]>
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 | ||
ccd979bd MF |
32 | #include <cluster/masklog.h> |
33 | ||
34 | #include "ocfs2.h" | |
35 | ||
36 | #include "alloc.h" | |
d6b32bbb | 37 | #include "blockcheck.h" |
ccd979bd MF |
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" | |
2f73e135 | 46 | #include "ocfs2_trace.h" |
ccd979bd MF |
47 | |
48 | #include "buffer_head_io.h" | |
49 | ||
ffda89a3 | 50 | #define NOT_ALLOC_NEW_GROUP 0 |
60ca81e8 TM |
51 | #define ALLOC_NEW_GROUP 0x1 |
52 | #define ALLOC_GROUPS_FROM_GLOBAL 0x2 | |
ffda89a3 | 53 | |
b89c5428 | 54 | #define OCFS2_MAX_TO_STEAL 1024 |
4d0ddb2c | 55 | |
7d1fe093 | 56 | struct ocfs2_suballoc_result { |
2b6cb576 JB |
57 | u64 sr_bg_blkno; /* The bg we allocated from. Set |
58 | to 0 when a block group is | |
59 | contiguous. */ | |
e49e2767 MF |
60 | u64 sr_bg_stable_blkno; /* |
61 | * Doesn't change, always | |
62 | * set to target block | |
63 | * group descriptor | |
64 | * block. | |
65 | */ | |
ba206635 | 66 | u64 sr_blkno; /* The first allocated block */ |
7d1fe093 JB |
67 | unsigned int sr_bit_offset; /* The bit in the bg */ |
68 | unsigned int sr_bits; /* How many bits we claimed */ | |
69 | }; | |
70 | ||
b2b6ebf5 MF |
71 | static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res) |
72 | { | |
73 | if (res->sr_blkno == 0) | |
74 | return 0; | |
75 | ||
76 | if (res->sr_bg_blkno) | |
77 | return res->sr_bg_blkno; | |
78 | ||
79 | return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset); | |
80 | } | |
81 | ||
ccd979bd MF |
82 | static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg); |
83 | static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe); | |
84 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl); | |
1fabe148 | 85 | static int ocfs2_block_group_fill(handle_t *handle, |
ccd979bd MF |
86 | struct inode *alloc_inode, |
87 | struct buffer_head *bg_bh, | |
88 | u64 group_blkno, | |
798db35f | 89 | unsigned int group_clusters, |
ccd979bd MF |
90 | u16 my_chain, |
91 | struct ocfs2_chain_list *cl); | |
92 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb, | |
93 | struct inode *alloc_inode, | |
1187c968 | 94 | struct buffer_head *bh, |
60ca81e8 | 95 | u64 max_block, |
feb473a6 | 96 | u64 *last_alloc_group, |
60ca81e8 | 97 | int flags); |
ccd979bd | 98 | |
ccd979bd MF |
99 | static int ocfs2_cluster_group_search(struct inode *inode, |
100 | struct buffer_head *group_bh, | |
101 | u32 bits_wanted, u32 min_bits, | |
1187c968 | 102 | u64 max_block, |
7d1fe093 | 103 | struct ocfs2_suballoc_result *res); |
ccd979bd MF |
104 | static int ocfs2_block_group_search(struct inode *inode, |
105 | struct buffer_head *group_bh, | |
106 | u32 bits_wanted, u32 min_bits, | |
1187c968 | 107 | u64 max_block, |
7d1fe093 | 108 | struct ocfs2_suballoc_result *res); |
aa8f8e93 | 109 | static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac, |
1fabe148 | 110 | handle_t *handle, |
ccd979bd MF |
111 | u32 bits_wanted, |
112 | u32 min_bits, | |
7d1fe093 | 113 | struct ocfs2_suballoc_result *res); |
ccd979bd MF |
114 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh, |
115 | int nr); | |
1fabe148 | 116 | static int ocfs2_relink_block_group(handle_t *handle, |
ccd979bd MF |
117 | struct inode *alloc_inode, |
118 | struct buffer_head *fe_bh, | |
119 | struct buffer_head *bg_bh, | |
120 | struct buffer_head *prev_bg_bh, | |
121 | u16 chain); | |
122 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg, | |
123 | u32 wanted); | |
ccd979bd MF |
124 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode, |
125 | u64 bg_blkno, | |
126 | u16 bg_bit_off); | |
ccd979bd MF |
127 | static inline void ocfs2_block_to_cluster_group(struct inode *inode, |
128 | u64 data_blkno, | |
129 | u64 *bg_blkno, | |
130 | u16 *bg_bit_off); | |
1187c968 JB |
131 | static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb, |
132 | u32 bits_wanted, u64 max_block, | |
60ca81e8 | 133 | int flags, |
1187c968 | 134 | struct ocfs2_alloc_context **ac); |
ccd979bd | 135 | |
9c7af40b | 136 | void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac) |
ccd979bd | 137 | { |
da5cbf2f MF |
138 | struct inode *inode = ac->ac_inode; |
139 | ||
140 | if (inode) { | |
141 | if (ac->ac_which != OCFS2_AC_USE_LOCAL) | |
e63aecb6 | 142 | ocfs2_inode_unlock(inode, 1); |
da5cbf2f | 143 | |
5955102c | 144 | inode_unlock(inode); |
da5cbf2f MF |
145 | |
146 | iput(inode); | |
4d0ddb2c | 147 | ac->ac_inode = NULL; |
da5cbf2f | 148 | } |
a81cb88b MF |
149 | brelse(ac->ac_bh); |
150 | ac->ac_bh = NULL; | |
e3b4a97d | 151 | ac->ac_resv = NULL; |
46359295 JQ |
152 | kfree(ac->ac_find_loc_priv); |
153 | ac->ac_find_loc_priv = NULL; | |
4d0ddb2c TM |
154 | } |
155 | ||
156 | void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac) | |
157 | { | |
158 | ocfs2_free_ac_resource(ac); | |
ccd979bd MF |
159 | kfree(ac); |
160 | } | |
161 | ||
162 | static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl) | |
163 | { | |
164 | return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc); | |
165 | } | |
166 | ||
57e3e797 | 167 | #define do_error(fmt, ...) \ |
7ecef14a JP |
168 | do { \ |
169 | if (resize) \ | |
170 | mlog(ML_ERROR, fmt, ##__VA_ARGS__); \ | |
171 | else \ | |
172 | return ocfs2_error(sb, fmt, ##__VA_ARGS__); \ | |
173 | } while (0) | |
57e3e797 | 174 | |
970e4936 JB |
175 | static int ocfs2_validate_gd_self(struct super_block *sb, |
176 | struct buffer_head *bh, | |
78c37eb0 | 177 | int resize) |
970e4936 JB |
178 | { |
179 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data; | |
180 | ||
7bf72ede | 181 | if (!OCFS2_IS_VALID_GROUP_DESC(gd)) { |
7ecef14a | 182 | do_error("Group descriptor #%llu has bad signature %.*s\n", |
68f64d47 | 183 | (unsigned long long)bh->b_blocknr, 7, |
57e3e797 | 184 | gd->bg_signature); |
7bf72ede MF |
185 | } |
186 | ||
68f64d47 | 187 | if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) { |
7ecef14a | 188 | do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n", |
68f64d47 JB |
189 | (unsigned long long)bh->b_blocknr, |
190 | (unsigned long long)le64_to_cpu(gd->bg_blkno)); | |
68f64d47 JB |
191 | } |
192 | ||
193 | if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) { | |
7ecef14a | 194 | do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n", |
68f64d47 JB |
195 | (unsigned long long)bh->b_blocknr, |
196 | le32_to_cpu(gd->bg_generation)); | |
68f64d47 JB |
197 | } |
198 | ||
970e4936 | 199 | if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) { |
7ecef14a | 200 | do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n", |
970e4936 JB |
201 | (unsigned long long)bh->b_blocknr, |
202 | le16_to_cpu(gd->bg_bits), | |
203 | le16_to_cpu(gd->bg_free_bits_count)); | |
970e4936 JB |
204 | } |
205 | ||
206 | if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) { | |
7ecef14a | 207 | do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n", |
970e4936 JB |
208 | (unsigned long long)bh->b_blocknr, |
209 | le16_to_cpu(gd->bg_bits), | |
210 | 8 * le16_to_cpu(gd->bg_size)); | |
970e4936 JB |
211 | } |
212 | ||
213 | return 0; | |
214 | } | |
215 | ||
216 | static int ocfs2_validate_gd_parent(struct super_block *sb, | |
217 | struct ocfs2_dinode *di, | |
218 | struct buffer_head *bh, | |
78c37eb0 | 219 | int resize) |
970e4936 JB |
220 | { |
221 | unsigned int max_bits; | |
222 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data; | |
223 | ||
7bf72ede | 224 | if (di->i_blkno != gd->bg_parent_dinode) { |
7ecef14a | 225 | do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n", |
68f64d47 | 226 | (unsigned long long)bh->b_blocknr, |
57e3e797 JB |
227 | (unsigned long long)le64_to_cpu(gd->bg_parent_dinode), |
228 | (unsigned long long)le64_to_cpu(di->i_blkno)); | |
7bf72ede MF |
229 | } |
230 | ||
231 | max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc); | |
232 | if (le16_to_cpu(gd->bg_bits) > max_bits) { | |
7ecef14a | 233 | do_error("Group descriptor #%llu has bit count of %u\n", |
68f64d47 | 234 | (unsigned long long)bh->b_blocknr, |
57e3e797 | 235 | le16_to_cpu(gd->bg_bits)); |
7bf72ede MF |
236 | } |
237 | ||
78c37eb0 TM |
238 | /* In resize, we may meet the case bg_chain == cl_next_free_rec. */ |
239 | if ((le16_to_cpu(gd->bg_chain) > | |
240 | le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) || | |
241 | ((le16_to_cpu(gd->bg_chain) == | |
242 | le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) { | |
7ecef14a | 243 | do_error("Group descriptor #%llu has bad chain %u\n", |
68f64d47 | 244 | (unsigned long long)bh->b_blocknr, |
57e3e797 | 245 | le16_to_cpu(gd->bg_chain)); |
7bf72ede MF |
246 | } |
247 | ||
970e4936 JB |
248 | return 0; |
249 | } | |
7bf72ede | 250 | |
57e3e797 | 251 | #undef do_error |
7bf72ede | 252 | |
970e4936 JB |
253 | /* |
254 | * This version only prints errors. It does not fail the filesystem, and | |
255 | * exists only for resize. | |
256 | */ | |
257 | int ocfs2_check_group_descriptor(struct super_block *sb, | |
258 | struct ocfs2_dinode *di, | |
259 | struct buffer_head *bh) | |
260 | { | |
261 | int rc; | |
d6b32bbb JB |
262 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data; |
263 | ||
264 | BUG_ON(!buffer_uptodate(bh)); | |
970e4936 | 265 | |
d6b32bbb JB |
266 | /* |
267 | * If the ecc fails, we return the error but otherwise | |
268 | * leave the filesystem running. We know any error is | |
269 | * local to this block. | |
270 | */ | |
271 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check); | |
13723d00 JB |
272 | if (rc) { |
273 | mlog(ML_ERROR, | |
274 | "Checksum failed for group descriptor %llu\n", | |
275 | (unsigned long long)bh->b_blocknr); | |
276 | } else | |
d6b32bbb | 277 | rc = ocfs2_validate_gd_self(sb, bh, 1); |
970e4936 JB |
278 | if (!rc) |
279 | rc = ocfs2_validate_gd_parent(sb, di, bh, 1); | |
280 | ||
281 | return rc; | |
282 | } | |
283 | ||
284 | static int ocfs2_validate_group_descriptor(struct super_block *sb, | |
285 | struct buffer_head *bh) | |
286 | { | |
d6b32bbb JB |
287 | int rc; |
288 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data; | |
289 | ||
2f73e135 TM |
290 | trace_ocfs2_validate_group_descriptor( |
291 | (unsigned long long)bh->b_blocknr); | |
970e4936 | 292 | |
d6b32bbb JB |
293 | BUG_ON(!buffer_uptodate(bh)); |
294 | ||
295 | /* | |
296 | * If the ecc fails, we return the error but otherwise | |
297 | * leave the filesystem running. We know any error is | |
298 | * local to this block. | |
299 | */ | |
300 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check); | |
301 | if (rc) | |
302 | return rc; | |
303 | ||
304 | /* | |
305 | * Errors after here are fatal. | |
306 | */ | |
307 | ||
970e4936 | 308 | return ocfs2_validate_gd_self(sb, bh, 0); |
7bf72ede MF |
309 | } |
310 | ||
68f64d47 JB |
311 | int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di, |
312 | u64 gd_blkno, struct buffer_head **bh) | |
313 | { | |
314 | int rc; | |
315 | struct buffer_head *tmp = *bh; | |
316 | ||
8cb471e8 | 317 | rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp, |
970e4936 | 318 | ocfs2_validate_group_descriptor); |
68f64d47 JB |
319 | if (rc) |
320 | goto out; | |
321 | ||
970e4936 | 322 | rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0); |
68f64d47 JB |
323 | if (rc) { |
324 | brelse(tmp); | |
325 | goto out; | |
326 | } | |
327 | ||
328 | /* If ocfs2_read_block() got us a new bh, pass it up. */ | |
329 | if (!*bh) | |
330 | *bh = tmp; | |
331 | ||
332 | out: | |
333 | return rc; | |
334 | } | |
335 | ||
798db35f JB |
336 | static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb, |
337 | struct ocfs2_group_desc *bg, | |
338 | struct ocfs2_chain_list *cl, | |
47dea423 | 339 | u64 p_blkno, unsigned int clusters) |
798db35f JB |
340 | { |
341 | struct ocfs2_extent_list *el = &bg->bg_list; | |
342 | struct ocfs2_extent_rec *rec; | |
343 | ||
4711954e | 344 | BUG_ON(!ocfs2_supports_discontig_bg(osb)); |
798db35f JB |
345 | if (!el->l_next_free_rec) |
346 | el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb)); | |
347 | rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)]; | |
4711954e | 348 | rec->e_blkno = cpu_to_le64(p_blkno); |
798db35f JB |
349 | rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) / |
350 | le16_to_cpu(cl->cl_bpc)); | |
47dea423 | 351 | rec->e_leaf_clusters = cpu_to_le16(clusters); |
798db35f | 352 | le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc)); |
4711954e TM |
353 | le16_add_cpu(&bg->bg_free_bits_count, |
354 | clusters * le16_to_cpu(cl->cl_bpc)); | |
798db35f JB |
355 | le16_add_cpu(&el->l_next_free_rec, 1); |
356 | } | |
357 | ||
1fabe148 | 358 | static int ocfs2_block_group_fill(handle_t *handle, |
ccd979bd MF |
359 | struct inode *alloc_inode, |
360 | struct buffer_head *bg_bh, | |
361 | u64 group_blkno, | |
798db35f | 362 | unsigned int group_clusters, |
ccd979bd MF |
363 | u16 my_chain, |
364 | struct ocfs2_chain_list *cl) | |
365 | { | |
366 | int status = 0; | |
798db35f | 367 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb); |
ccd979bd MF |
368 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; |
369 | struct super_block * sb = alloc_inode->i_sb; | |
370 | ||
ccd979bd | 371 | if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) { |
7ecef14a JP |
372 | status = ocfs2_error(alloc_inode->i_sb, |
373 | "group block (%llu) != b_blocknr (%llu)\n", | |
374 | (unsigned long long)group_blkno, | |
375 | (unsigned long long) bg_bh->b_blocknr); | |
ccd979bd MF |
376 | goto bail; |
377 | } | |
378 | ||
13723d00 | 379 | status = ocfs2_journal_access_gd(handle, |
0cf2f763 | 380 | INODE_CACHE(alloc_inode), |
13723d00 JB |
381 | bg_bh, |
382 | OCFS2_JOURNAL_ACCESS_CREATE); | |
ccd979bd MF |
383 | if (status < 0) { |
384 | mlog_errno(status); | |
385 | goto bail; | |
386 | } | |
387 | ||
388 | memset(bg, 0, sb->s_blocksize); | |
389 | strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE); | |
390 | bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation); | |
8571882c TM |
391 | bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1, |
392 | osb->s_feature_incompat)); | |
ccd979bd MF |
393 | bg->bg_chain = cpu_to_le16(my_chain); |
394 | bg->bg_next_group = cl->cl_recs[my_chain].c_blkno; | |
395 | bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno); | |
396 | bg->bg_blkno = cpu_to_le64(group_blkno); | |
798db35f JB |
397 | if (group_clusters == le16_to_cpu(cl->cl_cpg)) |
398 | bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl)); | |
399 | else | |
4711954e | 400 | ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno, |
798db35f JB |
401 | group_clusters); |
402 | ||
ccd979bd MF |
403 | /* set the 1st bit in the bitmap to account for the descriptor block */ |
404 | ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap); | |
405 | bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1); | |
406 | ||
ec20cec7 | 407 | ocfs2_journal_dirty(handle, bg_bh); |
ccd979bd MF |
408 | |
409 | /* There is no need to zero out or otherwise initialize the | |
410 | * other blocks in a group - All valid FS metadata in a block | |
411 | * group stores the superblock fs_generation value at | |
412 | * allocation time. */ | |
413 | ||
414 | bail: | |
c1e8d35e TM |
415 | if (status) |
416 | mlog_errno(status); | |
ccd979bd MF |
417 | return status; |
418 | } | |
419 | ||
420 | static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl) | |
421 | { | |
422 | u16 curr, best; | |
423 | ||
424 | best = curr = 0; | |
425 | while (curr < le16_to_cpu(cl->cl_count)) { | |
426 | if (le32_to_cpu(cl->cl_recs[best].c_total) > | |
427 | le32_to_cpu(cl->cl_recs[curr].c_total)) | |
428 | best = curr; | |
429 | curr++; | |
430 | } | |
431 | return best; | |
432 | } | |
433 | ||
798db35f JB |
434 | static struct buffer_head * |
435 | ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle, | |
436 | struct inode *alloc_inode, | |
437 | struct ocfs2_alloc_context *ac, | |
438 | struct ocfs2_chain_list *cl) | |
439 | { | |
440 | int status; | |
441 | u32 bit_off, num_bits; | |
442 | u64 bg_blkno; | |
443 | struct buffer_head *bg_bh; | |
444 | unsigned int alloc_rec = ocfs2_find_smallest_chain(cl); | |
445 | ||
1ed9b777 | 446 | status = ocfs2_claim_clusters(handle, ac, |
798db35f JB |
447 | le16_to_cpu(cl->cl_cpg), &bit_off, |
448 | &num_bits); | |
449 | if (status < 0) { | |
450 | if (status != -ENOSPC) | |
451 | mlog_errno(status); | |
452 | goto bail; | |
453 | } | |
454 | ||
455 | /* setup the group */ | |
456 | bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
2f73e135 TM |
457 | trace_ocfs2_block_group_alloc_contig( |
458 | (unsigned long long)bg_blkno, alloc_rec); | |
798db35f JB |
459 | |
460 | bg_bh = sb_getblk(osb->sb, bg_blkno); | |
461 | if (!bg_bh) { | |
7391a294 | 462 | status = -ENOMEM; |
798db35f JB |
463 | mlog_errno(status); |
464 | goto bail; | |
465 | } | |
466 | ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh); | |
467 | ||
468 | status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh, | |
469 | bg_blkno, num_bits, alloc_rec, cl); | |
470 | if (status < 0) { | |
471 | brelse(bg_bh); | |
472 | mlog_errno(status); | |
473 | } | |
474 | ||
475 | bail: | |
476 | return status ? ERR_PTR(status) : bg_bh; | |
477 | } | |
478 | ||
479 | static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb, | |
480 | handle_t *handle, | |
481 | struct ocfs2_alloc_context *ac, | |
482 | unsigned int min_bits, | |
483 | u32 *bit_off, u32 *num_bits) | |
484 | { | |
18d3a98f | 485 | int status = 0; |
798db35f JB |
486 | |
487 | while (min_bits) { | |
1ed9b777 | 488 | status = ocfs2_claim_clusters(handle, ac, min_bits, |
798db35f JB |
489 | bit_off, num_bits); |
490 | if (status != -ENOSPC) | |
491 | break; | |
492 | ||
493 | min_bits >>= 1; | |
494 | } | |
495 | ||
496 | return status; | |
497 | } | |
498 | ||
499 | static int ocfs2_block_group_grow_discontig(handle_t *handle, | |
500 | struct inode *alloc_inode, | |
501 | struct buffer_head *bg_bh, | |
502 | struct ocfs2_alloc_context *ac, | |
503 | struct ocfs2_chain_list *cl, | |
504 | unsigned int min_bits) | |
505 | { | |
506 | int status; | |
507 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb); | |
508 | struct ocfs2_group_desc *bg = | |
509 | (struct ocfs2_group_desc *)bg_bh->b_data; | |
4711954e TM |
510 | unsigned int needed = le16_to_cpu(cl->cl_cpg) - |
511 | le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc); | |
798db35f JB |
512 | u32 p_cpos, clusters; |
513 | u64 p_blkno; | |
514 | struct ocfs2_extent_list *el = &bg->bg_list; | |
515 | ||
516 | status = ocfs2_journal_access_gd(handle, | |
517 | INODE_CACHE(alloc_inode), | |
518 | bg_bh, | |
519 | OCFS2_JOURNAL_ACCESS_CREATE); | |
520 | if (status < 0) { | |
521 | mlog_errno(status); | |
522 | goto bail; | |
523 | } | |
524 | ||
525 | while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) < | |
526 | le16_to_cpu(el->l_count))) { | |
798db35f JB |
527 | if (min_bits > needed) |
528 | min_bits = needed; | |
529 | status = ocfs2_block_group_claim_bits(osb, handle, ac, | |
530 | min_bits, &p_cpos, | |
531 | &clusters); | |
532 | if (status < 0) { | |
533 | if (status != -ENOSPC) | |
534 | mlog_errno(status); | |
535 | goto bail; | |
536 | } | |
537 | p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos); | |
538 | ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno, | |
539 | clusters); | |
540 | ||
541 | min_bits = clusters; | |
4711954e TM |
542 | needed = le16_to_cpu(cl->cl_cpg) - |
543 | le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc); | |
798db35f JB |
544 | } |
545 | ||
546 | if (needed > 0) { | |
4711954e TM |
547 | /* |
548 | * We have used up all the extent rec but can't fill up | |
549 | * the cpg. So bail out. | |
550 | */ | |
551 | status = -ENOSPC; | |
552 | goto bail; | |
798db35f JB |
553 | } |
554 | ||
555 | ocfs2_journal_dirty(handle, bg_bh); | |
556 | ||
557 | bail: | |
558 | return status; | |
559 | } | |
560 | ||
8b06bc59 JB |
561 | static void ocfs2_bg_alloc_cleanup(handle_t *handle, |
562 | struct ocfs2_alloc_context *cluster_ac, | |
563 | struct inode *alloc_inode, | |
564 | struct buffer_head *bg_bh) | |
798db35f | 565 | { |
8b06bc59 | 566 | int i, ret; |
798db35f JB |
567 | struct ocfs2_group_desc *bg; |
568 | struct ocfs2_extent_list *el; | |
569 | struct ocfs2_extent_rec *rec; | |
570 | ||
571 | if (!bg_bh) | |
572 | return; | |
573 | ||
574 | bg = (struct ocfs2_group_desc *)bg_bh->b_data; | |
575 | el = &bg->bg_list; | |
576 | for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) { | |
577 | rec = &el->l_recs[i]; | |
8b06bc59 JB |
578 | ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode, |
579 | cluster_ac->ac_bh, | |
580 | le64_to_cpu(rec->e_blkno), | |
72094e43 | 581 | le16_to_cpu(rec->e_leaf_clusters)); |
8b06bc59 JB |
582 | if (ret) |
583 | mlog_errno(ret); | |
584 | /* Try all the clusters to free */ | |
798db35f JB |
585 | } |
586 | ||
587 | ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh); | |
588 | brelse(bg_bh); | |
589 | } | |
590 | ||
591 | static struct buffer_head * | |
592 | ocfs2_block_group_alloc_discontig(handle_t *handle, | |
593 | struct inode *alloc_inode, | |
594 | struct ocfs2_alloc_context *ac, | |
8b06bc59 | 595 | struct ocfs2_chain_list *cl) |
798db35f JB |
596 | { |
597 | int status; | |
598 | u32 bit_off, num_bits; | |
599 | u64 bg_blkno; | |
600 | unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1; | |
601 | struct buffer_head *bg_bh = NULL; | |
602 | unsigned int alloc_rec = ocfs2_find_smallest_chain(cl); | |
603 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb); | |
604 | ||
4711954e | 605 | if (!ocfs2_supports_discontig_bg(osb)) { |
798db35f JB |
606 | status = -ENOSPC; |
607 | goto bail; | |
608 | } | |
609 | ||
8b06bc59 JB |
610 | status = ocfs2_extend_trans(handle, |
611 | ocfs2_calc_bg_discontig_credits(osb->sb)); | |
612 | if (status) { | |
613 | mlog_errno(status); | |
614 | goto bail; | |
615 | } | |
616 | ||
95ec0adf JB |
617 | /* |
618 | * We're going to be grabbing from multiple cluster groups. | |
619 | * We don't have enough credits to relink them all, and the | |
620 | * cluster groups will be staying in cache for the duration of | |
621 | * this operation. | |
622 | */ | |
309a85b6 | 623 | ac->ac_disable_chain_relink = 1; |
95ec0adf | 624 | |
798db35f JB |
625 | /* Claim the first region */ |
626 | status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits, | |
627 | &bit_off, &num_bits); | |
628 | if (status < 0) { | |
629 | if (status != -ENOSPC) | |
630 | mlog_errno(status); | |
631 | goto bail; | |
632 | } | |
633 | min_bits = num_bits; | |
634 | ||
635 | /* setup the group */ | |
636 | bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
2f73e135 TM |
637 | trace_ocfs2_block_group_alloc_discontig( |
638 | (unsigned long long)bg_blkno, alloc_rec); | |
798db35f JB |
639 | |
640 | bg_bh = sb_getblk(osb->sb, bg_blkno); | |
641 | if (!bg_bh) { | |
7391a294 | 642 | status = -ENOMEM; |
798db35f JB |
643 | mlog_errno(status); |
644 | goto bail; | |
645 | } | |
646 | ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh); | |
647 | ||
648 | status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh, | |
649 | bg_blkno, num_bits, alloc_rec, cl); | |
650 | if (status < 0) { | |
651 | mlog_errno(status); | |
652 | goto bail; | |
653 | } | |
654 | ||
655 | status = ocfs2_block_group_grow_discontig(handle, alloc_inode, | |
656 | bg_bh, ac, cl, min_bits); | |
657 | if (status) | |
658 | mlog_errno(status); | |
659 | ||
660 | bail: | |
661 | if (status) | |
8b06bc59 | 662 | ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh); |
798db35f JB |
663 | return status ? ERR_PTR(status) : bg_bh; |
664 | } | |
665 | ||
ccd979bd MF |
666 | /* |
667 | * We expect the block group allocator to already be locked. | |
668 | */ | |
669 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb, | |
670 | struct inode *alloc_inode, | |
1187c968 | 671 | struct buffer_head *bh, |
60ca81e8 | 672 | u64 max_block, |
feb473a6 | 673 | u64 *last_alloc_group, |
60ca81e8 | 674 | int flags) |
ccd979bd MF |
675 | { |
676 | int status, credits; | |
677 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data; | |
678 | struct ocfs2_chain_list *cl; | |
679 | struct ocfs2_alloc_context *ac = NULL; | |
1fabe148 | 680 | handle_t *handle = NULL; |
4711954e | 681 | u16 alloc_rec; |
ccd979bd MF |
682 | struct buffer_head *bg_bh = NULL; |
683 | struct ocfs2_group_desc *bg; | |
684 | ||
685 | BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode)); | |
686 | ||
ccd979bd | 687 | cl = &fe->id2.i_chain; |
1187c968 JB |
688 | status = ocfs2_reserve_clusters_with_limit(osb, |
689 | le16_to_cpu(cl->cl_cpg), | |
60ca81e8 | 690 | max_block, flags, &ac); |
ccd979bd MF |
691 | if (status < 0) { |
692 | if (status != -ENOSPC) | |
693 | mlog_errno(status); | |
694 | goto bail; | |
695 | } | |
696 | ||
697 | credits = ocfs2_calc_group_alloc_credits(osb->sb, | |
698 | le16_to_cpu(cl->cl_cpg)); | |
65eff9cc | 699 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
700 | if (IS_ERR(handle)) { |
701 | status = PTR_ERR(handle); | |
702 | handle = NULL; | |
703 | mlog_errno(status); | |
704 | goto bail; | |
705 | } | |
706 | ||
feb473a6 | 707 | if (last_alloc_group && *last_alloc_group != 0) { |
2f73e135 TM |
708 | trace_ocfs2_block_group_alloc( |
709 | (unsigned long long)*last_alloc_group); | |
feb473a6 TM |
710 | ac->ac_last_group = *last_alloc_group; |
711 | } | |
798db35f JB |
712 | |
713 | bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode, | |
714 | ac, cl); | |
715 | if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC)) | |
716 | bg_bh = ocfs2_block_group_alloc_discontig(handle, | |
717 | alloc_inode, | |
8b06bc59 | 718 | ac, cl); |
798db35f JB |
719 | if (IS_ERR(bg_bh)) { |
720 | status = PTR_ERR(bg_bh); | |
721 | bg_bh = NULL; | |
ccd979bd MF |
722 | if (status != -ENOSPC) |
723 | mlog_errno(status); | |
724 | goto bail; | |
725 | } | |
ccd979bd MF |
726 | bg = (struct ocfs2_group_desc *) bg_bh->b_data; |
727 | ||
0cf2f763 | 728 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode), |
13723d00 | 729 | bh, OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
730 | if (status < 0) { |
731 | mlog_errno(status); | |
732 | goto bail; | |
733 | } | |
734 | ||
4711954e TM |
735 | alloc_rec = le16_to_cpu(bg->bg_chain); |
736 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_free, | |
ccd979bd | 737 | le16_to_cpu(bg->bg_free_bits_count)); |
4711954e | 738 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, |
798db35f | 739 | le16_to_cpu(bg->bg_bits)); |
0a463b74 | 740 | cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno; |
ccd979bd MF |
741 | if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count)) |
742 | le16_add_cpu(&cl->cl_next_free_rec, 1); | |
743 | ||
744 | le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) - | |
745 | le16_to_cpu(bg->bg_free_bits_count)); | |
746 | le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits)); | |
747 | le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg)); | |
748 | ||
ec20cec7 | 749 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
750 | |
751 | spin_lock(&OCFS2_I(alloc_inode)->ip_lock); | |
752 | OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
753 | fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb, | |
754 | le32_to_cpu(fe->i_clusters))); | |
755 | spin_unlock(&OCFS2_I(alloc_inode)->ip_lock); | |
756 | i_size_write(alloc_inode, le64_to_cpu(fe->i_size)); | |
8110b073 | 757 | alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode); |
6fdb702d | 758 | ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0); |
ccd979bd MF |
759 | |
760 | status = 0; | |
feb473a6 TM |
761 | |
762 | /* save the new last alloc group so that the caller can cache it. */ | |
763 | if (last_alloc_group) | |
764 | *last_alloc_group = ac->ac_last_group; | |
765 | ||
ccd979bd MF |
766 | bail: |
767 | if (handle) | |
02dc1af4 | 768 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
769 | |
770 | if (ac) | |
771 | ocfs2_free_alloc_context(ac); | |
772 | ||
a81cb88b | 773 | brelse(bg_bh); |
ccd979bd | 774 | |
c1e8d35e TM |
775 | if (status) |
776 | mlog_errno(status); | |
ccd979bd MF |
777 | return status; |
778 | } | |
779 | ||
780 | static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb, | |
da5cbf2f MF |
781 | struct ocfs2_alloc_context *ac, |
782 | int type, | |
ffda89a3 | 783 | u32 slot, |
feb473a6 | 784 | u64 *last_alloc_group, |
60ca81e8 | 785 | int flags) |
ccd979bd MF |
786 | { |
787 | int status; | |
788 | u32 bits_wanted = ac->ac_bits_wanted; | |
da5cbf2f | 789 | struct inode *alloc_inode; |
ccd979bd | 790 | struct buffer_head *bh = NULL; |
ccd979bd MF |
791 | struct ocfs2_dinode *fe; |
792 | u32 free_bits; | |
793 | ||
da5cbf2f MF |
794 | alloc_inode = ocfs2_get_system_file_inode(osb, type, slot); |
795 | if (!alloc_inode) { | |
796 | mlog_errno(-EINVAL); | |
797 | return -EINVAL; | |
798 | } | |
ccd979bd | 799 | |
5955102c | 800 | inode_lock(alloc_inode); |
da5cbf2f | 801 | |
e63aecb6 | 802 | status = ocfs2_inode_lock(alloc_inode, &bh, 1); |
ccd979bd | 803 | if (status < 0) { |
5955102c | 804 | inode_unlock(alloc_inode); |
da5cbf2f MF |
805 | iput(alloc_inode); |
806 | ||
ccd979bd | 807 | mlog_errno(status); |
da5cbf2f | 808 | return status; |
ccd979bd MF |
809 | } |
810 | ||
da5cbf2f | 811 | ac->ac_inode = alloc_inode; |
a4a48911 | 812 | ac->ac_alloc_slot = slot; |
da5cbf2f | 813 | |
ccd979bd | 814 | fe = (struct ocfs2_dinode *) bh->b_data; |
10995aa2 JB |
815 | |
816 | /* The bh was validated by the inode read inside | |
817 | * ocfs2_inode_lock(). Any corruption is a code bug. */ | |
818 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
819 | ||
ccd979bd | 820 | if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) { |
7ecef14a JP |
821 | status = ocfs2_error(alloc_inode->i_sb, |
822 | "Invalid chain allocator %llu\n", | |
823 | (unsigned long long)le64_to_cpu(fe->i_blkno)); | |
ccd979bd MF |
824 | goto bail; |
825 | } | |
826 | ||
827 | free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) - | |
828 | le32_to_cpu(fe->id1.bitmap1.i_used); | |
829 | ||
830 | if (bits_wanted > free_bits) { | |
831 | /* cluster bitmap never grows */ | |
832 | if (ocfs2_is_cluster_bitmap(alloc_inode)) { | |
2f73e135 TM |
833 | trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted, |
834 | free_bits); | |
ccd979bd MF |
835 | status = -ENOSPC; |
836 | goto bail; | |
837 | } | |
838 | ||
60ca81e8 | 839 | if (!(flags & ALLOC_NEW_GROUP)) { |
2f73e135 TM |
840 | trace_ocfs2_reserve_suballoc_bits_no_new_group( |
841 | slot, bits_wanted, free_bits); | |
ffda89a3 TM |
842 | status = -ENOSPC; |
843 | goto bail; | |
844 | } | |
845 | ||
1187c968 | 846 | status = ocfs2_block_group_alloc(osb, alloc_inode, bh, |
feb473a6 TM |
847 | ac->ac_max_block, |
848 | last_alloc_group, flags); | |
ccd979bd MF |
849 | if (status < 0) { |
850 | if (status != -ENOSPC) | |
851 | mlog_errno(status); | |
852 | goto bail; | |
853 | } | |
854 | atomic_inc(&osb->alloc_stats.bg_extends); | |
855 | ||
856 | /* You should never ask for this much metadata */ | |
857 | BUG_ON(bits_wanted > | |
858 | (le32_to_cpu(fe->id1.bitmap1.i_total) | |
859 | - le32_to_cpu(fe->id1.bitmap1.i_used))); | |
860 | } | |
861 | ||
862 | get_bh(bh); | |
863 | ac->ac_bh = bh; | |
864 | bail: | |
a81cb88b | 865 | brelse(bh); |
ccd979bd | 866 | |
c1e8d35e TM |
867 | if (status) |
868 | mlog_errno(status); | |
ccd979bd MF |
869 | return status; |
870 | } | |
871 | ||
b89c5428 TY |
872 | static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb) |
873 | { | |
874 | spin_lock(&osb->osb_lock); | |
875 | osb->s_inode_steal_slot = OCFS2_INVALID_SLOT; | |
876 | spin_unlock(&osb->osb_lock); | |
877 | atomic_set(&osb->s_num_inodes_stolen, 0); | |
878 | } | |
879 | ||
880 | static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb) | |
881 | { | |
882 | spin_lock(&osb->osb_lock); | |
883 | osb->s_meta_steal_slot = OCFS2_INVALID_SLOT; | |
884 | spin_unlock(&osb->osb_lock); | |
885 | atomic_set(&osb->s_num_meta_stolen, 0); | |
886 | } | |
887 | ||
888 | void ocfs2_init_steal_slots(struct ocfs2_super *osb) | |
889 | { | |
890 | ocfs2_init_inode_steal_slot(osb); | |
891 | ocfs2_init_meta_steal_slot(osb); | |
892 | } | |
893 | ||
894 | static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type) | |
895 | { | |
896 | spin_lock(&osb->osb_lock); | |
897 | if (type == INODE_ALLOC_SYSTEM_INODE) | |
898 | osb->s_inode_steal_slot = slot; | |
899 | else if (type == EXTENT_ALLOC_SYSTEM_INODE) | |
900 | osb->s_meta_steal_slot = slot; | |
901 | spin_unlock(&osb->osb_lock); | |
902 | } | |
903 | ||
904 | static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type) | |
905 | { | |
906 | int slot = OCFS2_INVALID_SLOT; | |
907 | ||
908 | spin_lock(&osb->osb_lock); | |
909 | if (type == INODE_ALLOC_SYSTEM_INODE) | |
910 | slot = osb->s_inode_steal_slot; | |
911 | else if (type == EXTENT_ALLOC_SYSTEM_INODE) | |
912 | slot = osb->s_meta_steal_slot; | |
913 | spin_unlock(&osb->osb_lock); | |
914 | ||
915 | return slot; | |
916 | } | |
917 | ||
918 | static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb) | |
919 | { | |
920 | return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE); | |
921 | } | |
922 | ||
923 | static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb) | |
924 | { | |
925 | return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE); | |
926 | } | |
927 | ||
928 | static int ocfs2_steal_resource(struct ocfs2_super *osb, | |
929 | struct ocfs2_alloc_context *ac, | |
930 | int type) | |
931 | { | |
932 | int i, status = -ENOSPC; | |
933 | int slot = __ocfs2_get_steal_slot(osb, type); | |
934 | ||
935 | /* Start to steal resource from the first slot after ours. */ | |
936 | if (slot == OCFS2_INVALID_SLOT) | |
937 | slot = osb->slot_num + 1; | |
938 | ||
939 | for (i = 0; i < osb->max_slots; i++, slot++) { | |
940 | if (slot == osb->max_slots) | |
941 | slot = 0; | |
942 | ||
943 | if (slot == osb->slot_num) | |
944 | continue; | |
945 | ||
946 | status = ocfs2_reserve_suballoc_bits(osb, ac, | |
947 | type, | |
948 | (u32)slot, NULL, | |
949 | NOT_ALLOC_NEW_GROUP); | |
950 | if (status >= 0) { | |
951 | __ocfs2_set_steal_slot(osb, slot, type); | |
952 | break; | |
953 | } | |
954 | ||
955 | ocfs2_free_ac_resource(ac); | |
956 | } | |
957 | ||
958 | return status; | |
959 | } | |
960 | ||
961 | static int ocfs2_steal_inode(struct ocfs2_super *osb, | |
962 | struct ocfs2_alloc_context *ac) | |
963 | { | |
964 | return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE); | |
965 | } | |
966 | ||
967 | static int ocfs2_steal_meta(struct ocfs2_super *osb, | |
968 | struct ocfs2_alloc_context *ac) | |
969 | { | |
970 | return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE); | |
971 | } | |
972 | ||
cf1d6c76 TY |
973 | int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb, |
974 | int blocks, | |
975 | struct ocfs2_alloc_context **ac) | |
ccd979bd MF |
976 | { |
977 | int status; | |
b89c5428 | 978 | int slot = ocfs2_get_meta_steal_slot(osb); |
ccd979bd | 979 | |
cd861280 | 980 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
981 | if (!(*ac)) { |
982 | status = -ENOMEM; | |
983 | mlog_errno(status); | |
984 | goto bail; | |
985 | } | |
986 | ||
cf1d6c76 | 987 | (*ac)->ac_bits_wanted = blocks; |
ccd979bd | 988 | (*ac)->ac_which = OCFS2_AC_USE_META; |
ccd979bd MF |
989 | (*ac)->ac_group_search = ocfs2_block_group_search; |
990 | ||
b89c5428 TY |
991 | if (slot != OCFS2_INVALID_SLOT && |
992 | atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL) | |
993 | goto extent_steal; | |
994 | ||
995 | atomic_set(&osb->s_num_meta_stolen, 0); | |
da5cbf2f | 996 | status = ocfs2_reserve_suballoc_bits(osb, (*ac), |
ffda89a3 | 997 | EXTENT_ALLOC_SYSTEM_INODE, |
b89c5428 | 998 | (u32)osb->slot_num, NULL, |
33d5d380 | 999 | ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP); |
b89c5428 TY |
1000 | |
1001 | ||
1002 | if (status >= 0) { | |
1003 | status = 0; | |
1004 | if (slot != OCFS2_INVALID_SLOT) | |
1005 | ocfs2_init_meta_steal_slot(osb); | |
1006 | goto bail; | |
1007 | } else if (status < 0 && status != -ENOSPC) { | |
1008 | mlog_errno(status); | |
1009 | goto bail; | |
1010 | } | |
1011 | ||
1012 | ocfs2_free_ac_resource(*ac); | |
1013 | ||
1014 | extent_steal: | |
1015 | status = ocfs2_steal_meta(osb, *ac); | |
1016 | atomic_inc(&osb->s_num_meta_stolen); | |
ccd979bd MF |
1017 | if (status < 0) { |
1018 | if (status != -ENOSPC) | |
1019 | mlog_errno(status); | |
1020 | goto bail; | |
1021 | } | |
1022 | ||
1023 | status = 0; | |
1024 | bail: | |
1025 | if ((status < 0) && *ac) { | |
1026 | ocfs2_free_alloc_context(*ac); | |
1027 | *ac = NULL; | |
1028 | } | |
1029 | ||
c1e8d35e TM |
1030 | if (status) |
1031 | mlog_errno(status); | |
ccd979bd MF |
1032 | return status; |
1033 | } | |
1034 | ||
cf1d6c76 TY |
1035 | int ocfs2_reserve_new_metadata(struct ocfs2_super *osb, |
1036 | struct ocfs2_extent_list *root_el, | |
1037 | struct ocfs2_alloc_context **ac) | |
1038 | { | |
1039 | return ocfs2_reserve_new_metadata_blocks(osb, | |
1040 | ocfs2_extend_meta_needed(root_el), | |
1041 | ac); | |
1042 | } | |
1043 | ||
ccd979bd | 1044 | int ocfs2_reserve_new_inode(struct ocfs2_super *osb, |
ccd979bd MF |
1045 | struct ocfs2_alloc_context **ac) |
1046 | { | |
1047 | int status; | |
b89c5428 | 1048 | int slot = ocfs2_get_inode_steal_slot(osb); |
feb473a6 | 1049 | u64 alloc_group; |
ccd979bd | 1050 | |
cd861280 | 1051 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
1052 | if (!(*ac)) { |
1053 | status = -ENOMEM; | |
1054 | mlog_errno(status); | |
1055 | goto bail; | |
1056 | } | |
1057 | ||
1058 | (*ac)->ac_bits_wanted = 1; | |
ccd979bd MF |
1059 | (*ac)->ac_which = OCFS2_AC_USE_INODE; |
1060 | ||
ccd979bd MF |
1061 | (*ac)->ac_group_search = ocfs2_block_group_search; |
1062 | ||
1187c968 JB |
1063 | /* |
1064 | * stat(2) can't handle i_ino > 32bits, so we tell the | |
1065 | * lower levels not to allocate us a block group past that | |
12462f1d | 1066 | * limit. The 'inode64' mount option avoids this behavior. |
1187c968 | 1067 | */ |
12462f1d JB |
1068 | if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64)) |
1069 | (*ac)->ac_max_block = (u32)~0U; | |
1187c968 | 1070 | |
4d0ddb2c TM |
1071 | /* |
1072 | * slot is set when we successfully steal inode from other nodes. | |
1073 | * It is reset in 3 places: | |
1074 | * 1. when we flush the truncate log | |
1075 | * 2. when we complete local alloc recovery. | |
1076 | * 3. when we successfully allocate from our own slot. | |
1077 | * After it is set, we will go on stealing inodes until we find the | |
1078 | * need to check our slots to see whether there is some space for us. | |
1079 | */ | |
1080 | if (slot != OCFS2_INVALID_SLOT && | |
b89c5428 | 1081 | atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL) |
4d0ddb2c TM |
1082 | goto inode_steal; |
1083 | ||
1084 | atomic_set(&osb->s_num_inodes_stolen, 0); | |
feb473a6 | 1085 | alloc_group = osb->osb_inode_alloc_group; |
da5cbf2f MF |
1086 | status = ocfs2_reserve_suballoc_bits(osb, *ac, |
1087 | INODE_ALLOC_SYSTEM_INODE, | |
b89c5428 | 1088 | (u32)osb->slot_num, |
feb473a6 | 1089 | &alloc_group, |
60ca81e8 TM |
1090 | ALLOC_NEW_GROUP | |
1091 | ALLOC_GROUPS_FROM_GLOBAL); | |
4d0ddb2c TM |
1092 | if (status >= 0) { |
1093 | status = 0; | |
1094 | ||
feb473a6 TM |
1095 | spin_lock(&osb->osb_lock); |
1096 | osb->osb_inode_alloc_group = alloc_group; | |
1097 | spin_unlock(&osb->osb_lock); | |
2f73e135 TM |
1098 | trace_ocfs2_reserve_new_inode_new_group( |
1099 | (unsigned long long)alloc_group); | |
feb473a6 | 1100 | |
4d0ddb2c TM |
1101 | /* |
1102 | * Some inodes must be freed by us, so try to allocate | |
1103 | * from our own next time. | |
1104 | */ | |
1105 | if (slot != OCFS2_INVALID_SLOT) | |
1106 | ocfs2_init_inode_steal_slot(osb); | |
1107 | goto bail; | |
1108 | } else if (status < 0 && status != -ENOSPC) { | |
1109 | mlog_errno(status); | |
1110 | goto bail; | |
1111 | } | |
1112 | ||
1113 | ocfs2_free_ac_resource(*ac); | |
1114 | ||
1115 | inode_steal: | |
b89c5428 | 1116 | status = ocfs2_steal_inode(osb, *ac); |
4d0ddb2c | 1117 | atomic_inc(&osb->s_num_inodes_stolen); |
ccd979bd MF |
1118 | if (status < 0) { |
1119 | if (status != -ENOSPC) | |
1120 | mlog_errno(status); | |
1121 | goto bail; | |
1122 | } | |
1123 | ||
1124 | status = 0; | |
1125 | bail: | |
1126 | if ((status < 0) && *ac) { | |
1127 | ocfs2_free_alloc_context(*ac); | |
1128 | *ac = NULL; | |
1129 | } | |
1130 | ||
c1e8d35e TM |
1131 | if (status) |
1132 | mlog_errno(status); | |
ccd979bd MF |
1133 | return status; |
1134 | } | |
1135 | ||
1136 | /* local alloc code has to do the same thing, so rather than do this | |
1137 | * twice.. */ | |
1138 | int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb, | |
1139 | struct ocfs2_alloc_context *ac) | |
1140 | { | |
1141 | int status; | |
1142 | ||
ccd979bd MF |
1143 | ac->ac_which = OCFS2_AC_USE_MAIN; |
1144 | ac->ac_group_search = ocfs2_cluster_group_search; | |
1145 | ||
da5cbf2f MF |
1146 | status = ocfs2_reserve_suballoc_bits(osb, ac, |
1147 | GLOBAL_BITMAP_SYSTEM_INODE, | |
feb473a6 | 1148 | OCFS2_INVALID_SLOT, NULL, |
ffda89a3 | 1149 | ALLOC_NEW_GROUP); |
47ee9d89 | 1150 | if (status < 0 && status != -ENOSPC) |
ccd979bd | 1151 | mlog_errno(status); |
da5cbf2f | 1152 | |
ccd979bd MF |
1153 | return status; |
1154 | } | |
1155 | ||
1156 | /* Callers don't need to care which bitmap (local alloc or main) to | |
1157 | * use so we figure it out for them, but unfortunately this clutters | |
1158 | * things a bit. */ | |
1187c968 JB |
1159 | static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb, |
1160 | u32 bits_wanted, u64 max_block, | |
60ca81e8 | 1161 | int flags, |
1187c968 | 1162 | struct ocfs2_alloc_context **ac) |
ccd979bd | 1163 | { |
2070ad1a ER |
1164 | int status, ret = 0; |
1165 | int retried = 0; | |
ccd979bd | 1166 | |
cd861280 | 1167 | *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL); |
ccd979bd MF |
1168 | if (!(*ac)) { |
1169 | status = -ENOMEM; | |
1170 | mlog_errno(status); | |
1171 | goto bail; | |
1172 | } | |
1173 | ||
1174 | (*ac)->ac_bits_wanted = bits_wanted; | |
1187c968 | 1175 | (*ac)->ac_max_block = max_block; |
ccd979bd MF |
1176 | |
1177 | status = -ENOSPC; | |
60ca81e8 TM |
1178 | if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) && |
1179 | ocfs2_alloc_should_use_local(osb, bits_wanted)) { | |
ccd979bd | 1180 | status = ocfs2_reserve_local_alloc_bits(osb, |
ccd979bd MF |
1181 | bits_wanted, |
1182 | *ac); | |
a57c8fd2 | 1183 | if ((status < 0) && (status != -ENOSPC)) { |
ccd979bd MF |
1184 | mlog_errno(status); |
1185 | goto bail; | |
ccd979bd MF |
1186 | } |
1187 | } | |
1188 | ||
1189 | if (status == -ENOSPC) { | |
2070ad1a | 1190 | retry: |
ccd979bd | 1191 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac); |
2070ad1a ER |
1192 | /* Retry if there is sufficient space cached in truncate log */ |
1193 | if (status == -ENOSPC && !retried) { | |
1194 | retried = 1; | |
1195 | ocfs2_inode_unlock((*ac)->ac_inode, 1); | |
1196 | inode_unlock((*ac)->ac_inode); | |
1197 | ||
1198 | ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted); | |
3bb8b653 JQ |
1199 | if (ret == 1) { |
1200 | iput((*ac)->ac_inode); | |
1201 | (*ac)->ac_inode = NULL; | |
2070ad1a | 1202 | goto retry; |
3bb8b653 | 1203 | } |
2070ad1a ER |
1204 | |
1205 | if (ret < 0) | |
1206 | mlog_errno(ret); | |
1207 | ||
1208 | inode_lock((*ac)->ac_inode); | |
3bb8b653 JQ |
1209 | ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1); |
1210 | if (ret < 0) { | |
1211 | mlog_errno(ret); | |
1212 | inode_unlock((*ac)->ac_inode); | |
1213 | iput((*ac)->ac_inode); | |
1214 | (*ac)->ac_inode = NULL; | |
1215 | goto bail; | |
1216 | } | |
2070ad1a | 1217 | } |
ccd979bd MF |
1218 | if (status < 0) { |
1219 | if (status != -ENOSPC) | |
1220 | mlog_errno(status); | |
1221 | goto bail; | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | status = 0; | |
1226 | bail: | |
1227 | if ((status < 0) && *ac) { | |
1228 | ocfs2_free_alloc_context(*ac); | |
1229 | *ac = NULL; | |
1230 | } | |
1231 | ||
c1e8d35e TM |
1232 | if (status) |
1233 | mlog_errno(status); | |
ccd979bd MF |
1234 | return status; |
1235 | } | |
1236 | ||
1187c968 JB |
1237 | int ocfs2_reserve_clusters(struct ocfs2_super *osb, |
1238 | u32 bits_wanted, | |
1239 | struct ocfs2_alloc_context **ac) | |
1240 | { | |
60ca81e8 TM |
1241 | return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0, |
1242 | ALLOC_NEW_GROUP, ac); | |
1187c968 JB |
1243 | } |
1244 | ||
ccd979bd MF |
1245 | /* |
1246 | * More or less lifted from ext3. I'll leave their description below: | |
1247 | * | |
1248 | * "For ext3 allocations, we must not reuse any blocks which are | |
1249 | * allocated in the bitmap buffer's "last committed data" copy. This | |
1250 | * prevents deletes from freeing up the page for reuse until we have | |
1251 | * committed the delete transaction. | |
1252 | * | |
1253 | * If we didn't do this, then deleting something and reallocating it as | |
1254 | * data would allow the old block to be overwritten before the | |
1255 | * transaction committed (because we force data to disk before commit). | |
1256 | * This would lead to corruption if we crashed between overwriting the | |
1257 | * data and committing the delete. | |
1258 | * | |
1259 | * @@@ We may want to make this allocation behaviour conditional on | |
1260 | * data-writes at some point, and disable it for metadata allocations or | |
1261 | * sync-data inodes." | |
1262 | * | |
1263 | * Note: OCFS2 already does this differently for metadata vs data | |
c78bad11 | 1264 | * allocations, as those bitmaps are separate and undo access is never |
ccd979bd MF |
1265 | * called on a metadata group descriptor. |
1266 | */ | |
1267 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh, | |
1268 | int nr) | |
1269 | { | |
1270 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
94e41ecf | 1271 | int ret; |
ccd979bd MF |
1272 | |
1273 | if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap)) | |
1274 | return 0; | |
94e41ecf SM |
1275 | |
1276 | if (!buffer_jbd(bg_bh)) | |
ccd979bd MF |
1277 | return 1; |
1278 | ||
94e41ecf | 1279 | jbd_lock_bh_state(bg_bh); |
ccd979bd | 1280 | bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data; |
94e41ecf SM |
1281 | if (bg) |
1282 | ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap); | |
1283 | else | |
1284 | ret = 1; | |
1285 | jbd_unlock_bh_state(bg_bh); | |
1286 | ||
1287 | return ret; | |
ccd979bd MF |
1288 | } |
1289 | ||
1290 | static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb, | |
1291 | struct buffer_head *bg_bh, | |
1292 | unsigned int bits_wanted, | |
7bf72ede | 1293 | unsigned int total_bits, |
7d1fe093 | 1294 | struct ocfs2_suballoc_result *res) |
ccd979bd MF |
1295 | { |
1296 | void *bitmap; | |
1297 | u16 best_offset, best_size; | |
1298 | int offset, start, found, status = 0; | |
1299 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
1300 | ||
42035306 JB |
1301 | /* Callers got this descriptor from |
1302 | * ocfs2_read_group_descriptor(). Any corruption is a code bug. */ | |
1303 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg)); | |
ccd979bd MF |
1304 | |
1305 | found = start = best_offset = best_size = 0; | |
1306 | bitmap = bg->bg_bitmap; | |
1307 | ||
7bf72ede MF |
1308 | while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) { |
1309 | if (offset == total_bits) | |
ccd979bd MF |
1310 | break; |
1311 | ||
1312 | if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) { | |
1313 | /* We found a zero, but we can't use it as it | |
1314 | * hasn't been put to disk yet! */ | |
1315 | found = 0; | |
1316 | start = offset + 1; | |
1317 | } else if (offset == start) { | |
1318 | /* we found a zero */ | |
1319 | found++; | |
1320 | /* move start to the next bit to test */ | |
1321 | start++; | |
1322 | } else { | |
1323 | /* got a zero after some ones */ | |
1324 | found = 1; | |
1325 | start = offset + 1; | |
1326 | } | |
1327 | if (found > best_size) { | |
1328 | best_size = found; | |
1329 | best_offset = start - found; | |
1330 | } | |
1331 | /* we got everything we needed */ | |
1332 | if (found == bits_wanted) { | |
1333 | /* mlog(0, "Found it all!\n"); */ | |
1334 | break; | |
1335 | } | |
1336 | } | |
1337 | ||
7d1fe093 JB |
1338 | if (best_size) { |
1339 | res->sr_bit_offset = best_offset; | |
1340 | res->sr_bits = best_size; | |
ccd979bd MF |
1341 | } else { |
1342 | status = -ENOSPC; | |
1343 | /* No error log here -- see the comment above | |
1344 | * ocfs2_test_bg_bit_allocatable */ | |
1345 | } | |
1346 | ||
1347 | return status; | |
1348 | } | |
1349 | ||
0a2fcd89 | 1350 | int ocfs2_block_group_set_bits(handle_t *handle, |
ccd979bd MF |
1351 | struct inode *alloc_inode, |
1352 | struct ocfs2_group_desc *bg, | |
1353 | struct buffer_head *group_bh, | |
1354 | unsigned int bit_off, | |
1355 | unsigned int num_bits) | |
1356 | { | |
1357 | int status; | |
1358 | void *bitmap = bg->bg_bitmap; | |
1359 | int journal_type = OCFS2_JOURNAL_ACCESS_WRITE; | |
1360 | ||
42035306 JB |
1361 | /* All callers get the descriptor via |
1362 | * ocfs2_read_group_descriptor(). Any corruption is a code bug. */ | |
1363 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg)); | |
ccd979bd MF |
1364 | BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits); |
1365 | ||
2f73e135 | 1366 | trace_ocfs2_block_group_set_bits(bit_off, num_bits); |
ccd979bd MF |
1367 | |
1368 | if (ocfs2_is_cluster_bitmap(alloc_inode)) | |
1369 | journal_type = OCFS2_JOURNAL_ACCESS_UNDO; | |
1370 | ||
13723d00 | 1371 | status = ocfs2_journal_access_gd(handle, |
0cf2f763 | 1372 | INODE_CACHE(alloc_inode), |
13723d00 JB |
1373 | group_bh, |
1374 | journal_type); | |
ccd979bd MF |
1375 | if (status < 0) { |
1376 | mlog_errno(status); | |
1377 | goto bail; | |
1378 | } | |
1379 | ||
1380 | le16_add_cpu(&bg->bg_free_bits_count, -num_bits); | |
9b5cd10e | 1381 | if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) { |
7ecef14a JP |
1382 | return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n", |
1383 | (unsigned long long)le64_to_cpu(bg->bg_blkno), | |
1384 | le16_to_cpu(bg->bg_bits), | |
1385 | le16_to_cpu(bg->bg_free_bits_count), | |
1386 | num_bits); | |
9b5cd10e | 1387 | } |
ccd979bd MF |
1388 | while(num_bits--) |
1389 | ocfs2_set_bit(bit_off++, bitmap); | |
1390 | ||
ec20cec7 | 1391 | ocfs2_journal_dirty(handle, group_bh); |
ccd979bd MF |
1392 | |
1393 | bail: | |
ccd979bd MF |
1394 | return status; |
1395 | } | |
1396 | ||
1397 | /* find the one with the most empty bits */ | |
1398 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl) | |
1399 | { | |
1400 | u16 curr, best; | |
1401 | ||
1402 | BUG_ON(!cl->cl_next_free_rec); | |
1403 | ||
1404 | best = curr = 0; | |
1405 | while (curr < le16_to_cpu(cl->cl_next_free_rec)) { | |
1406 | if (le32_to_cpu(cl->cl_recs[curr].c_free) > | |
1407 | le32_to_cpu(cl->cl_recs[best].c_free)) | |
1408 | best = curr; | |
1409 | curr++; | |
1410 | } | |
1411 | ||
1412 | BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec)); | |
1413 | return best; | |
1414 | } | |
1415 | ||
1fabe148 | 1416 | static int ocfs2_relink_block_group(handle_t *handle, |
ccd979bd MF |
1417 | struct inode *alloc_inode, |
1418 | struct buffer_head *fe_bh, | |
1419 | struct buffer_head *bg_bh, | |
1420 | struct buffer_head *prev_bg_bh, | |
1421 | u16 chain) | |
1422 | { | |
1423 | int status; | |
1424 | /* there is a really tiny chance the journal calls could fail, | |
1425 | * but we wouldn't want inconsistent blocks in *any* case. */ | |
49309841 | 1426 | u64 bg_ptr, prev_bg_ptr; |
ccd979bd MF |
1427 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; |
1428 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
1429 | struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data; | |
1430 | ||
42035306 JB |
1431 | /* The caller got these descriptors from |
1432 | * ocfs2_read_group_descriptor(). Any corruption is a code bug. */ | |
1433 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg)); | |
1434 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg)); | |
ccd979bd | 1435 | |
2f73e135 TM |
1436 | trace_ocfs2_relink_block_group( |
1437 | (unsigned long long)le64_to_cpu(fe->i_blkno), chain, | |
1438 | (unsigned long long)le64_to_cpu(bg->bg_blkno), | |
1439 | (unsigned long long)le64_to_cpu(prev_bg->bg_blkno)); | |
ccd979bd | 1440 | |
ccd979bd MF |
1441 | bg_ptr = le64_to_cpu(bg->bg_next_group); |
1442 | prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group); | |
1443 | ||
0cf2f763 JB |
1444 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode), |
1445 | prev_bg_bh, | |
13723d00 | 1446 | OCFS2_JOURNAL_ACCESS_WRITE); |
25e28921 | 1447 | if (status < 0) |
49309841 | 1448 | goto out; |
ccd979bd MF |
1449 | |
1450 | prev_bg->bg_next_group = bg->bg_next_group; | |
ec20cec7 | 1451 | ocfs2_journal_dirty(handle, prev_bg_bh); |
ccd979bd | 1452 | |
0cf2f763 JB |
1453 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode), |
1454 | bg_bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
25e28921 | 1455 | if (status < 0) |
49309841 | 1456 | goto out_rollback_prev_bg; |
ccd979bd MF |
1457 | |
1458 | bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno; | |
ec20cec7 | 1459 | ocfs2_journal_dirty(handle, bg_bh); |
ccd979bd | 1460 | |
0cf2f763 JB |
1461 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode), |
1462 | fe_bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
25e28921 | 1463 | if (status < 0) |
49309841 | 1464 | goto out_rollback_bg; |
ccd979bd MF |
1465 | |
1466 | fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno; | |
ec20cec7 | 1467 | ocfs2_journal_dirty(handle, fe_bh); |
ccd979bd | 1468 | |
49309841 | 1469 | out: |
25e28921 AM |
1470 | if (status < 0) |
1471 | mlog_errno(status); | |
49309841 | 1472 | return status; |
42035306 | 1473 | |
49309841 JL |
1474 | out_rollback_bg: |
1475 | bg->bg_next_group = cpu_to_le64(bg_ptr); | |
1476 | out_rollback_prev_bg: | |
1477 | prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr); | |
25e28921 | 1478 | goto out; |
ccd979bd MF |
1479 | } |
1480 | ||
1481 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg, | |
1482 | u32 wanted) | |
1483 | { | |
1484 | return le16_to_cpu(bg->bg_free_bits_count) > wanted; | |
1485 | } | |
1486 | ||
1487 | /* return 0 on success, -ENOSPC to keep searching and any other < 0 | |
1488 | * value on error. */ | |
1489 | static int ocfs2_cluster_group_search(struct inode *inode, | |
1490 | struct buffer_head *group_bh, | |
1491 | u32 bits_wanted, u32 min_bits, | |
1187c968 | 1492 | u64 max_block, |
7d1fe093 | 1493 | struct ocfs2_suballoc_result *res) |
ccd979bd MF |
1494 | { |
1495 | int search = -ENOSPC; | |
1496 | int ret; | |
1187c968 | 1497 | u64 blkoff; |
7bf72ede | 1498 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data; |
9c7af40b | 1499 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
7bf72ede | 1500 | unsigned int max_bits, gd_cluster_off; |
ccd979bd MF |
1501 | |
1502 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
1503 | ||
7bf72ede MF |
1504 | if (gd->bg_free_bits_count) { |
1505 | max_bits = le16_to_cpu(gd->bg_bits); | |
1506 | ||
1507 | /* Tail groups in cluster bitmaps which aren't cpg | |
25985edc | 1508 | * aligned are prone to partial extension by a failed |
7bf72ede MF |
1509 | * fs resize. If the file system resize never got to |
1510 | * update the dinode cluster count, then we don't want | |
1511 | * to trust any clusters past it, regardless of what | |
1512 | * the group descriptor says. */ | |
1513 | gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb, | |
1514 | le64_to_cpu(gd->bg_blkno)); | |
1515 | if ((gd_cluster_off + max_bits) > | |
1516 | OCFS2_I(inode)->ip_clusters) { | |
1517 | max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off; | |
2f73e135 TM |
1518 | trace_ocfs2_cluster_group_search_wrong_max_bits( |
1519 | (unsigned long long)le64_to_cpu(gd->bg_blkno), | |
1520 | le16_to_cpu(gd->bg_bits), | |
1521 | OCFS2_I(inode)->ip_clusters, max_bits); | |
7bf72ede MF |
1522 | } |
1523 | ||
ccd979bd MF |
1524 | ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb), |
1525 | group_bh, bits_wanted, | |
7d1fe093 | 1526 | max_bits, res); |
ccd979bd MF |
1527 | if (ret) |
1528 | return ret; | |
1529 | ||
1187c968 JB |
1530 | if (max_block) { |
1531 | blkoff = ocfs2_clusters_to_blocks(inode->i_sb, | |
1532 | gd_cluster_off + | |
7d1fe093 JB |
1533 | res->sr_bit_offset + |
1534 | res->sr_bits); | |
2f73e135 TM |
1535 | trace_ocfs2_cluster_group_search_max_block( |
1536 | (unsigned long long)blkoff, | |
1537 | (unsigned long long)max_block); | |
1187c968 JB |
1538 | if (blkoff > max_block) |
1539 | return -ENOSPC; | |
1540 | } | |
1541 | ||
ccd979bd MF |
1542 | /* ocfs2_block_group_find_clear_bits() might |
1543 | * return success, but we still want to return | |
1544 | * -ENOSPC unless it found the minimum number | |
1545 | * of bits. */ | |
7d1fe093 | 1546 | if (min_bits <= res->sr_bits) |
ccd979bd | 1547 | search = 0; /* success */ |
7d1fe093 | 1548 | else if (res->sr_bits) { |
9c7af40b MF |
1549 | /* |
1550 | * Don't show bits which we'll be returning | |
1551 | * for allocation to the local alloc bitmap. | |
1552 | */ | |
7d1fe093 | 1553 | ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits); |
ccd979bd MF |
1554 | } |
1555 | } | |
1556 | ||
1557 | return search; | |
1558 | } | |
1559 | ||
1560 | static int ocfs2_block_group_search(struct inode *inode, | |
1561 | struct buffer_head *group_bh, | |
1562 | u32 bits_wanted, u32 min_bits, | |
1187c968 | 1563 | u64 max_block, |
7d1fe093 | 1564 | struct ocfs2_suballoc_result *res) |
ccd979bd MF |
1565 | { |
1566 | int ret = -ENOSPC; | |
1187c968 | 1567 | u64 blkoff; |
ccd979bd MF |
1568 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data; |
1569 | ||
1570 | BUG_ON(min_bits != 1); | |
1571 | BUG_ON(ocfs2_is_cluster_bitmap(inode)); | |
1572 | ||
1187c968 | 1573 | if (bg->bg_free_bits_count) { |
ccd979bd MF |
1574 | ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb), |
1575 | group_bh, bits_wanted, | |
7bf72ede | 1576 | le16_to_cpu(bg->bg_bits), |
7d1fe093 | 1577 | res); |
1187c968 | 1578 | if (!ret && max_block) { |
7d1fe093 JB |
1579 | blkoff = le64_to_cpu(bg->bg_blkno) + |
1580 | res->sr_bit_offset + res->sr_bits; | |
2f73e135 TM |
1581 | trace_ocfs2_block_group_search_max_block( |
1582 | (unsigned long long)blkoff, | |
1583 | (unsigned long long)max_block); | |
1187c968 JB |
1584 | if (blkoff > max_block) |
1585 | ret = -ENOSPC; | |
1586 | } | |
1587 | } | |
ccd979bd MF |
1588 | |
1589 | return ret; | |
1590 | } | |
1591 | ||
0a2fcd89 | 1592 | int ocfs2_alloc_dinode_update_counts(struct inode *inode, |
1fabe148 | 1593 | handle_t *handle, |
883d4cae MF |
1594 | struct buffer_head *di_bh, |
1595 | u32 num_bits, | |
1596 | u16 chain) | |
1597 | { | |
1598 | int ret; | |
1599 | u32 tmp_used; | |
1600 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
1601 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain; | |
1602 | ||
0cf2f763 | 1603 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, |
13723d00 | 1604 | OCFS2_JOURNAL_ACCESS_WRITE); |
883d4cae MF |
1605 | if (ret < 0) { |
1606 | mlog_errno(ret); | |
1607 | goto out; | |
1608 | } | |
1609 | ||
1610 | tmp_used = le32_to_cpu(di->id1.bitmap1.i_used); | |
1611 | di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used); | |
1612 | le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits); | |
ec20cec7 | 1613 | ocfs2_journal_dirty(handle, di_bh); |
883d4cae MF |
1614 | |
1615 | out: | |
1616 | return ret; | |
1617 | } | |
1618 | ||
db66c715 YL |
1619 | void ocfs2_rollback_alloc_dinode_counts(struct inode *inode, |
1620 | struct buffer_head *di_bh, | |
1621 | u32 num_bits, | |
1622 | u16 chain) | |
1623 | { | |
1624 | u32 tmp_used; | |
1625 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data; | |
1626 | struct ocfs2_chain_list *cl; | |
1627 | ||
1628 | cl = (struct ocfs2_chain_list *)&di->id2.i_chain; | |
1629 | tmp_used = le32_to_cpu(di->id1.bitmap1.i_used); | |
1630 | di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits); | |
1631 | le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits); | |
1632 | } | |
1633 | ||
ba206635 JB |
1634 | static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res, |
1635 | struct ocfs2_extent_rec *rec, | |
1636 | struct ocfs2_chain_list *cl) | |
13e434cf JB |
1637 | { |
1638 | unsigned int bpc = le16_to_cpu(cl->cl_bpc); | |
1639 | unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc; | |
72094e43 | 1640 | unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc; |
13e434cf JB |
1641 | |
1642 | if (res->sr_bit_offset < bitoff) | |
1643 | return 0; | |
1644 | if (res->sr_bit_offset >= (bitoff + bitcount)) | |
1645 | return 0; | |
ba206635 JB |
1646 | res->sr_blkno = le64_to_cpu(rec->e_blkno) + |
1647 | (res->sr_bit_offset - bitoff); | |
13e434cf JB |
1648 | if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount)) |
1649 | res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset; | |
1650 | return 1; | |
1651 | } | |
1652 | ||
ba206635 JB |
1653 | static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac, |
1654 | struct ocfs2_group_desc *bg, | |
1655 | struct ocfs2_suballoc_result *res) | |
13e434cf JB |
1656 | { |
1657 | int i; | |
2b6cb576 | 1658 | u64 bg_blkno = res->sr_bg_blkno; /* Save off */ |
13e434cf JB |
1659 | struct ocfs2_extent_rec *rec; |
1660 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data; | |
1661 | struct ocfs2_chain_list *cl = &di->id2.i_chain; | |
1662 | ||
ba206635 JB |
1663 | if (ocfs2_is_cluster_bitmap(ac->ac_inode)) { |
1664 | res->sr_blkno = 0; | |
13e434cf | 1665 | return; |
ba206635 | 1666 | } |
13e434cf | 1667 | |
ba206635 | 1668 | res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset; |
2b6cb576 | 1669 | res->sr_bg_blkno = 0; /* Clear it for contig block groups */ |
4711954e | 1670 | if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) || |
ba206635 | 1671 | !bg->bg_list.l_next_free_rec) |
13e434cf JB |
1672 | return; |
1673 | ||
1674 | for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) { | |
1675 | rec = &bg->bg_list.l_recs[i]; | |
2b6cb576 JB |
1676 | if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) { |
1677 | res->sr_bg_blkno = bg_blkno; /* Restore */ | |
13e434cf | 1678 | break; |
2b6cb576 | 1679 | } |
13e434cf JB |
1680 | } |
1681 | } | |
1682 | ||
883d4cae | 1683 | static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac, |
1fabe148 | 1684 | handle_t *handle, |
883d4cae MF |
1685 | u32 bits_wanted, |
1686 | u32 min_bits, | |
7d1fe093 | 1687 | struct ocfs2_suballoc_result *res, |
883d4cae MF |
1688 | u16 *bits_left) |
1689 | { | |
1690 | int ret; | |
883d4cae MF |
1691 | struct buffer_head *group_bh = NULL; |
1692 | struct ocfs2_group_desc *gd; | |
68f64d47 | 1693 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data; |
883d4cae | 1694 | struct inode *alloc_inode = ac->ac_inode; |
883d4cae | 1695 | |
7d1fe093 JB |
1696 | ret = ocfs2_read_group_descriptor(alloc_inode, di, |
1697 | res->sr_bg_blkno, &group_bh); | |
883d4cae MF |
1698 | if (ret < 0) { |
1699 | mlog_errno(ret); | |
1700 | return ret; | |
1701 | } | |
1702 | ||
1703 | gd = (struct ocfs2_group_desc *) group_bh->b_data; | |
883d4cae | 1704 | ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits, |
7d1fe093 | 1705 | ac->ac_max_block, res); |
883d4cae MF |
1706 | if (ret < 0) { |
1707 | if (ret != -ENOSPC) | |
1708 | mlog_errno(ret); | |
1709 | goto out; | |
1710 | } | |
1711 | ||
13e434cf | 1712 | if (!ret) |
ba206635 | 1713 | ocfs2_bg_discontig_fix_result(ac, gd, res); |
13e434cf | 1714 | |
e49e2767 MF |
1715 | /* |
1716 | * sr_bg_blkno might have been changed by | |
1717 | * ocfs2_bg_discontig_fix_result | |
1718 | */ | |
1719 | res->sr_bg_stable_blkno = group_bh->b_blocknr; | |
1720 | ||
1721 | if (ac->ac_find_loc_only) | |
1722 | goto out_loc_only; | |
1723 | ||
883d4cae | 1724 | ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh, |
7d1fe093 | 1725 | res->sr_bits, |
883d4cae MF |
1726 | le16_to_cpu(gd->bg_chain)); |
1727 | if (ret < 0) { | |
1728 | mlog_errno(ret); | |
1729 | goto out; | |
1730 | } | |
1731 | ||
1732 | ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh, | |
7d1fe093 | 1733 | res->sr_bit_offset, res->sr_bits); |
db66c715 YL |
1734 | if (ret < 0) { |
1735 | ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh, | |
1736 | res->sr_bits, | |
1737 | le16_to_cpu(gd->bg_chain)); | |
883d4cae | 1738 | mlog_errno(ret); |
db66c715 | 1739 | } |
883d4cae | 1740 | |
e49e2767 | 1741 | out_loc_only: |
883d4cae MF |
1742 | *bits_left = le16_to_cpu(gd->bg_free_bits_count); |
1743 | ||
1744 | out: | |
1745 | brelse(group_bh); | |
1746 | ||
1747 | return ret; | |
1748 | } | |
1749 | ||
ccd979bd | 1750 | static int ocfs2_search_chain(struct ocfs2_alloc_context *ac, |
1fabe148 | 1751 | handle_t *handle, |
ccd979bd MF |
1752 | u32 bits_wanted, |
1753 | u32 min_bits, | |
7d1fe093 | 1754 | struct ocfs2_suballoc_result *res, |
883d4cae | 1755 | u16 *bits_left) |
ccd979bd MF |
1756 | { |
1757 | int status; | |
7d1fe093 | 1758 | u16 chain; |
ccd979bd | 1759 | u64 next_group; |
ccd979bd MF |
1760 | struct inode *alloc_inode = ac->ac_inode; |
1761 | struct buffer_head *group_bh = NULL; | |
1762 | struct buffer_head *prev_group_bh = NULL; | |
1763 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data; | |
1764 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain; | |
1765 | struct ocfs2_group_desc *bg; | |
1766 | ||
1767 | chain = ac->ac_chain; | |
2f73e135 TM |
1768 | trace_ocfs2_search_chain_begin( |
1769 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, | |
1770 | bits_wanted, chain); | |
ccd979bd | 1771 | |
68f64d47 JB |
1772 | status = ocfs2_read_group_descriptor(alloc_inode, fe, |
1773 | le64_to_cpu(cl->cl_recs[chain].c_blkno), | |
1774 | &group_bh); | |
ccd979bd MF |
1775 | if (status < 0) { |
1776 | mlog_errno(status); | |
1777 | goto bail; | |
1778 | } | |
1779 | bg = (struct ocfs2_group_desc *) group_bh->b_data; | |
ccd979bd MF |
1780 | |
1781 | status = -ENOSPC; | |
1782 | /* for now, the chain search is a bit simplistic. We just use | |
1783 | * the 1st group with any empty bits. */ | |
1187c968 JB |
1784 | while ((status = ac->ac_group_search(alloc_inode, group_bh, |
1785 | bits_wanted, min_bits, | |
7d1fe093 JB |
1786 | ac->ac_max_block, |
1787 | res)) == -ENOSPC) { | |
ccd979bd MF |
1788 | if (!bg->bg_next_group) |
1789 | break; | |
a81cb88b MF |
1790 | |
1791 | brelse(prev_group_bh); | |
1792 | prev_group_bh = NULL; | |
1793 | ||
ccd979bd MF |
1794 | next_group = le64_to_cpu(bg->bg_next_group); |
1795 | prev_group_bh = group_bh; | |
1796 | group_bh = NULL; | |
68f64d47 JB |
1797 | status = ocfs2_read_group_descriptor(alloc_inode, fe, |
1798 | next_group, &group_bh); | |
ccd979bd MF |
1799 | if (status < 0) { |
1800 | mlog_errno(status); | |
1801 | goto bail; | |
1802 | } | |
1803 | bg = (struct ocfs2_group_desc *) group_bh->b_data; | |
ccd979bd MF |
1804 | } |
1805 | if (status < 0) { | |
1806 | if (status != -ENOSPC) | |
1807 | mlog_errno(status); | |
1808 | goto bail; | |
1809 | } | |
1810 | ||
2f73e135 TM |
1811 | trace_ocfs2_search_chain_succ( |
1812 | (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits); | |
ccd979bd | 1813 | |
7d1fe093 | 1814 | res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno); |
ccd979bd | 1815 | |
7d1fe093 | 1816 | BUG_ON(res->sr_bits == 0); |
13e434cf | 1817 | if (!status) |
ba206635 | 1818 | ocfs2_bg_discontig_fix_result(ac, bg, res); |
13e434cf | 1819 | |
e49e2767 MF |
1820 | /* |
1821 | * sr_bg_blkno might have been changed by | |
1822 | * ocfs2_bg_discontig_fix_result | |
1823 | */ | |
1824 | res->sr_bg_stable_blkno = group_bh->b_blocknr; | |
ccd979bd MF |
1825 | |
1826 | /* | |
1827 | * Keep track of previous block descriptor read. When | |
1828 | * we find a target, if we have read more than X | |
1829 | * number of descriptors, and the target is reasonably | |
1830 | * empty, relink him to top of his chain. | |
1831 | * | |
1832 | * We've read 0 extra blocks and only send one more to | |
1833 | * the transaction, yet the next guy to search has a | |
1834 | * much easier time. | |
1835 | * | |
1836 | * Do this *after* figuring out how many bits we're taking out | |
1837 | * of our target group. | |
1838 | */ | |
309a85b6 | 1839 | if (!ac->ac_disable_chain_relink && |
ccd979bd | 1840 | (prev_group_bh) && |
7d1fe093 | 1841 | (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) { |
ccd979bd MF |
1842 | status = ocfs2_relink_block_group(handle, alloc_inode, |
1843 | ac->ac_bh, group_bh, | |
1844 | prev_group_bh, chain); | |
1845 | if (status < 0) { | |
1846 | mlog_errno(status); | |
1847 | goto bail; | |
1848 | } | |
1849 | } | |
1850 | ||
e49e2767 MF |
1851 | if (ac->ac_find_loc_only) |
1852 | goto out_loc_only; | |
1853 | ||
d5134982 MF |
1854 | status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, |
1855 | ac->ac_bh, res->sr_bits, | |
1856 | chain); | |
1857 | if (status) { | |
ccd979bd MF |
1858 | mlog_errno(status); |
1859 | goto bail; | |
1860 | } | |
1861 | ||
ccd979bd MF |
1862 | status = ocfs2_block_group_set_bits(handle, |
1863 | alloc_inode, | |
1864 | bg, | |
1865 | group_bh, | |
7d1fe093 JB |
1866 | res->sr_bit_offset, |
1867 | res->sr_bits); | |
ccd979bd | 1868 | if (status < 0) { |
db66c715 YL |
1869 | ocfs2_rollback_alloc_dinode_counts(alloc_inode, |
1870 | ac->ac_bh, res->sr_bits, chain); | |
ccd979bd MF |
1871 | mlog_errno(status); |
1872 | goto bail; | |
1873 | } | |
1874 | ||
2f73e135 TM |
1875 | trace_ocfs2_search_chain_end( |
1876 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
1877 | res->sr_bits); | |
ccd979bd | 1878 | |
e49e2767 | 1879 | out_loc_only: |
883d4cae | 1880 | *bits_left = le16_to_cpu(bg->bg_free_bits_count); |
ccd979bd | 1881 | bail: |
a81cb88b MF |
1882 | brelse(group_bh); |
1883 | brelse(prev_group_bh); | |
ccd979bd | 1884 | |
c1e8d35e TM |
1885 | if (status) |
1886 | mlog_errno(status); | |
ccd979bd MF |
1887 | return status; |
1888 | } | |
1889 | ||
1890 | /* will give out up to bits_wanted contiguous bits. */ | |
aa8f8e93 | 1891 | static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac, |
1fabe148 | 1892 | handle_t *handle, |
ccd979bd MF |
1893 | u32 bits_wanted, |
1894 | u32 min_bits, | |
7d1fe093 | 1895 | struct ocfs2_suballoc_result *res) |
ccd979bd MF |
1896 | { |
1897 | int status; | |
1898 | u16 victim, i; | |
883d4cae | 1899 | u16 bits_left = 0; |
b2b6ebf5 | 1900 | u64 hint = ac->ac_last_group; |
ccd979bd MF |
1901 | struct ocfs2_chain_list *cl; |
1902 | struct ocfs2_dinode *fe; | |
1903 | ||
ccd979bd MF |
1904 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted); |
1905 | BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given)); | |
1906 | BUG_ON(!ac->ac_bh); | |
1907 | ||
1908 | fe = (struct ocfs2_dinode *) ac->ac_bh->b_data; | |
10995aa2 JB |
1909 | |
1910 | /* The bh was validated by the inode read during | |
1911 | * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */ | |
1912 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
1913 | ||
ccd979bd MF |
1914 | if (le32_to_cpu(fe->id1.bitmap1.i_used) >= |
1915 | le32_to_cpu(fe->id1.bitmap1.i_total)) { | |
17a5b9ab | 1916 | status = ocfs2_error(ac->ac_inode->i_sb, |
7ecef14a JP |
1917 | "Chain allocator dinode %llu has %u used bits but only %u total\n", |
1918 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
1919 | le32_to_cpu(fe->id1.bitmap1.i_used), | |
1920 | le32_to_cpu(fe->id1.bitmap1.i_total)); | |
ccd979bd MF |
1921 | goto bail; |
1922 | } | |
1923 | ||
b2b6ebf5 | 1924 | res->sr_bg_blkno = hint; |
7d1fe093 | 1925 | if (res->sr_bg_blkno) { |
883d4cae MF |
1926 | /* Attempt to short-circuit the usual search mechanism |
1927 | * by jumping straight to the most recently used | |
b595076a | 1928 | * allocation group. This helps us maintain some |
883d4cae | 1929 | * contiguousness across allocations. */ |
da5cbf2f | 1930 | status = ocfs2_search_one_group(ac, handle, bits_wanted, |
7d1fe093 JB |
1931 | min_bits, res, &bits_left); |
1932 | if (!status) | |
883d4cae | 1933 | goto set_hint; |
883d4cae MF |
1934 | if (status < 0 && status != -ENOSPC) { |
1935 | mlog_errno(status); | |
1936 | goto bail; | |
1937 | } | |
1938 | } | |
1939 | ||
ccd979bd MF |
1940 | cl = (struct ocfs2_chain_list *) &fe->id2.i_chain; |
1941 | ||
1942 | victim = ocfs2_find_victim_chain(cl); | |
1943 | ac->ac_chain = victim; | |
ccd979bd | 1944 | |
7d1fe093 JB |
1945 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, |
1946 | res, &bits_left); | |
b2b6ebf5 | 1947 | if (!status) { |
1d1aff8c JQ |
1948 | if (ocfs2_is_cluster_bitmap(ac->ac_inode)) |
1949 | hint = res->sr_bg_blkno; | |
1950 | else | |
1951 | hint = ocfs2_group_from_res(res); | |
883d4cae | 1952 | goto set_hint; |
b2b6ebf5 | 1953 | } |
ccd979bd MF |
1954 | if (status < 0 && status != -ENOSPC) { |
1955 | mlog_errno(status); | |
1956 | goto bail; | |
1957 | } | |
1958 | ||
2f73e135 | 1959 | trace_ocfs2_claim_suballoc_bits(victim); |
ccd979bd MF |
1960 | |
1961 | /* If we didn't pick a good victim, then just default to | |
1962 | * searching each chain in order. Don't allow chain relinking | |
1963 | * because we only calculate enough journal credits for one | |
1964 | * relink per alloc. */ | |
309a85b6 | 1965 | ac->ac_disable_chain_relink = 1; |
ccd979bd MF |
1966 | for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) { |
1967 | if (i == victim) | |
1968 | continue; | |
1969 | if (!cl->cl_recs[i].c_free) | |
1970 | continue; | |
1971 | ||
1972 | ac->ac_chain = i; | |
da5cbf2f | 1973 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, |
7d1fe093 | 1974 | res, &bits_left); |
b2b6ebf5 MF |
1975 | if (!status) { |
1976 | hint = ocfs2_group_from_res(res); | |
ccd979bd | 1977 | break; |
b2b6ebf5 | 1978 | } |
ccd979bd MF |
1979 | if (status < 0 && status != -ENOSPC) { |
1980 | mlog_errno(status); | |
1981 | goto bail; | |
1982 | } | |
1983 | } | |
ccd979bd | 1984 | |
883d4cae MF |
1985 | set_hint: |
1986 | if (status != -ENOSPC) { | |
1987 | /* If the next search of this group is not likely to | |
1988 | * yield a suitable extent, then we reset the last | |
1989 | * group hint so as to not waste a disk read */ | |
1990 | if (bits_left < min_bits) | |
1991 | ac->ac_last_group = 0; | |
1992 | else | |
b2b6ebf5 | 1993 | ac->ac_last_group = hint; |
883d4cae MF |
1994 | } |
1995 | ||
1996 | bail: | |
c1e8d35e TM |
1997 | if (status) |
1998 | mlog_errno(status); | |
ccd979bd MF |
1999 | return status; |
2000 | } | |
2001 | ||
1ed9b777 | 2002 | int ocfs2_claim_metadata(handle_t *handle, |
ccd979bd MF |
2003 | struct ocfs2_alloc_context *ac, |
2004 | u32 bits_wanted, | |
2b6cb576 | 2005 | u64 *suballoc_loc, |
ccd979bd MF |
2006 | u16 *suballoc_bit_start, |
2007 | unsigned int *num_bits, | |
2008 | u64 *blkno_start) | |
2009 | { | |
2010 | int status; | |
ba206635 | 2011 | struct ocfs2_suballoc_result res = { .sr_blkno = 0, }; |
ccd979bd MF |
2012 | |
2013 | BUG_ON(!ac); | |
2014 | BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted)); | |
2015 | BUG_ON(ac->ac_which != OCFS2_AC_USE_META); | |
ccd979bd | 2016 | |
aa8f8e93 | 2017 | status = ocfs2_claim_suballoc_bits(ac, |
da5cbf2f | 2018 | handle, |
ccd979bd MF |
2019 | bits_wanted, |
2020 | 1, | |
7d1fe093 | 2021 | &res); |
ccd979bd MF |
2022 | if (status < 0) { |
2023 | mlog_errno(status); | |
2024 | goto bail; | |
2025 | } | |
1ed9b777 | 2026 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs); |
ccd979bd | 2027 | |
2b6cb576 | 2028 | *suballoc_loc = res.sr_bg_blkno; |
7d1fe093 | 2029 | *suballoc_bit_start = res.sr_bit_offset; |
ba206635 | 2030 | *blkno_start = res.sr_blkno; |
7d1fe093 JB |
2031 | ac->ac_bits_given += res.sr_bits; |
2032 | *num_bits = res.sr_bits; | |
ccd979bd MF |
2033 | status = 0; |
2034 | bail: | |
c1e8d35e TM |
2035 | if (status) |
2036 | mlog_errno(status); | |
ccd979bd MF |
2037 | return status; |
2038 | } | |
2039 | ||
13821151 | 2040 | static void ocfs2_init_inode_ac_group(struct inode *dir, |
abf1b3cb | 2041 | struct buffer_head *parent_di_bh, |
13821151 TM |
2042 | struct ocfs2_alloc_context *ac) |
2043 | { | |
abf1b3cb | 2044 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data; |
13821151 TM |
2045 | /* |
2046 | * Try to allocate inodes from some specific group. | |
2047 | * | |
2048 | * If the parent dir has recorded the last group used in allocation, | |
2049 | * cool, use it. Otherwise if we try to allocate new inode from the | |
2050 | * same slot the parent dir belongs to, use the same chunk. | |
2051 | * | |
2052 | * We are very careful here to avoid the mistake of setting | |
2053 | * ac_last_group to a group descriptor from a different (unlocked) slot. | |
2054 | */ | |
2055 | if (OCFS2_I(dir)->ip_last_used_group && | |
2056 | OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot) | |
2057 | ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group; | |
abf1b3cb TM |
2058 | else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) { |
2059 | if (di->i_suballoc_loc) | |
2060 | ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc); | |
2061 | else | |
2062 | ac->ac_last_group = ocfs2_which_suballoc_group( | |
2063 | le64_to_cpu(di->i_blkno), | |
2064 | le16_to_cpu(di->i_suballoc_bit)); | |
2065 | } | |
13821151 TM |
2066 | } |
2067 | ||
2068 | static inline void ocfs2_save_inode_ac_group(struct inode *dir, | |
2069 | struct ocfs2_alloc_context *ac) | |
2070 | { | |
2071 | OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group; | |
2072 | OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot; | |
2073 | } | |
2074 | ||
e49e2767 MF |
2075 | int ocfs2_find_new_inode_loc(struct inode *dir, |
2076 | struct buffer_head *parent_fe_bh, | |
2077 | struct ocfs2_alloc_context *ac, | |
2078 | u64 *fe_blkno) | |
2079 | { | |
2080 | int ret; | |
2081 | handle_t *handle = NULL; | |
2082 | struct ocfs2_suballoc_result *res; | |
2083 | ||
2084 | BUG_ON(!ac); | |
2085 | BUG_ON(ac->ac_bits_given != 0); | |
2086 | BUG_ON(ac->ac_bits_wanted != 1); | |
2087 | BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE); | |
2088 | ||
2089 | res = kzalloc(sizeof(*res), GFP_NOFS); | |
2090 | if (res == NULL) { | |
2091 | ret = -ENOMEM; | |
2092 | mlog_errno(ret); | |
2093 | goto out; | |
2094 | } | |
2095 | ||
2096 | ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac); | |
2097 | ||
2098 | /* | |
2099 | * The handle started here is for chain relink. Alternatively, | |
2100 | * we could just disable relink for these calls. | |
2101 | */ | |
2102 | handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC); | |
2103 | if (IS_ERR(handle)) { | |
2104 | ret = PTR_ERR(handle); | |
2105 | handle = NULL; | |
2106 | mlog_errno(ret); | |
2107 | goto out; | |
2108 | } | |
2109 | ||
2110 | /* | |
2111 | * This will instruct ocfs2_claim_suballoc_bits and | |
2112 | * ocfs2_search_one_group to search but save actual allocation | |
2113 | * for later. | |
2114 | */ | |
2115 | ac->ac_find_loc_only = 1; | |
2116 | ||
2117 | ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res); | |
2118 | if (ret < 0) { | |
2119 | mlog_errno(ret); | |
2120 | goto out; | |
2121 | } | |
2122 | ||
2123 | ac->ac_find_loc_priv = res; | |
2124 | *fe_blkno = res->sr_blkno; | |
6fdb702d | 2125 | ocfs2_update_inode_fsync_trans(handle, dir, 0); |
e49e2767 MF |
2126 | out: |
2127 | if (handle) | |
2128 | ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle); | |
2129 | ||
2130 | if (ret) | |
2131 | kfree(res); | |
2132 | ||
2133 | return ret; | |
2134 | } | |
2135 | ||
2136 | int ocfs2_claim_new_inode_at_loc(handle_t *handle, | |
2137 | struct inode *dir, | |
2138 | struct ocfs2_alloc_context *ac, | |
2139 | u64 *suballoc_loc, | |
2140 | u16 *suballoc_bit, | |
2141 | u64 di_blkno) | |
2142 | { | |
2143 | int ret; | |
2144 | u16 chain; | |
2145 | struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv; | |
2146 | struct buffer_head *bg_bh = NULL; | |
2147 | struct ocfs2_group_desc *bg; | |
2148 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data; | |
2149 | ||
2150 | /* | |
2151 | * Since di_blkno is being passed back in, we check for any | |
2152 | * inconsistencies which may have happened between | |
2153 | * calls. These are code bugs as di_blkno is not expected to | |
2154 | * change once returned from ocfs2_find_new_inode_loc() | |
2155 | */ | |
2156 | BUG_ON(res->sr_blkno != di_blkno); | |
2157 | ||
2158 | ret = ocfs2_read_group_descriptor(ac->ac_inode, di, | |
2159 | res->sr_bg_stable_blkno, &bg_bh); | |
2160 | if (ret) { | |
2161 | mlog_errno(ret); | |
2162 | goto out; | |
2163 | } | |
2164 | ||
2165 | bg = (struct ocfs2_group_desc *) bg_bh->b_data; | |
2166 | chain = le16_to_cpu(bg->bg_chain); | |
2167 | ||
2168 | ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle, | |
2169 | ac->ac_bh, res->sr_bits, | |
2170 | chain); | |
2171 | if (ret) { | |
2172 | mlog_errno(ret); | |
2173 | goto out; | |
2174 | } | |
2175 | ||
2176 | ret = ocfs2_block_group_set_bits(handle, | |
2177 | ac->ac_inode, | |
2178 | bg, | |
2179 | bg_bh, | |
2180 | res->sr_bit_offset, | |
2181 | res->sr_bits); | |
2182 | if (ret < 0) { | |
db66c715 YL |
2183 | ocfs2_rollback_alloc_dinode_counts(ac->ac_inode, |
2184 | ac->ac_bh, res->sr_bits, chain); | |
e49e2767 MF |
2185 | mlog_errno(ret); |
2186 | goto out; | |
2187 | } | |
2188 | ||
2f73e135 TM |
2189 | trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno, |
2190 | res->sr_bits); | |
e49e2767 MF |
2191 | |
2192 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs); | |
2193 | ||
2194 | BUG_ON(res->sr_bits != 1); | |
2195 | ||
2196 | *suballoc_loc = res->sr_bg_blkno; | |
2197 | *suballoc_bit = res->sr_bit_offset; | |
2198 | ac->ac_bits_given++; | |
2199 | ocfs2_save_inode_ac_group(dir, ac); | |
2200 | ||
2201 | out: | |
2202 | brelse(bg_bh); | |
2203 | ||
2204 | return ret; | |
2205 | } | |
2206 | ||
1ed9b777 | 2207 | int ocfs2_claim_new_inode(handle_t *handle, |
13821151 TM |
2208 | struct inode *dir, |
2209 | struct buffer_head *parent_fe_bh, | |
ccd979bd | 2210 | struct ocfs2_alloc_context *ac, |
2b6cb576 | 2211 | u64 *suballoc_loc, |
ccd979bd MF |
2212 | u16 *suballoc_bit, |
2213 | u64 *fe_blkno) | |
2214 | { | |
2215 | int status; | |
2b6cb576 | 2216 | struct ocfs2_suballoc_result res; |
ccd979bd | 2217 | |
ccd979bd MF |
2218 | BUG_ON(!ac); |
2219 | BUG_ON(ac->ac_bits_given != 0); | |
2220 | BUG_ON(ac->ac_bits_wanted != 1); | |
2221 | BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE); | |
ccd979bd | 2222 | |
13821151 TM |
2223 | ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac); |
2224 | ||
aa8f8e93 | 2225 | status = ocfs2_claim_suballoc_bits(ac, |
da5cbf2f | 2226 | handle, |
ccd979bd MF |
2227 | 1, |
2228 | 1, | |
7d1fe093 | 2229 | &res); |
ccd979bd MF |
2230 | if (status < 0) { |
2231 | mlog_errno(status); | |
2232 | goto bail; | |
2233 | } | |
1ed9b777 | 2234 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs); |
ccd979bd | 2235 | |
7d1fe093 | 2236 | BUG_ON(res.sr_bits != 1); |
ccd979bd | 2237 | |
2b6cb576 | 2238 | *suballoc_loc = res.sr_bg_blkno; |
7d1fe093 | 2239 | *suballoc_bit = res.sr_bit_offset; |
ba206635 | 2240 | *fe_blkno = res.sr_blkno; |
ccd979bd | 2241 | ac->ac_bits_given++; |
13821151 | 2242 | ocfs2_save_inode_ac_group(dir, ac); |
ccd979bd MF |
2243 | status = 0; |
2244 | bail: | |
c1e8d35e TM |
2245 | if (status) |
2246 | mlog_errno(status); | |
ccd979bd MF |
2247 | return status; |
2248 | } | |
2249 | ||
2250 | /* translate a group desc. blkno and it's bitmap offset into | |
2251 | * disk cluster offset. */ | |
2252 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode, | |
2253 | u64 bg_blkno, | |
2254 | u16 bg_bit_off) | |
2255 | { | |
2256 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2257 | u32 cluster = 0; | |
2258 | ||
2259 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
2260 | ||
2261 | if (bg_blkno != osb->first_cluster_group_blkno) | |
2262 | cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno); | |
2263 | cluster += (u32) bg_bit_off; | |
2264 | return cluster; | |
2265 | } | |
2266 | ||
2267 | /* given a cluster offset, calculate which block group it belongs to | |
2268 | * and return that block offset. */ | |
d659072f | 2269 | u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster) |
ccd979bd MF |
2270 | { |
2271 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2272 | u32 group_no; | |
2273 | ||
2274 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
2275 | ||
2276 | group_no = cluster / osb->bitmap_cpg; | |
2277 | if (!group_no) | |
2278 | return osb->first_cluster_group_blkno; | |
2279 | return ocfs2_clusters_to_blocks(inode->i_sb, | |
2280 | group_no * osb->bitmap_cpg); | |
2281 | } | |
2282 | ||
2283 | /* given the block number of a cluster start, calculate which cluster | |
2284 | * group and descriptor bitmap offset that corresponds to. */ | |
2285 | static inline void ocfs2_block_to_cluster_group(struct inode *inode, | |
2286 | u64 data_blkno, | |
2287 | u64 *bg_blkno, | |
2288 | u16 *bg_bit_off) | |
2289 | { | |
2290 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2291 | u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno); | |
2292 | ||
2293 | BUG_ON(!ocfs2_is_cluster_bitmap(inode)); | |
2294 | ||
2295 | *bg_blkno = ocfs2_which_cluster_group(inode, | |
2296 | data_cluster); | |
2297 | ||
2298 | if (*bg_blkno == osb->first_cluster_group_blkno) | |
2299 | *bg_bit_off = (u16) data_cluster; | |
2300 | else | |
2301 | *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb, | |
2302 | data_blkno - *bg_blkno); | |
2303 | } | |
2304 | ||
2305 | /* | |
2306 | * min_bits - minimum contiguous chunk from this total allocation we | |
2307 | * can handle. set to what we asked for originally for a full | |
2308 | * contig. allocation, set to '1' to indicate we can deal with extents | |
2309 | * of any size. | |
2310 | */ | |
1ed9b777 | 2311 | int __ocfs2_claim_clusters(handle_t *handle, |
415cb800 MF |
2312 | struct ocfs2_alloc_context *ac, |
2313 | u32 min_clusters, | |
2314 | u32 max_clusters, | |
2315 | u32 *cluster_start, | |
2316 | u32 *num_clusters) | |
ccd979bd MF |
2317 | { |
2318 | int status; | |
415cb800 | 2319 | unsigned int bits_wanted = max_clusters; |
ba206635 | 2320 | struct ocfs2_suballoc_result res = { .sr_blkno = 0, }; |
1ed9b777 | 2321 | struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb); |
ccd979bd | 2322 | |
ccd979bd MF |
2323 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted); |
2324 | ||
2325 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL | |
2326 | && ac->ac_which != OCFS2_AC_USE_MAIN); | |
ccd979bd MF |
2327 | |
2328 | if (ac->ac_which == OCFS2_AC_USE_LOCAL) { | |
33d5d380 MF |
2329 | WARN_ON(min_clusters > 1); |
2330 | ||
ccd979bd MF |
2331 | status = ocfs2_claim_local_alloc_bits(osb, |
2332 | handle, | |
2333 | ac, | |
2334 | bits_wanted, | |
2335 | cluster_start, | |
2336 | num_clusters); | |
2337 | if (!status) | |
2338 | atomic_inc(&osb->alloc_stats.local_data); | |
2339 | } else { | |
2340 | if (min_clusters > (osb->bitmap_cpg - 1)) { | |
2341 | /* The only paths asking for contiguousness | |
2342 | * should know about this already. */ | |
2fbe8d1e SM |
2343 | mlog(ML_ERROR, "minimum allocation requested %u exceeds " |
2344 | "group bitmap size %u!\n", min_clusters, | |
2345 | osb->bitmap_cpg); | |
ccd979bd MF |
2346 | status = -ENOSPC; |
2347 | goto bail; | |
2348 | } | |
2349 | /* clamp the current request down to a realistic size. */ | |
2350 | if (bits_wanted > (osb->bitmap_cpg - 1)) | |
2351 | bits_wanted = osb->bitmap_cpg - 1; | |
2352 | ||
aa8f8e93 | 2353 | status = ocfs2_claim_suballoc_bits(ac, |
da5cbf2f | 2354 | handle, |
ccd979bd MF |
2355 | bits_wanted, |
2356 | min_clusters, | |
7d1fe093 | 2357 | &res); |
ccd979bd | 2358 | if (!status) { |
ba206635 | 2359 | BUG_ON(res.sr_blkno); /* cluster alloc can't set */ |
ccd979bd MF |
2360 | *cluster_start = |
2361 | ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode, | |
7d1fe093 JB |
2362 | res.sr_bg_blkno, |
2363 | res.sr_bit_offset); | |
ccd979bd | 2364 | atomic_inc(&osb->alloc_stats.bitmap_data); |
4711954e | 2365 | *num_clusters = res.sr_bits; |
ccd979bd MF |
2366 | } |
2367 | } | |
2368 | if (status < 0) { | |
2369 | if (status != -ENOSPC) | |
2370 | mlog_errno(status); | |
2371 | goto bail; | |
2372 | } | |
2373 | ||
4711954e | 2374 | ac->ac_bits_given += *num_clusters; |
ccd979bd MF |
2375 | |
2376 | bail: | |
c1e8d35e TM |
2377 | if (status) |
2378 | mlog_errno(status); | |
ccd979bd MF |
2379 | return status; |
2380 | } | |
2381 | ||
1ed9b777 | 2382 | int ocfs2_claim_clusters(handle_t *handle, |
415cb800 MF |
2383 | struct ocfs2_alloc_context *ac, |
2384 | u32 min_clusters, | |
2385 | u32 *cluster_start, | |
2386 | u32 *num_clusters) | |
2387 | { | |
2388 | unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given; | |
2389 | ||
1ed9b777 | 2390 | return __ocfs2_claim_clusters(handle, ac, min_clusters, |
415cb800 MF |
2391 | bits_wanted, cluster_start, num_clusters); |
2392 | } | |
2393 | ||
b4414eea MF |
2394 | static int ocfs2_block_group_clear_bits(handle_t *handle, |
2395 | struct inode *alloc_inode, | |
2396 | struct ocfs2_group_desc *bg, | |
2397 | struct buffer_head *group_bh, | |
2398 | unsigned int bit_off, | |
2399 | unsigned int num_bits, | |
2400 | void (*undo_fn)(unsigned int bit, | |
2401 | unsigned long *bmap)) | |
ccd979bd MF |
2402 | { |
2403 | int status; | |
2404 | unsigned int tmp; | |
ccd979bd MF |
2405 | struct ocfs2_group_desc *undo_bg = NULL; |
2406 | ||
42035306 JB |
2407 | /* The caller got this descriptor from |
2408 | * ocfs2_read_group_descriptor(). Any corruption is a code bug. */ | |
2409 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg)); | |
ccd979bd | 2410 | |
2f73e135 | 2411 | trace_ocfs2_block_group_clear_bits(bit_off, num_bits); |
ccd979bd | 2412 | |
b4414eea | 2413 | BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode)); |
0cf2f763 | 2414 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode), |
b4414eea MF |
2415 | group_bh, |
2416 | undo_fn ? | |
2417 | OCFS2_JOURNAL_ACCESS_UNDO : | |
2418 | OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
2419 | if (status < 0) { |
2420 | mlog_errno(status); | |
2421 | goto bail; | |
2422 | } | |
2423 | ||
b4414eea | 2424 | if (undo_fn) { |
94e41ecf SM |
2425 | jbd_lock_bh_state(group_bh); |
2426 | undo_bg = (struct ocfs2_group_desc *) | |
2427 | bh2jh(group_bh)->b_committed_data; | |
2428 | BUG_ON(!undo_bg); | |
2429 | } | |
ccd979bd MF |
2430 | |
2431 | tmp = num_bits; | |
2432 | while(tmp--) { | |
2433 | ocfs2_clear_bit((bit_off + tmp), | |
2434 | (unsigned long *) bg->bg_bitmap); | |
b4414eea MF |
2435 | if (undo_fn) |
2436 | undo_fn(bit_off + tmp, | |
2437 | (unsigned long *) undo_bg->bg_bitmap); | |
ccd979bd MF |
2438 | } |
2439 | le16_add_cpu(&bg->bg_free_bits_count, num_bits); | |
9b5cd10e | 2440 | if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) { |
7ecef14a JP |
2441 | return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n", |
2442 | (unsigned long long)le64_to_cpu(bg->bg_blkno), | |
2443 | le16_to_cpu(bg->bg_bits), | |
2444 | le16_to_cpu(bg->bg_free_bits_count), | |
2445 | num_bits); | |
9b5cd10e | 2446 | } |
ccd979bd | 2447 | |
b4414eea | 2448 | if (undo_fn) |
94e41ecf SM |
2449 | jbd_unlock_bh_state(group_bh); |
2450 | ||
ec20cec7 | 2451 | ocfs2_journal_dirty(handle, group_bh); |
ccd979bd MF |
2452 | bail: |
2453 | return status; | |
2454 | } | |
2455 | ||
2456 | /* | |
2457 | * expects the suballoc inode to already be locked. | |
2458 | */ | |
b4414eea MF |
2459 | static int _ocfs2_free_suballoc_bits(handle_t *handle, |
2460 | struct inode *alloc_inode, | |
2461 | struct buffer_head *alloc_bh, | |
2462 | unsigned int start_bit, | |
2463 | u64 bg_blkno, | |
2464 | unsigned int count, | |
2465 | void (*undo_fn)(unsigned int bit, | |
2466 | unsigned long *bitmap)) | |
ccd979bd MF |
2467 | { |
2468 | int status = 0; | |
2469 | u32 tmp_used; | |
ccd979bd MF |
2470 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data; |
2471 | struct ocfs2_chain_list *cl = &fe->id2.i_chain; | |
2472 | struct buffer_head *group_bh = NULL; | |
2473 | struct ocfs2_group_desc *group; | |
2474 | ||
10995aa2 JB |
2475 | /* The alloc_bh comes from ocfs2_free_dinode() or |
2476 | * ocfs2_free_clusters(). The callers have all locked the | |
2477 | * allocator and gotten alloc_bh from the lock call. This | |
25985edc | 2478 | * validates the dinode buffer. Any corruption that has happened |
10995aa2 JB |
2479 | * is a code bug. */ |
2480 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); | |
ccd979bd MF |
2481 | BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl)); |
2482 | ||
2f73e135 TM |
2483 | trace_ocfs2_free_suballoc_bits( |
2484 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, | |
2485 | (unsigned long long)bg_blkno, | |
2486 | start_bit, count); | |
ccd979bd | 2487 | |
68f64d47 JB |
2488 | status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno, |
2489 | &group_bh); | |
ccd979bd MF |
2490 | if (status < 0) { |
2491 | mlog_errno(status); | |
2492 | goto bail; | |
2493 | } | |
ccd979bd | 2494 | group = (struct ocfs2_group_desc *) group_bh->b_data; |
68f64d47 | 2495 | |
ccd979bd MF |
2496 | BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits)); |
2497 | ||
2498 | status = ocfs2_block_group_clear_bits(handle, alloc_inode, | |
2499 | group, group_bh, | |
b4414eea | 2500 | start_bit, count, undo_fn); |
ccd979bd MF |
2501 | if (status < 0) { |
2502 | mlog_errno(status); | |
2503 | goto bail; | |
2504 | } | |
2505 | ||
0cf2f763 JB |
2506 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode), |
2507 | alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE); | |
ccd979bd MF |
2508 | if (status < 0) { |
2509 | mlog_errno(status); | |
a47726bc JQ |
2510 | ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh, |
2511 | start_bit, count); | |
ccd979bd MF |
2512 | goto bail; |
2513 | } | |
2514 | ||
2515 | le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free, | |
2516 | count); | |
2517 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used); | |
2518 | fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count); | |
ec20cec7 | 2519 | ocfs2_journal_dirty(handle, alloc_bh); |
ccd979bd MF |
2520 | |
2521 | bail: | |
a81cb88b | 2522 | brelse(group_bh); |
ccd979bd | 2523 | |
c1e8d35e TM |
2524 | if (status) |
2525 | mlog_errno(status); | |
ccd979bd MF |
2526 | return status; |
2527 | } | |
2528 | ||
b4414eea MF |
2529 | int ocfs2_free_suballoc_bits(handle_t *handle, |
2530 | struct inode *alloc_inode, | |
2531 | struct buffer_head *alloc_bh, | |
2532 | unsigned int start_bit, | |
2533 | u64 bg_blkno, | |
2534 | unsigned int count) | |
2535 | { | |
2536 | return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh, | |
2537 | start_bit, bg_blkno, count, NULL); | |
2538 | } | |
2539 | ||
1fabe148 | 2540 | int ocfs2_free_dinode(handle_t *handle, |
ccd979bd MF |
2541 | struct inode *inode_alloc_inode, |
2542 | struct buffer_head *inode_alloc_bh, | |
2543 | struct ocfs2_dinode *di) | |
2544 | { | |
2545 | u64 blk = le64_to_cpu(di->i_blkno); | |
2546 | u16 bit = le16_to_cpu(di->i_suballoc_bit); | |
2547 | u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit); | |
2548 | ||
74380c47 TM |
2549 | if (di->i_suballoc_loc) |
2550 | bg_blkno = le64_to_cpu(di->i_suballoc_loc); | |
ccd979bd MF |
2551 | return ocfs2_free_suballoc_bits(handle, inode_alloc_inode, |
2552 | inode_alloc_bh, bit, bg_blkno, 1); | |
2553 | } | |
2554 | ||
b4414eea MF |
2555 | static int _ocfs2_free_clusters(handle_t *handle, |
2556 | struct inode *bitmap_inode, | |
2557 | struct buffer_head *bitmap_bh, | |
2558 | u64 start_blk, | |
2559 | unsigned int num_clusters, | |
2560 | void (*undo_fn)(unsigned int bit, | |
2561 | unsigned long *bitmap)) | |
ccd979bd MF |
2562 | { |
2563 | int status; | |
2564 | u16 bg_start_bit; | |
2565 | u64 bg_blkno; | |
2566 | struct ocfs2_dinode *fe; | |
2567 | ||
2568 | /* You can't ever have a contiguous set of clusters | |
2569 | * bigger than a block group bitmap so we never have to worry | |
ef6b689b TM |
2570 | * about looping on them. |
2571 | * This is expensive. We can safely remove once this stuff has | |
ccd979bd MF |
2572 | * gotten tested really well. */ |
2573 | BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk))); | |
2574 | ||
2575 | fe = (struct ocfs2_dinode *) bitmap_bh->b_data; | |
2576 | ||
2577 | ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno, | |
2578 | &bg_start_bit); | |
2579 | ||
2f73e135 TM |
2580 | trace_ocfs2_free_clusters((unsigned long long)bg_blkno, |
2581 | (unsigned long long)start_blk, | |
2582 | bg_start_bit, num_clusters); | |
ccd979bd | 2583 | |
b4414eea MF |
2584 | status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh, |
2585 | bg_start_bit, bg_blkno, | |
2586 | num_clusters, undo_fn); | |
9c7af40b | 2587 | if (status < 0) { |
ccd979bd | 2588 | mlog_errno(status); |
9c7af40b MF |
2589 | goto out; |
2590 | } | |
2591 | ||
2592 | ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb), | |
2593 | num_clusters); | |
ccd979bd | 2594 | |
9c7af40b | 2595 | out: |
c1e8d35e TM |
2596 | if (status) |
2597 | mlog_errno(status); | |
ccd979bd MF |
2598 | return status; |
2599 | } | |
2600 | ||
b4414eea MF |
2601 | int ocfs2_free_clusters(handle_t *handle, |
2602 | struct inode *bitmap_inode, | |
2603 | struct buffer_head *bitmap_bh, | |
2604 | u64 start_blk, | |
2605 | unsigned int num_clusters) | |
2606 | { | |
2607 | return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh, | |
2608 | start_blk, num_clusters, | |
2609 | _ocfs2_set_bit); | |
2610 | } | |
2611 | ||
2612 | /* | |
2613 | * Give never-used clusters back to the global bitmap. We don't need | |
2614 | * to protect these bits in the undo buffer. | |
2615 | */ | |
2616 | int ocfs2_release_clusters(handle_t *handle, | |
2617 | struct inode *bitmap_inode, | |
2618 | struct buffer_head *bitmap_bh, | |
2619 | u64 start_blk, | |
2620 | unsigned int num_clusters) | |
2621 | { | |
2622 | return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh, | |
2623 | start_blk, num_clusters, | |
2624 | _ocfs2_clear_bit); | |
2625 | } | |
2626 | ||
ccd979bd MF |
2627 | static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg) |
2628 | { | |
2629 | printk("Block Group:\n"); | |
2630 | printk("bg_signature: %s\n", bg->bg_signature); | |
2631 | printk("bg_size: %u\n", bg->bg_size); | |
2632 | printk("bg_bits: %u\n", bg->bg_bits); | |
2633 | printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count); | |
2634 | printk("bg_chain: %u\n", bg->bg_chain); | |
2635 | printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation)); | |
b0697053 MF |
2636 | printk("bg_next_group: %llu\n", |
2637 | (unsigned long long)bg->bg_next_group); | |
2638 | printk("bg_parent_dinode: %llu\n", | |
2639 | (unsigned long long)bg->bg_parent_dinode); | |
2640 | printk("bg_blkno: %llu\n", | |
2641 | (unsigned long long)bg->bg_blkno); | |
ccd979bd MF |
2642 | } |
2643 | ||
2644 | static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe) | |
2645 | { | |
2646 | int i; | |
2647 | ||
b0697053 | 2648 | printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno); |
ccd979bd | 2649 | printk("i_signature: %s\n", fe->i_signature); |
b0697053 MF |
2650 | printk("i_size: %llu\n", |
2651 | (unsigned long long)fe->i_size); | |
ccd979bd MF |
2652 | printk("i_clusters: %u\n", fe->i_clusters); |
2653 | printk("i_generation: %u\n", | |
2654 | le32_to_cpu(fe->i_generation)); | |
2655 | printk("id1.bitmap1.i_used: %u\n", | |
2656 | le32_to_cpu(fe->id1.bitmap1.i_used)); | |
2657 | printk("id1.bitmap1.i_total: %u\n", | |
2658 | le32_to_cpu(fe->id1.bitmap1.i_total)); | |
2659 | printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg); | |
2660 | printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc); | |
2661 | printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count); | |
2662 | printk("id2.i_chain.cl_next_free_rec: %u\n", | |
2663 | fe->id2.i_chain.cl_next_free_rec); | |
2664 | for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) { | |
2665 | printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i, | |
2666 | fe->id2.i_chain.cl_recs[i].c_free); | |
2667 | printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i, | |
2668 | fe->id2.i_chain.cl_recs[i].c_total); | |
b0697053 MF |
2669 | printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i, |
2670 | (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno); | |
ccd979bd MF |
2671 | } |
2672 | } | |
e7d4cb6b TM |
2673 | |
2674 | /* | |
2675 | * For a given allocation, determine which allocators will need to be | |
2676 | * accessed, and lock them, reserving the appropriate number of bits. | |
2677 | * | |
2678 | * Sparse file systems call this from ocfs2_write_begin_nolock() | |
2679 | * and ocfs2_allocate_unwritten_extents(). | |
2680 | * | |
2681 | * File systems which don't support holes call this from | |
2682 | * ocfs2_extend_allocation(). | |
2683 | */ | |
f99b9b7c JB |
2684 | int ocfs2_lock_allocators(struct inode *inode, |
2685 | struct ocfs2_extent_tree *et, | |
e7d4cb6b TM |
2686 | u32 clusters_to_add, u32 extents_to_split, |
2687 | struct ocfs2_alloc_context **data_ac, | |
f99b9b7c | 2688 | struct ocfs2_alloc_context **meta_ac) |
e7d4cb6b TM |
2689 | { |
2690 | int ret = 0, num_free_extents; | |
2691 | unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split; | |
2692 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
2693 | ||
2694 | *meta_ac = NULL; | |
2695 | if (data_ac) | |
2696 | *data_ac = NULL; | |
2697 | ||
2698 | BUG_ON(clusters_to_add != 0 && data_ac == NULL); | |
2699 | ||
964f14a0 | 2700 | num_free_extents = ocfs2_num_free_extents(et); |
e7d4cb6b TM |
2701 | if (num_free_extents < 0) { |
2702 | ret = num_free_extents; | |
2703 | mlog_errno(ret); | |
2704 | goto out; | |
2705 | } | |
2706 | ||
2707 | /* | |
2708 | * Sparse allocation file systems need to be more conservative | |
2709 | * with reserving room for expansion - the actual allocation | |
2710 | * happens while we've got a journal handle open so re-taking | |
2711 | * a cluster lock (because we ran out of room for another | |
2712 | * extent) will violate ordering rules. | |
2713 | * | |
2714 | * Most of the time we'll only be seeing this 1 cluster at a time | |
2715 | * anyway. | |
2716 | * | |
2717 | * Always lock for any unwritten extents - we might want to | |
2718 | * add blocks during a split. | |
2719 | */ | |
2720 | if (!num_free_extents || | |
2721 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) { | |
f99b9b7c | 2722 | ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac); |
e7d4cb6b TM |
2723 | if (ret < 0) { |
2724 | if (ret != -ENOSPC) | |
2725 | mlog_errno(ret); | |
2726 | goto out; | |
2727 | } | |
2728 | } | |
2729 | ||
2730 | if (clusters_to_add == 0) | |
2731 | goto out; | |
2732 | ||
2733 | ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac); | |
2734 | if (ret < 0) { | |
2735 | if (ret != -ENOSPC) | |
2736 | mlog_errno(ret); | |
2737 | goto out; | |
2738 | } | |
2739 | ||
2740 | out: | |
2741 | if (ret) { | |
2742 | if (*meta_ac) { | |
2743 | ocfs2_free_alloc_context(*meta_ac); | |
2744 | *meta_ac = NULL; | |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * We cannot have an error and a non null *data_ac. | |
2749 | */ | |
2750 | } | |
2751 | ||
2752 | return ret; | |
2753 | } | |
6ca497a8 | 2754 | |
2755 | /* | |
2756 | * Read the inode specified by blkno to get suballoc_slot and | |
2757 | * suballoc_bit. | |
2758 | */ | |
2759 | static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno, | |
889f004a TM |
2760 | u16 *suballoc_slot, u64 *group_blkno, |
2761 | u16 *suballoc_bit) | |
6ca497a8 | 2762 | { |
2763 | int status; | |
2764 | struct buffer_head *inode_bh = NULL; | |
2765 | struct ocfs2_dinode *inode_fe; | |
2766 | ||
2f73e135 | 2767 | trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno); |
6ca497a8 | 2768 | |
2769 | /* dirty read disk */ | |
2770 | status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh); | |
2771 | if (status < 0) { | |
5b09b507 JB |
2772 | mlog(ML_ERROR, "read block %llu failed %d\n", |
2773 | (unsigned long long)blkno, status); | |
6ca497a8 | 2774 | goto bail; |
2775 | } | |
2776 | ||
2777 | inode_fe = (struct ocfs2_dinode *) inode_bh->b_data; | |
2778 | if (!OCFS2_IS_VALID_DINODE(inode_fe)) { | |
5b09b507 JB |
2779 | mlog(ML_ERROR, "invalid inode %llu requested\n", |
2780 | (unsigned long long)blkno); | |
6ca497a8 | 2781 | status = -EINVAL; |
2782 | goto bail; | |
2783 | } | |
2784 | ||
0fba8137 | 2785 | if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT && |
6ca497a8 | 2786 | (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) { |
2787 | mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n", | |
5b09b507 JB |
2788 | (unsigned long long)blkno, |
2789 | (u32)le16_to_cpu(inode_fe->i_suballoc_slot)); | |
6ca497a8 | 2790 | status = -EINVAL; |
2791 | goto bail; | |
2792 | } | |
2793 | ||
2794 | if (suballoc_slot) | |
2795 | *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot); | |
2796 | if (suballoc_bit) | |
2797 | *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit); | |
889f004a TM |
2798 | if (group_blkno) |
2799 | *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc); | |
6ca497a8 | 2800 | |
2801 | bail: | |
2802 | brelse(inode_bh); | |
2803 | ||
c1e8d35e TM |
2804 | if (status) |
2805 | mlog_errno(status); | |
6ca497a8 | 2806 | return status; |
2807 | } | |
2808 | ||
2809 | /* | |
2810 | * test whether bit is SET in allocator bitmap or not. on success, 0 | |
2811 | * is returned and *res is 1 for SET; 0 otherwise. when fails, errno | |
2812 | * is returned and *res is meaningless. Call this after you have | |
2813 | * cluster locked against suballoc, or you may get a result based on | |
2814 | * non-up2date contents | |
2815 | */ | |
2816 | static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb, | |
2817 | struct inode *suballoc, | |
889f004a TM |
2818 | struct buffer_head *alloc_bh, |
2819 | u64 group_blkno, u64 blkno, | |
6ca497a8 | 2820 | u16 bit, int *res) |
2821 | { | |
abf1b3cb | 2822 | struct ocfs2_dinode *alloc_di; |
6ca497a8 | 2823 | struct ocfs2_group_desc *group; |
2824 | struct buffer_head *group_bh = NULL; | |
2825 | u64 bg_blkno; | |
2826 | int status; | |
2827 | ||
2f73e135 TM |
2828 | trace_ocfs2_test_suballoc_bit((unsigned long long)blkno, |
2829 | (unsigned int)bit); | |
6ca497a8 | 2830 | |
abf1b3cb TM |
2831 | alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data; |
2832 | if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) { | |
6ca497a8 | 2833 | mlog(ML_ERROR, "suballoc bit %u out of range of %u\n", |
2834 | (unsigned int)bit, | |
abf1b3cb | 2835 | ocfs2_bits_per_group(&alloc_di->id2.i_chain)); |
6ca497a8 | 2836 | status = -EINVAL; |
2837 | goto bail; | |
2838 | } | |
2839 | ||
889f004a TM |
2840 | bg_blkno = group_blkno ? group_blkno : |
2841 | ocfs2_which_suballoc_group(blkno, bit); | |
abf1b3cb | 2842 | status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno, |
6ca497a8 | 2843 | &group_bh); |
2844 | if (status < 0) { | |
5b09b507 JB |
2845 | mlog(ML_ERROR, "read group %llu failed %d\n", |
2846 | (unsigned long long)bg_blkno, status); | |
6ca497a8 | 2847 | goto bail; |
2848 | } | |
2849 | ||
2850 | group = (struct ocfs2_group_desc *) group_bh->b_data; | |
2851 | *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap); | |
2852 | ||
2853 | bail: | |
2854 | brelse(group_bh); | |
2855 | ||
c1e8d35e TM |
2856 | if (status) |
2857 | mlog_errno(status); | |
6ca497a8 | 2858 | return status; |
2859 | } | |
2860 | ||
2861 | /* | |
2862 | * Test if the bit representing this inode (blkno) is set in the | |
2863 | * suballocator. | |
2864 | * | |
2865 | * On success, 0 is returned and *res is 1 for SET; 0 otherwise. | |
2866 | * | |
2867 | * In the event of failure, a negative value is returned and *res is | |
2868 | * meaningless. | |
2869 | * | |
2870 | * Callers must make sure to hold nfs_sync_lock to prevent | |
2871 | * ocfs2_delete_inode() on another node from accessing the same | |
2872 | * suballocator concurrently. | |
2873 | */ | |
2874 | int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res) | |
2875 | { | |
2876 | int status; | |
889f004a | 2877 | u64 group_blkno = 0; |
6ca497a8 | 2878 | u16 suballoc_bit = 0, suballoc_slot = 0; |
2879 | struct inode *inode_alloc_inode; | |
2880 | struct buffer_head *alloc_bh = NULL; | |
2881 | ||
2f73e135 | 2882 | trace_ocfs2_test_inode_bit((unsigned long long)blkno); |
6ca497a8 | 2883 | |
2884 | status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot, | |
889f004a | 2885 | &group_blkno, &suballoc_bit); |
6ca497a8 | 2886 | if (status < 0) { |
2887 | mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status); | |
2888 | goto bail; | |
2889 | } | |
2890 | ||
2891 | inode_alloc_inode = | |
2892 | ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE, | |
2893 | suballoc_slot); | |
2894 | if (!inode_alloc_inode) { | |
2895 | /* the error code could be inaccurate, but we are not able to | |
2896 | * get the correct one. */ | |
2897 | status = -EINVAL; | |
2898 | mlog(ML_ERROR, "unable to get alloc inode in slot %u\n", | |
2899 | (u32)suballoc_slot); | |
2900 | goto bail; | |
2901 | } | |
2902 | ||
5955102c | 2903 | inode_lock(inode_alloc_inode); |
6ca497a8 | 2904 | status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0); |
2905 | if (status < 0) { | |
5955102c | 2906 | inode_unlock(inode_alloc_inode); |
7dc3e839 | 2907 | iput(inode_alloc_inode); |
6ca497a8 | 2908 | mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n", |
2909 | (u32)suballoc_slot, status); | |
2910 | goto bail; | |
2911 | } | |
2912 | ||
2913 | status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh, | |
889f004a | 2914 | group_blkno, blkno, suballoc_bit, res); |
6ca497a8 | 2915 | if (status < 0) |
2916 | mlog(ML_ERROR, "test suballoc bit failed %d\n", status); | |
2917 | ||
2918 | ocfs2_inode_unlock(inode_alloc_inode, 0); | |
5955102c | 2919 | inode_unlock(inode_alloc_inode); |
6ca497a8 | 2920 | |
2921 | iput(inode_alloc_inode); | |
2922 | brelse(alloc_bh); | |
2923 | bail: | |
c1e8d35e TM |
2924 | if (status) |
2925 | mlog_errno(status); | |
6ca497a8 | 2926 | return status; |
2927 | } |