Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * suballoc.c | |
5 | * | |
6 | * metadata alloc and free | |
7 | * Inspired by ext3 block groups. | |
8 | * | |
9 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
10 | * | |
11 | * This program is free software; you can redistribute it and/or | |
12 | * modify it under the terms of the GNU General Public | |
13 | * License as published by the Free Software Foundation; either | |
14 | * version 2 of the License, or (at your option) any later version. | |
15 | * | |
16 | * This program is distributed in the hope that it will be useful, | |
17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
19 | * General Public License for more details. | |
20 | * | |
21 | * You should have received a copy of the GNU General Public | |
22 | * License along with this program; if not, write to the | |
23 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
24 | * Boston, MA 021110-1307, USA. | |
25 | */ | |
26 | ||
27 | #include <linux/fs.h> | |
28 | #include <linux/types.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/highmem.h> | |
31 | ||
32 | #define MLOG_MASK_PREFIX ML_DISK_ALLOC | |
33 | #include <cluster/masklog.h> | |
34 | ||
35 | #include "ocfs2.h" | |
36 | ||
37 | #include "alloc.h" | |
38 | #include "dlmglue.h" | |
39 | #include "inode.h" | |
40 | #include "journal.h" | |
41 | #include "localalloc.h" | |
42 | #include "suballoc.h" | |
43 | #include "super.h" | |
44 | #include "sysfile.h" | |
45 | #include "uptodate.h" | |
46 | ||
47 | #include "buffer_head_io.h" | |
48 | ||
ffda89a3 TM |
49 | #define NOT_ALLOC_NEW_GROUP 0 |
50 | #define ALLOC_NEW_GROUP 1 | |
51 | ||
4d0ddb2c TM |
52 | #define OCFS2_MAX_INODES_TO_STEAL 1024 |
53 | ||
ccd979bd MF |
54 | static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg); |
55 | static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe); | |
56 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl); | |
1fabe148 | 57 | static int ocfs2_block_group_fill(handle_t *handle, |
ccd979bd MF |
58 | struct inode *alloc_inode, |
59 | struct buffer_head *bg_bh, | |
60 | u64 group_blkno, | |
61 | u16 my_chain, | |
62 | struct ocfs2_chain_list *cl); | |
63 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb, | |
64 | struct inode *alloc_inode, | |
65 | struct buffer_head *bh); | |
66 | ||
ccd979bd MF |
67 | static int ocfs2_cluster_group_search(struct inode *inode, |
68 | struct buffer_head *group_bh, | |
69 | u32 bits_wanted, u32 min_bits, | |
70 | u16 *bit_off, u16 *bits_found); | |
71 | static int ocfs2_block_group_search(struct inode *inode, | |
72 | struct buffer_head *group_bh, | |
73 | u32 bits_wanted, u32 min_bits, | |
74 | u16 *bit_off, u16 *bits_found); | |
ccd979bd MF |
75 | static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb, |
76 | struct ocfs2_alloc_context *ac, | |
1fabe148 | 77 | handle_t *handle, |
ccd979bd MF |
78 | u32 bits_wanted, |
79 | u32 min_bits, | |
80 | u16 *bit_off, | |
81 | unsigned int *num_bits, | |
82 | u64 *bg_blkno); | |
83 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh, | |
84 | int nr); | |
1fabe148 | 85 | static inline int ocfs2_block_group_set_bits(handle_t *handle, |
ccd979bd MF |
86 | struct inode *alloc_inode, |
87 | struct ocfs2_group_desc *bg, | |
88 | struct buffer_head *group_bh, | |
89 | unsigned int bit_off, | |
90 | unsigned int num_bits); | |
1fabe148 | 91 | static inline int ocfs2_block_group_clear_bits(handle_t *handle, |
ccd979bd MF |
92 | struct inode *alloc_inode, |
93 | struct ocfs2_group_desc *bg, | |
94 | struct buffer_head *group_bh, | |
95 | unsigned int bit_off, | |
96 | unsigned int num_bits); | |
97 | ||
1fabe148 | 98 | static int ocfs2_relink_block_group(handle_t *handle, |
ccd979bd MF |
99 | struct inode *alloc_inode, |
100 | struct buffer_head *fe_bh, | |
101 | struct buffer_head *bg_bh, | |
102 | struct buffer_head *prev_bg_bh, | |
103 | u16 chain); | |
104 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg, | |
105 | u32 wanted); | |
ccd979bd MF |
106 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode, |
107 | u64 bg_blkno, | |
108 | u16 bg_bit_off); | |
ccd979bd MF |
109 | static inline void ocfs2_block_to_cluster_group(struct inode *inode, |
110 | u64 data_blkno, | |
111 | u64 *bg_blkno, | |
112 | u16 *bg_bit_off); | |
113 | ||
9c7af40b | 114 | void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac) |
ccd979bd | 115 | { |
da5cbf2f MF |
116 | struct inode *inode = ac->ac_inode; |
117 | ||
118 | if (inode) { | |
119 | if (ac->ac_which != OCFS2_AC_USE_LOCAL) | |
e63aecb6 | 120 | ocfs2_inode_unlock(inode, 1); |
da5cbf2f MF |
121 | |
122 | mutex_unlock(&inode->i_mutex); | |
123 | ||
124 | iput(inode); | |
4d0ddb2c | 125 | ac->ac_inode = NULL; |
da5cbf2f | 126 | } |
4d0ddb2c | 127 | if (ac->ac_bh) { |
ccd979bd | 128 | brelse(ac->ac_bh); |
4d0ddb2c TM |
129 | ac->ac_bh = NULL; |
130 | } | |
131 | } | |
132 | ||
133 | void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac) | |
134 | { | |
135 | ocfs2_free_ac_resource(ac); | |
ccd979bd MF |
136 | kfree(ac); |
137 | } | |
138 | ||
139 | static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl) | |
140 | { | |
141 | return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc); | |
142 | } | |
143 | ||
7bf72ede | 144 | /* somewhat more expensive than our other checks, so use sparingly. */ |
d659072f TM |
145 | int ocfs2_check_group_descriptor(struct super_block *sb, |
146 | struct ocfs2_dinode *di, | |
147 | struct ocfs2_group_desc *gd) | |
7bf72ede MF |
148 | { |
149 | unsigned int max_bits; | |
150 | ||
151 | if (!OCFS2_IS_VALID_GROUP_DESC(gd)) { | |
152 | OCFS2_RO_ON_INVALID_GROUP_DESC(sb, gd); | |
153 | return -EIO; | |
154 | } | |
155 | ||
156 | if (di->i_blkno != gd->bg_parent_dinode) { | |
157 | ocfs2_error(sb, "Group descriptor # %llu has bad parent " | |
158 | "pointer (%llu, expected %llu)", | |
159 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
160 | (unsigned long long)le64_to_cpu(gd->bg_parent_dinode), | |
161 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
162 | return -EIO; | |
163 | } | |
164 | ||
165 | max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc); | |
166 | if (le16_to_cpu(gd->bg_bits) > max_bits) { | |
167 | ocfs2_error(sb, "Group descriptor # %llu has bit count of %u", | |
168 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
169 | le16_to_cpu(gd->bg_bits)); | |
170 | return -EIO; | |
171 | } | |
172 | ||
173 | if (le16_to_cpu(gd->bg_chain) >= | |
174 | le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) { | |
175 | ocfs2_error(sb, "Group descriptor # %llu has bad chain %u", | |
176 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
177 | le16_to_cpu(gd->bg_chain)); | |
178 | return -EIO; | |
179 | } | |
180 | ||
181 | if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) { | |
182 | ocfs2_error(sb, "Group descriptor # %llu has bit count %u but " | |
183 | "claims that %u are free", | |
184 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
185 | le16_to_cpu(gd->bg_bits), | |
186 | le16_to_cpu(gd->bg_free_bits_count)); | |
187 | return -EIO; | |
188 | } | |
189 | ||
190 | if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) { | |
191 | ocfs2_error(sb, "Group descriptor # %llu has bit count %u but " | |
192 | "max bitmap bits of %u", | |
193 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
194 | le16_to_cpu(gd->bg_bits), | |
195 | 8 * le16_to_cpu(gd->bg_size)); | |
196 | return -EIO; | |
197 | } | |
198 | ||
199 | return 0; | |
200 | } | |
201 | ||
1fabe148 | 202 | static int ocfs2_block_group_fill(handle_t *handle, |
ccd979bd MF |
203 | struct inode *alloc_inode, |
204 | struct buffer_head *bg_bh, | |
205 | u64 group_blkno, | |
206 | u16 my_chain, | |
207 | struct ocfs2_chain_list *cl) | |
208 | { | |
209 | int status = 0; | |
210 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
211 | struct super_block * sb = alloc_inode->i_sb; | |
212 | ||
213 | mlog_entry_void(); | |
214 | ||
215 | if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) { | |
b0697053 MF |
216 | ocfs2_error(alloc_inode->i_sb, "group block (%llu) != " |
217 | "b_blocknr (%llu)", | |
218 | (unsigned long long)group_blkno, | |
ccd979bd MF |
219 | (unsigned long long) bg_bh->b_blocknr); |
220 | status = -EIO; | |
221 | goto bail; | |
222 | } | |
223 | ||
224 | status = ocfs2_journal_access(handle, | |
225 | alloc_inode, | |
226 | bg_bh, | |
227 | OCFS2_JOURNAL_ACCESS_CREATE); | |
228 | if (status < 0) { | |
229 | mlog_errno(status); | |
230 | goto bail; | |
231 | } | |
232 | ||
233 | memset(bg, 0, sb->s_blocksize); | |
234 | strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE); | |
235 | bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation); | |
236 | bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb)); | |
237 | bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl)); | |
238 | bg->bg_chain = cpu_to_le16(my_chain); | |
239 | bg->bg_next_group = cl->cl_recs[my_chain].c_blkno; | |
240 | bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno); | |
241 | bg->bg_blkno = cpu_to_le64(group_blkno); | |
242 | /* set the 1st bit in the bitmap to account for the descriptor block */ | |
243 | ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap); | |
244 | bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1); | |
245 | ||
246 | status = ocfs2_journal_dirty(handle, bg_bh); | |
247 | if (status < 0) | |
248 | mlog_errno(status); | |
249 | ||
250 | /* There is no need to zero out or otherwise initialize the | |
251 | * other blocks in a group - All valid FS metadata in a block | |
252 | * group stores the superblock fs_generation value at | |
253 | * allocation time. */ | |
254 | ||
255 | bail: | |
256 | mlog_exit(status); | |
257 | return status; | |
258 | } | |
259 | ||
260 | static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl) | |
261 | { | |
262 | u16 curr, best; | |
263 | ||
264 | best = curr = 0; | |
265 | while (curr < le16_to_cpu(cl->cl_count)) { | |
266 | if (le32_to_cpu(cl->cl_recs[best].c_total) > | |
267 | le32_to_cpu(cl->cl_recs[curr].c_total)) | |
268 | best = curr; | |
269 | curr++; | |
270 | } | |
271 | return best; | |
272 | } | |
273 | ||
274 | /* | |
275 | * We expect the block group allocator to already be locked. | |
276 | */ | |
277 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb, | |
278 | struct inode *alloc_inode, | |
279 | struct buffer_head *bh) | |
280 | { | |
281 | int status, credits; | |
282 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
283 | struct ocfs2_chain_list *cl; | |
284 | struct ocfs2_alloc_context *ac = NULL; | |
1fabe148 | 285 | handle_t *handle = NULL; |
ccd979bd MF |
286 | u32 bit_off, num_bits; |
287 | u16 alloc_rec; | |
288 | u64 bg_blkno; | |
289 | struct buffer_head *bg_bh = NULL; | |
290 | struct ocfs2_group_desc *bg; | |
291 | ||
292 | BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode)); | |
293 | ||
294 | mlog_entry_void(); | |
295 | ||
ccd979bd MF |
296 | cl = &fe->id2.i_chain; |
297 | status = ocfs2_reserve_clusters(osb, | |
ccd979bd MF |
298 | le16_to_cpu(cl->cl_cpg), |
299 | &ac); | |
300 | if (status < 0) { | |
301 | if (status != -ENOSPC) | |
302 | mlog_errno(status); | |
303 | goto bail; | |
304 | } | |
305 | ||
306 | credits = ocfs2_calc_group_alloc_credits(osb->sb, | |
307 | le16_to_cpu(cl->cl_cpg)); | |
65eff9cc | 308 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
309 | if (IS_ERR(handle)) { |
310 | status = PTR_ERR(handle); | |
311 | handle = NULL; | |
312 | mlog_errno(status); | |
313 | goto bail; | |
314 | } | |
315 | ||
316 | status = ocfs2_claim_clusters(osb, | |
317 | handle, | |
318 | ac, | |
319 | le16_to_cpu(cl->cl_cpg), | |
320 | &bit_off, | |
321 | &num_bits); | |
322 | if (status < 0) { | |
323 | if (status != -ENOSPC) | |
324 | mlog_errno(status); | |
325 | goto bail; | |
326 | } | |
327 | ||
328 | alloc_rec = ocfs2_find_smallest_chain(cl); | |
329 | ||
330 | /* setup the group */ | |
331 | bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
b0697053 MF |
332 | mlog(0, "new descriptor, record %u, at block %llu\n", |
333 | alloc_rec, (unsigned long long)bg_blkno); | |
ccd979bd MF |
334 | |
335 | bg_bh = sb_getblk(osb->sb, bg_blkno); | |
336 | if (!bg_bh) { | |
337 | status = -EIO; | |
338 | mlog_errno(status); | |
339 | goto bail; | |
340 | } | |
341 | ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh); | |
342 | ||
343 | status = ocfs2_block_group_fill(handle, | |
344 | alloc_inode, | |
345 | bg_bh, | |
346 | bg_blkno, | |
347 | alloc_rec, | |
348 | cl); | |
349 | if (status < 0) { | |
350 | mlog_errno(status); | |
351 | goto bail; | |
352 | } | |
353 | ||
354 | bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
355 | ||
356 | status = ocfs2_journal_access(handle, alloc_inode, | |
357 | bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
358 | if (status < 0) { | |
359 | mlog_errno(status); | |
360 | goto bail; | |
361 | } | |
362 | ||
363 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_free, | |
364 | le16_to_cpu(bg->bg_free_bits_count)); | |
365 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits)); | |
366 | cl->cl_recs[alloc_rec].c_blkno = cpu_to_le64(bg_blkno); | |
367 | if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count)) | |
368 | le16_add_cpu(&cl->cl_next_free_rec, 1); | |
369 | ||
370 | le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) - | |
371 | le16_to_cpu(bg->bg_free_bits_count)); | |
372 | le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits)); | |
373 | le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg)); | |
374 | ||
375 | status = ocfs2_journal_dirty(handle, bh); | |
376 | if (status < 0) { | |
377 | mlog_errno(status); | |
378 | goto bail; | |
379 | } | |
380 | ||
381 | spin_lock(&OCFS2_I(alloc_inode)->ip_lock); | |
382 | OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
383 | fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb, | |
384 | le32_to_cpu(fe->i_clusters))); | |
385 | spin_unlock(&OCFS2_I(alloc_inode)->ip_lock); | |
386 | i_size_write(alloc_inode, le64_to_cpu(fe->i_size)); | |
8110b073 | 387 | alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode); |
ccd979bd MF |
388 | |
389 | status = 0; | |
390 | bail: | |
391 | if (handle) | |
02dc1af4 | 392 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
393 | |
394 | if (ac) | |
395 | ocfs2_free_alloc_context(ac); | |
396 | ||
397 | if (bg_bh) | |
398 | brelse(bg_bh); | |
399 | ||
400 | mlog_exit(status); | |
401 | return status; | |
402 | } | |
403 | ||
404 | static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb, | |
da5cbf2f MF |
405 | struct ocfs2_alloc_context *ac, |
406 | int type, | |
ffda89a3 TM |
407 | u32 slot, |
408 | int alloc_new_group) | |
ccd979bd MF |
409 | { |
410 | int status; | |
411 | u32 bits_wanted = ac->ac_bits_wanted; | |
da5cbf2f | 412 | struct inode *alloc_inode; |
ccd979bd | 413 | struct buffer_head *bh = NULL; |
ccd979bd MF |
414 | struct ocfs2_dinode *fe; |
415 | u32 free_bits; | |
416 | ||
417 | mlog_entry_void(); | |
418 | ||
da5cbf2f MF |
419 | alloc_inode = ocfs2_get_system_file_inode(osb, type, slot); |
420 | if (!alloc_inode) { | |
421 | mlog_errno(-EINVAL); | |
422 | return -EINVAL; | |
423 | } | |
ccd979bd | 424 | |
da5cbf2f MF |
425 | mutex_lock(&alloc_inode->i_mutex); |
426 | ||
e63aecb6 | 427 | status = ocfs2_inode_lock(alloc_inode, &bh, 1); |
ccd979bd | 428 | if (status < 0) { |
da5cbf2f MF |
429 | mutex_unlock(&alloc_inode->i_mutex); |
430 | iput(alloc_inode); | |
431 | ||
ccd979bd | 432 | mlog_errno(status); |
da5cbf2f | 433 | return status; |
ccd979bd MF |
434 | } |
435 | ||
da5cbf2f | 436 | ac->ac_inode = alloc_inode; |
a4a48911 | 437 | ac->ac_alloc_slot = slot; |
da5cbf2f | 438 | |
ccd979bd MF |
439 | fe = (struct ocfs2_dinode *) bh->b_data; |
440 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
441 | OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe); | |
442 | status = -EIO; | |
443 | goto bail; | |
444 | } | |
445 | if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) { | |
b0697053 MF |
446 | ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu", |
447 | (unsigned long long)le64_to_cpu(fe->i_blkno)); | |
ccd979bd MF |
448 | status = -EIO; |
449 | goto bail; | |
450 | } | |
451 | ||
452 | free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) - | |
453 | le32_to_cpu(fe->id1.bitmap1.i_used); | |
454 | ||
455 | if (bits_wanted > free_bits) { | |
456 | /* cluster bitmap never grows */ | |
457 | if (ocfs2_is_cluster_bitmap(alloc_inode)) { | |
458 | mlog(0, "Disk Full: wanted=%u, free_bits=%u\n", | |
459 | bits_wanted, free_bits); | |
460 | status = -ENOSPC; | |
461 | goto bail; | |
462 | } | |
463 | ||
ffda89a3 TM |
464 | if (alloc_new_group != ALLOC_NEW_GROUP) { |
465 | mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, " | |
466 | "and we don't alloc a new group for it.\n", | |
467 | slot, bits_wanted, free_bits); | |
468 | status = -ENOSPC; | |
469 | goto bail; | |
470 | } | |
471 | ||
ccd979bd MF |
472 | status = ocfs2_block_group_alloc(osb, alloc_inode, bh); |
473 | if (status < 0) { | |
474 | if (status != -ENOSPC) | |
475 | mlog_errno(status); | |
476 | goto bail; | |
477 | } | |
478 | atomic_inc(&osb->alloc_stats.bg_extends); | |
479 | ||
480 | /* You should never ask for this much metadata */ | |
481 | BUG_ON(bits_wanted > | |
482 | (le32_to_cpu(fe->id1.bitmap1.i_total) | |
483 | - le32_to_cpu(fe->id1.bitmap1.i_used))); | |
484 | } | |
485 | ||
486 | get_bh(bh); | |
487 | ac->ac_bh = bh; | |
488 | bail: | |
489 | if (bh) | |
490 | brelse(bh); | |
491 | ||
492 | mlog_exit(status); | |
493 | return status; | |
494 | } | |
495 | ||
496 | int ocfs2_reserve_new_metadata(struct ocfs2_super *osb, | |
811f933d | 497 | struct ocfs2_extent_list *root_el, |
ccd979bd MF |
498 | struct ocfs2_alloc_context **ac) |
499 | { | |
500 | int status; | |
da5cbf2f | 501 | u32 slot; |
ccd979bd | 502 | |
cd861280 | 503 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
504 | if (!(*ac)) { |
505 | status = -ENOMEM; | |
506 | mlog_errno(status); | |
507 | goto bail; | |
508 | } | |
509 | ||
811f933d | 510 | (*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(root_el); |
ccd979bd | 511 | (*ac)->ac_which = OCFS2_AC_USE_META; |
da5cbf2f | 512 | slot = osb->slot_num; |
ccd979bd MF |
513 | (*ac)->ac_group_search = ocfs2_block_group_search; |
514 | ||
da5cbf2f | 515 | status = ocfs2_reserve_suballoc_bits(osb, (*ac), |
ffda89a3 TM |
516 | EXTENT_ALLOC_SYSTEM_INODE, |
517 | slot, ALLOC_NEW_GROUP); | |
ccd979bd MF |
518 | if (status < 0) { |
519 | if (status != -ENOSPC) | |
520 | mlog_errno(status); | |
521 | goto bail; | |
522 | } | |
523 | ||
524 | status = 0; | |
525 | bail: | |
526 | if ((status < 0) && *ac) { | |
527 | ocfs2_free_alloc_context(*ac); | |
528 | *ac = NULL; | |
529 | } | |
530 | ||
ccd979bd MF |
531 | mlog_exit(status); |
532 | return status; | |
533 | } | |
534 | ||
4d0ddb2c TM |
535 | static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb, |
536 | struct ocfs2_alloc_context *ac) | |
537 | { | |
538 | int i, status = -ENOSPC; | |
539 | s16 slot = ocfs2_get_inode_steal_slot(osb); | |
540 | ||
541 | /* Start to steal inodes from the first slot after ours. */ | |
542 | if (slot == OCFS2_INVALID_SLOT) | |
543 | slot = osb->slot_num + 1; | |
544 | ||
545 | for (i = 0; i < osb->max_slots; i++, slot++) { | |
546 | if (slot == osb->max_slots) | |
547 | slot = 0; | |
548 | ||
549 | if (slot == osb->slot_num) | |
550 | continue; | |
551 | ||
552 | status = ocfs2_reserve_suballoc_bits(osb, ac, | |
553 | INODE_ALLOC_SYSTEM_INODE, | |
554 | slot, NOT_ALLOC_NEW_GROUP); | |
555 | if (status >= 0) { | |
556 | ocfs2_set_inode_steal_slot(osb, slot); | |
557 | break; | |
558 | } | |
559 | ||
560 | ocfs2_free_ac_resource(ac); | |
561 | } | |
562 | ||
563 | return status; | |
564 | } | |
565 | ||
ccd979bd | 566 | int ocfs2_reserve_new_inode(struct ocfs2_super *osb, |
ccd979bd MF |
567 | struct ocfs2_alloc_context **ac) |
568 | { | |
569 | int status; | |
4d0ddb2c | 570 | s16 slot = ocfs2_get_inode_steal_slot(osb); |
ccd979bd | 571 | |
cd861280 | 572 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
573 | if (!(*ac)) { |
574 | status = -ENOMEM; | |
575 | mlog_errno(status); | |
576 | goto bail; | |
577 | } | |
578 | ||
579 | (*ac)->ac_bits_wanted = 1; | |
ccd979bd MF |
580 | (*ac)->ac_which = OCFS2_AC_USE_INODE; |
581 | ||
ccd979bd MF |
582 | (*ac)->ac_group_search = ocfs2_block_group_search; |
583 | ||
4d0ddb2c TM |
584 | /* |
585 | * slot is set when we successfully steal inode from other nodes. | |
586 | * It is reset in 3 places: | |
587 | * 1. when we flush the truncate log | |
588 | * 2. when we complete local alloc recovery. | |
589 | * 3. when we successfully allocate from our own slot. | |
590 | * After it is set, we will go on stealing inodes until we find the | |
591 | * need to check our slots to see whether there is some space for us. | |
592 | */ | |
593 | if (slot != OCFS2_INVALID_SLOT && | |
594 | atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL) | |
595 | goto inode_steal; | |
596 | ||
597 | atomic_set(&osb->s_num_inodes_stolen, 0); | |
da5cbf2f MF |
598 | status = ocfs2_reserve_suballoc_bits(osb, *ac, |
599 | INODE_ALLOC_SYSTEM_INODE, | |
ffda89a3 | 600 | osb->slot_num, ALLOC_NEW_GROUP); |
4d0ddb2c TM |
601 | if (status >= 0) { |
602 | status = 0; | |
603 | ||
604 | /* | |
605 | * Some inodes must be freed by us, so try to allocate | |
606 | * from our own next time. | |
607 | */ | |
608 | if (slot != OCFS2_INVALID_SLOT) | |
609 | ocfs2_init_inode_steal_slot(osb); | |
610 | goto bail; | |
611 | } else if (status < 0 && status != -ENOSPC) { | |
612 | mlog_errno(status); | |
613 | goto bail; | |
614 | } | |
615 | ||
616 | ocfs2_free_ac_resource(*ac); | |
617 | ||
618 | inode_steal: | |
619 | status = ocfs2_steal_inode_from_other_nodes(osb, *ac); | |
620 | atomic_inc(&osb->s_num_inodes_stolen); | |
ccd979bd MF |
621 | if (status < 0) { |
622 | if (status != -ENOSPC) | |
623 | mlog_errno(status); | |
624 | goto bail; | |
625 | } | |
626 | ||
627 | status = 0; | |
628 | bail: | |
629 | if ((status < 0) && *ac) { | |
630 | ocfs2_free_alloc_context(*ac); | |
631 | *ac = NULL; | |
632 | } | |
633 | ||
ccd979bd MF |
634 | mlog_exit(status); |
635 | return status; | |
636 | } | |
637 | ||
638 | /* local alloc code has to do the same thing, so rather than do this | |
639 | * twice.. */ | |
640 | int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb, | |
641 | struct ocfs2_alloc_context *ac) | |
642 | { | |
643 | int status; | |
644 | ||
ccd979bd MF |
645 | ac->ac_which = OCFS2_AC_USE_MAIN; |
646 | ac->ac_group_search = ocfs2_cluster_group_search; | |
647 | ||
da5cbf2f MF |
648 | status = ocfs2_reserve_suballoc_bits(osb, ac, |
649 | GLOBAL_BITMAP_SYSTEM_INODE, | |
ffda89a3 TM |
650 | OCFS2_INVALID_SLOT, |
651 | ALLOC_NEW_GROUP); | |
da5cbf2f | 652 | if (status < 0 && status != -ENOSPC) { |
ccd979bd | 653 | mlog_errno(status); |
da5cbf2f MF |
654 | goto bail; |
655 | } | |
656 | ||
ccd979bd MF |
657 | bail: |
658 | return status; | |
659 | } | |
660 | ||
661 | /* Callers don't need to care which bitmap (local alloc or main) to | |
662 | * use so we figure it out for them, but unfortunately this clutters | |
663 | * things a bit. */ | |
664 | int ocfs2_reserve_clusters(struct ocfs2_super *osb, | |
ccd979bd MF |
665 | u32 bits_wanted, |
666 | struct ocfs2_alloc_context **ac) | |
667 | { | |
668 | int status; | |
669 | ||
670 | mlog_entry_void(); | |
671 | ||
cd861280 | 672 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
673 | if (!(*ac)) { |
674 | status = -ENOMEM; | |
675 | mlog_errno(status); | |
676 | goto bail; | |
677 | } | |
678 | ||
679 | (*ac)->ac_bits_wanted = bits_wanted; | |
ccd979bd MF |
680 | |
681 | status = -ENOSPC; | |
682 | if (ocfs2_alloc_should_use_local(osb, bits_wanted)) { | |
683 | status = ocfs2_reserve_local_alloc_bits(osb, | |
ccd979bd MF |
684 | bits_wanted, |
685 | *ac); | |
686 | if ((status < 0) && (status != -ENOSPC)) { | |
687 | mlog_errno(status); | |
688 | goto bail; | |
ccd979bd MF |
689 | } |
690 | } | |
691 | ||
692 | if (status == -ENOSPC) { | |
693 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); | |
694 | if (status < 0) { | |
695 | if (status != -ENOSPC) | |
696 | mlog_errno(status); | |
697 | goto bail; | |
698 | } | |
699 | } | |
700 | ||
701 | status = 0; | |
702 | bail: | |
703 | if ((status < 0) && *ac) { | |
704 | ocfs2_free_alloc_context(*ac); | |
705 | *ac = NULL; | |
706 | } | |
707 | ||
708 | mlog_exit(status); | |
709 | return status; | |
710 | } | |
711 | ||
712 | /* | |
713 | * More or less lifted from ext3. I'll leave their description below: | |
714 | * | |
715 | * "For ext3 allocations, we must not reuse any blocks which are | |
716 | * allocated in the bitmap buffer's "last committed data" copy. This | |
717 | * prevents deletes from freeing up the page for reuse until we have | |
718 | * committed the delete transaction. | |
719 | * | |
720 | * If we didn't do this, then deleting something and reallocating it as | |
721 | * data would allow the old block to be overwritten before the | |
722 | * transaction committed (because we force data to disk before commit). | |
723 | * This would lead to corruption if we crashed between overwriting the | |
724 | * data and committing the delete. | |
725 | * | |
726 | * @@@ We may want to make this allocation behaviour conditional on | |
727 | * data-writes at some point, and disable it for metadata allocations or | |
728 | * sync-data inodes." | |
729 | * | |
730 | * Note: OCFS2 already does this differently for metadata vs data | |
c78bad11 | 731 | * allocations, as those bitmaps are separate and undo access is never |
ccd979bd MF |
732 | * called on a metadata group descriptor. |
733 | */ | |
734 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh, | |
735 | int nr) | |
736 | { | |
737 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
738 | ||
739 | if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap)) | |
740 | return 0; | |
741 | if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data) | |
742 | return 1; | |
743 | ||
744 | bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data; | |
745 | return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap); | |
746 | } | |
747 | ||
748 | static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb, | |
749 | struct buffer_head *bg_bh, | |
750 | unsigned int bits_wanted, | |
7bf72ede | 751 | unsigned int total_bits, |
ccd979bd MF |
752 | u16 *bit_off, |
753 | u16 *bits_found) | |
754 | { | |
755 | void *bitmap; | |
756 | u16 best_offset, best_size; | |
757 | int offset, start, found, status = 0; | |
758 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
759 | ||
760 | if (!OCFS2_IS_VALID_GROUP_DESC(bg)) { | |
761 | OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg); | |
762 | return -EIO; | |
763 | } | |
764 | ||
765 | found = start = best_offset = best_size = 0; | |
766 | bitmap = bg->bg_bitmap; | |
767 | ||
7bf72ede MF |
768 | while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) { |
769 | if (offset == total_bits) | |
ccd979bd MF |
770 | break; |
771 | ||
772 | if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) { | |
773 | /* We found a zero, but we can't use it as it | |
774 | * hasn't been put to disk yet! */ | |
775 | found = 0; | |
776 | start = offset + 1; | |
777 | } else if (offset == start) { | |
778 | /* we found a zero */ | |
779 | found++; | |
780 | /* move start to the next bit to test */ | |
781 | start++; | |
782 | } else { | |
783 | /* got a zero after some ones */ | |
784 | found = 1; | |
785 | start = offset + 1; | |
786 | } | |
787 | if (found > best_size) { | |
788 | best_size = found; | |
789 | best_offset = start - found; | |
790 | } | |
791 | /* we got everything we needed */ | |
792 | if (found == bits_wanted) { | |
793 | /* mlog(0, "Found it all!\n"); */ | |
794 | break; | |
795 | } | |
796 | } | |
797 | ||
798 | /* XXX: I think the first clause is equivalent to the second | |
799 | * - jlbec */ | |
800 | if (found == bits_wanted) { | |
801 | *bit_off = start - found; | |
802 | *bits_found = found; | |
803 | } else if (best_size) { | |
804 | *bit_off = best_offset; | |
805 | *bits_found = best_size; | |
806 | } else { | |
807 | status = -ENOSPC; | |
808 | /* No error log here -- see the comment above | |
809 | * ocfs2_test_bg_bit_allocatable */ | |
810 | } | |
811 | ||
812 | return status; | |
813 | } | |
814 | ||
1fabe148 | 815 | static inline int ocfs2_block_group_set_bits(handle_t *handle, |
ccd979bd MF |
816 | struct inode *alloc_inode, |
817 | struct ocfs2_group_desc *bg, | |
818 | struct buffer_head *group_bh, | |
819 | unsigned int bit_off, | |
820 | unsigned int num_bits) | |
821 | { | |
822 | int status; | |
823 | void *bitmap = bg->bg_bitmap; | |
824 | int journal_type = OCFS2_JOURNAL_ACCESS_WRITE; | |
825 | ||
826 | mlog_entry_void(); | |
827 | ||
828 | if (!OCFS2_IS_VALID_GROUP_DESC(bg)) { | |
829 | OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg); | |
830 | status = -EIO; | |
831 | goto bail; | |
832 | } | |
833 | BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits); | |
834 | ||
835 | mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off, | |
836 | num_bits); | |
837 | ||
838 | if (ocfs2_is_cluster_bitmap(alloc_inode)) | |
839 | journal_type = OCFS2_JOURNAL_ACCESS_UNDO; | |
840 | ||
841 | status = ocfs2_journal_access(handle, | |
842 | alloc_inode, | |
843 | group_bh, | |
844 | journal_type); | |
845 | if (status < 0) { | |
846 | mlog_errno(status); | |
847 | goto bail; | |
848 | } | |
849 | ||
850 | le16_add_cpu(&bg->bg_free_bits_count, -num_bits); | |
851 | ||
852 | while(num_bits--) | |
853 | ocfs2_set_bit(bit_off++, bitmap); | |
854 | ||
855 | status = ocfs2_journal_dirty(handle, | |
856 | group_bh); | |
857 | if (status < 0) { | |
858 | mlog_errno(status); | |
859 | goto bail; | |
860 | } | |
861 | ||
862 | bail: | |
863 | mlog_exit(status); | |
864 | return status; | |
865 | } | |
866 | ||
867 | /* find the one with the most empty bits */ | |
868 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl) | |
869 | { | |
870 | u16 curr, best; | |
871 | ||
872 | BUG_ON(!cl->cl_next_free_rec); | |
873 | ||
874 | best = curr = 0; | |
875 | while (curr < le16_to_cpu(cl->cl_next_free_rec)) { | |
876 | if (le32_to_cpu(cl->cl_recs[curr].c_free) > | |
877 | le32_to_cpu(cl->cl_recs[best].c_free)) | |
878 | best = curr; | |
879 | curr++; | |
880 | } | |
881 | ||
882 | BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec)); | |
883 | return best; | |
884 | } | |
885 | ||
1fabe148 | 886 | static int ocfs2_relink_block_group(handle_t *handle, |
ccd979bd MF |
887 | struct inode *alloc_inode, |
888 | struct buffer_head *fe_bh, | |
889 | struct buffer_head *bg_bh, | |
890 | struct buffer_head *prev_bg_bh, | |
891 | u16 chain) | |
892 | { | |
893 | int status; | |
894 | /* there is a really tiny chance the journal calls could fail, | |
895 | * but we wouldn't want inconsistent blocks in *any* case. */ | |
896 | u64 fe_ptr, bg_ptr, prev_bg_ptr; | |
897 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
898 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
899 | struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data; | |
900 | ||
901 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
902 | OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe); | |
903 | status = -EIO; | |
904 | goto out; | |
905 | } | |
906 | if (!OCFS2_IS_VALID_GROUP_DESC(bg)) { | |
907 | OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg); | |
908 | status = -EIO; | |
909 | goto out; | |
910 | } | |
911 | if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) { | |
912 | OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg); | |
913 | status = -EIO; | |
914 | goto out; | |
915 | } | |
916 | ||
b0697053 | 917 | mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n", |
1ca1a111 MF |
918 | (unsigned long long)le64_to_cpu(fe->i_blkno), chain, |
919 | (unsigned long long)le64_to_cpu(bg->bg_blkno), | |
920 | (unsigned long long)le64_to_cpu(prev_bg->bg_blkno)); | |
ccd979bd MF |
921 | |
922 | fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno); | |
923 | bg_ptr = le64_to_cpu(bg->bg_next_group); | |
924 | prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group); | |
925 | ||
926 | status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh, | |
927 | OCFS2_JOURNAL_ACCESS_WRITE); | |
928 | if (status < 0) { | |
929 | mlog_errno(status); | |
930 | goto out_rollback; | |
931 | } | |
932 | ||
933 | prev_bg->bg_next_group = bg->bg_next_group; | |
934 | ||
935 | status = ocfs2_journal_dirty(handle, prev_bg_bh); | |
936 | if (status < 0) { | |
937 | mlog_errno(status); | |
938 | goto out_rollback; | |
939 | } | |
940 | ||
941 | status = ocfs2_journal_access(handle, alloc_inode, bg_bh, | |
942 | OCFS2_JOURNAL_ACCESS_WRITE); | |
943 | if (status < 0) { | |
944 | mlog_errno(status); | |
945 | goto out_rollback; | |
946 | } | |
947 | ||
948 | bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno; | |
949 | ||
950 | status = ocfs2_journal_dirty(handle, bg_bh); | |
951 | if (status < 0) { | |
952 | mlog_errno(status); | |
953 | goto out_rollback; | |
954 | } | |
955 | ||
956 | status = ocfs2_journal_access(handle, alloc_inode, fe_bh, | |
957 | OCFS2_JOURNAL_ACCESS_WRITE); | |
958 | if (status < 0) { | |
959 | mlog_errno(status); | |
960 | goto out_rollback; | |
961 | } | |
962 | ||
963 | fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno; | |
964 | ||
965 | status = ocfs2_journal_dirty(handle, fe_bh); | |
966 | if (status < 0) { | |
967 | mlog_errno(status); | |
968 | goto out_rollback; | |
969 | } | |
970 | ||
971 | status = 0; | |
972 | out_rollback: | |
973 | if (status < 0) { | |
974 | fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr); | |
975 | bg->bg_next_group = cpu_to_le64(bg_ptr); | |
976 | prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr); | |
977 | } | |
978 | out: | |
979 | mlog_exit(status); | |
980 | return status; | |
981 | } | |
982 | ||
983 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg, | |
984 | u32 wanted) | |
985 | { | |
986 | return le16_to_cpu(bg->bg_free_bits_count) > wanted; | |
987 | } | |
988 | ||
989 | /* return 0 on success, -ENOSPC to keep searching and any other < 0 | |
990 | * value on error. */ | |
991 | static int ocfs2_cluster_group_search(struct inode *inode, | |
992 | struct buffer_head *group_bh, | |
993 | u32 bits_wanted, u32 min_bits, | |
994 | u16 *bit_off, u16 *bits_found) | |
995 | { | |
996 | int search = -ENOSPC; | |
997 | int ret; | |
7bf72ede | 998 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data; |
9c7af40b | 999 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
ccd979bd | 1000 | u16 tmp_off, tmp_found; |
7bf72ede | 1001 | unsigned int max_bits, gd_cluster_off; |
ccd979bd MF |
1002 | |
1003 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
1004 | ||
7bf72ede MF |
1005 | if (gd->bg_free_bits_count) { |
1006 | max_bits = le16_to_cpu(gd->bg_bits); | |
1007 | ||
1008 | /* Tail groups in cluster bitmaps which aren't cpg | |
1009 | * aligned are prone to partial extention by a failed | |
1010 | * fs resize. If the file system resize never got to | |
1011 | * update the dinode cluster count, then we don't want | |
1012 | * to trust any clusters past it, regardless of what | |
1013 | * the group descriptor says. */ | |
1014 | gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb, | |
1015 | le64_to_cpu(gd->bg_blkno)); | |
1016 | if ((gd_cluster_off + max_bits) > | |
1017 | OCFS2_I(inode)->ip_clusters) { | |
1018 | max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off; | |
1019 | mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n", | |
1020 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
1021 | le16_to_cpu(gd->bg_bits), | |
1022 | OCFS2_I(inode)->ip_clusters, max_bits); | |
1023 | } | |
1024 | ||
ccd979bd MF |
1025 | ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb), |
1026 | group_bh, bits_wanted, | |
7bf72ede | 1027 | max_bits, |
ccd979bd MF |
1028 | &tmp_off, &tmp_found); |
1029 | if (ret) | |
1030 | return ret; | |
1031 | ||
1032 | /* ocfs2_block_group_find_clear_bits() might | |
1033 | * return success, but we still want to return | |
1034 | * -ENOSPC unless it found the minimum number | |
1035 | * of bits. */ | |
1036 | if (min_bits <= tmp_found) { | |
1037 | *bit_off = tmp_off; | |
1038 | *bits_found = tmp_found; | |
1039 | search = 0; /* success */ | |
9c7af40b MF |
1040 | } else if (tmp_found) { |
1041 | /* | |
1042 | * Don't show bits which we'll be returning | |
1043 | * for allocation to the local alloc bitmap. | |
1044 | */ | |
1045 | ocfs2_local_alloc_seen_free_bits(osb, tmp_found); | |
ccd979bd MF |
1046 | } |
1047 | } | |
1048 | ||
1049 | return search; | |
1050 | } | |
1051 | ||
1052 | static int ocfs2_block_group_search(struct inode *inode, | |
1053 | struct buffer_head *group_bh, | |
1054 | u32 bits_wanted, u32 min_bits, | |
1055 | u16 *bit_off, u16 *bits_found) | |
1056 | { | |
1057 | int ret = -ENOSPC; | |
1058 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data; | |
1059 | ||
1060 | BUG_ON(min_bits != 1); | |
1061 | BUG_ON(ocfs2_is_cluster_bitmap(inode)); | |
1062 | ||
1063 | if (bg->bg_free_bits_count) | |
1064 | ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb), | |
1065 | group_bh, bits_wanted, | |
7bf72ede | 1066 | le16_to_cpu(bg->bg_bits), |
ccd979bd MF |
1067 | bit_off, bits_found); |
1068 | ||
1069 | return ret; | |
1070 | } | |
1071 | ||
883d4cae | 1072 | static int ocfs2_alloc_dinode_update_counts(struct inode *inode, |
1fabe148 | 1073 | handle_t *handle, |
883d4cae MF |
1074 | struct buffer_head *di_bh, |
1075 | u32 num_bits, | |
1076 | u16 chain) | |
1077 | { | |
1078 | int ret; | |
1079 | u32 tmp_used; | |
1080 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
1081 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain; | |
1082 | ||
1083 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
1084 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1085 | if (ret < 0) { | |
1086 | mlog_errno(ret); | |
1087 | goto out; | |
1088 | } | |
1089 | ||
1090 | tmp_used = le32_to_cpu(di->id1.bitmap1.i_used); | |
1091 | di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used); | |
1092 | le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits); | |
1093 | ||
1094 | ret = ocfs2_journal_dirty(handle, di_bh); | |
1095 | if (ret < 0) | |
1096 | mlog_errno(ret); | |
1097 | ||
1098 | out: | |
1099 | return ret; | |
1100 | } | |
1101 | ||
1102 | static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac, | |
1fabe148 | 1103 | handle_t *handle, |
883d4cae MF |
1104 | u32 bits_wanted, |
1105 | u32 min_bits, | |
1106 | u16 *bit_off, | |
1107 | unsigned int *num_bits, | |
1108 | u64 gd_blkno, | |
1109 | u16 *bits_left) | |
1110 | { | |
1111 | int ret; | |
1112 | u16 found; | |
1113 | struct buffer_head *group_bh = NULL; | |
1114 | struct ocfs2_group_desc *gd; | |
1115 | struct inode *alloc_inode = ac->ac_inode; | |
883d4cae MF |
1116 | |
1117 | ret = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb), gd_blkno, | |
1118 | &group_bh, OCFS2_BH_CACHED, alloc_inode); | |
1119 | if (ret < 0) { | |
1120 | mlog_errno(ret); | |
1121 | return ret; | |
1122 | } | |
1123 | ||
1124 | gd = (struct ocfs2_group_desc *) group_bh->b_data; | |
1125 | if (!OCFS2_IS_VALID_GROUP_DESC(gd)) { | |
1126 | OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, gd); | |
1127 | ret = -EIO; | |
1128 | goto out; | |
1129 | } | |
1130 | ||
1131 | ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits, | |
1132 | bit_off, &found); | |
1133 | if (ret < 0) { | |
1134 | if (ret != -ENOSPC) | |
1135 | mlog_errno(ret); | |
1136 | goto out; | |
1137 | } | |
1138 | ||
1139 | *num_bits = found; | |
1140 | ||
1141 | ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh, | |
1142 | *num_bits, | |
1143 | le16_to_cpu(gd->bg_chain)); | |
1144 | if (ret < 0) { | |
1145 | mlog_errno(ret); | |
1146 | goto out; | |
1147 | } | |
1148 | ||
1149 | ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh, | |
1150 | *bit_off, *num_bits); | |
1151 | if (ret < 0) | |
1152 | mlog_errno(ret); | |
1153 | ||
1154 | *bits_left = le16_to_cpu(gd->bg_free_bits_count); | |
1155 | ||
1156 | out: | |
1157 | brelse(group_bh); | |
1158 | ||
1159 | return ret; | |
1160 | } | |
1161 | ||
ccd979bd | 1162 | static int ocfs2_search_chain(struct ocfs2_alloc_context *ac, |
1fabe148 | 1163 | handle_t *handle, |
ccd979bd MF |
1164 | u32 bits_wanted, |
1165 | u32 min_bits, | |
1166 | u16 *bit_off, | |
1167 | unsigned int *num_bits, | |
883d4cae MF |
1168 | u64 *bg_blkno, |
1169 | u16 *bits_left) | |
ccd979bd MF |
1170 | { |
1171 | int status; | |
1172 | u16 chain, tmp_bits; | |
1173 | u32 tmp_used; | |
1174 | u64 next_group; | |
ccd979bd MF |
1175 | struct inode *alloc_inode = ac->ac_inode; |
1176 | struct buffer_head *group_bh = NULL; | |
1177 | struct buffer_head *prev_group_bh = NULL; | |
1178 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data; | |
1179 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain; | |
1180 | struct ocfs2_group_desc *bg; | |
1181 | ||
1182 | chain = ac->ac_chain; | |
b0697053 MF |
1183 | mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n", |
1184 | bits_wanted, chain, | |
1185 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno); | |
ccd979bd MF |
1186 | |
1187 | status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb), | |
1188 | le64_to_cpu(cl->cl_recs[chain].c_blkno), | |
1189 | &group_bh, OCFS2_BH_CACHED, alloc_inode); | |
1190 | if (status < 0) { | |
1191 | mlog_errno(status); | |
1192 | goto bail; | |
1193 | } | |
1194 | bg = (struct ocfs2_group_desc *) group_bh->b_data; | |
7bf72ede MF |
1195 | status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg); |
1196 | if (status) { | |
1197 | mlog_errno(status); | |
ccd979bd MF |
1198 | goto bail; |
1199 | } | |
1200 | ||
1201 | status = -ENOSPC; | |
1202 | /* for now, the chain search is a bit simplistic. We just use | |
1203 | * the 1st group with any empty bits. */ | |
9c7af40b MF |
1204 | while ((status = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, |
1205 | min_bits, bit_off, &tmp_bits)) == -ENOSPC) { | |
ccd979bd MF |
1206 | if (!bg->bg_next_group) |
1207 | break; | |
1208 | ||
1209 | if (prev_group_bh) { | |
1210 | brelse(prev_group_bh); | |
1211 | prev_group_bh = NULL; | |
1212 | } | |
1213 | next_group = le64_to_cpu(bg->bg_next_group); | |
1214 | prev_group_bh = group_bh; | |
1215 | group_bh = NULL; | |
1216 | status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb), | |
1217 | next_group, &group_bh, | |
1218 | OCFS2_BH_CACHED, alloc_inode); | |
1219 | if (status < 0) { | |
1220 | mlog_errno(status); | |
1221 | goto bail; | |
1222 | } | |
1223 | bg = (struct ocfs2_group_desc *) group_bh->b_data; | |
7bf72ede MF |
1224 | status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg); |
1225 | if (status) { | |
1226 | mlog_errno(status); | |
ccd979bd MF |
1227 | goto bail; |
1228 | } | |
1229 | } | |
1230 | if (status < 0) { | |
1231 | if (status != -ENOSPC) | |
1232 | mlog_errno(status); | |
1233 | goto bail; | |
1234 | } | |
1235 | ||
b0697053 | 1236 | mlog(0, "alloc succeeds: we give %u bits from block group %llu\n", |
1ca1a111 | 1237 | tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno)); |
ccd979bd MF |
1238 | |
1239 | *num_bits = tmp_bits; | |
1240 | ||
1241 | BUG_ON(*num_bits == 0); | |
1242 | ||
1243 | /* | |
1244 | * Keep track of previous block descriptor read. When | |
1245 | * we find a target, if we have read more than X | |
1246 | * number of descriptors, and the target is reasonably | |
1247 | * empty, relink him to top of his chain. | |
1248 | * | |
1249 | * We've read 0 extra blocks and only send one more to | |
1250 | * the transaction, yet the next guy to search has a | |
1251 | * much easier time. | |
1252 | * | |
1253 | * Do this *after* figuring out how many bits we're taking out | |
1254 | * of our target group. | |
1255 | */ | |
1256 | if (ac->ac_allow_chain_relink && | |
1257 | (prev_group_bh) && | |
1258 | (ocfs2_block_group_reasonably_empty(bg, *num_bits))) { | |
1259 | status = ocfs2_relink_block_group(handle, alloc_inode, | |
1260 | ac->ac_bh, group_bh, | |
1261 | prev_group_bh, chain); | |
1262 | if (status < 0) { | |
1263 | mlog_errno(status); | |
1264 | goto bail; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | /* Ok, claim our bits now: set the info on dinode, chainlist | |
1269 | * and then the group */ | |
1270 | status = ocfs2_journal_access(handle, | |
1271 | alloc_inode, | |
1272 | ac->ac_bh, | |
1273 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1274 | if (status < 0) { | |
1275 | mlog_errno(status); | |
1276 | goto bail; | |
1277 | } | |
1278 | ||
1279 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used); | |
1280 | fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used); | |
1281 | le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits)); | |
1282 | ||
1283 | status = ocfs2_journal_dirty(handle, | |
1284 | ac->ac_bh); | |
1285 | if (status < 0) { | |
1286 | mlog_errno(status); | |
1287 | goto bail; | |
1288 | } | |
1289 | ||
1290 | status = ocfs2_block_group_set_bits(handle, | |
1291 | alloc_inode, | |
1292 | bg, | |
1293 | group_bh, | |
1294 | *bit_off, | |
1295 | *num_bits); | |
1296 | if (status < 0) { | |
1297 | mlog_errno(status); | |
1298 | goto bail; | |
1299 | } | |
1300 | ||
b0697053 | 1301 | mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits, |
1ca1a111 | 1302 | (unsigned long long)le64_to_cpu(fe->i_blkno)); |
ccd979bd MF |
1303 | |
1304 | *bg_blkno = le64_to_cpu(bg->bg_blkno); | |
883d4cae | 1305 | *bits_left = le16_to_cpu(bg->bg_free_bits_count); |
ccd979bd MF |
1306 | bail: |
1307 | if (group_bh) | |
1308 | brelse(group_bh); | |
1309 | if (prev_group_bh) | |
1310 | brelse(prev_group_bh); | |
1311 | ||
1312 | mlog_exit(status); | |
1313 | return status; | |
1314 | } | |
1315 | ||
1316 | /* will give out up to bits_wanted contiguous bits. */ | |
1317 | static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb, | |
1318 | struct ocfs2_alloc_context *ac, | |
1fabe148 | 1319 | handle_t *handle, |
ccd979bd MF |
1320 | u32 bits_wanted, |
1321 | u32 min_bits, | |
1322 | u16 *bit_off, | |
1323 | unsigned int *num_bits, | |
1324 | u64 *bg_blkno) | |
1325 | { | |
1326 | int status; | |
1327 | u16 victim, i; | |
883d4cae MF |
1328 | u16 bits_left = 0; |
1329 | u64 hint_blkno = ac->ac_last_group; | |
ccd979bd MF |
1330 | struct ocfs2_chain_list *cl; |
1331 | struct ocfs2_dinode *fe; | |
1332 | ||
1333 | mlog_entry_void(); | |
1334 | ||
1335 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted); | |
1336 | BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given)); | |
1337 | BUG_ON(!ac->ac_bh); | |
1338 | ||
1339 | fe = (struct ocfs2_dinode *) ac->ac_bh->b_data; | |
1340 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
1341 | OCFS2_RO_ON_INVALID_DINODE(osb->sb, fe); | |
1342 | status = -EIO; | |
1343 | goto bail; | |
1344 | } | |
1345 | if (le32_to_cpu(fe->id1.bitmap1.i_used) >= | |
1346 | le32_to_cpu(fe->id1.bitmap1.i_total)) { | |
b0697053 MF |
1347 | ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used " |
1348 | "bits but only %u total.", | |
1349 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
ccd979bd MF |
1350 | le32_to_cpu(fe->id1.bitmap1.i_used), |
1351 | le32_to_cpu(fe->id1.bitmap1.i_total)); | |
1352 | status = -EIO; | |
1353 | goto bail; | |
1354 | } | |
1355 | ||
883d4cae MF |
1356 | if (hint_blkno) { |
1357 | /* Attempt to short-circuit the usual search mechanism | |
1358 | * by jumping straight to the most recently used | |
1359 | * allocation group. This helps us mantain some | |
1360 | * contiguousness across allocations. */ | |
da5cbf2f MF |
1361 | status = ocfs2_search_one_group(ac, handle, bits_wanted, |
1362 | min_bits, bit_off, num_bits, | |
883d4cae MF |
1363 | hint_blkno, &bits_left); |
1364 | if (!status) { | |
1365 | /* Be careful to update *bg_blkno here as the | |
1366 | * caller is expecting it to be filled in, and | |
1367 | * ocfs2_search_one_group() won't do that for | |
1368 | * us. */ | |
1369 | *bg_blkno = hint_blkno; | |
1370 | goto set_hint; | |
1371 | } | |
1372 | if (status < 0 && status != -ENOSPC) { | |
1373 | mlog_errno(status); | |
1374 | goto bail; | |
1375 | } | |
1376 | } | |
1377 | ||
ccd979bd MF |
1378 | cl = (struct ocfs2_chain_list *) &fe->id2.i_chain; |
1379 | ||
1380 | victim = ocfs2_find_victim_chain(cl); | |
1381 | ac->ac_chain = victim; | |
1382 | ac->ac_allow_chain_relink = 1; | |
1383 | ||
da5cbf2f | 1384 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off, |
883d4cae | 1385 | num_bits, bg_blkno, &bits_left); |
ccd979bd | 1386 | if (!status) |
883d4cae | 1387 | goto set_hint; |
ccd979bd MF |
1388 | if (status < 0 && status != -ENOSPC) { |
1389 | mlog_errno(status); | |
1390 | goto bail; | |
1391 | } | |
1392 | ||
1393 | mlog(0, "Search of victim chain %u came up with nothing, " | |
1394 | "trying all chains now.\n", victim); | |
1395 | ||
1396 | /* If we didn't pick a good victim, then just default to | |
1397 | * searching each chain in order. Don't allow chain relinking | |
1398 | * because we only calculate enough journal credits for one | |
1399 | * relink per alloc. */ | |
1400 | ac->ac_allow_chain_relink = 0; | |
1401 | for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) { | |
1402 | if (i == victim) | |
1403 | continue; | |
1404 | if (!cl->cl_recs[i].c_free) | |
1405 | continue; | |
1406 | ||
1407 | ac->ac_chain = i; | |
da5cbf2f | 1408 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, |
883d4cae MF |
1409 | bit_off, num_bits, bg_blkno, |
1410 | &bits_left); | |
ccd979bd MF |
1411 | if (!status) |
1412 | break; | |
1413 | if (status < 0 && status != -ENOSPC) { | |
1414 | mlog_errno(status); | |
1415 | goto bail; | |
1416 | } | |
1417 | } | |
ccd979bd | 1418 | |
883d4cae MF |
1419 | set_hint: |
1420 | if (status != -ENOSPC) { | |
1421 | /* If the next search of this group is not likely to | |
1422 | * yield a suitable extent, then we reset the last | |
1423 | * group hint so as to not waste a disk read */ | |
1424 | if (bits_left < min_bits) | |
1425 | ac->ac_last_group = 0; | |
1426 | else | |
1427 | ac->ac_last_group = *bg_blkno; | |
1428 | } | |
1429 | ||
1430 | bail: | |
ccd979bd MF |
1431 | mlog_exit(status); |
1432 | return status; | |
1433 | } | |
1434 | ||
1435 | int ocfs2_claim_metadata(struct ocfs2_super *osb, | |
1fabe148 | 1436 | handle_t *handle, |
ccd979bd MF |
1437 | struct ocfs2_alloc_context *ac, |
1438 | u32 bits_wanted, | |
1439 | u16 *suballoc_bit_start, | |
1440 | unsigned int *num_bits, | |
1441 | u64 *blkno_start) | |
1442 | { | |
1443 | int status; | |
1444 | u64 bg_blkno; | |
1445 | ||
1446 | BUG_ON(!ac); | |
1447 | BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted)); | |
1448 | BUG_ON(ac->ac_which != OCFS2_AC_USE_META); | |
ccd979bd MF |
1449 | |
1450 | status = ocfs2_claim_suballoc_bits(osb, | |
1451 | ac, | |
da5cbf2f | 1452 | handle, |
ccd979bd MF |
1453 | bits_wanted, |
1454 | 1, | |
1455 | suballoc_bit_start, | |
1456 | num_bits, | |
1457 | &bg_blkno); | |
1458 | if (status < 0) { | |
1459 | mlog_errno(status); | |
1460 | goto bail; | |
1461 | } | |
1462 | atomic_inc(&osb->alloc_stats.bg_allocs); | |
1463 | ||
1464 | *blkno_start = bg_blkno + (u64) *suballoc_bit_start; | |
1465 | ac->ac_bits_given += (*num_bits); | |
1466 | status = 0; | |
1467 | bail: | |
1468 | mlog_exit(status); | |
1469 | return status; | |
1470 | } | |
1471 | ||
1472 | int ocfs2_claim_new_inode(struct ocfs2_super *osb, | |
1fabe148 | 1473 | handle_t *handle, |
ccd979bd MF |
1474 | struct ocfs2_alloc_context *ac, |
1475 | u16 *suballoc_bit, | |
1476 | u64 *fe_blkno) | |
1477 | { | |
1478 | int status; | |
1479 | unsigned int num_bits; | |
1480 | u64 bg_blkno; | |
1481 | ||
1482 | mlog_entry_void(); | |
1483 | ||
1484 | BUG_ON(!ac); | |
1485 | BUG_ON(ac->ac_bits_given != 0); | |
1486 | BUG_ON(ac->ac_bits_wanted != 1); | |
1487 | BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE); | |
ccd979bd MF |
1488 | |
1489 | status = ocfs2_claim_suballoc_bits(osb, | |
1490 | ac, | |
da5cbf2f | 1491 | handle, |
ccd979bd MF |
1492 | 1, |
1493 | 1, | |
1494 | suballoc_bit, | |
1495 | &num_bits, | |
1496 | &bg_blkno); | |
1497 | if (status < 0) { | |
1498 | mlog_errno(status); | |
1499 | goto bail; | |
1500 | } | |
1501 | atomic_inc(&osb->alloc_stats.bg_allocs); | |
1502 | ||
1503 | BUG_ON(num_bits != 1); | |
1504 | ||
1505 | *fe_blkno = bg_blkno + (u64) (*suballoc_bit); | |
1506 | ac->ac_bits_given++; | |
1507 | status = 0; | |
1508 | bail: | |
1509 | mlog_exit(status); | |
1510 | return status; | |
1511 | } | |
1512 | ||
1513 | /* translate a group desc. blkno and it's bitmap offset into | |
1514 | * disk cluster offset. */ | |
1515 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode, | |
1516 | u64 bg_blkno, | |
1517 | u16 bg_bit_off) | |
1518 | { | |
1519 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1520 | u32 cluster = 0; | |
1521 | ||
1522 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
1523 | ||
1524 | if (bg_blkno != osb->first_cluster_group_blkno) | |
1525 | cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno); | |
1526 | cluster += (u32) bg_bit_off; | |
1527 | return cluster; | |
1528 | } | |
1529 | ||
1530 | /* given a cluster offset, calculate which block group it belongs to | |
1531 | * and return that block offset. */ | |
d659072f | 1532 | u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster) |
ccd979bd MF |
1533 | { |
1534 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1535 | u32 group_no; | |
1536 | ||
1537 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
1538 | ||
1539 | group_no = cluster / osb->bitmap_cpg; | |
1540 | if (!group_no) | |
1541 | return osb->first_cluster_group_blkno; | |
1542 | return ocfs2_clusters_to_blocks(inode->i_sb, | |
1543 | group_no * osb->bitmap_cpg); | |
1544 | } | |
1545 | ||
1546 | /* given the block number of a cluster start, calculate which cluster | |
1547 | * group and descriptor bitmap offset that corresponds to. */ | |
1548 | static inline void ocfs2_block_to_cluster_group(struct inode *inode, | |
1549 | u64 data_blkno, | |
1550 | u64 *bg_blkno, | |
1551 | u16 *bg_bit_off) | |
1552 | { | |
1553 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1554 | u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno); | |
1555 | ||
1556 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
1557 | ||
1558 | *bg_blkno = ocfs2_which_cluster_group(inode, | |
1559 | data_cluster); | |
1560 | ||
1561 | if (*bg_blkno == osb->first_cluster_group_blkno) | |
1562 | *bg_bit_off = (u16) data_cluster; | |
1563 | else | |
1564 | *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb, | |
1565 | data_blkno - *bg_blkno); | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * min_bits - minimum contiguous chunk from this total allocation we | |
1570 | * can handle. set to what we asked for originally for a full | |
1571 | * contig. allocation, set to '1' to indicate we can deal with extents | |
1572 | * of any size. | |
1573 | */ | |
415cb800 MF |
1574 | int __ocfs2_claim_clusters(struct ocfs2_super *osb, |
1575 | handle_t *handle, | |
1576 | struct ocfs2_alloc_context *ac, | |
1577 | u32 min_clusters, | |
1578 | u32 max_clusters, | |
1579 | u32 *cluster_start, | |
1580 | u32 *num_clusters) | |
ccd979bd MF |
1581 | { |
1582 | int status; | |
415cb800 | 1583 | unsigned int bits_wanted = max_clusters; |
883d4cae | 1584 | u64 bg_blkno = 0; |
ccd979bd MF |
1585 | u16 bg_bit_off; |
1586 | ||
1587 | mlog_entry_void(); | |
1588 | ||
ccd979bd MF |
1589 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted); |
1590 | ||
1591 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL | |
1592 | && ac->ac_which != OCFS2_AC_USE_MAIN); | |
ccd979bd MF |
1593 | |
1594 | if (ac->ac_which == OCFS2_AC_USE_LOCAL) { | |
1595 | status = ocfs2_claim_local_alloc_bits(osb, | |
1596 | handle, | |
1597 | ac, | |
1598 | bits_wanted, | |
1599 | cluster_start, | |
1600 | num_clusters); | |
1601 | if (!status) | |
1602 | atomic_inc(&osb->alloc_stats.local_data); | |
1603 | } else { | |
1604 | if (min_clusters > (osb->bitmap_cpg - 1)) { | |
1605 | /* The only paths asking for contiguousness | |
1606 | * should know about this already. */ | |
2fbe8d1e SM |
1607 | mlog(ML_ERROR, "minimum allocation requested %u exceeds " |
1608 | "group bitmap size %u!\n", min_clusters, | |
1609 | osb->bitmap_cpg); | |
ccd979bd MF |
1610 | status = -ENOSPC; |
1611 | goto bail; | |
1612 | } | |
1613 | /* clamp the current request down to a realistic size. */ | |
1614 | if (bits_wanted > (osb->bitmap_cpg - 1)) | |
1615 | bits_wanted = osb->bitmap_cpg - 1; | |
1616 | ||
1617 | status = ocfs2_claim_suballoc_bits(osb, | |
1618 | ac, | |
da5cbf2f | 1619 | handle, |
ccd979bd MF |
1620 | bits_wanted, |
1621 | min_clusters, | |
1622 | &bg_bit_off, | |
1623 | num_clusters, | |
1624 | &bg_blkno); | |
1625 | if (!status) { | |
1626 | *cluster_start = | |
1627 | ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode, | |
1628 | bg_blkno, | |
1629 | bg_bit_off); | |
1630 | atomic_inc(&osb->alloc_stats.bitmap_data); | |
1631 | } | |
1632 | } | |
1633 | if (status < 0) { | |
1634 | if (status != -ENOSPC) | |
1635 | mlog_errno(status); | |
1636 | goto bail; | |
1637 | } | |
1638 | ||
1639 | ac->ac_bits_given += *num_clusters; | |
1640 | ||
1641 | bail: | |
1642 | mlog_exit(status); | |
1643 | return status; | |
1644 | } | |
1645 | ||
415cb800 MF |
1646 | int ocfs2_claim_clusters(struct ocfs2_super *osb, |
1647 | handle_t *handle, | |
1648 | struct ocfs2_alloc_context *ac, | |
1649 | u32 min_clusters, | |
1650 | u32 *cluster_start, | |
1651 | u32 *num_clusters) | |
1652 | { | |
1653 | unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given; | |
1654 | ||
1655 | return __ocfs2_claim_clusters(osb, handle, ac, min_clusters, | |
1656 | bits_wanted, cluster_start, num_clusters); | |
1657 | } | |
1658 | ||
1fabe148 | 1659 | static inline int ocfs2_block_group_clear_bits(handle_t *handle, |
ccd979bd MF |
1660 | struct inode *alloc_inode, |
1661 | struct ocfs2_group_desc *bg, | |
1662 | struct buffer_head *group_bh, | |
1663 | unsigned int bit_off, | |
1664 | unsigned int num_bits) | |
1665 | { | |
1666 | int status; | |
1667 | unsigned int tmp; | |
1668 | int journal_type = OCFS2_JOURNAL_ACCESS_WRITE; | |
1669 | struct ocfs2_group_desc *undo_bg = NULL; | |
1670 | ||
1671 | mlog_entry_void(); | |
1672 | ||
1673 | if (!OCFS2_IS_VALID_GROUP_DESC(bg)) { | |
1674 | OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg); | |
1675 | status = -EIO; | |
1676 | goto bail; | |
1677 | } | |
1678 | ||
1679 | mlog(0, "off = %u, num = %u\n", bit_off, num_bits); | |
1680 | ||
1681 | if (ocfs2_is_cluster_bitmap(alloc_inode)) | |
1682 | journal_type = OCFS2_JOURNAL_ACCESS_UNDO; | |
1683 | ||
1684 | status = ocfs2_journal_access(handle, alloc_inode, group_bh, | |
1685 | journal_type); | |
1686 | if (status < 0) { | |
1687 | mlog_errno(status); | |
1688 | goto bail; | |
1689 | } | |
1690 | ||
1691 | if (ocfs2_is_cluster_bitmap(alloc_inode)) | |
1692 | undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data; | |
1693 | ||
1694 | tmp = num_bits; | |
1695 | while(tmp--) { | |
1696 | ocfs2_clear_bit((bit_off + tmp), | |
1697 | (unsigned long *) bg->bg_bitmap); | |
1698 | if (ocfs2_is_cluster_bitmap(alloc_inode)) | |
1699 | ocfs2_set_bit(bit_off + tmp, | |
1700 | (unsigned long *) undo_bg->bg_bitmap); | |
1701 | } | |
1702 | le16_add_cpu(&bg->bg_free_bits_count, num_bits); | |
1703 | ||
1704 | status = ocfs2_journal_dirty(handle, group_bh); | |
1705 | if (status < 0) | |
1706 | mlog_errno(status); | |
1707 | bail: | |
1708 | return status; | |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * expects the suballoc inode to already be locked. | |
1713 | */ | |
2b604351 MF |
1714 | int ocfs2_free_suballoc_bits(handle_t *handle, |
1715 | struct inode *alloc_inode, | |
1716 | struct buffer_head *alloc_bh, | |
1717 | unsigned int start_bit, | |
1718 | u64 bg_blkno, | |
1719 | unsigned int count) | |
ccd979bd MF |
1720 | { |
1721 | int status = 0; | |
1722 | u32 tmp_used; | |
1723 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb); | |
1724 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data; | |
1725 | struct ocfs2_chain_list *cl = &fe->id2.i_chain; | |
1726 | struct buffer_head *group_bh = NULL; | |
1727 | struct ocfs2_group_desc *group; | |
1728 | ||
1729 | mlog_entry_void(); | |
1730 | ||
1731 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
1732 | OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe); | |
1733 | status = -EIO; | |
1734 | goto bail; | |
1735 | } | |
1736 | BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl)); | |
1737 | ||
b0697053 MF |
1738 | mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n", |
1739 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count, | |
1740 | (unsigned long long)bg_blkno, start_bit); | |
ccd979bd MF |
1741 | |
1742 | status = ocfs2_read_block(osb, bg_blkno, &group_bh, OCFS2_BH_CACHED, | |
1743 | alloc_inode); | |
1744 | if (status < 0) { | |
1745 | mlog_errno(status); | |
1746 | goto bail; | |
1747 | } | |
1748 | ||
1749 | group = (struct ocfs2_group_desc *) group_bh->b_data; | |
7bf72ede MF |
1750 | status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, group); |
1751 | if (status) { | |
1752 | mlog_errno(status); | |
ccd979bd MF |
1753 | goto bail; |
1754 | } | |
1755 | BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits)); | |
1756 | ||
1757 | status = ocfs2_block_group_clear_bits(handle, alloc_inode, | |
1758 | group, group_bh, | |
1759 | start_bit, count); | |
1760 | if (status < 0) { | |
1761 | mlog_errno(status); | |
1762 | goto bail; | |
1763 | } | |
1764 | ||
1765 | status = ocfs2_journal_access(handle, alloc_inode, alloc_bh, | |
1766 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1767 | if (status < 0) { | |
1768 | mlog_errno(status); | |
1769 | goto bail; | |
1770 | } | |
1771 | ||
1772 | le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free, | |
1773 | count); | |
1774 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used); | |
1775 | fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count); | |
1776 | ||
1777 | status = ocfs2_journal_dirty(handle, alloc_bh); | |
1778 | if (status < 0) { | |
1779 | mlog_errno(status); | |
1780 | goto bail; | |
1781 | } | |
1782 | ||
1783 | bail: | |
1784 | if (group_bh) | |
1785 | brelse(group_bh); | |
1786 | ||
1787 | mlog_exit(status); | |
1788 | return status; | |
1789 | } | |
1790 | ||
1fabe148 | 1791 | int ocfs2_free_dinode(handle_t *handle, |
ccd979bd MF |
1792 | struct inode *inode_alloc_inode, |
1793 | struct buffer_head *inode_alloc_bh, | |
1794 | struct ocfs2_dinode *di) | |
1795 | { | |
1796 | u64 blk = le64_to_cpu(di->i_blkno); | |
1797 | u16 bit = le16_to_cpu(di->i_suballoc_bit); | |
1798 | u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit); | |
1799 | ||
1800 | return ocfs2_free_suballoc_bits(handle, inode_alloc_inode, | |
1801 | inode_alloc_bh, bit, bg_blkno, 1); | |
1802 | } | |
1803 | ||
1fabe148 | 1804 | int ocfs2_free_clusters(handle_t *handle, |
ccd979bd MF |
1805 | struct inode *bitmap_inode, |
1806 | struct buffer_head *bitmap_bh, | |
1807 | u64 start_blk, | |
1808 | unsigned int num_clusters) | |
1809 | { | |
1810 | int status; | |
1811 | u16 bg_start_bit; | |
1812 | u64 bg_blkno; | |
1813 | struct ocfs2_dinode *fe; | |
1814 | ||
1815 | /* You can't ever have a contiguous set of clusters | |
1816 | * bigger than a block group bitmap so we never have to worry | |
1817 | * about looping on them. */ | |
1818 | ||
1819 | mlog_entry_void(); | |
1820 | ||
1821 | /* This is expensive. We can safely remove once this stuff has | |
1822 | * gotten tested really well. */ | |
1823 | BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk))); | |
1824 | ||
1825 | fe = (struct ocfs2_dinode *) bitmap_bh->b_data; | |
1826 | ||
1827 | ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno, | |
1828 | &bg_start_bit); | |
1829 | ||
b0697053 MF |
1830 | mlog(0, "want to free %u clusters starting at block %llu\n", |
1831 | num_clusters, (unsigned long long)start_blk); | |
1832 | mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n", | |
1833 | (unsigned long long)bg_blkno, bg_start_bit); | |
ccd979bd MF |
1834 | |
1835 | status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh, | |
1836 | bg_start_bit, bg_blkno, | |
1837 | num_clusters); | |
9c7af40b | 1838 | if (status < 0) { |
ccd979bd | 1839 | mlog_errno(status); |
9c7af40b MF |
1840 | goto out; |
1841 | } | |
1842 | ||
1843 | ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb), | |
1844 | num_clusters); | |
ccd979bd | 1845 | |
9c7af40b | 1846 | out: |
ccd979bd MF |
1847 | mlog_exit(status); |
1848 | return status; | |
1849 | } | |
1850 | ||
1851 | static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg) | |
1852 | { | |
1853 | printk("Block Group:\n"); | |
1854 | printk("bg_signature: %s\n", bg->bg_signature); | |
1855 | printk("bg_size: %u\n", bg->bg_size); | |
1856 | printk("bg_bits: %u\n", bg->bg_bits); | |
1857 | printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count); | |
1858 | printk("bg_chain: %u\n", bg->bg_chain); | |
1859 | printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation)); | |
b0697053 MF |
1860 | printk("bg_next_group: %llu\n", |
1861 | (unsigned long long)bg->bg_next_group); | |
1862 | printk("bg_parent_dinode: %llu\n", | |
1863 | (unsigned long long)bg->bg_parent_dinode); | |
1864 | printk("bg_blkno: %llu\n", | |
1865 | (unsigned long long)bg->bg_blkno); | |
ccd979bd MF |
1866 | } |
1867 | ||
1868 | static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe) | |
1869 | { | |
1870 | int i; | |
1871 | ||
b0697053 | 1872 | printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno); |
ccd979bd | 1873 | printk("i_signature: %s\n", fe->i_signature); |
b0697053 MF |
1874 | printk("i_size: %llu\n", |
1875 | (unsigned long long)fe->i_size); | |
ccd979bd MF |
1876 | printk("i_clusters: %u\n", fe->i_clusters); |
1877 | printk("i_generation: %u\n", | |
1878 | le32_to_cpu(fe->i_generation)); | |
1879 | printk("id1.bitmap1.i_used: %u\n", | |
1880 | le32_to_cpu(fe->id1.bitmap1.i_used)); | |
1881 | printk("id1.bitmap1.i_total: %u\n", | |
1882 | le32_to_cpu(fe->id1.bitmap1.i_total)); | |
1883 | printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg); | |
1884 | printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc); | |
1885 | printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count); | |
1886 | printk("id2.i_chain.cl_next_free_rec: %u\n", | |
1887 | fe->id2.i_chain.cl_next_free_rec); | |
1888 | for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) { | |
1889 | printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i, | |
1890 | fe->id2.i_chain.cl_recs[i].c_free); | |
1891 | printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i, | |
1892 | fe->id2.i_chain.cl_recs[i].c_total); | |
b0697053 MF |
1893 | printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i, |
1894 | (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno); | |
ccd979bd MF |
1895 | } |
1896 | } |