTIG_NONE) | |
4111 | ret = contig_type; | |
4112 | } | |
4113 | ||
ad5a4d70 TM |
4114 | out: |
4115 | if (left_path) | |
4116 | ocfs2_free_path(left_path); | |
4117 | if (right_path) | |
4118 | ocfs2_free_path(right_path); | |
4119 | ||
328d5752 MF |
4120 | return ret; |
4121 | } | |
4122 | ||
dcd0538f MF |
4123 | static void ocfs2_figure_contig_type(struct inode *inode, |
4124 | struct ocfs2_insert_type *insert, | |
4125 | struct ocfs2_extent_list *el, | |
ca12b7c4 TM |
4126 | struct ocfs2_extent_rec *insert_rec, |
4127 | struct ocfs2_extent_tree *et) | |
dcd0538f MF |
4128 | { |
4129 | int i; | |
4130 | enum ocfs2_contig_type contig_type = CONTIG_NONE; | |
4131 | ||
e48edee2 MF |
4132 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
4133 | ||
dcd0538f MF |
4134 | for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { |
4135 | contig_type = ocfs2_extent_contig(inode, &el->l_recs[i], | |
4136 | insert_rec); | |
4137 | if (contig_type != CONTIG_NONE) { | |
4138 | insert->ins_contig_index = i; | |
4139 | break; | |
4140 | } | |
4141 | } | |
4142 | insert->ins_contig = contig_type; | |
ca12b7c4 TM |
4143 | |
4144 | if (insert->ins_contig != CONTIG_NONE) { | |
4145 | struct ocfs2_extent_rec *rec = | |
4146 | &el->l_recs[insert->ins_contig_index]; | |
4147 | unsigned int len = le16_to_cpu(rec->e_leaf_clusters) + | |
4148 | le16_to_cpu(insert_rec->e_leaf_clusters); | |
4149 | ||
4150 | /* | |
4151 | * Caller might want us to limit the size of extents, don't | |
4152 | * calculate contiguousness if we might exceed that limit. | |
4153 | */ | |
ce1d9ea6 JB |
4154 | if (et->et_max_leaf_clusters && |
4155 | (len > et->et_max_leaf_clusters)) | |
ca12b7c4 TM |
4156 | insert->ins_contig = CONTIG_NONE; |
4157 | } | |
dcd0538f MF |
4158 | } |
4159 | ||
4160 | /* | |
4161 | * This should only be called against the righmost leaf extent list. | |
4162 | * | |
4163 | * ocfs2_figure_appending_type() will figure out whether we'll have to | |
4164 | * insert at the tail of the rightmost leaf. | |
4165 | * | |
e7d4cb6b TM |
4166 | * This should also work against the root extent list for tree's with 0 |
4167 | * depth. If we consider the root extent list to be the rightmost leaf node | |
dcd0538f MF |
4168 | * then the logic here makes sense. |
4169 | */ | |
4170 | static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert, | |
4171 | struct ocfs2_extent_list *el, | |
4172 | struct ocfs2_extent_rec *insert_rec) | |
4173 | { | |
4174 | int i; | |
4175 | u32 cpos = le32_to_cpu(insert_rec->e_cpos); | |
4176 | struct ocfs2_extent_rec *rec; | |
4177 | ||
4178 | insert->ins_appending = APPEND_NONE; | |
4179 | ||
e48edee2 | 4180 | BUG_ON(le16_to_cpu(el->l_tree_depth) != 0); |
dcd0538f MF |
4181 | |
4182 | if (!el->l_next_free_rec) | |
4183 | goto set_tail_append; | |
4184 | ||
4185 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
4186 | /* Were all records empty? */ | |
4187 | if (le16_to_cpu(el->l_next_free_rec) == 1) | |
4188 | goto set_tail_append; | |
4189 | } | |
4190 | ||
4191 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
4192 | rec = &el->l_recs[i]; | |
4193 | ||
e48edee2 MF |
4194 | if (cpos >= |
4195 | (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters))) | |
dcd0538f MF |
4196 | goto set_tail_append; |
4197 | ||
4198 | return; | |
4199 | ||
4200 | set_tail_append: | |
4201 | insert->ins_appending = APPEND_TAIL; | |
4202 | } | |
4203 | ||
4204 | /* | |
4205 | * Helper function called at the begining of an insert. | |
4206 | * | |
4207 | * This computes a few things that are commonly used in the process of | |
4208 | * inserting into the btree: | |
4209 | * - Whether the new extent is contiguous with an existing one. | |
4210 | * - The current tree depth. | |
4211 | * - Whether the insert is an appending one. | |
4212 | * - The total # of free records in the tree. | |
4213 | * | |
4214 | * All of the information is stored on the ocfs2_insert_type | |
4215 | * structure. | |
4216 | */ | |
4217 | static int ocfs2_figure_insert_type(struct inode *inode, | |
e7d4cb6b | 4218 | struct ocfs2_extent_tree *et, |
dcd0538f MF |
4219 | struct buffer_head **last_eb_bh, |
4220 | struct ocfs2_extent_rec *insert_rec, | |
c77534f6 | 4221 | int *free_records, |
dcd0538f MF |
4222 | struct ocfs2_insert_type *insert) |
4223 | { | |
4224 | int ret; | |
dcd0538f MF |
4225 | struct ocfs2_extent_block *eb; |
4226 | struct ocfs2_extent_list *el; | |
4227 | struct ocfs2_path *path = NULL; | |
4228 | struct buffer_head *bh = NULL; | |
4229 | ||
328d5752 MF |
4230 | insert->ins_split = SPLIT_NONE; |
4231 | ||
ce1d9ea6 | 4232 | el = et->et_root_el; |
dcd0538f MF |
4233 | insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth); |
4234 | ||
4235 | if (el->l_tree_depth) { | |
4236 | /* | |
4237 | * If we have tree depth, we read in the | |
4238 | * rightmost extent block ahead of time as | |
4239 | * ocfs2_figure_insert_type() and ocfs2_add_branch() | |
4240 | * may want it later. | |
4241 | */ | |
4242 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4243 | ocfs2_et_get_last_eb_blk(et), &bh, |
dcd0538f MF |
4244 | OCFS2_BH_CACHED, inode); |
4245 | if (ret) { | |
4246 | mlog_exit(ret); | |
4247 | goto out; | |
4248 | } | |
ccd979bd | 4249 | eb = (struct ocfs2_extent_block *) bh->b_data; |
ccd979bd | 4250 | el = &eb->h_list; |
dcd0538f | 4251 | } |
ccd979bd | 4252 | |
dcd0538f MF |
4253 | /* |
4254 | * Unless we have a contiguous insert, we'll need to know if | |
4255 | * there is room left in our allocation tree for another | |
4256 | * extent record. | |
4257 | * | |
4258 | * XXX: This test is simplistic, we can search for empty | |
4259 | * extent records too. | |
4260 | */ | |
c77534f6 | 4261 | *free_records = le16_to_cpu(el->l_count) - |
dcd0538f MF |
4262 | le16_to_cpu(el->l_next_free_rec); |
4263 | ||
4264 | if (!insert->ins_tree_depth) { | |
ca12b7c4 | 4265 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4266 | ocfs2_figure_appending_type(insert, el, insert_rec); |
4267 | return 0; | |
ccd979bd MF |
4268 | } |
4269 | ||
ce1d9ea6 | 4270 | path = ocfs2_new_path(et->et_root_bh, et->et_root_el); |
dcd0538f MF |
4271 | if (!path) { |
4272 | ret = -ENOMEM; | |
4273 | mlog_errno(ret); | |
4274 | goto out; | |
4275 | } | |
ccd979bd | 4276 | |
dcd0538f MF |
4277 | /* |
4278 | * In the case that we're inserting past what the tree | |
4279 | * currently accounts for, ocfs2_find_path() will return for | |
4280 | * us the rightmost tree path. This is accounted for below in | |
4281 | * the appending code. | |
4282 | */ | |
4283 | ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos)); | |
4284 | if (ret) { | |
4285 | mlog_errno(ret); | |
4286 | goto out; | |
4287 | } | |
ccd979bd | 4288 | |
dcd0538f MF |
4289 | el = path_leaf_el(path); |
4290 | ||
4291 | /* | |
4292 | * Now that we have the path, there's two things we want to determine: | |
4293 | * 1) Contiguousness (also set contig_index if this is so) | |
4294 | * | |
4295 | * 2) Are we doing an append? We can trivially break this up | |
4296 | * into two types of appends: simple record append, or a | |
4297 | * rotate inside the tail leaf. | |
4298 | */ | |
ca12b7c4 | 4299 | ocfs2_figure_contig_type(inode, insert, el, insert_rec, et); |
dcd0538f MF |
4300 | |
4301 | /* | |
4302 | * The insert code isn't quite ready to deal with all cases of | |
4303 | * left contiguousness. Specifically, if it's an insert into | |
4304 | * the 1st record in a leaf, it will require the adjustment of | |
e48edee2 | 4305 | * cluster count on the last record of the path directly to it's |
dcd0538f MF |
4306 | * left. For now, just catch that case and fool the layers |
4307 | * above us. This works just fine for tree_depth == 0, which | |
4308 | * is why we allow that above. | |
4309 | */ | |
4310 | if (insert->ins_contig == CONTIG_LEFT && | |
4311 | insert->ins_contig_index == 0) | |
4312 | insert->ins_contig = CONTIG_NONE; | |
4313 | ||
4314 | /* | |
4315 | * Ok, so we can simply compare against last_eb to figure out | |
4316 | * whether the path doesn't exist. This will only happen in | |
4317 | * the case that we're doing a tail append, so maybe we can | |
4318 | * take advantage of that information somehow. | |
4319 | */ | |
35dc0aa3 | 4320 | if (ocfs2_et_get_last_eb_blk(et) == |
e7d4cb6b | 4321 | path_leaf_bh(path)->b_blocknr) { |
dcd0538f MF |
4322 | /* |
4323 | * Ok, ocfs2_find_path() returned us the rightmost | |
4324 | * tree path. This might be an appending insert. There are | |
4325 | * two cases: | |
4326 | * 1) We're doing a true append at the tail: | |
4327 | * -This might even be off the end of the leaf | |
4328 | * 2) We're "appending" by rotating in the tail | |
4329 | */ | |
4330 | ocfs2_figure_appending_type(insert, el, insert_rec); | |
4331 | } | |
4332 | ||
4333 | out: | |
4334 | ocfs2_free_path(path); | |
4335 | ||
4336 | if (ret == 0) | |
4337 | *last_eb_bh = bh; | |
4338 | else | |
4339 | brelse(bh); | |
4340 | return ret; | |
ccd979bd MF |
4341 | } |
4342 | ||
dcd0538f MF |
4343 | /* |
4344 | * Insert an extent into an inode btree. | |
4345 | * | |
4346 | * The caller needs to update fe->i_clusters | |
4347 | */ | |
f56654c4 TM |
4348 | static int ocfs2_insert_extent(struct ocfs2_super *osb, |
4349 | handle_t *handle, | |
4350 | struct inode *inode, | |
4351 | struct buffer_head *root_bh, | |
4352 | u32 cpos, | |
4353 | u64 start_blk, | |
4354 | u32 new_clusters, | |
4355 | u8 flags, | |
4356 | struct ocfs2_alloc_context *meta_ac, | |
4357 | struct ocfs2_extent_tree *et) | |
ccd979bd | 4358 | { |
c3afcbb3 | 4359 | int status; |
c77534f6 | 4360 | int uninitialized_var(free_records); |
ccd979bd | 4361 | struct buffer_head *last_eb_bh = NULL; |
dcd0538f MF |
4362 | struct ocfs2_insert_type insert = {0, }; |
4363 | struct ocfs2_extent_rec rec; | |
4364 | ||
1afc32b9 MF |
4365 | BUG_ON(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL); |
4366 | ||
dcd0538f MF |
4367 | mlog(0, "add %u clusters at position %u to inode %llu\n", |
4368 | new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4369 | ||
4370 | mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) && | |
4371 | (OCFS2_I(inode)->ip_clusters != cpos), | |
4372 | "Device %s, asking for sparse allocation: inode %llu, " | |
4373 | "cpos %u, clusters %u\n", | |
4374 | osb->dev_str, | |
4375 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, | |
4376 | OCFS2_I(inode)->ip_clusters); | |
4377 | ||
e48edee2 | 4378 | memset(&rec, 0, sizeof(rec)); |
dcd0538f MF |
4379 | rec.e_cpos = cpu_to_le32(cpos); |
4380 | rec.e_blkno = cpu_to_le64(start_blk); | |
e48edee2 | 4381 | rec.e_leaf_clusters = cpu_to_le16(new_clusters); |
2ae99a60 | 4382 | rec.e_flags = flags; |
dcd0538f | 4383 | |
e7d4cb6b | 4384 | status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec, |
c77534f6 | 4385 | &free_records, &insert); |
dcd0538f MF |
4386 | if (status < 0) { |
4387 | mlog_errno(status); | |
4388 | goto bail; | |
ccd979bd MF |
4389 | } |
4390 | ||
dcd0538f MF |
4391 | mlog(0, "Insert.appending: %u, Insert.Contig: %u, " |
4392 | "Insert.contig_index: %d, Insert.free_records: %d, " | |
4393 | "Insert.tree_depth: %d\n", | |
4394 | insert.ins_appending, insert.ins_contig, insert.ins_contig_index, | |
c77534f6 | 4395 | free_records, insert.ins_tree_depth); |
ccd979bd | 4396 | |
c77534f6 | 4397 | if (insert.ins_contig == CONTIG_NONE && free_records == 0) { |
e7d4cb6b | 4398 | status = ocfs2_grow_tree(inode, handle, et, |
328d5752 | 4399 | &insert.ins_tree_depth, &last_eb_bh, |
c3afcbb3 MF |
4400 | meta_ac); |
4401 | if (status) { | |
ccd979bd MF |
4402 | mlog_errno(status); |
4403 | goto bail; | |
4404 | } | |
ccd979bd MF |
4405 | } |
4406 | ||
dcd0538f | 4407 | /* Finally, we can add clusters. This might rotate the tree for us. */ |
e7d4cb6b | 4408 | status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert); |
ccd979bd MF |
4409 | if (status < 0) |
4410 | mlog_errno(status); | |
ce1d9ea6 | 4411 | else if (et->et_type == OCFS2_DINODE_EXTENT) |
83418978 | 4412 | ocfs2_extent_map_insert_rec(inode, &rec); |
ccd979bd MF |
4413 | |
4414 | bail: | |
ccd979bd MF |
4415 | if (last_eb_bh) |
4416 | brelse(last_eb_bh); | |
4417 | ||
f56654c4 TM |
4418 | mlog_exit(status); |
4419 | return status; | |
4420 | } | |
4421 | ||
4422 | int ocfs2_dinode_insert_extent(struct ocfs2_super *osb, | |
4423 | handle_t *handle, | |
4424 | struct inode *inode, | |
4425 | struct buffer_head *root_bh, | |
4426 | u32 cpos, | |
4427 | u64 start_blk, | |
4428 | u32 new_clusters, | |
4429 | u8 flags, | |
4430 | struct ocfs2_alloc_context *meta_ac) | |
4431 | { | |
4432 | int status; | |
dc0ce61a | 4433 | struct ocfs2_extent_tree et; |
f56654c4 | 4434 | |
dc0ce61a JB |
4435 | ocfs2_get_extent_tree(&et, inode, root_bh, OCFS2_DINODE_EXTENT, |
4436 | NULL); | |
f56654c4 TM |
4437 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, |
4438 | cpos, start_blk, new_clusters, | |
dc0ce61a JB |
4439 | flags, meta_ac, &et); |
4440 | ocfs2_put_extent_tree(&et); | |
f56654c4 | 4441 | |
f56654c4 TM |
4442 | return status; |
4443 | } | |
4444 | ||
4445 | int ocfs2_xattr_value_insert_extent(struct ocfs2_super *osb, | |
4446 | handle_t *handle, | |
4447 | struct inode *inode, | |
4448 | struct buffer_head *root_bh, | |
4449 | u32 cpos, | |
4450 | u64 start_blk, | |
4451 | u32 new_clusters, | |
4452 | u8 flags, | |
4453 | struct ocfs2_alloc_context *meta_ac, | |
ea5efa15 | 4454 | void *obj) |
f56654c4 TM |
4455 | { |
4456 | int status; | |
dc0ce61a | 4457 | struct ocfs2_extent_tree et; |
f56654c4 | 4458 | |
dc0ce61a | 4459 | ocfs2_get_extent_tree(&et, inode, root_bh, |
ea5efa15 | 4460 | OCFS2_XATTR_VALUE_EXTENT, obj); |
f56654c4 TM |
4461 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, |
4462 | cpos, start_blk, new_clusters, | |
dc0ce61a JB |
4463 | flags, meta_ac, &et); |
4464 | ocfs2_put_extent_tree(&et); | |
f56654c4 | 4465 | |
ccd979bd MF |
4466 | return status; |
4467 | } | |
4468 | ||
ba492615 TM |
4469 | int ocfs2_xattr_tree_insert_extent(struct ocfs2_super *osb, |
4470 | handle_t *handle, | |
4471 | struct inode *inode, | |
4472 | struct buffer_head *root_bh, | |
4473 | u32 cpos, | |
4474 | u64 start_blk, | |
4475 | u32 new_clusters, | |
4476 | u8 flags, | |
4477 | struct ocfs2_alloc_context *meta_ac) | |
4478 | { | |
4479 | int status; | |
dc0ce61a | 4480 | struct ocfs2_extent_tree et; |
ba492615 | 4481 | |
dc0ce61a JB |
4482 | ocfs2_get_extent_tree(&et, inode, root_bh, OCFS2_XATTR_TREE_EXTENT, |
4483 | NULL); | |
ba492615 TM |
4484 | status = ocfs2_insert_extent(osb, handle, inode, root_bh, |
4485 | cpos, start_blk, new_clusters, | |
dc0ce61a JB |
4486 | flags, meta_ac, &et); |
4487 | ocfs2_put_extent_tree(&et); | |
ba492615 | 4488 | |
ba492615 TM |
4489 | return status; |
4490 | } | |
4491 | ||
0eb8d47e TM |
4492 | /* |
4493 | * Allcate and add clusters into the extent b-tree. | |
4494 | * The new clusters(clusters_to_add) will be inserted at logical_offset. | |
4495 | * The extent b-tree's root is root_el and it should be in root_bh, and | |
4496 | * it is not limited to the file storage. Any extent tree can use this | |
4497 | * function if it implements the proper ocfs2_extent_tree. | |
4498 | */ | |
4499 | int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb, | |
4500 | struct inode *inode, | |
4501 | u32 *logical_offset, | |
4502 | u32 clusters_to_add, | |
4503 | int mark_unwritten, | |
4504 | struct buffer_head *root_bh, | |
4505 | struct ocfs2_extent_list *root_el, | |
4506 | handle_t *handle, | |
4507 | struct ocfs2_alloc_context *data_ac, | |
4508 | struct ocfs2_alloc_context *meta_ac, | |
4509 | enum ocfs2_alloc_restarted *reason_ret, | |
f56654c4 | 4510 | enum ocfs2_extent_tree_type type, |
ea5efa15 | 4511 | void *obj) |
0eb8d47e TM |
4512 | { |
4513 | int status = 0; | |
4514 | int free_extents; | |
4515 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
4516 | u32 bit_off, num_bits; | |
4517 | u64 block; | |
4518 | u8 flags = 0; | |
4519 | ||
4520 | BUG_ON(!clusters_to_add); | |
4521 | ||
4522 | if (mark_unwritten) | |
4523 | flags = OCFS2_EXT_UNWRITTEN; | |
4524 | ||
f56654c4 | 4525 | free_extents = ocfs2_num_free_extents(osb, inode, root_bh, type, |
ea5efa15 | 4526 | obj); |
0eb8d47e TM |
4527 | if (free_extents < 0) { |
4528 | status = free_extents; | |
4529 | mlog_errno(status); | |
4530 | goto leave; | |
4531 | } | |
4532 | ||
4533 | /* there are two cases which could cause us to EAGAIN in the | |
4534 | * we-need-more-metadata case: | |
4535 | * 1) we haven't reserved *any* | |
4536 | * 2) we are so fragmented, we've needed to add metadata too | |
4537 | * many times. */ | |
4538 | if (!free_extents && !meta_ac) { | |
4539 | mlog(0, "we haven't reserved any metadata!\n"); | |
4540 | status = -EAGAIN; | |
4541 | reason = RESTART_META; | |
4542 | goto leave; | |
4543 | } else if ((!free_extents) | |
4544 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
4545 | < ocfs2_extend_meta_needed(root_el))) { | |
4546 | mlog(0, "filesystem is really fragmented...\n"); | |
4547 | status = -EAGAIN; | |
4548 | reason = RESTART_META; | |
4549 | goto leave; | |
4550 | } | |
4551 | ||
4552 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
4553 | clusters_to_add, &bit_off, &num_bits); | |
4554 | if (status < 0) { | |
4555 | if (status != -ENOSPC) | |
4556 | mlog_errno(status); | |
4557 | goto leave; | |
4558 | } | |
4559 | ||
4560 | BUG_ON(num_bits > clusters_to_add); | |
4561 | ||
4562 | /* reserve our write early -- insert_extent may update the inode */ | |
4563 | status = ocfs2_journal_access(handle, inode, root_bh, | |
4564 | OCFS2_JOURNAL_ACCESS_WRITE); | |
4565 | if (status < 0) { | |
4566 | mlog_errno(status); | |
4567 | goto leave; | |
4568 | } | |
4569 | ||
4570 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
4571 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", | |
4572 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
f56654c4 TM |
4573 | if (type == OCFS2_DINODE_EXTENT) |
4574 | status = ocfs2_dinode_insert_extent(osb, handle, inode, root_bh, | |
4575 | *logical_offset, block, | |
4576 | num_bits, flags, meta_ac); | |
ba492615 TM |
4577 | else if (type == OCFS2_XATTR_TREE_EXTENT) |
4578 | status = ocfs2_xattr_tree_insert_extent(osb, handle, | |
4579 | inode, root_bh, | |
4580 | *logical_offset, | |
4581 | block, num_bits, flags, | |
4582 | meta_ac); | |
f56654c4 TM |
4583 | else |
4584 | status = ocfs2_xattr_value_insert_extent(osb, handle, | |
4585 | inode, root_bh, | |
4586 | *logical_offset, | |
4587 | block, num_bits, flags, | |
ea5efa15 | 4588 | meta_ac, obj); |
0eb8d47e TM |
4589 | if (status < 0) { |
4590 | mlog_errno(status); | |
4591 | goto leave; | |
4592 | } | |
4593 | ||
4594 | status = ocfs2_journal_dirty(handle, root_bh); | |
4595 | if (status < 0) { | |
4596 | mlog_errno(status); | |
4597 | goto leave; | |
4598 | } | |
4599 | ||
4600 | clusters_to_add -= num_bits; | |
4601 | *logical_offset += num_bits; | |
4602 | ||
4603 | if (clusters_to_add) { | |
4604 | mlog(0, "need to alloc once more, wanted = %u\n", | |
4605 | clusters_to_add); | |
4606 | status = -EAGAIN; | |
4607 | reason = RESTART_TRANS; | |
4608 | } | |
4609 | ||
4610 | leave: | |
4611 | mlog_exit(status); | |
4612 | if (reason_ret) | |
4613 | *reason_ret = reason; | |
4614 | return status; | |
4615 | } | |
4616 | ||
328d5752 MF |
4617 | static void ocfs2_make_right_split_rec(struct super_block *sb, |
4618 | struct ocfs2_extent_rec *split_rec, | |
4619 | u32 cpos, | |
4620 | struct ocfs2_extent_rec *rec) | |
4621 | { | |
4622 | u32 rec_cpos = le32_to_cpu(rec->e_cpos); | |
4623 | u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters); | |
4624 | ||
4625 | memset(split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4626 | ||
4627 | split_rec->e_cpos = cpu_to_le32(cpos); | |
4628 | split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos); | |
4629 | ||
4630 | split_rec->e_blkno = rec->e_blkno; | |
4631 | le64_add_cpu(&split_rec->e_blkno, | |
4632 | ocfs2_clusters_to_blocks(sb, cpos - rec_cpos)); | |
4633 | ||
4634 | split_rec->e_flags = rec->e_flags; | |
4635 | } | |
4636 | ||
4637 | static int ocfs2_split_and_insert(struct inode *inode, | |
4638 | handle_t *handle, | |
4639 | struct ocfs2_path *path, | |
e7d4cb6b | 4640 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4641 | struct buffer_head **last_eb_bh, |
4642 | int split_index, | |
4643 | struct ocfs2_extent_rec *orig_split_rec, | |
4644 | struct ocfs2_alloc_context *meta_ac) | |
4645 | { | |
4646 | int ret = 0, depth; | |
4647 | unsigned int insert_range, rec_range, do_leftright = 0; | |
4648 | struct ocfs2_extent_rec tmprec; | |
4649 | struct ocfs2_extent_list *rightmost_el; | |
4650 | struct ocfs2_extent_rec rec; | |
4651 | struct ocfs2_extent_rec split_rec = *orig_split_rec; | |
4652 | struct ocfs2_insert_type insert; | |
4653 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4654 | |
4655 | leftright: | |
4656 | /* | |
4657 | * Store a copy of the record on the stack - it might move | |
4658 | * around as the tree is manipulated below. | |
4659 | */ | |
4660 | rec = path_leaf_el(path)->l_recs[split_index]; | |
4661 | ||
ce1d9ea6 | 4662 | rightmost_el = et->et_root_el; |
328d5752 MF |
4663 | |
4664 | depth = le16_to_cpu(rightmost_el->l_tree_depth); | |
4665 | if (depth) { | |
4666 | BUG_ON(!(*last_eb_bh)); | |
4667 | eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data; | |
4668 | rightmost_el = &eb->h_list; | |
4669 | } | |
4670 | ||
4671 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4672 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b TM |
4673 | ret = ocfs2_grow_tree(inode, handle, et, |
4674 | &depth, last_eb_bh, meta_ac); | |
328d5752 MF |
4675 | if (ret) { |
4676 | mlog_errno(ret); | |
4677 | goto out; | |
4678 | } | |
328d5752 MF |
4679 | } |
4680 | ||
4681 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4682 | insert.ins_appending = APPEND_NONE; | |
4683 | insert.ins_contig = CONTIG_NONE; | |
328d5752 MF |
4684 | insert.ins_tree_depth = depth; |
4685 | ||
4686 | insert_range = le32_to_cpu(split_rec.e_cpos) + | |
4687 | le16_to_cpu(split_rec.e_leaf_clusters); | |
4688 | rec_range = le32_to_cpu(rec.e_cpos) + | |
4689 | le16_to_cpu(rec.e_leaf_clusters); | |
4690 | ||
4691 | if (split_rec.e_cpos == rec.e_cpos) { | |
4692 | insert.ins_split = SPLIT_LEFT; | |
4693 | } else if (insert_range == rec_range) { | |
4694 | insert.ins_split = SPLIT_RIGHT; | |
4695 | } else { | |
4696 | /* | |
4697 | * Left/right split. We fake this as a right split | |
4698 | * first and then make a second pass as a left split. | |
4699 | */ | |
4700 | insert.ins_split = SPLIT_RIGHT; | |
4701 | ||
4702 | ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range, | |
4703 | &rec); | |
4704 | ||
4705 | split_rec = tmprec; | |
4706 | ||
4707 | BUG_ON(do_leftright); | |
4708 | do_leftright = 1; | |
4709 | } | |
4710 | ||
e7d4cb6b | 4711 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
328d5752 MF |
4712 | if (ret) { |
4713 | mlog_errno(ret); | |
4714 | goto out; | |
4715 | } | |
4716 | ||
4717 | if (do_leftright == 1) { | |
4718 | u32 cpos; | |
4719 | struct ocfs2_extent_list *el; | |
4720 | ||
4721 | do_leftright++; | |
4722 | split_rec = *orig_split_rec; | |
4723 | ||
4724 | ocfs2_reinit_path(path, 1); | |
4725 | ||
4726 | cpos = le32_to_cpu(split_rec.e_cpos); | |
4727 | ret = ocfs2_find_path(inode, path, cpos); | |
4728 | if (ret) { | |
4729 | mlog_errno(ret); | |
4730 | goto out; | |
4731 | } | |
4732 | ||
4733 | el = path_leaf_el(path); | |
4734 | split_index = ocfs2_search_extent_list(el, cpos); | |
4735 | goto leftright; | |
4736 | } | |
4737 | out: | |
4738 | ||
4739 | return ret; | |
4740 | } | |
4741 | ||
4742 | /* | |
4743 | * Mark part or all of the extent record at split_index in the leaf | |
4744 | * pointed to by path as written. This removes the unwritten | |
4745 | * extent flag. | |
4746 | * | |
4747 | * Care is taken to handle contiguousness so as to not grow the tree. | |
4748 | * | |
4749 | * meta_ac is not strictly necessary - we only truly need it if growth | |
4750 | * of the tree is required. All other cases will degrade into a less | |
4751 | * optimal tree layout. | |
4752 | * | |
e7d4cb6b TM |
4753 | * last_eb_bh should be the rightmost leaf block for any extent |
4754 | * btree. Since a split may grow the tree or a merge might shrink it, | |
4755 | * the caller cannot trust the contents of that buffer after this call. | |
328d5752 MF |
4756 | * |
4757 | * This code is optimized for readability - several passes might be | |
4758 | * made over certain portions of the tree. All of those blocks will | |
4759 | * have been brought into cache (and pinned via the journal), so the | |
4760 | * extra overhead is not expressed in terms of disk reads. | |
4761 | */ | |
4762 | static int __ocfs2_mark_extent_written(struct inode *inode, | |
e7d4cb6b | 4763 | struct ocfs2_extent_tree *et, |
328d5752 MF |
4764 | handle_t *handle, |
4765 | struct ocfs2_path *path, | |
4766 | int split_index, | |
4767 | struct ocfs2_extent_rec *split_rec, | |
4768 | struct ocfs2_alloc_context *meta_ac, | |
4769 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
4770 | { | |
4771 | int ret = 0; | |
4772 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
e8aed345 | 4773 | struct buffer_head *last_eb_bh = NULL; |
328d5752 MF |
4774 | struct ocfs2_extent_rec *rec = &el->l_recs[split_index]; |
4775 | struct ocfs2_merge_ctxt ctxt; | |
4776 | struct ocfs2_extent_list *rightmost_el; | |
4777 | ||
3cf0c507 | 4778 | if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) { |
328d5752 MF |
4779 | ret = -EIO; |
4780 | mlog_errno(ret); | |
4781 | goto out; | |
4782 | } | |
4783 | ||
4784 | if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) || | |
4785 | ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) < | |
4786 | (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) { | |
4787 | ret = -EIO; | |
4788 | mlog_errno(ret); | |
4789 | goto out; | |
4790 | } | |
4791 | ||
ad5a4d70 | 4792 | ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, path, el, |
328d5752 MF |
4793 | split_index, |
4794 | split_rec); | |
4795 | ||
4796 | /* | |
4797 | * The core merge / split code wants to know how much room is | |
4798 | * left in this inodes allocation tree, so we pass the | |
4799 | * rightmost extent list. | |
4800 | */ | |
4801 | if (path->p_tree_depth) { | |
4802 | struct ocfs2_extent_block *eb; | |
328d5752 MF |
4803 | |
4804 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4805 | ocfs2_et_get_last_eb_blk(et), |
328d5752 MF |
4806 | &last_eb_bh, OCFS2_BH_CACHED, inode); |
4807 | if (ret) { | |
4808 | mlog_exit(ret); | |
4809 | goto out; | |
4810 | } | |
4811 | ||
4812 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4813 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
4814 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
4815 | ret = -EROFS; | |
4816 | goto out; | |
4817 | } | |
4818 | ||
4819 | rightmost_el = &eb->h_list; | |
4820 | } else | |
4821 | rightmost_el = path_root_el(path); | |
4822 | ||
328d5752 MF |
4823 | if (rec->e_cpos == split_rec->e_cpos && |
4824 | rec->e_leaf_clusters == split_rec->e_leaf_clusters) | |
4825 | ctxt.c_split_covers_rec = 1; | |
4826 | else | |
4827 | ctxt.c_split_covers_rec = 0; | |
4828 | ||
4829 | ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]); | |
4830 | ||
015452b1 MF |
4831 | mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n", |
4832 | split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent, | |
4833 | ctxt.c_split_covers_rec); | |
328d5752 MF |
4834 | |
4835 | if (ctxt.c_contig_type == CONTIG_NONE) { | |
4836 | if (ctxt.c_split_covers_rec) | |
4837 | el->l_recs[split_index] = *split_rec; | |
4838 | else | |
e7d4cb6b | 4839 | ret = ocfs2_split_and_insert(inode, handle, path, et, |
328d5752 MF |
4840 | &last_eb_bh, split_index, |
4841 | split_rec, meta_ac); | |
4842 | if (ret) | |
4843 | mlog_errno(ret); | |
4844 | } else { | |
4845 | ret = ocfs2_try_to_merge_extent(inode, handle, path, | |
4846 | split_index, split_rec, | |
e7d4cb6b | 4847 | dealloc, &ctxt, et); |
328d5752 MF |
4848 | if (ret) |
4849 | mlog_errno(ret); | |
4850 | } | |
4851 | ||
328d5752 MF |
4852 | out: |
4853 | brelse(last_eb_bh); | |
4854 | return ret; | |
4855 | } | |
4856 | ||
4857 | /* | |
4858 | * Mark the already-existing extent at cpos as written for len clusters. | |
4859 | * | |
4860 | * If the existing extent is larger than the request, initiate a | |
4861 | * split. An attempt will be made at merging with adjacent extents. | |
4862 | * | |
4863 | * The caller is responsible for passing down meta_ac if we'll need it. | |
4864 | */ | |
e7d4cb6b | 4865 | int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh, |
328d5752 MF |
4866 | handle_t *handle, u32 cpos, u32 len, u32 phys, |
4867 | struct ocfs2_alloc_context *meta_ac, | |
e7d4cb6b | 4868 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
f56654c4 | 4869 | enum ocfs2_extent_tree_type et_type, |
ea5efa15 | 4870 | void *obj) |
328d5752 MF |
4871 | { |
4872 | int ret, index; | |
4873 | u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys); | |
4874 | struct ocfs2_extent_rec split_rec; | |
4875 | struct ocfs2_path *left_path = NULL; | |
4876 | struct ocfs2_extent_list *el; | |
dc0ce61a | 4877 | struct ocfs2_extent_tree et; |
328d5752 MF |
4878 | |
4879 | mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n", | |
4880 | inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno); | |
4881 | ||
ea5efa15 | 4882 | ocfs2_get_extent_tree(&et, inode, root_bh, et_type, obj); |
dc0ce61a | 4883 | |
328d5752 MF |
4884 | if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) { |
4885 | ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents " | |
4886 | "that are being written to, but the feature bit " | |
4887 | "is not set in the super block.", | |
4888 | (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
4889 | ret = -EROFS; | |
4890 | goto out; | |
4891 | } | |
4892 | ||
4893 | /* | |
4894 | * XXX: This should be fixed up so that we just re-insert the | |
4895 | * next extent records. | |
4896 | */ | |
e7d4cb6b TM |
4897 | if (et_type == OCFS2_DINODE_EXTENT) |
4898 | ocfs2_extent_map_trunc(inode, 0); | |
328d5752 | 4899 | |
dc0ce61a | 4900 | left_path = ocfs2_new_path(et.et_root_bh, et.et_root_el); |
328d5752 MF |
4901 | if (!left_path) { |
4902 | ret = -ENOMEM; | |
4903 | mlog_errno(ret); | |
4904 | goto out; | |
4905 | } | |
4906 | ||
4907 | ret = ocfs2_find_path(inode, left_path, cpos); | |
4908 | if (ret) { | |
4909 | mlog_errno(ret); | |
4910 | goto out; | |
4911 | } | |
4912 | el = path_leaf_el(left_path); | |
4913 | ||
4914 | index = ocfs2_search_extent_list(el, cpos); | |
4915 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
4916 | ocfs2_error(inode->i_sb, | |
4917 | "Inode %llu has an extent at cpos %u which can no " | |
4918 | "longer be found.\n", | |
4919 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
4920 | ret = -EROFS; | |
4921 | goto out; | |
4922 | } | |
4923 | ||
4924 | memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec)); | |
4925 | split_rec.e_cpos = cpu_to_le32(cpos); | |
4926 | split_rec.e_leaf_clusters = cpu_to_le16(len); | |
4927 | split_rec.e_blkno = cpu_to_le64(start_blkno); | |
4928 | split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags; | |
4929 | split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN; | |
4930 | ||
dc0ce61a | 4931 | ret = __ocfs2_mark_extent_written(inode, &et, handle, left_path, |
e7d4cb6b TM |
4932 | index, &split_rec, meta_ac, |
4933 | dealloc); | |
328d5752 MF |
4934 | if (ret) |
4935 | mlog_errno(ret); | |
4936 | ||
4937 | out: | |
4938 | ocfs2_free_path(left_path); | |
dc0ce61a | 4939 | ocfs2_put_extent_tree(&et); |
328d5752 MF |
4940 | return ret; |
4941 | } | |
4942 | ||
e7d4cb6b | 4943 | static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et, |
d0c7d708 MF |
4944 | handle_t *handle, struct ocfs2_path *path, |
4945 | int index, u32 new_range, | |
4946 | struct ocfs2_alloc_context *meta_ac) | |
4947 | { | |
4948 | int ret, depth, credits = handle->h_buffer_credits; | |
d0c7d708 MF |
4949 | struct buffer_head *last_eb_bh = NULL; |
4950 | struct ocfs2_extent_block *eb; | |
4951 | struct ocfs2_extent_list *rightmost_el, *el; | |
4952 | struct ocfs2_extent_rec split_rec; | |
4953 | struct ocfs2_extent_rec *rec; | |
4954 | struct ocfs2_insert_type insert; | |
4955 | ||
4956 | /* | |
4957 | * Setup the record to split before we grow the tree. | |
4958 | */ | |
4959 | el = path_leaf_el(path); | |
4960 | rec = &el->l_recs[index]; | |
4961 | ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec); | |
4962 | ||
4963 | depth = path->p_tree_depth; | |
4964 | if (depth > 0) { | |
4965 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
35dc0aa3 | 4966 | ocfs2_et_get_last_eb_blk(et), |
d0c7d708 MF |
4967 | &last_eb_bh, OCFS2_BH_CACHED, inode); |
4968 | if (ret < 0) { | |
4969 | mlog_errno(ret); | |
4970 | goto out; | |
4971 | } | |
4972 | ||
4973 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
4974 | rightmost_el = &eb->h_list; | |
4975 | } else | |
4976 | rightmost_el = path_leaf_el(path); | |
4977 | ||
811f933d | 4978 | credits += path->p_tree_depth + |
ce1d9ea6 | 4979 | ocfs2_extend_meta_needed(et->et_root_el); |
d0c7d708 MF |
4980 | ret = ocfs2_extend_trans(handle, credits); |
4981 | if (ret) { | |
4982 | mlog_errno(ret); | |
4983 | goto out; | |
4984 | } | |
4985 | ||
4986 | if (le16_to_cpu(rightmost_el->l_next_free_rec) == | |
4987 | le16_to_cpu(rightmost_el->l_count)) { | |
e7d4cb6b | 4988 | ret = ocfs2_grow_tree(inode, handle, et, &depth, &last_eb_bh, |
d0c7d708 MF |
4989 | meta_ac); |
4990 | if (ret) { | |
4991 | mlog_errno(ret); | |
4992 | goto out; | |
4993 | } | |
d0c7d708 MF |
4994 | } |
4995 | ||
4996 | memset(&insert, 0, sizeof(struct ocfs2_insert_type)); | |
4997 | insert.ins_appending = APPEND_NONE; | |
4998 | insert.ins_contig = CONTIG_NONE; | |
4999 | insert.ins_split = SPLIT_RIGHT; | |
d0c7d708 MF |
5000 | insert.ins_tree_depth = depth; |
5001 | ||
e7d4cb6b | 5002 | ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert); |
d0c7d708 MF |
5003 | if (ret) |
5004 | mlog_errno(ret); | |
5005 | ||
5006 | out: | |
5007 | brelse(last_eb_bh); | |
5008 | return ret; | |
5009 | } | |
5010 | ||
5011 | static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle, | |
5012 | struct ocfs2_path *path, int index, | |
5013 | struct ocfs2_cached_dealloc_ctxt *dealloc, | |
e7d4cb6b TM |
5014 | u32 cpos, u32 len, |
5015 | struct ocfs2_extent_tree *et) | |
d0c7d708 MF |
5016 | { |
5017 | int ret; | |
5018 | u32 left_cpos, rec_range, trunc_range; | |
5019 | int wants_rotate = 0, is_rightmost_tree_rec = 0; | |
5020 | struct super_block *sb = inode->i_sb; | |
5021 | struct ocfs2_path *left_path = NULL; | |
5022 | struct ocfs2_extent_list *el = path_leaf_el(path); | |
5023 | struct ocfs2_extent_rec *rec; | |
5024 | struct ocfs2_extent_block *eb; | |
5025 | ||
5026 | if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) { | |
e7d4cb6b | 5027 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
5028 | if (ret) { |
5029 | mlog_errno(ret); | |
5030 | goto out; | |
5031 | } | |
5032 | ||
5033 | index--; | |
5034 | } | |
5035 | ||
5036 | if (index == (le16_to_cpu(el->l_next_free_rec) - 1) && | |
5037 | path->p_tree_depth) { | |
5038 | /* | |
5039 | * Check whether this is the rightmost tree record. If | |
5040 | * we remove all of this record or part of its right | |
5041 | * edge then an update of the record lengths above it | |
5042 | * will be required. | |
5043 | */ | |
5044 | eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data; | |
5045 | if (eb->h_next_leaf_blk == 0) | |
5046 | is_rightmost_tree_rec = 1; | |
5047 | } | |
5048 | ||
5049 | rec = &el->l_recs[index]; | |
5050 | if (index == 0 && path->p_tree_depth && | |
5051 | le32_to_cpu(rec->e_cpos) == cpos) { | |
5052 | /* | |
5053 | * Changing the leftmost offset (via partial or whole | |
5054 | * record truncate) of an interior (or rightmost) path | |
5055 | * means we have to update the subtree that is formed | |
5056 | * by this leaf and the one to it's left. | |
5057 | * | |
5058 | * There are two cases we can skip: | |
5059 | * 1) Path is the leftmost one in our inode tree. | |
5060 | * 2) The leaf is rightmost and will be empty after | |
5061 | * we remove the extent record - the rotate code | |
5062 | * knows how to update the newly formed edge. | |
5063 | */ | |
5064 | ||
5065 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, | |
5066 | &left_cpos); | |
5067 | if (ret) { | |
5068 | mlog_errno(ret); | |
5069 | goto out; | |
5070 | } | |
5071 | ||
5072 | if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) { | |
5073 | left_path = ocfs2_new_path(path_root_bh(path), | |
5074 | path_root_el(path)); | |
5075 | if (!left_path) { | |
5076 | ret = -ENOMEM; | |
5077 | mlog_errno(ret); | |
5078 | goto out; | |
5079 | } | |
5080 | ||
5081 | ret = ocfs2_find_path(inode, left_path, left_cpos); | |
5082 | if (ret) { | |
5083 | mlog_errno(ret); | |
5084 | goto out; | |
5085 | } | |
5086 | } | |
5087 | } | |
5088 | ||
5089 | ret = ocfs2_extend_rotate_transaction(handle, 0, | |
5090 | handle->h_buffer_credits, | |
5091 | path); | |
5092 | if (ret) { | |
5093 | mlog_errno(ret); | |
5094 | goto out; | |
5095 | } | |
5096 | ||
5097 | ret = ocfs2_journal_access_path(inode, handle, path); | |
5098 | if (ret) { | |
5099 | mlog_errno(ret); | |
5100 | goto out; | |
5101 | } | |
5102 | ||
5103 | ret = ocfs2_journal_access_path(inode, handle, left_path); | |
5104 | if (ret) { | |
5105 | mlog_errno(ret); | |
5106 | goto out; | |
5107 | } | |
5108 | ||
5109 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5110 | trunc_range = cpos + len; | |
5111 | ||
5112 | if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) { | |
5113 | int next_free; | |
5114 | ||
5115 | memset(rec, 0, sizeof(*rec)); | |
5116 | ocfs2_cleanup_merge(el, index); | |
5117 | wants_rotate = 1; | |
5118 | ||
5119 | next_free = le16_to_cpu(el->l_next_free_rec); | |
5120 | if (is_rightmost_tree_rec && next_free > 1) { | |
5121 | /* | |
5122 | * We skip the edge update if this path will | |
5123 | * be deleted by the rotate code. | |
5124 | */ | |
5125 | rec = &el->l_recs[next_free - 1]; | |
5126 | ocfs2_adjust_rightmost_records(inode, handle, path, | |
5127 | rec); | |
5128 | } | |
5129 | } else if (le32_to_cpu(rec->e_cpos) == cpos) { | |
5130 | /* Remove leftmost portion of the record. */ | |
5131 | le32_add_cpu(&rec->e_cpos, len); | |
5132 | le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len)); | |
5133 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5134 | } else if (rec_range == trunc_range) { | |
5135 | /* Remove rightmost portion of the record */ | |
5136 | le16_add_cpu(&rec->e_leaf_clusters, -len); | |
5137 | if (is_rightmost_tree_rec) | |
5138 | ocfs2_adjust_rightmost_records(inode, handle, path, rec); | |
5139 | } else { | |
5140 | /* Caller should have trapped this. */ | |
5141 | mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) " | |
5142 | "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5143 | le32_to_cpu(rec->e_cpos), | |
5144 | le16_to_cpu(rec->e_leaf_clusters), cpos, len); | |
5145 | BUG(); | |
5146 | } | |
5147 | ||
5148 | if (left_path) { | |
5149 | int subtree_index; | |
5150 | ||
5151 | subtree_index = ocfs2_find_subtree_root(inode, left_path, path); | |
5152 | ocfs2_complete_edge_insert(inode, handle, left_path, path, | |
5153 | subtree_index); | |
5154 | } | |
5155 | ||
5156 | ocfs2_journal_dirty(handle, path_leaf_bh(path)); | |
5157 | ||
e7d4cb6b | 5158 | ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et); |
d0c7d708 MF |
5159 | if (ret) { |
5160 | mlog_errno(ret); | |
5161 | goto out; | |
5162 | } | |
5163 | ||
5164 | out: | |
5165 | ocfs2_free_path(left_path); | |
5166 | return ret; | |
5167 | } | |
5168 | ||
e7d4cb6b | 5169 | int ocfs2_remove_extent(struct inode *inode, struct buffer_head *root_bh, |
063c4561 MF |
5170 | u32 cpos, u32 len, handle_t *handle, |
5171 | struct ocfs2_alloc_context *meta_ac, | |
e7d4cb6b | 5172 | struct ocfs2_cached_dealloc_ctxt *dealloc, |
f56654c4 | 5173 | enum ocfs2_extent_tree_type et_type, |
ea5efa15 | 5174 | void *obj) |
d0c7d708 MF |
5175 | { |
5176 | int ret, index; | |
5177 | u32 rec_range, trunc_range; | |
5178 | struct ocfs2_extent_rec *rec; | |
5179 | struct ocfs2_extent_list *el; | |
e7d4cb6b | 5180 | struct ocfs2_path *path = NULL; |
dc0ce61a | 5181 | struct ocfs2_extent_tree et; |
e7d4cb6b | 5182 | |
ea5efa15 | 5183 | ocfs2_get_extent_tree(&et, inode, root_bh, et_type, obj); |
d0c7d708 MF |
5184 | |
5185 | ocfs2_extent_map_trunc(inode, 0); | |
5186 | ||
dc0ce61a | 5187 | path = ocfs2_new_path(et.et_root_bh, et.et_root_el); |
d0c7d708 MF |
5188 | if (!path) { |
5189 | ret = -ENOMEM; | |
5190 | mlog_errno(ret); | |
5191 | goto out; | |
5192 | } | |
5193 | ||
5194 | ret = ocfs2_find_path(inode, path, cpos); | |
5195 | if (ret) { | |
5196 | mlog_errno(ret); | |
5197 | goto out; | |
5198 | } | |
5199 | ||
5200 | el = path_leaf_el(path); | |
5201 | index = ocfs2_search_extent_list(el, cpos); | |
5202 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5203 | ocfs2_error(inode->i_sb, | |
5204 | "Inode %llu has an extent at cpos %u which can no " | |
5205 | "longer be found.\n", | |
5206 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos); | |
5207 | ret = -EROFS; | |
5208 | goto out; | |
5209 | } | |
5210 | ||
5211 | /* | |
5212 | * We have 3 cases of extent removal: | |
5213 | * 1) Range covers the entire extent rec | |
5214 | * 2) Range begins or ends on one edge of the extent rec | |
5215 | * 3) Range is in the middle of the extent rec (no shared edges) | |
5216 | * | |
5217 | * For case 1 we remove the extent rec and left rotate to | |
5218 | * fill the hole. | |
5219 | * | |
5220 | * For case 2 we just shrink the existing extent rec, with a | |
5221 | * tree update if the shrinking edge is also the edge of an | |
5222 | * extent block. | |
5223 | * | |
5224 | * For case 3 we do a right split to turn the extent rec into | |
5225 | * something case 2 can handle. | |
5226 | */ | |
5227 | rec = &el->l_recs[index]; | |
5228 | rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
5229 | trunc_range = cpos + len; | |
5230 | ||
5231 | BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range); | |
5232 | ||
5233 | mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d " | |
5234 | "(cpos %u, len %u)\n", | |
5235 | (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index, | |
5236 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec)); | |
5237 | ||
5238 | if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) { | |
5239 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
dc0ce61a | 5240 | cpos, len, &et); |
d0c7d708 MF |
5241 | if (ret) { |
5242 | mlog_errno(ret); | |
5243 | goto out; | |
5244 | } | |
5245 | } else { | |
dc0ce61a | 5246 | ret = ocfs2_split_tree(inode, &et, handle, path, index, |
d0c7d708 MF |
5247 | trunc_range, meta_ac); |
5248 | if (ret) { | |
5249 | mlog_errno(ret); | |
5250 | goto out; | |
5251 | } | |
5252 | ||
5253 | /* | |
5254 | * The split could have manipulated the tree enough to | |
5255 | * move the record location, so we have to look for it again. | |
5256 | */ | |
5257 | ocfs2_reinit_path(path, 1); | |
5258 | ||
5259 | ret = ocfs2_find_path(inode, path, cpos); | |
5260 | if (ret) { | |
5261 | mlog_errno(ret); | |
5262 | goto out; | |
5263 | } | |
5264 | ||
5265 | el = path_leaf_el(path); | |
5266 | index = ocfs2_search_extent_list(el, cpos); | |
5267 | if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) { | |
5268 | ocfs2_error(inode->i_sb, | |
5269 | "Inode %llu: split at cpos %u lost record.", | |
5270 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5271 | cpos); | |
5272 | ret = -EROFS; | |
5273 | goto out; | |
5274 | } | |
5275 | ||
5276 | /* | |
5277 | * Double check our values here. If anything is fishy, | |
5278 | * it's easier to catch it at the top level. | |
5279 | */ | |
5280 | rec = &el->l_recs[index]; | |
5281 | rec_range = le32_to_cpu(rec->e_cpos) + | |
5282 | ocfs2_rec_clusters(el, rec); | |
5283 | if (rec_range != trunc_range) { | |
5284 | ocfs2_error(inode->i_sb, | |
5285 | "Inode %llu: error after split at cpos %u" | |
5286 | "trunc len %u, existing record is (%u,%u)", | |
5287 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
5288 | cpos, len, le32_to_cpu(rec->e_cpos), | |
5289 | ocfs2_rec_clusters(el, rec)); | |
5290 | ret = -EROFS; | |
5291 | goto out; | |
5292 | } | |
5293 | ||
5294 | ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc, | |
dc0ce61a | 5295 | cpos, len, &et); |
d0c7d708 MF |
5296 | if (ret) { |
5297 | mlog_errno(ret); | |
5298 | goto out; | |
5299 | } | |
5300 | } | |
5301 | ||
5302 | out: | |
5303 | ocfs2_free_path(path); | |
dc0ce61a | 5304 | ocfs2_put_extent_tree(&et); |
d0c7d708 MF |
5305 | return ret; |
5306 | } | |
5307 | ||
063c4561 | 5308 | int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb) |
ccd979bd MF |
5309 | { |
5310 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5311 | struct ocfs2_dinode *di; | |
5312 | struct ocfs2_truncate_log *tl; | |
5313 | ||
5314 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5315 | tl = &di->id2.i_dealloc; | |
5316 | ||
5317 | mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count), | |
5318 | "slot %d, invalid truncate log parameters: used = " | |
5319 | "%u, count = %u\n", osb->slot_num, | |
5320 | le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count)); | |
5321 | return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count); | |
5322 | } | |
5323 | ||
5324 | static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl, | |
5325 | unsigned int new_start) | |
5326 | { | |
5327 | unsigned int tail_index; | |
5328 | unsigned int current_tail; | |
5329 | ||
5330 | /* No records, nothing to coalesce */ | |
5331 | if (!le16_to_cpu(tl->tl_used)) | |
5332 | return 0; | |
5333 | ||
5334 | tail_index = le16_to_cpu(tl->tl_used) - 1; | |
5335 | current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start); | |
5336 | current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters); | |
5337 | ||
5338 | return current_tail == new_start; | |
5339 | } | |
5340 | ||
063c4561 MF |
5341 | int ocfs2_truncate_log_append(struct ocfs2_super *osb, |
5342 | handle_t *handle, | |
5343 | u64 start_blk, | |
5344 | unsigned int num_clusters) | |
ccd979bd MF |
5345 | { |
5346 | int status, index; | |
5347 | unsigned int start_cluster, tl_count; | |
5348 | struct inode *tl_inode = osb->osb_tl_inode; | |
5349 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5350 | struct ocfs2_dinode *di; | |
5351 | struct ocfs2_truncate_log *tl; | |
5352 | ||
b0697053 MF |
5353 | mlog_entry("start_blk = %llu, num_clusters = %u\n", |
5354 | (unsigned long long)start_blk, num_clusters); | |
ccd979bd | 5355 | |
1b1dcc1b | 5356 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5357 | |
5358 | start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk); | |
5359 | ||
5360 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5361 | tl = &di->id2.i_dealloc; | |
5362 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5363 | OCFS2_RO_ON_INVALID_DINODE(osb->sb, di); | |
5364 | status = -EIO; | |
5365 | goto bail; | |
5366 | } | |
5367 | ||
5368 | tl_count = le16_to_cpu(tl->tl_count); | |
5369 | mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) || | |
5370 | tl_count == 0, | |
b0697053 MF |
5371 | "Truncate record count on #%llu invalid " |
5372 | "wanted %u, actual %u\n", | |
5373 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, | |
ccd979bd MF |
5374 | ocfs2_truncate_recs_per_inode(osb->sb), |
5375 | le16_to_cpu(tl->tl_count)); | |
5376 | ||
5377 | /* Caller should have known to flush before calling us. */ | |
5378 | index = le16_to_cpu(tl->tl_used); | |
5379 | if (index >= tl_count) { | |
5380 | status = -ENOSPC; | |
5381 | mlog_errno(status); | |
5382 | goto bail; | |
5383 | } | |
5384 | ||
5385 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5386 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5387 | if (status < 0) { | |
5388 | mlog_errno(status); | |
5389 | goto bail; | |
5390 | } | |
5391 | ||
5392 | mlog(0, "Log truncate of %u clusters starting at cluster %u to " | |
b0697053 MF |
5393 | "%llu (index = %d)\n", num_clusters, start_cluster, |
5394 | (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index); | |
ccd979bd MF |
5395 | |
5396 | if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) { | |
5397 | /* | |
5398 | * Move index back to the record we are coalescing with. | |
5399 | * ocfs2_truncate_log_can_coalesce() guarantees nonzero | |
5400 | */ | |
5401 | index--; | |
5402 | ||
5403 | num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters); | |
5404 | mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n", | |
5405 | index, le32_to_cpu(tl->tl_recs[index].t_start), | |
5406 | num_clusters); | |
5407 | } else { | |
5408 | tl->tl_recs[index].t_start = cpu_to_le32(start_cluster); | |
5409 | tl->tl_used = cpu_to_le16(index + 1); | |
5410 | } | |
5411 | tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters); | |
5412 | ||
5413 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5414 | if (status < 0) { | |
5415 | mlog_errno(status); | |
5416 | goto bail; | |
5417 | } | |
5418 | ||
5419 | bail: | |
5420 | mlog_exit(status); | |
5421 | return status; | |
5422 | } | |
5423 | ||
5424 | static int ocfs2_replay_truncate_records(struct ocfs2_super *osb, | |
1fabe148 | 5425 | handle_t *handle, |
ccd979bd MF |
5426 | struct inode *data_alloc_inode, |
5427 | struct buffer_head *data_alloc_bh) | |
5428 | { | |
5429 | int status = 0; | |
5430 | int i; | |
5431 | unsigned int num_clusters; | |
5432 | u64 start_blk; | |
5433 | struct ocfs2_truncate_rec rec; | |
5434 | struct ocfs2_dinode *di; | |
5435 | struct ocfs2_truncate_log *tl; | |
5436 | struct inode *tl_inode = osb->osb_tl_inode; | |
5437 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5438 | ||
5439 | mlog_entry_void(); | |
5440 | ||
5441 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5442 | tl = &di->id2.i_dealloc; | |
5443 | i = le16_to_cpu(tl->tl_used) - 1; | |
5444 | while (i >= 0) { | |
5445 | /* Caller has given us at least enough credits to | |
5446 | * update the truncate log dinode */ | |
5447 | status = ocfs2_journal_access(handle, tl_inode, tl_bh, | |
5448 | OCFS2_JOURNAL_ACCESS_WRITE); | |
5449 | if (status < 0) { | |
5450 | mlog_errno(status); | |
5451 | goto bail; | |
5452 | } | |
5453 | ||
5454 | tl->tl_used = cpu_to_le16(i); | |
5455 | ||
5456 | status = ocfs2_journal_dirty(handle, tl_bh); | |
5457 | if (status < 0) { | |
5458 | mlog_errno(status); | |
5459 | goto bail; | |
5460 | } | |
5461 | ||
5462 | /* TODO: Perhaps we can calculate the bulk of the | |
5463 | * credits up front rather than extending like | |
5464 | * this. */ | |
5465 | status = ocfs2_extend_trans(handle, | |
5466 | OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC); | |
5467 | if (status < 0) { | |
5468 | mlog_errno(status); | |
5469 | goto bail; | |
5470 | } | |
5471 | ||
5472 | rec = tl->tl_recs[i]; | |
5473 | start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb, | |
5474 | le32_to_cpu(rec.t_start)); | |
5475 | num_clusters = le32_to_cpu(rec.t_clusters); | |
5476 | ||
5477 | /* if start_blk is not set, we ignore the record as | |
5478 | * invalid. */ | |
5479 | if (start_blk) { | |
5480 | mlog(0, "free record %d, start = %u, clusters = %u\n", | |
5481 | i, le32_to_cpu(rec.t_start), num_clusters); | |
5482 | ||
5483 | status = ocfs2_free_clusters(handle, data_alloc_inode, | |
5484 | data_alloc_bh, start_blk, | |
5485 | num_clusters); | |
5486 | if (status < 0) { | |
5487 | mlog_errno(status); | |
5488 | goto bail; | |
5489 | } | |
5490 | } | |
5491 | i--; | |
5492 | } | |
5493 | ||
5494 | bail: | |
5495 | mlog_exit(status); | |
5496 | return status; | |
5497 | } | |
5498 | ||
1b1dcc1b | 5499 | /* Expects you to already be holding tl_inode->i_mutex */ |
063c4561 | 5500 | int __ocfs2_flush_truncate_log(struct ocfs2_super *osb) |
ccd979bd MF |
5501 | { |
5502 | int status; | |
5503 | unsigned int num_to_flush; | |
1fabe148 | 5504 | handle_t *handle; |
ccd979bd MF |
5505 | struct inode *tl_inode = osb->osb_tl_inode; |
5506 | struct inode *data_alloc_inode = NULL; | |
5507 | struct buffer_head *tl_bh = osb->osb_tl_bh; | |
5508 | struct buffer_head *data_alloc_bh = NULL; | |
5509 | struct ocfs2_dinode *di; | |
5510 | struct ocfs2_truncate_log *tl; | |
5511 | ||
5512 | mlog_entry_void(); | |
5513 | ||
1b1dcc1b | 5514 | BUG_ON(mutex_trylock(&tl_inode->i_mutex)); |
ccd979bd MF |
5515 | |
5516 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5517 | tl = &di->id2.i_dealloc; | |
5518 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5519 | OCFS2_RO_ON_INVALID_DINODE(osb->sb, di); | |
5520 | status = -EIO; | |
e08dc8b9 | 5521 | goto out; |
ccd979bd MF |
5522 | } |
5523 | ||
5524 | num_to_flush = le16_to_cpu(tl->tl_used); | |
b0697053 MF |
5525 | mlog(0, "Flush %u records from truncate log #%llu\n", |
5526 | num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno); | |
ccd979bd MF |
5527 | if (!num_to_flush) { |
5528 | status = 0; | |
e08dc8b9 | 5529 | goto out; |
ccd979bd MF |
5530 | } |
5531 | ||
5532 | data_alloc_inode = ocfs2_get_system_file_inode(osb, | |
5533 | GLOBAL_BITMAP_SYSTEM_INODE, | |
5534 | OCFS2_INVALID_SLOT); | |
5535 | if (!data_alloc_inode) { | |
5536 | status = -EINVAL; | |
5537 | mlog(ML_ERROR, "Could not get bitmap inode!\n"); | |
e08dc8b9 | 5538 | goto out; |
ccd979bd MF |
5539 | } |
5540 | ||
e08dc8b9 MF |
5541 | mutex_lock(&data_alloc_inode->i_mutex); |
5542 | ||
e63aecb6 | 5543 | status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1); |
ccd979bd MF |
5544 | if (status < 0) { |
5545 | mlog_errno(status); | |
e08dc8b9 | 5546 | goto out_mutex; |
ccd979bd MF |
5547 | } |
5548 | ||
65eff9cc | 5549 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5550 | if (IS_ERR(handle)) { |
5551 | status = PTR_ERR(handle); | |
ccd979bd | 5552 | mlog_errno(status); |
e08dc8b9 | 5553 | goto out_unlock; |
ccd979bd MF |
5554 | } |
5555 | ||
5556 | status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode, | |
5557 | data_alloc_bh); | |
e08dc8b9 | 5558 | if (status < 0) |
ccd979bd | 5559 | mlog_errno(status); |
ccd979bd | 5560 | |
02dc1af4 | 5561 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 5562 | |
e08dc8b9 MF |
5563 | out_unlock: |
5564 | brelse(data_alloc_bh); | |
e63aecb6 | 5565 | ocfs2_inode_unlock(data_alloc_inode, 1); |
ccd979bd | 5566 | |
e08dc8b9 MF |
5567 | out_mutex: |
5568 | mutex_unlock(&data_alloc_inode->i_mutex); | |
5569 | iput(data_alloc_inode); | |
ccd979bd | 5570 | |
e08dc8b9 | 5571 | out: |
ccd979bd MF |
5572 | mlog_exit(status); |
5573 | return status; | |
5574 | } | |
5575 | ||
5576 | int ocfs2_flush_truncate_log(struct ocfs2_super *osb) | |
5577 | { | |
5578 | int status; | |
5579 | struct inode *tl_inode = osb->osb_tl_inode; | |
5580 | ||
1b1dcc1b | 5581 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd | 5582 | status = __ocfs2_flush_truncate_log(osb); |
1b1dcc1b | 5583 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5584 | |
5585 | return status; | |
5586 | } | |
5587 | ||
c4028958 | 5588 | static void ocfs2_truncate_log_worker(struct work_struct *work) |
ccd979bd MF |
5589 | { |
5590 | int status; | |
c4028958 DH |
5591 | struct ocfs2_super *osb = |
5592 | container_of(work, struct ocfs2_super, | |
5593 | osb_truncate_log_wq.work); | |
ccd979bd MF |
5594 | |
5595 | mlog_entry_void(); | |
5596 | ||
5597 | status = ocfs2_flush_truncate_log(osb); | |
5598 | if (status < 0) | |
5599 | mlog_errno(status); | |
4d0ddb2c TM |
5600 | else |
5601 | ocfs2_init_inode_steal_slot(osb); | |
ccd979bd MF |
5602 | |
5603 | mlog_exit(status); | |
5604 | } | |
5605 | ||
5606 | #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ) | |
5607 | void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb, | |
5608 | int cancel) | |
5609 | { | |
5610 | if (osb->osb_tl_inode) { | |
5611 | /* We want to push off log flushes while truncates are | |
5612 | * still running. */ | |
5613 | if (cancel) | |
5614 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5615 | ||
5616 | queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq, | |
5617 | OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL); | |
5618 | } | |
5619 | } | |
5620 | ||
5621 | static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb, | |
5622 | int slot_num, | |
5623 | struct inode **tl_inode, | |
5624 | struct buffer_head **tl_bh) | |
5625 | { | |
5626 | int status; | |
5627 | struct inode *inode = NULL; | |
5628 | struct buffer_head *bh = NULL; | |
5629 | ||
5630 | inode = ocfs2_get_system_file_inode(osb, | |
5631 | TRUNCATE_LOG_SYSTEM_INODE, | |
5632 | slot_num); | |
5633 | if (!inode) { | |
5634 | status = -EINVAL; | |
5635 | mlog(ML_ERROR, "Could not get load truncate log inode!\n"); | |
5636 | goto bail; | |
5637 | } | |
5638 | ||
5639 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, | |
5640 | OCFS2_BH_CACHED, inode); | |
5641 | if (status < 0) { | |
5642 | iput(inode); | |
5643 | mlog_errno(status); | |
5644 | goto bail; | |
5645 | } | |
5646 | ||
5647 | *tl_inode = inode; | |
5648 | *tl_bh = bh; | |
5649 | bail: | |
5650 | mlog_exit(status); | |
5651 | return status; | |
5652 | } | |
5653 | ||
5654 | /* called during the 1st stage of node recovery. we stamp a clean | |
5655 | * truncate log and pass back a copy for processing later. if the | |
5656 | * truncate log does not require processing, a *tl_copy is set to | |
5657 | * NULL. */ | |
5658 | int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb, | |
5659 | int slot_num, | |
5660 | struct ocfs2_dinode **tl_copy) | |
5661 | { | |
5662 | int status; | |
5663 | struct inode *tl_inode = NULL; | |
5664 | struct buffer_head *tl_bh = NULL; | |
5665 | struct ocfs2_dinode *di; | |
5666 | struct ocfs2_truncate_log *tl; | |
5667 | ||
5668 | *tl_copy = NULL; | |
5669 | ||
5670 | mlog(0, "recover truncate log from slot %d\n", slot_num); | |
5671 | ||
5672 | status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh); | |
5673 | if (status < 0) { | |
5674 | mlog_errno(status); | |
5675 | goto bail; | |
5676 | } | |
5677 | ||
5678 | di = (struct ocfs2_dinode *) tl_bh->b_data; | |
5679 | tl = &di->id2.i_dealloc; | |
5680 | if (!OCFS2_IS_VALID_DINODE(di)) { | |
5681 | OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di); | |
5682 | status = -EIO; | |
5683 | goto bail; | |
5684 | } | |
5685 | ||
5686 | if (le16_to_cpu(tl->tl_used)) { | |
5687 | mlog(0, "We'll have %u logs to recover\n", | |
5688 | le16_to_cpu(tl->tl_used)); | |
5689 | ||
5690 | *tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL); | |
5691 | if (!(*tl_copy)) { | |
5692 | status = -ENOMEM; | |
5693 | mlog_errno(status); | |
5694 | goto bail; | |
5695 | } | |
5696 | ||
5697 | /* Assuming the write-out below goes well, this copy | |
5698 | * will be passed back to recovery for processing. */ | |
5699 | memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size); | |
5700 | ||
5701 | /* All we need to do to clear the truncate log is set | |
5702 | * tl_used. */ | |
5703 | tl->tl_used = 0; | |
5704 | ||
5705 | status = ocfs2_write_block(osb, tl_bh, tl_inode); | |
5706 | if (status < 0) { | |
5707 | mlog_errno(status); | |
5708 | goto bail; | |
5709 | } | |
5710 | } | |
5711 | ||
5712 | bail: | |
5713 | if (tl_inode) | |
5714 | iput(tl_inode); | |
5715 | if (tl_bh) | |
5716 | brelse(tl_bh); | |
5717 | ||
5718 | if (status < 0 && (*tl_copy)) { | |
5719 | kfree(*tl_copy); | |
5720 | *tl_copy = NULL; | |
5721 | } | |
5722 | ||
5723 | mlog_exit(status); | |
5724 | return status; | |
5725 | } | |
5726 | ||
5727 | int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb, | |
5728 | struct ocfs2_dinode *tl_copy) | |
5729 | { | |
5730 | int status = 0; | |
5731 | int i; | |
5732 | unsigned int clusters, num_recs, start_cluster; | |
5733 | u64 start_blk; | |
1fabe148 | 5734 | handle_t *handle; |
ccd979bd MF |
5735 | struct inode *tl_inode = osb->osb_tl_inode; |
5736 | struct ocfs2_truncate_log *tl; | |
5737 | ||
5738 | mlog_entry_void(); | |
5739 | ||
5740 | if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) { | |
5741 | mlog(ML_ERROR, "Asked to recover my own truncate log!\n"); | |
5742 | return -EINVAL; | |
5743 | } | |
5744 | ||
5745 | tl = &tl_copy->id2.i_dealloc; | |
5746 | num_recs = le16_to_cpu(tl->tl_used); | |
b0697053 | 5747 | mlog(0, "cleanup %u records from %llu\n", num_recs, |
1ca1a111 | 5748 | (unsigned long long)le64_to_cpu(tl_copy->i_blkno)); |
ccd979bd | 5749 | |
1b1dcc1b | 5750 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
5751 | for(i = 0; i < num_recs; i++) { |
5752 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
5753 | status = __ocfs2_flush_truncate_log(osb); | |
5754 | if (status < 0) { | |
5755 | mlog_errno(status); | |
5756 | goto bail_up; | |
5757 | } | |
5758 | } | |
5759 | ||
65eff9cc | 5760 | handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE); |
ccd979bd MF |
5761 | if (IS_ERR(handle)) { |
5762 | status = PTR_ERR(handle); | |
5763 | mlog_errno(status); | |
5764 | goto bail_up; | |
5765 | } | |
5766 | ||
5767 | clusters = le32_to_cpu(tl->tl_recs[i].t_clusters); | |
5768 | start_cluster = le32_to_cpu(tl->tl_recs[i].t_start); | |
5769 | start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster); | |
5770 | ||
5771 | status = ocfs2_truncate_log_append(osb, handle, | |
5772 | start_blk, clusters); | |
02dc1af4 | 5773 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
5774 | if (status < 0) { |
5775 | mlog_errno(status); | |
5776 | goto bail_up; | |
5777 | } | |
5778 | } | |
5779 | ||
5780 | bail_up: | |
1b1dcc1b | 5781 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
5782 | |
5783 | mlog_exit(status); | |
5784 | return status; | |
5785 | } | |
5786 | ||
5787 | void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb) | |
5788 | { | |
5789 | int status; | |
5790 | struct inode *tl_inode = osb->osb_tl_inode; | |
5791 | ||
5792 | mlog_entry_void(); | |
5793 | ||
5794 | if (tl_inode) { | |
5795 | cancel_delayed_work(&osb->osb_truncate_log_wq); | |
5796 | flush_workqueue(ocfs2_wq); | |
5797 | ||
5798 | status = ocfs2_flush_truncate_log(osb); | |
5799 | if (status < 0) | |
5800 | mlog_errno(status); | |
5801 | ||
5802 | brelse(osb->osb_tl_bh); | |
5803 | iput(osb->osb_tl_inode); | |
5804 | } | |
5805 | ||
5806 | mlog_exit_void(); | |
5807 | } | |
5808 | ||
5809 | int ocfs2_truncate_log_init(struct ocfs2_super *osb) | |
5810 | { | |
5811 | int status; | |
5812 | struct inode *tl_inode = NULL; | |
5813 | struct buffer_head *tl_bh = NULL; | |
5814 | ||
5815 | mlog_entry_void(); | |
5816 | ||
5817 | status = ocfs2_get_truncate_log_info(osb, | |
5818 | osb->slot_num, | |
5819 | &tl_inode, | |
5820 | &tl_bh); | |
5821 | if (status < 0) | |
5822 | mlog_errno(status); | |
5823 | ||
5824 | /* ocfs2_truncate_log_shutdown keys on the existence of | |
5825 | * osb->osb_tl_inode so we don't set any of the osb variables | |
5826 | * until we're sure all is well. */ | |
c4028958 DH |
5827 | INIT_DELAYED_WORK(&osb->osb_truncate_log_wq, |
5828 | ocfs2_truncate_log_worker); | |
ccd979bd MF |
5829 | osb->osb_tl_bh = tl_bh; |
5830 | osb->osb_tl_inode = tl_inode; | |
5831 | ||
5832 | mlog_exit(status); | |
5833 | return status; | |
5834 | } | |
5835 | ||
2b604351 MF |
5836 | /* |
5837 | * Delayed de-allocation of suballocator blocks. | |
5838 | * | |
5839 | * Some sets of block de-allocations might involve multiple suballocator inodes. | |
5840 | * | |
5841 | * The locking for this can get extremely complicated, especially when | |
5842 | * the suballocator inodes to delete from aren't known until deep | |
5843 | * within an unrelated codepath. | |
5844 | * | |
5845 | * ocfs2_extent_block structures are a good example of this - an inode | |
5846 | * btree could have been grown by any number of nodes each allocating | |
5847 | * out of their own suballoc inode. | |
5848 | * | |
5849 | * These structures allow the delay of block de-allocation until a | |
5850 | * later time, when locking of multiple cluster inodes won't cause | |
5851 | * deadlock. | |
5852 | */ | |
5853 | ||
5854 | /* | |
5855 | * Describes a single block free from a suballocator | |
5856 | */ | |
5857 | struct ocfs2_cached_block_free { | |
5858 | struct ocfs2_cached_block_free *free_next; | |
5859 | u64 free_blk; | |
5860 | unsigned int free_bit; | |
5861 | }; | |
5862 | ||
5863 | struct ocfs2_per_slot_free_list { | |
5864 | struct ocfs2_per_slot_free_list *f_next_suballocator; | |
5865 | int f_inode_type; | |
5866 | int f_slot; | |
5867 | struct ocfs2_cached_block_free *f_first; | |
5868 | }; | |
5869 | ||
5870 | static int ocfs2_free_cached_items(struct ocfs2_super *osb, | |
5871 | int sysfile_type, | |
5872 | int slot, | |
5873 | struct ocfs2_cached_block_free *head) | |
5874 | { | |
5875 | int ret; | |
5876 | u64 bg_blkno; | |
5877 | handle_t *handle; | |
5878 | struct inode *inode; | |
5879 | struct buffer_head *di_bh = NULL; | |
5880 | struct ocfs2_cached_block_free *tmp; | |
5881 | ||
5882 | inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot); | |
5883 | if (!inode) { | |
5884 | ret = -EINVAL; | |
5885 | mlog_errno(ret); | |
5886 | goto out; | |
5887 | } | |
5888 | ||
5889 | mutex_lock(&inode->i_mutex); | |
5890 | ||
e63aecb6 | 5891 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
2b604351 MF |
5892 | if (ret) { |
5893 | mlog_errno(ret); | |
5894 | goto out_mutex; | |
5895 | } | |
5896 | ||
5897 | handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE); | |
5898 | if (IS_ERR(handle)) { | |
5899 | ret = PTR_ERR(handle); | |
5900 | mlog_errno(ret); | |
5901 | goto out_unlock; | |
5902 | } | |
5903 | ||
5904 | while (head) { | |
5905 | bg_blkno = ocfs2_which_suballoc_group(head->free_blk, | |
5906 | head->free_bit); | |
5907 | mlog(0, "Free bit: (bit %u, blkno %llu)\n", | |
5908 | head->free_bit, (unsigned long long)head->free_blk); | |
5909 | ||
5910 | ret = ocfs2_free_suballoc_bits(handle, inode, di_bh, | |
5911 | head->free_bit, bg_blkno, 1); | |
5912 | if (ret) { | |
5913 | mlog_errno(ret); | |
5914 | goto out_journal; | |
5915 | } | |
5916 | ||
5917 | ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE); | |
5918 | if (ret) { | |
5919 | mlog_errno(ret); | |
5920 | goto out_journal; | |
5921 | } | |
5922 | ||
5923 | tmp = head; | |
5924 | head = head->free_next; | |
5925 | kfree(tmp); | |
5926 | } | |
5927 | ||
5928 | out_journal: | |
5929 | ocfs2_commit_trans(osb, handle); | |
5930 | ||
5931 | out_unlock: | |
e63aecb6 | 5932 | ocfs2_inode_unlock(inode, 1); |
2b604351 MF |
5933 | brelse(di_bh); |
5934 | out_mutex: | |
5935 | mutex_unlock(&inode->i_mutex); | |
5936 | iput(inode); | |
5937 | out: | |
5938 | while(head) { | |
5939 | /* Premature exit may have left some dangling items. */ | |
5940 | tmp = head; | |
5941 | head = head->free_next; | |
5942 | kfree(tmp); | |
5943 | } | |
5944 | ||
5945 | return ret; | |
5946 | } | |
5947 | ||
5948 | int ocfs2_run_deallocs(struct ocfs2_super *osb, | |
5949 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
5950 | { | |
5951 | int ret = 0, ret2; | |
5952 | struct ocfs2_per_slot_free_list *fl; | |
5953 | ||
5954 | if (!ctxt) | |
5955 | return 0; | |
5956 | ||
5957 | while (ctxt->c_first_suballocator) { | |
5958 | fl = ctxt->c_first_suballocator; | |
5959 | ||
5960 | if (fl->f_first) { | |
5961 | mlog(0, "Free items: (type %u, slot %d)\n", | |
5962 | fl->f_inode_type, fl->f_slot); | |
5963 | ret2 = ocfs2_free_cached_items(osb, fl->f_inode_type, | |
5964 | fl->f_slot, fl->f_first); | |
5965 | if (ret2) | |
5966 | mlog_errno(ret2); | |
5967 | if (!ret) | |
5968 | ret = ret2; | |
5969 | } | |
5970 | ||
5971 | ctxt->c_first_suballocator = fl->f_next_suballocator; | |
5972 | kfree(fl); | |
5973 | } | |
5974 | ||
5975 | return ret; | |
5976 | } | |
5977 | ||
5978 | static struct ocfs2_per_slot_free_list * | |
5979 | ocfs2_find_per_slot_free_list(int type, | |
5980 | int slot, | |
5981 | struct ocfs2_cached_dealloc_ctxt *ctxt) | |
5982 | { | |
5983 | struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator; | |
5984 | ||
5985 | while (fl) { | |
5986 | if (fl->f_inode_type == type && fl->f_slot == slot) | |
5987 | return fl; | |
5988 | ||
5989 | fl = fl->f_next_suballocator; | |
5990 | } | |
5991 | ||
5992 | fl = kmalloc(sizeof(*fl), GFP_NOFS); | |
5993 | if (fl) { | |
5994 | fl->f_inode_type = type; | |
5995 | fl->f_slot = slot; | |
5996 | fl->f_first = NULL; | |
5997 | fl->f_next_suballocator = ctxt->c_first_suballocator; | |
5998 | ||
5999 | ctxt->c_first_suballocator = fl; | |
6000 | } | |
6001 | return fl; | |
6002 | } | |
6003 | ||
6004 | static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt, | |
6005 | int type, int slot, u64 blkno, | |
6006 | unsigned int bit) | |
6007 | { | |
6008 | int ret; | |
6009 | struct ocfs2_per_slot_free_list *fl; | |
6010 | struct ocfs2_cached_block_free *item; | |
6011 | ||
6012 | fl = ocfs2_find_per_slot_free_list(type, slot, ctxt); | |
6013 | if (fl == NULL) { | |
6014 | ret = -ENOMEM; | |
6015 | mlog_errno(ret); | |
6016 | goto out; | |
6017 | } | |
6018 | ||
6019 | item = kmalloc(sizeof(*item), GFP_NOFS); | |
6020 | if (item == NULL) { | |
6021 | ret = -ENOMEM; | |
6022 | mlog_errno(ret); | |
6023 | goto out; | |
6024 | } | |
6025 | ||
6026 | mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n", | |
6027 | type, slot, bit, (unsigned long long)blkno); | |
6028 | ||
6029 | item->free_blk = blkno; | |
6030 | item->free_bit = bit; | |
6031 | item->free_next = fl->f_first; | |
6032 | ||
6033 | fl->f_first = item; | |
6034 | ||
6035 | ret = 0; | |
6036 | out: | |
6037 | return ret; | |
6038 | } | |
6039 | ||
59a5e416 MF |
6040 | static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt, |
6041 | struct ocfs2_extent_block *eb) | |
6042 | { | |
6043 | return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE, | |
6044 | le16_to_cpu(eb->h_suballoc_slot), | |
6045 | le64_to_cpu(eb->h_blkno), | |
6046 | le16_to_cpu(eb->h_suballoc_bit)); | |
6047 | } | |
6048 | ||
ccd979bd MF |
6049 | /* This function will figure out whether the currently last extent |
6050 | * block will be deleted, and if it will, what the new last extent | |
6051 | * block will be so we can update his h_next_leaf_blk field, as well | |
6052 | * as the dinodes i_last_eb_blk */ | |
dcd0538f | 6053 | static int ocfs2_find_new_last_ext_blk(struct inode *inode, |
3a0782d0 | 6054 | unsigned int clusters_to_del, |
dcd0538f | 6055 | struct ocfs2_path *path, |
ccd979bd MF |
6056 | struct buffer_head **new_last_eb) |
6057 | { | |
3a0782d0 | 6058 | int next_free, ret = 0; |
dcd0538f | 6059 | u32 cpos; |
3a0782d0 | 6060 | struct ocfs2_extent_rec *rec; |
ccd979bd MF |
6061 | struct ocfs2_extent_block *eb; |
6062 | struct ocfs2_extent_list *el; | |
6063 | struct buffer_head *bh = NULL; | |
6064 | ||
6065 | *new_last_eb = NULL; | |
6066 | ||
ccd979bd | 6067 | /* we have no tree, so of course, no last_eb. */ |
dcd0538f MF |
6068 | if (!path->p_tree_depth) |
6069 | goto out; | |
ccd979bd MF |
6070 | |
6071 | /* trunc to zero special case - this makes tree_depth = 0 | |
6072 | * regardless of what it is. */ | |
3a0782d0 | 6073 | if (OCFS2_I(inode)->ip_clusters == clusters_to_del) |
dcd0538f | 6074 | goto out; |
ccd979bd | 6075 | |
dcd0538f | 6076 | el = path_leaf_el(path); |
ccd979bd MF |
6077 | BUG_ON(!el->l_next_free_rec); |
6078 | ||
3a0782d0 MF |
6079 | /* |
6080 | * Make sure that this extent list will actually be empty | |
6081 | * after we clear away the data. We can shortcut out if | |
6082 | * there's more than one non-empty extent in the | |
6083 | * list. Otherwise, a check of the remaining extent is | |
6084 | * necessary. | |
6085 | */ | |
6086 | next_free = le16_to_cpu(el->l_next_free_rec); | |
6087 | rec = NULL; | |
dcd0538f | 6088 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { |
3a0782d0 | 6089 | if (next_free > 2) |
dcd0538f | 6090 | goto out; |
3a0782d0 MF |
6091 | |
6092 | /* We may have a valid extent in index 1, check it. */ | |
6093 | if (next_free == 2) | |
6094 | rec = &el->l_recs[1]; | |
6095 | ||
6096 | /* | |
6097 | * Fall through - no more nonempty extents, so we want | |
6098 | * to delete this leaf. | |
6099 | */ | |
6100 | } else { | |
6101 | if (next_free > 1) | |
6102 | goto out; | |
6103 | ||
6104 | rec = &el->l_recs[0]; | |
6105 | } | |
6106 | ||
6107 | if (rec) { | |
6108 | /* | |
6109 | * Check it we'll only be trimming off the end of this | |
6110 | * cluster. | |
6111 | */ | |
e48edee2 | 6112 | if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del) |
3a0782d0 MF |
6113 | goto out; |
6114 | } | |
ccd979bd | 6115 | |
dcd0538f MF |
6116 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos); |
6117 | if (ret) { | |
6118 | mlog_errno(ret); | |
6119 | goto out; | |
6120 | } | |
ccd979bd | 6121 | |
dcd0538f MF |
6122 | ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh); |
6123 | if (ret) { | |
6124 | mlog_errno(ret); | |
6125 | goto out; | |
6126 | } | |
ccd979bd | 6127 | |
dcd0538f MF |
6128 | eb = (struct ocfs2_extent_block *) bh->b_data; |
6129 | el = &eb->h_list; | |
6130 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
6131 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
6132 | ret = -EROFS; | |
6133 | goto out; | |
6134 | } | |
ccd979bd MF |
6135 | |
6136 | *new_last_eb = bh; | |
6137 | get_bh(*new_last_eb); | |
dcd0538f MF |
6138 | mlog(0, "returning block %llu, (cpos: %u)\n", |
6139 | (unsigned long long)le64_to_cpu(eb->h_blkno), cpos); | |
6140 | out: | |
6141 | brelse(bh); | |
ccd979bd | 6142 | |
dcd0538f | 6143 | return ret; |
ccd979bd MF |
6144 | } |
6145 | ||
3a0782d0 MF |
6146 | /* |
6147 | * Trim some clusters off the rightmost edge of a tree. Only called | |
6148 | * during truncate. | |
6149 | * | |
6150 | * The caller needs to: | |
6151 | * - start journaling of each path component. | |
6152 | * - compute and fully set up any new last ext block | |
6153 | */ | |
6154 | static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path, | |
6155 | handle_t *handle, struct ocfs2_truncate_context *tc, | |
6156 | u32 clusters_to_del, u64 *delete_start) | |
6157 | { | |
6158 | int ret, i, index = path->p_tree_depth; | |
6159 | u32 new_edge = 0; | |
6160 | u64 deleted_eb = 0; | |
6161 | struct buffer_head *bh; | |
6162 | struct ocfs2_extent_list *el; | |
6163 | struct ocfs2_extent_rec *rec; | |
6164 | ||
6165 | *delete_start = 0; | |
6166 | ||
6167 | while (index >= 0) { | |
6168 | bh = path->p_node[index].bh; | |
6169 | el = path->p_node[index].el; | |
6170 | ||
6171 | mlog(0, "traveling tree (index = %d, block = %llu)\n", | |
6172 | index, (unsigned long long)bh->b_blocknr); | |
6173 | ||
6174 | BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0); | |
6175 | ||
6176 | if (index != | |
6177 | (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) { | |
6178 | ocfs2_error(inode->i_sb, | |
6179 | "Inode %lu has invalid ext. block %llu", | |
6180 | inode->i_ino, | |
6181 | (unsigned long long)bh->b_blocknr); | |
6182 | ret = -EROFS; | |
6183 | goto out; | |
6184 | } | |
6185 | ||
6186 | find_tail_record: | |
6187 | i = le16_to_cpu(el->l_next_free_rec) - 1; | |
6188 | rec = &el->l_recs[i]; | |
6189 | ||
6190 | mlog(0, "Extent list before: record %d: (%u, %u, %llu), " | |
6191 | "next = %u\n", i, le32_to_cpu(rec->e_cpos), | |
e48edee2 | 6192 | ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6193 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6194 | le16_to_cpu(el->l_next_free_rec)); | |
6195 | ||
e48edee2 | 6196 | BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del); |
3a0782d0 MF |
6197 | |
6198 | if (le16_to_cpu(el->l_tree_depth) == 0) { | |
6199 | /* | |
6200 | * If the leaf block contains a single empty | |
6201 | * extent and no records, we can just remove | |
6202 | * the block. | |
6203 | */ | |
6204 | if (i == 0 && ocfs2_is_empty_extent(rec)) { | |
6205 | memset(rec, 0, | |
6206 | sizeof(struct ocfs2_extent_rec)); | |
6207 | el->l_next_free_rec = cpu_to_le16(0); | |
6208 | ||
6209 | goto delete; | |
6210 | } | |
6211 | ||
6212 | /* | |
6213 | * Remove any empty extents by shifting things | |
6214 | * left. That should make life much easier on | |
6215 | * the code below. This condition is rare | |
6216 | * enough that we shouldn't see a performance | |
6217 | * hit. | |
6218 | */ | |
6219 | if (ocfs2_is_empty_extent(&el->l_recs[0])) { | |
6220 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6221 | ||
6222 | for(i = 0; | |
6223 | i < le16_to_cpu(el->l_next_free_rec); i++) | |
6224 | el->l_recs[i] = el->l_recs[i + 1]; | |
6225 | ||
6226 | memset(&el->l_recs[i], 0, | |
6227 | sizeof(struct ocfs2_extent_rec)); | |
6228 | ||
6229 | /* | |
6230 | * We've modified our extent list. The | |
6231 | * simplest way to handle this change | |
6232 | * is to being the search from the | |
6233 | * start again. | |
6234 | */ | |
6235 | goto find_tail_record; | |
6236 | } | |
6237 | ||
e48edee2 | 6238 | le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del); |
3a0782d0 MF |
6239 | |
6240 | /* | |
6241 | * We'll use "new_edge" on our way back up the | |
6242 | * tree to know what our rightmost cpos is. | |
6243 | */ | |
e48edee2 | 6244 | new_edge = le16_to_cpu(rec->e_leaf_clusters); |
3a0782d0 MF |
6245 | new_edge += le32_to_cpu(rec->e_cpos); |
6246 | ||
6247 | /* | |
6248 | * The caller will use this to delete data blocks. | |
6249 | */ | |
6250 | *delete_start = le64_to_cpu(rec->e_blkno) | |
6251 | + ocfs2_clusters_to_blocks(inode->i_sb, | |
e48edee2 | 6252 | le16_to_cpu(rec->e_leaf_clusters)); |
3a0782d0 MF |
6253 | |
6254 | /* | |
6255 | * If it's now empty, remove this record. | |
6256 | */ | |
e48edee2 | 6257 | if (le16_to_cpu(rec->e_leaf_clusters) == 0) { |
3a0782d0 MF |
6258 | memset(rec, 0, |
6259 | sizeof(struct ocfs2_extent_rec)); | |
6260 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6261 | } | |
6262 | } else { | |
6263 | if (le64_to_cpu(rec->e_blkno) == deleted_eb) { | |
6264 | memset(rec, 0, | |
6265 | sizeof(struct ocfs2_extent_rec)); | |
6266 | le16_add_cpu(&el->l_next_free_rec, -1); | |
6267 | ||
6268 | goto delete; | |
6269 | } | |
6270 | ||
6271 | /* Can this actually happen? */ | |
6272 | if (le16_to_cpu(el->l_next_free_rec) == 0) | |
6273 | goto delete; | |
6274 | ||
6275 | /* | |
6276 | * We never actually deleted any clusters | |
6277 | * because our leaf was empty. There's no | |
6278 | * reason to adjust the rightmost edge then. | |
6279 | */ | |
6280 | if (new_edge == 0) | |
6281 | goto delete; | |
6282 | ||
e48edee2 MF |
6283 | rec->e_int_clusters = cpu_to_le32(new_edge); |
6284 | le32_add_cpu(&rec->e_int_clusters, | |
3a0782d0 MF |
6285 | -le32_to_cpu(rec->e_cpos)); |
6286 | ||
6287 | /* | |
6288 | * A deleted child record should have been | |
6289 | * caught above. | |
6290 | */ | |
e48edee2 | 6291 | BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0); |
3a0782d0 MF |
6292 | } |
6293 | ||
6294 | delete: | |
6295 | ret = ocfs2_journal_dirty(handle, bh); | |
6296 | if (ret) { | |
6297 | mlog_errno(ret); | |
6298 | goto out; | |
6299 | } | |
6300 | ||
6301 | mlog(0, "extent list container %llu, after: record %d: " | |
6302 | "(%u, %u, %llu), next = %u.\n", | |
6303 | (unsigned long long)bh->b_blocknr, i, | |
e48edee2 | 6304 | le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec), |
3a0782d0 MF |
6305 | (unsigned long long)le64_to_cpu(rec->e_blkno), |
6306 | le16_to_cpu(el->l_next_free_rec)); | |
6307 | ||
6308 | /* | |
6309 | * We must be careful to only attempt delete of an | |
6310 | * extent block (and not the root inode block). | |
6311 | */ | |
6312 | if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) { | |
6313 | struct ocfs2_extent_block *eb = | |
6314 | (struct ocfs2_extent_block *)bh->b_data; | |
6315 | ||
6316 | /* | |
6317 | * Save this for use when processing the | |
6318 | * parent block. | |
6319 | */ | |
6320 | deleted_eb = le64_to_cpu(eb->h_blkno); | |
6321 | ||
6322 | mlog(0, "deleting this extent block.\n"); | |
6323 | ||
6324 | ocfs2_remove_from_cache(inode, bh); | |
6325 | ||
e48edee2 | 6326 | BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0])); |
3a0782d0 MF |
6327 | BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos)); |
6328 | BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno)); | |
6329 | ||
59a5e416 MF |
6330 | ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb); |
6331 | /* An error here is not fatal. */ | |
6332 | if (ret < 0) | |
6333 | mlog_errno(ret); | |
3a0782d0 MF |
6334 | } else { |
6335 | deleted_eb = 0; | |
6336 | } | |
6337 | ||
6338 | index--; | |
6339 | } | |
6340 | ||
6341 | ret = 0; | |
6342 | out: | |
6343 | return ret; | |
6344 | } | |
6345 | ||
ccd979bd MF |
6346 | static int ocfs2_do_truncate(struct ocfs2_super *osb, |
6347 | unsigned int clusters_to_del, | |
6348 | struct inode *inode, | |
6349 | struct buffer_head *fe_bh, | |
1fabe148 | 6350 | handle_t *handle, |
dcd0538f MF |
6351 | struct ocfs2_truncate_context *tc, |
6352 | struct ocfs2_path *path) | |
ccd979bd | 6353 | { |
3a0782d0 | 6354 | int status; |
ccd979bd | 6355 | struct ocfs2_dinode *fe; |
ccd979bd MF |
6356 | struct ocfs2_extent_block *last_eb = NULL; |
6357 | struct ocfs2_extent_list *el; | |
ccd979bd | 6358 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
6359 | u64 delete_blk = 0; |
6360 | ||
6361 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
6362 | ||
3a0782d0 | 6363 | status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del, |
dcd0538f | 6364 | path, &last_eb_bh); |
ccd979bd MF |
6365 | if (status < 0) { |
6366 | mlog_errno(status); | |
6367 | goto bail; | |
6368 | } | |
dcd0538f MF |
6369 | |
6370 | /* | |
6371 | * Each component will be touched, so we might as well journal | |
6372 | * here to avoid having to handle errors later. | |
6373 | */ | |
3a0782d0 MF |
6374 | status = ocfs2_journal_access_path(inode, handle, path); |
6375 | if (status < 0) { | |
6376 | mlog_errno(status); | |
6377 | goto bail; | |
dcd0538f MF |
6378 | } |
6379 | ||
6380 | if (last_eb_bh) { | |
6381 | status = ocfs2_journal_access(handle, inode, last_eb_bh, | |
6382 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6383 | if (status < 0) { | |
6384 | mlog_errno(status); | |
6385 | goto bail; | |
6386 | } | |
6387 | ||
ccd979bd | 6388 | last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; |
dcd0538f | 6389 | } |
ccd979bd | 6390 | |
dcd0538f MF |
6391 | el = &(fe->id2.i_list); |
6392 | ||
6393 | /* | |
6394 | * Lower levels depend on this never happening, but it's best | |
6395 | * to check it up here before changing the tree. | |
6396 | */ | |
e48edee2 | 6397 | if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) { |
dcd0538f MF |
6398 | ocfs2_error(inode->i_sb, |
6399 | "Inode %lu has an empty extent record, depth %u\n", | |
6400 | inode->i_ino, le16_to_cpu(el->l_tree_depth)); | |
3a0782d0 | 6401 | status = -EROFS; |
ccd979bd MF |
6402 | goto bail; |
6403 | } | |
ccd979bd MF |
6404 | |
6405 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
6406 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) - | |
6407 | clusters_to_del; | |
6408 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
6409 | le32_add_cpu(&fe->i_clusters, -clusters_to_del); | |
e535e2ef | 6410 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd | 6411 | |
3a0782d0 MF |
6412 | status = ocfs2_trim_tree(inode, path, handle, tc, |
6413 | clusters_to_del, &delete_blk); | |
6414 | if (status) { | |
6415 | mlog_errno(status); | |
6416 | goto bail; | |
ccd979bd MF |
6417 | } |
6418 | ||
dcd0538f | 6419 | if (le32_to_cpu(fe->i_clusters) == 0) { |
ccd979bd MF |
6420 | /* trunc to zero is a special case. */ |
6421 | el->l_tree_depth = 0; | |
6422 | fe->i_last_eb_blk = 0; | |
6423 | } else if (last_eb) | |
6424 | fe->i_last_eb_blk = last_eb->h_blkno; | |
6425 | ||
6426 | status = ocfs2_journal_dirty(handle, fe_bh); | |
6427 | if (status < 0) { | |
6428 | mlog_errno(status); | |
6429 | goto bail; | |
6430 | } | |
6431 | ||
6432 | if (last_eb) { | |
6433 | /* If there will be a new last extent block, then by | |
6434 | * definition, there cannot be any leaves to the right of | |
6435 | * him. */ | |
ccd979bd MF |
6436 | last_eb->h_next_leaf_blk = 0; |
6437 | status = ocfs2_journal_dirty(handle, last_eb_bh); | |
6438 | if (status < 0) { | |
6439 | mlog_errno(status); | |
6440 | goto bail; | |
6441 | } | |
6442 | } | |
6443 | ||
3a0782d0 MF |
6444 | if (delete_blk) { |
6445 | status = ocfs2_truncate_log_append(osb, handle, delete_blk, | |
6446 | clusters_to_del); | |
ccd979bd MF |
6447 | if (status < 0) { |
6448 | mlog_errno(status); | |
6449 | goto bail; | |
6450 | } | |
ccd979bd MF |
6451 | } |
6452 | status = 0; | |
6453 | bail: | |
dcd0538f | 6454 | |
ccd979bd MF |
6455 | mlog_exit(status); |
6456 | return status; | |
6457 | } | |
6458 | ||
60b11392 MF |
6459 | static int ocfs2_writeback_zero_func(handle_t *handle, struct buffer_head *bh) |
6460 | { | |
6461 | set_buffer_uptodate(bh); | |
6462 | mark_buffer_dirty(bh); | |
6463 | return 0; | |
6464 | } | |
6465 | ||
6466 | static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh) | |
6467 | { | |
6468 | set_buffer_uptodate(bh); | |
6469 | mark_buffer_dirty(bh); | |
6470 | return ocfs2_journal_dirty_data(handle, bh); | |
6471 | } | |
6472 | ||
1d410a6e MF |
6473 | static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle, |
6474 | unsigned int from, unsigned int to, | |
6475 | struct page *page, int zero, u64 *phys) | |
6476 | { | |
6477 | int ret, partial = 0; | |
6478 | ||
6479 | ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0); | |
6480 | if (ret) | |
6481 | mlog_errno(ret); | |
6482 | ||
6483 | if (zero) | |
eebd2aa3 | 6484 | zero_user_segment(page, from, to); |
1d410a6e MF |
6485 | |
6486 | /* | |
6487 | * Need to set the buffers we zero'd into uptodate | |
6488 | * here if they aren't - ocfs2_map_page_blocks() | |
6489 | * might've skipped some | |
6490 | */ | |
6491 | if (ocfs2_should_order_data(inode)) { | |
6492 | ret = walk_page_buffers(handle, | |
6493 | page_buffers(page), | |
6494 | from, to, &partial, | |
6495 | ocfs2_ordered_zero_func); | |
6496 | if (ret < 0) | |
6497 | mlog_errno(ret); | |
6498 | } else { | |
6499 | ret = walk_page_buffers(handle, page_buffers(page), | |
6500 | from, to, &partial, | |
6501 | ocfs2_writeback_zero_func); | |
6502 | if (ret < 0) | |
6503 | mlog_errno(ret); | |
6504 | } | |
6505 | ||
6506 | if (!partial) | |
6507 | SetPageUptodate(page); | |
6508 | ||
6509 | flush_dcache_page(page); | |
6510 | } | |
6511 | ||
35edec1d MF |
6512 | static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start, |
6513 | loff_t end, struct page **pages, | |
6514 | int numpages, u64 phys, handle_t *handle) | |
60b11392 | 6515 | { |
1d410a6e | 6516 | int i; |
60b11392 MF |
6517 | struct page *page; |
6518 | unsigned int from, to = PAGE_CACHE_SIZE; | |
6519 | struct super_block *sb = inode->i_sb; | |
6520 | ||
6521 | BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb))); | |
6522 | ||
6523 | if (numpages == 0) | |
6524 | goto out; | |
6525 | ||
35edec1d | 6526 | to = PAGE_CACHE_SIZE; |
60b11392 MF |
6527 | for(i = 0; i < numpages; i++) { |
6528 | page = pages[i]; | |
6529 | ||
35edec1d MF |
6530 | from = start & (PAGE_CACHE_SIZE - 1); |
6531 | if ((end >> PAGE_CACHE_SHIFT) == page->index) | |
6532 | to = end & (PAGE_CACHE_SIZE - 1); | |
6533 | ||
60b11392 MF |
6534 | BUG_ON(from > PAGE_CACHE_SIZE); |
6535 | BUG_ON(to > PAGE_CACHE_SIZE); | |
6536 | ||
1d410a6e MF |
6537 | ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1, |
6538 | &phys); | |
60b11392 | 6539 | |
35edec1d | 6540 | start = (page->index + 1) << PAGE_CACHE_SHIFT; |
60b11392 MF |
6541 | } |
6542 | out: | |
1d410a6e MF |
6543 | if (pages) |
6544 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6545 | } |
6546 | ||
35edec1d | 6547 | static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end, |
1d410a6e | 6548 | struct page **pages, int *num) |
60b11392 | 6549 | { |
1d410a6e | 6550 | int numpages, ret = 0; |
60b11392 MF |
6551 | struct super_block *sb = inode->i_sb; |
6552 | struct address_space *mapping = inode->i_mapping; | |
6553 | unsigned long index; | |
35edec1d | 6554 | loff_t last_page_bytes; |
60b11392 | 6555 | |
35edec1d | 6556 | BUG_ON(start > end); |
60b11392 | 6557 | |
35edec1d MF |
6558 | BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits != |
6559 | (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits); | |
6560 | ||
1d410a6e | 6561 | numpages = 0; |
35edec1d MF |
6562 | last_page_bytes = PAGE_ALIGN(end); |
6563 | index = start >> PAGE_CACHE_SHIFT; | |
60b11392 MF |
6564 | do { |
6565 | pages[numpages] = grab_cache_page(mapping, index); | |
6566 | if (!pages[numpages]) { | |
6567 | ret = -ENOMEM; | |
6568 | mlog_errno(ret); | |
6569 | goto out; | |
6570 | } | |
6571 | ||
6572 | numpages++; | |
6573 | index++; | |
35edec1d | 6574 | } while (index < (last_page_bytes >> PAGE_CACHE_SHIFT)); |
60b11392 MF |
6575 | |
6576 | out: | |
6577 | if (ret != 0) { | |
1d410a6e MF |
6578 | if (pages) |
6579 | ocfs2_unlock_and_free_pages(pages, numpages); | |
60b11392 MF |
6580 | numpages = 0; |
6581 | } | |
6582 | ||
6583 | *num = numpages; | |
6584 | ||
6585 | return ret; | |
6586 | } | |
6587 | ||
6588 | /* | |
6589 | * Zero the area past i_size but still within an allocated | |
6590 | * cluster. This avoids exposing nonzero data on subsequent file | |
6591 | * extends. | |
6592 | * | |
6593 | * We need to call this before i_size is updated on the inode because | |
6594 | * otherwise block_write_full_page() will skip writeout of pages past | |
6595 | * i_size. The new_i_size parameter is passed for this reason. | |
6596 | */ | |
35edec1d MF |
6597 | int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle, |
6598 | u64 range_start, u64 range_end) | |
60b11392 | 6599 | { |
1d410a6e | 6600 | int ret = 0, numpages; |
60b11392 MF |
6601 | struct page **pages = NULL; |
6602 | u64 phys; | |
1d410a6e MF |
6603 | unsigned int ext_flags; |
6604 | struct super_block *sb = inode->i_sb; | |
60b11392 MF |
6605 | |
6606 | /* | |
6607 | * File systems which don't support sparse files zero on every | |
6608 | * extend. | |
6609 | */ | |
1d410a6e | 6610 | if (!ocfs2_sparse_alloc(OCFS2_SB(sb))) |
60b11392 MF |
6611 | return 0; |
6612 | ||
1d410a6e | 6613 | pages = kcalloc(ocfs2_pages_per_cluster(sb), |
60b11392 MF |
6614 | sizeof(struct page *), GFP_NOFS); |
6615 | if (pages == NULL) { | |
6616 | ret = -ENOMEM; | |
6617 | mlog_errno(ret); | |
6618 | goto out; | |
6619 | } | |
6620 | ||
1d410a6e MF |
6621 | if (range_start == range_end) |
6622 | goto out; | |
6623 | ||
6624 | ret = ocfs2_extent_map_get_blocks(inode, | |
6625 | range_start >> sb->s_blocksize_bits, | |
6626 | &phys, NULL, &ext_flags); | |
60b11392 MF |
6627 | if (ret) { |
6628 | mlog_errno(ret); | |
6629 | goto out; | |
6630 | } | |
6631 | ||
1d410a6e MF |
6632 | /* |
6633 | * Tail is a hole, or is marked unwritten. In either case, we | |
6634 | * can count on read and write to return/push zero's. | |
6635 | */ | |
6636 | if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN) | |
60b11392 MF |
6637 | goto out; |
6638 | ||
1d410a6e MF |
6639 | ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages, |
6640 | &numpages); | |
6641 | if (ret) { | |
6642 | mlog_errno(ret); | |
6643 | goto out; | |
6644 | } | |
6645 | ||
35edec1d MF |
6646 | ocfs2_zero_cluster_pages(inode, range_start, range_end, pages, |
6647 | numpages, phys, handle); | |
60b11392 MF |
6648 | |
6649 | /* | |
6650 | * Initiate writeout of the pages we zero'd here. We don't | |
6651 | * wait on them - the truncate_inode_pages() call later will | |
6652 | * do that for us. | |
6653 | */ | |
35edec1d MF |
6654 | ret = do_sync_mapping_range(inode->i_mapping, range_start, |
6655 | range_end - 1, SYNC_FILE_RANGE_WRITE); | |
60b11392 MF |
6656 | if (ret) |
6657 | mlog_errno(ret); | |
6658 | ||
6659 | out: | |
6660 | if (pages) | |
6661 | kfree(pages); | |
6662 | ||
6663 | return ret; | |
6664 | } | |
6665 | ||
fdd77704 TY |
6666 | static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode, |
6667 | struct ocfs2_dinode *di) | |
1afc32b9 MF |
6668 | { |
6669 | unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits; | |
fdd77704 | 6670 | unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size); |
1afc32b9 | 6671 | |
fdd77704 TY |
6672 | if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL) |
6673 | memset(&di->id2, 0, blocksize - | |
6674 | offsetof(struct ocfs2_dinode, id2) - | |
6675 | xattrsize); | |
6676 | else | |
6677 | memset(&di->id2, 0, blocksize - | |
6678 | offsetof(struct ocfs2_dinode, id2)); | |
1afc32b9 MF |
6679 | } |
6680 | ||
5b6a3a2b MF |
6681 | void ocfs2_dinode_new_extent_list(struct inode *inode, |
6682 | struct ocfs2_dinode *di) | |
6683 | { | |
fdd77704 | 6684 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
5b6a3a2b MF |
6685 | di->id2.i_list.l_tree_depth = 0; |
6686 | di->id2.i_list.l_next_free_rec = 0; | |
fdd77704 TY |
6687 | di->id2.i_list.l_count = cpu_to_le16( |
6688 | ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di)); | |
5b6a3a2b MF |
6689 | } |
6690 | ||
1afc32b9 MF |
6691 | void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di) |
6692 | { | |
6693 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6694 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
6695 | ||
6696 | spin_lock(&oi->ip_lock); | |
6697 | oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL; | |
6698 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6699 | spin_unlock(&oi->ip_lock); | |
6700 | ||
6701 | /* | |
6702 | * We clear the entire i_data structure here so that all | |
6703 | * fields can be properly initialized. | |
6704 | */ | |
fdd77704 | 6705 | ocfs2_zero_dinode_id2_with_xattr(inode, di); |
1afc32b9 | 6706 | |
fdd77704 TY |
6707 | idata->id_count = cpu_to_le16( |
6708 | ocfs2_max_inline_data_with_xattr(inode->i_sb, di)); | |
1afc32b9 MF |
6709 | } |
6710 | ||
6711 | int ocfs2_convert_inline_data_to_extents(struct inode *inode, | |
6712 | struct buffer_head *di_bh) | |
6713 | { | |
6714 | int ret, i, has_data, num_pages = 0; | |
6715 | handle_t *handle; | |
6716 | u64 uninitialized_var(block); | |
6717 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
6718 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
6719 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1afc32b9 MF |
6720 | struct ocfs2_alloc_context *data_ac = NULL; |
6721 | struct page **pages = NULL; | |
6722 | loff_t end = osb->s_clustersize; | |
6723 | ||
6724 | has_data = i_size_read(inode) ? 1 : 0; | |
6725 | ||
6726 | if (has_data) { | |
6727 | pages = kcalloc(ocfs2_pages_per_cluster(osb->sb), | |
6728 | sizeof(struct page *), GFP_NOFS); | |
6729 | if (pages == NULL) { | |
6730 | ret = -ENOMEM; | |
6731 | mlog_errno(ret); | |
6732 | goto out; | |
6733 | } | |
6734 | ||
6735 | ret = ocfs2_reserve_clusters(osb, 1, &data_ac); | |
6736 | if (ret) { | |
6737 | mlog_errno(ret); | |
6738 | goto out; | |
6739 | } | |
6740 | } | |
6741 | ||
6742 | handle = ocfs2_start_trans(osb, OCFS2_INLINE_TO_EXTENTS_CREDITS); | |
6743 | if (IS_ERR(handle)) { | |
6744 | ret = PTR_ERR(handle); | |
6745 | mlog_errno(ret); | |
6746 | goto out_unlock; | |
6747 | } | |
6748 | ||
6749 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
6750 | OCFS2_JOURNAL_ACCESS_WRITE); | |
6751 | if (ret) { | |
6752 | mlog_errno(ret); | |
6753 | goto out_commit; | |
6754 | } | |
6755 | ||
6756 | if (has_data) { | |
6757 | u32 bit_off, num; | |
6758 | unsigned int page_end; | |
6759 | u64 phys; | |
6760 | ||
6761 | ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, | |
6762 | &num); | |
6763 | if (ret) { | |
6764 | mlog_errno(ret); | |
6765 | goto out_commit; | |
6766 | } | |
6767 | ||
6768 | /* | |
6769 | * Save two copies, one for insert, and one that can | |
6770 | * be changed by ocfs2_map_and_dirty_page() below. | |
6771 | */ | |
6772 | block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off); | |
6773 | ||
6774 | /* | |
6775 | * Non sparse file systems zero on extend, so no need | |
6776 | * to do that now. | |
6777 | */ | |
6778 | if (!ocfs2_sparse_alloc(osb) && | |
6779 | PAGE_CACHE_SIZE < osb->s_clustersize) | |
6780 | end = PAGE_CACHE_SIZE; | |
6781 | ||
6782 | ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages); | |
6783 | if (ret) { | |
6784 | mlog_errno(ret); | |
6785 | goto out_commit; | |
6786 | } | |
6787 | ||
6788 | /* | |
6789 | * This should populate the 1st page for us and mark | |
6790 | * it up to date. | |
6791 | */ | |
6792 | ret = ocfs2_read_inline_data(inode, pages[0], di_bh); | |
6793 | if (ret) { | |
6794 | mlog_errno(ret); | |
6795 | goto out_commit; | |
6796 | } | |
6797 | ||
6798 | page_end = PAGE_CACHE_SIZE; | |
6799 | if (PAGE_CACHE_SIZE > osb->s_clustersize) | |
6800 | page_end = osb->s_clustersize; | |
6801 | ||
6802 | for (i = 0; i < num_pages; i++) | |
6803 | ocfs2_map_and_dirty_page(inode, handle, 0, page_end, | |
6804 | pages[i], i > 0, &phys); | |
6805 | } | |
6806 | ||
6807 | spin_lock(&oi->ip_lock); | |
6808 | oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL; | |
6809 | di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features); | |
6810 | spin_unlock(&oi->ip_lock); | |
6811 | ||
5b6a3a2b | 6812 | ocfs2_dinode_new_extent_list(inode, di); |
1afc32b9 MF |
6813 | |
6814 | ocfs2_journal_dirty(handle, di_bh); | |
6815 | ||
6816 | if (has_data) { | |
6817 | /* | |
6818 | * An error at this point should be extremely rare. If | |
6819 | * this proves to be false, we could always re-build | |
6820 | * the in-inode data from our pages. | |
6821 | */ | |
f56654c4 TM |
6822 | ret = ocfs2_dinode_insert_extent(osb, handle, inode, di_bh, |
6823 | 0, block, 1, 0, NULL); | |
1afc32b9 MF |
6824 | if (ret) { |
6825 | mlog_errno(ret); | |
6826 | goto out_commit; | |
6827 | } | |
6828 | ||
6829 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
6830 | } | |
6831 | ||
6832 | out_commit: | |
6833 | ocfs2_commit_trans(osb, handle); | |
6834 | ||
6835 | out_unlock: | |
6836 | if (data_ac) | |
6837 | ocfs2_free_alloc_context(data_ac); | |
6838 | ||
6839 | out: | |
6840 | if (pages) { | |
6841 | ocfs2_unlock_and_free_pages(pages, num_pages); | |
6842 | kfree(pages); | |
6843 | } | |
6844 | ||
6845 | return ret; | |
6846 | } | |
6847 | ||
ccd979bd MF |
6848 | /* |
6849 | * It is expected, that by the time you call this function, | |
6850 | * inode->i_size and fe->i_size have been adjusted. | |
6851 | * | |
6852 | * WARNING: This will kfree the truncate context | |
6853 | */ | |
6854 | int ocfs2_commit_truncate(struct ocfs2_super *osb, | |
6855 | struct inode *inode, | |
6856 | struct buffer_head *fe_bh, | |
6857 | struct ocfs2_truncate_context *tc) | |
6858 | { | |
6859 | int status, i, credits, tl_sem = 0; | |
dcd0538f | 6860 | u32 clusters_to_del, new_highest_cpos, range; |
ccd979bd | 6861 | struct ocfs2_extent_list *el; |
1fabe148 | 6862 | handle_t *handle = NULL; |
ccd979bd | 6863 | struct inode *tl_inode = osb->osb_tl_inode; |
dcd0538f | 6864 | struct ocfs2_path *path = NULL; |
e7d4cb6b | 6865 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data; |
ccd979bd MF |
6866 | |
6867 | mlog_entry_void(); | |
6868 | ||
dcd0538f | 6869 | new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb, |
ccd979bd MF |
6870 | i_size_read(inode)); |
6871 | ||
e7d4cb6b | 6872 | path = ocfs2_new_path(fe_bh, &di->id2.i_list); |
dcd0538f MF |
6873 | if (!path) { |
6874 | status = -ENOMEM; | |
6875 | mlog_errno(status); | |
6876 | goto bail; | |
6877 | } | |
83418978 MF |
6878 | |
6879 | ocfs2_extent_map_trunc(inode, new_highest_cpos); | |
6880 | ||
ccd979bd | 6881 | start: |
3a0782d0 MF |
6882 | /* |
6883 | * Check that we still have allocation to delete. | |
6884 | */ | |
6885 | if (OCFS2_I(inode)->ip_clusters == 0) { | |
6886 | status = 0; | |
6887 | goto bail; | |
6888 | } | |
6889 | ||
dcd0538f MF |
6890 | /* |
6891 | * Truncate always works against the rightmost tree branch. | |
6892 | */ | |
6893 | status = ocfs2_find_path(inode, path, UINT_MAX); | |
6894 | if (status) { | |
6895 | mlog_errno(status); | |
6896 | goto bail; | |
ccd979bd MF |
6897 | } |
6898 | ||
dcd0538f MF |
6899 | mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n", |
6900 | OCFS2_I(inode)->ip_clusters, path->p_tree_depth); | |
6901 | ||
6902 | /* | |
6903 | * By now, el will point to the extent list on the bottom most | |
6904 | * portion of this tree. Only the tail record is considered in | |
6905 | * each pass. | |
6906 | * | |
6907 | * We handle the following cases, in order: | |
6908 | * - empty extent: delete the remaining branch | |
6909 | * - remove the entire record | |
6910 | * - remove a partial record | |
6911 | * - no record needs to be removed (truncate has completed) | |
6912 | */ | |
6913 | el = path_leaf_el(path); | |
3a0782d0 MF |
6914 | if (le16_to_cpu(el->l_next_free_rec) == 0) { |
6915 | ocfs2_error(inode->i_sb, | |
6916 | "Inode %llu has empty extent block at %llu\n", | |
6917 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
6918 | (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
6919 | status = -EROFS; | |
6920 | goto bail; | |
6921 | } | |
6922 | ||
ccd979bd | 6923 | i = le16_to_cpu(el->l_next_free_rec) - 1; |
dcd0538f | 6924 | range = le32_to_cpu(el->l_recs[i].e_cpos) + |
e48edee2 | 6925 | ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f MF |
6926 | if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) { |
6927 | clusters_to_del = 0; | |
6928 | } else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) { | |
e48edee2 | 6929 | clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]); |
dcd0538f | 6930 | } else if (range > new_highest_cpos) { |
e48edee2 | 6931 | clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) + |
ccd979bd | 6932 | le32_to_cpu(el->l_recs[i].e_cpos)) - |
dcd0538f MF |
6933 | new_highest_cpos; |
6934 | } else { | |
6935 | status = 0; | |
6936 | goto bail; | |
6937 | } | |
ccd979bd | 6938 | |
dcd0538f MF |
6939 | mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n", |
6940 | clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr); | |
6941 | ||
1b1dcc1b | 6942 | mutex_lock(&tl_inode->i_mutex); |
ccd979bd MF |
6943 | tl_sem = 1; |
6944 | /* ocfs2_truncate_log_needs_flush guarantees us at least one | |
6945 | * record is free for use. If there isn't any, we flush to get | |
6946 | * an empty truncate log. */ | |
6947 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
6948 | status = __ocfs2_flush_truncate_log(osb); | |
6949 | if (status < 0) { | |
6950 | mlog_errno(status); | |
6951 | goto bail; | |
6952 | } | |
6953 | } | |
6954 | ||
6955 | credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del, | |
dcd0538f MF |
6956 | (struct ocfs2_dinode *)fe_bh->b_data, |
6957 | el); | |
65eff9cc | 6958 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
6959 | if (IS_ERR(handle)) { |
6960 | status = PTR_ERR(handle); | |
6961 | handle = NULL; | |
6962 | mlog_errno(status); | |
6963 | goto bail; | |
6964 | } | |
6965 | ||
dcd0538f MF |
6966 | status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle, |
6967 | tc, path); | |
ccd979bd MF |
6968 | if (status < 0) { |
6969 | mlog_errno(status); | |
6970 | goto bail; | |
6971 | } | |
6972 | ||
1b1dcc1b | 6973 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6974 | tl_sem = 0; |
6975 | ||
02dc1af4 | 6976 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
6977 | handle = NULL; |
6978 | ||
dcd0538f MF |
6979 | ocfs2_reinit_path(path, 1); |
6980 | ||
6981 | /* | |
3a0782d0 MF |
6982 | * The check above will catch the case where we've truncated |
6983 | * away all allocation. | |
dcd0538f | 6984 | */ |
3a0782d0 MF |
6985 | goto start; |
6986 | ||
ccd979bd | 6987 | bail: |
ccd979bd MF |
6988 | |
6989 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
6990 | ||
6991 | if (tl_sem) | |
1b1dcc1b | 6992 | mutex_unlock(&tl_inode->i_mutex); |
ccd979bd MF |
6993 | |
6994 | if (handle) | |
02dc1af4 | 6995 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 6996 | |
59a5e416 MF |
6997 | ocfs2_run_deallocs(osb, &tc->tc_dealloc); |
6998 | ||
dcd0538f | 6999 | ocfs2_free_path(path); |
ccd979bd MF |
7000 | |
7001 | /* This will drop the ext_alloc cluster lock for us */ | |
7002 | ocfs2_free_truncate_context(tc); | |
7003 | ||
7004 | mlog_exit(status); | |
7005 | return status; | |
7006 | } | |
7007 | ||
ccd979bd | 7008 | /* |
59a5e416 | 7009 | * Expects the inode to already be locked. |
ccd979bd MF |
7010 | */ |
7011 | int ocfs2_prepare_truncate(struct ocfs2_super *osb, | |
7012 | struct inode *inode, | |
7013 | struct buffer_head *fe_bh, | |
7014 | struct ocfs2_truncate_context **tc) | |
7015 | { | |
59a5e416 | 7016 | int status; |
ccd979bd MF |
7017 | unsigned int new_i_clusters; |
7018 | struct ocfs2_dinode *fe; | |
7019 | struct ocfs2_extent_block *eb; | |
ccd979bd | 7020 | struct buffer_head *last_eb_bh = NULL; |
ccd979bd MF |
7021 | |
7022 | mlog_entry_void(); | |
7023 | ||
7024 | *tc = NULL; | |
7025 | ||
7026 | new_i_clusters = ocfs2_clusters_for_bytes(osb->sb, | |
7027 | i_size_read(inode)); | |
7028 | fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
7029 | ||
7030 | mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size =" | |
1ca1a111 MF |
7031 | "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters, |
7032 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 7033 | |
cd861280 | 7034 | *tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL); |
ccd979bd MF |
7035 | if (!(*tc)) { |
7036 | status = -ENOMEM; | |
7037 | mlog_errno(status); | |
7038 | goto bail; | |
7039 | } | |
59a5e416 | 7040 | ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc); |
ccd979bd | 7041 | |
ccd979bd | 7042 | if (fe->id2.i_list.l_tree_depth) { |
ccd979bd MF |
7043 | status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk), |
7044 | &last_eb_bh, OCFS2_BH_CACHED, inode); | |
7045 | if (status < 0) { | |
7046 | mlog_errno(status); | |
7047 | goto bail; | |
7048 | } | |
7049 | eb = (struct ocfs2_extent_block *) last_eb_bh->b_data; | |
7050 | if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) { | |
7051 | OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb); | |
7052 | ||
7053 | brelse(last_eb_bh); | |
7054 | status = -EIO; | |
7055 | goto bail; | |
7056 | } | |
ccd979bd MF |
7057 | } |
7058 | ||
7059 | (*tc)->tc_last_eb_bh = last_eb_bh; | |
7060 | ||
ccd979bd MF |
7061 | status = 0; |
7062 | bail: | |
7063 | if (status < 0) { | |
7064 | if (*tc) | |
7065 | ocfs2_free_truncate_context(*tc); | |
7066 | *tc = NULL; | |
7067 | } | |
7068 | mlog_exit_void(); | |
7069 | return status; | |
7070 | } | |
7071 | ||
1afc32b9 MF |
7072 | /* |
7073 | * 'start' is inclusive, 'end' is not. | |
7074 | */ | |
7075 | int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh, | |
7076 | unsigned int start, unsigned int end, int trunc) | |
7077 | { | |
7078 | int ret; | |
7079 | unsigned int numbytes; | |
7080 | handle_t *handle; | |
7081 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
7082 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
7083 | struct ocfs2_inline_data *idata = &di->id2.i_data; | |
7084 | ||
7085 | if (end > i_size_read(inode)) | |
7086 | end = i_size_read(inode); | |
7087 | ||
7088 | BUG_ON(start >= end); | |
7089 | ||
7090 | if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) || | |
7091 | !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) || | |
7092 | !ocfs2_supports_inline_data(osb)) { | |
7093 | ocfs2_error(inode->i_sb, | |
7094 | "Inline data flags for inode %llu don't agree! " | |
7095 | "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n", | |
7096 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
7097 | le16_to_cpu(di->i_dyn_features), | |
7098 | OCFS2_I(inode)->ip_dyn_features, | |
7099 | osb->s_feature_incompat); | |
7100 | ret = -EROFS; | |
7101 | goto out; | |
7102 | } | |
7103 | ||
7104 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
7105 | if (IS_ERR(handle)) { | |
7106 | ret = PTR_ERR(handle); | |
7107 | mlog_errno(ret); | |
7108 | goto out; | |
7109 | } | |
7110 | ||
7111 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
7112 | OCFS2_JOURNAL_ACCESS_WRITE); | |
7113 | if (ret) { | |
7114 | mlog_errno(ret); | |
7115 | goto out_commit; | |
7116 | } | |
7117 | ||
7118 | numbytes = end - start; | |
7119 | memset(idata->id_data + start, 0, numbytes); | |
7120 | ||
7121 | /* | |
7122 | * No need to worry about the data page here - it's been | |
7123 | * truncated already and inline data doesn't need it for | |
7124 | * pushing zero's to disk, so we'll let readpage pick it up | |
7125 | * later. | |
7126 | */ | |
7127 | if (trunc) { | |
7128 | i_size_write(inode, start); | |
7129 | di->i_size = cpu_to_le64(start); | |
7130 | } | |
7131 | ||
7132 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
7133 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
7134 | ||
7135 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
7136 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
7137 | ||
7138 | ocfs2_journal_dirty(handle, di_bh); | |
7139 | ||
7140 | out_commit: | |
7141 | ocfs2_commit_trans(osb, handle); | |
7142 | ||
7143 | out: | |
7144 | return ret; | |
7145 | } | |
7146 | ||
ccd979bd MF |
7147 | static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc) |
7148 | { | |
59a5e416 MF |
7149 | /* |
7150 | * The caller is responsible for completing deallocation | |
7151 | * before freeing the context. | |
7152 | */ | |
7153 | if (tc->tc_dealloc.c_first_suballocator != NULL) | |
7154 | mlog(ML_NOTICE, | |
7155 | "Truncate completion has non-empty dealloc context\n"); | |
ccd979bd MF |
7156 | |
7157 | if (tc->tc_last_eb_bh) | |
7158 | brelse(tc->tc_last_eb_bh); | |
7159 | ||
7160 | kfree(tc); | |
7161 | } |