Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * io.c | |
5 | * | |
6 | * Buffer cache handling | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
26 | #include <linux/fs.h> | |
27 | #include <linux/types.h> | |
ccd979bd | 28 | #include <linux/highmem.h> |
2f8b5444 | 29 | #include <linux/bio.h> |
ccd979bd MF |
30 | |
31 | #include <cluster/masklog.h> | |
32 | ||
33 | #include "ocfs2.h" | |
34 | ||
35 | #include "alloc.h" | |
36 | #include "inode.h" | |
37 | #include "journal.h" | |
38 | #include "uptodate.h" | |
ccd979bd | 39 | #include "buffer_head_io.h" |
15057e98 | 40 | #include "ocfs2_trace.h" |
ccd979bd | 41 | |
970e4936 JB |
42 | /* |
43 | * Bits on bh->b_state used by ocfs2. | |
44 | * | |
b86c86fa | 45 | * These MUST be after the JBD2 bits. Hence, we use BH_JBDPrivateStart. |
970e4936 JB |
46 | */ |
47 | enum ocfs2_state_bits { | |
b86c86fa | 48 | BH_NeedsValidate = BH_JBDPrivateStart, |
970e4936 JB |
49 | }; |
50 | ||
51 | /* Expand the magic b_state functions */ | |
52 | BUFFER_FNS(NeedsValidate, needs_validate); | |
53 | ||
ccd979bd | 54 | int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh, |
8cb471e8 | 55 | struct ocfs2_caching_info *ci) |
ccd979bd MF |
56 | { |
57 | int ret = 0; | |
58 | ||
15057e98 | 59 | trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci); |
ccd979bd MF |
60 | |
61 | BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO); | |
62 | BUG_ON(buffer_jbd(bh)); | |
63 | ||
64 | /* No need to check for a soft readonly file system here. non | |
65 | * journalled writes are only ever done on system files which | |
66 | * can get modified during recovery even if read-only. */ | |
67 | if (ocfs2_is_hard_readonly(osb)) { | |
68 | ret = -EROFS; | |
c1e8d35e | 69 | mlog_errno(ret); |
ccd979bd MF |
70 | goto out; |
71 | } | |
72 | ||
8cb471e8 | 73 | ocfs2_metadata_cache_io_lock(ci); |
ccd979bd MF |
74 | |
75 | lock_buffer(bh); | |
76 | set_buffer_uptodate(bh); | |
77 | ||
78 | /* remove from dirty list before I/O. */ | |
79 | clear_buffer_dirty(bh); | |
80 | ||
da1e9098 | 81 | get_bh(bh); /* for end_buffer_write_sync() */ |
ccd979bd | 82 | bh->b_end_io = end_buffer_write_sync; |
2a222ca9 | 83 | submit_bh(REQ_OP_WRITE, 0, bh); |
ccd979bd MF |
84 | |
85 | wait_on_buffer(bh); | |
86 | ||
87 | if (buffer_uptodate(bh)) { | |
8cb471e8 | 88 | ocfs2_set_buffer_uptodate(ci, bh); |
ccd979bd MF |
89 | } else { |
90 | /* We don't need to remove the clustered uptodate | |
91 | * information for this bh as it's not marked locally | |
92 | * uptodate. */ | |
93 | ret = -EIO; | |
c1e8d35e | 94 | mlog_errno(ret); |
ccd979bd MF |
95 | } |
96 | ||
8cb471e8 | 97 | ocfs2_metadata_cache_io_unlock(ci); |
ccd979bd | 98 | out: |
ccd979bd MF |
99 | return ret; |
100 | } | |
101 | ||
da1e9098 JB |
102 | int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block, |
103 | unsigned int nr, struct buffer_head *bhs[]) | |
104 | { | |
105 | int status = 0; | |
106 | unsigned int i; | |
107 | struct buffer_head *bh; | |
108 | ||
15057e98 TM |
109 | trace_ocfs2_read_blocks_sync((unsigned long long)block, nr); |
110 | ||
111 | if (!nr) | |
da1e9098 | 112 | goto bail; |
da1e9098 JB |
113 | |
114 | for (i = 0 ; i < nr ; i++) { | |
115 | if (bhs[i] == NULL) { | |
116 | bhs[i] = sb_getblk(osb->sb, block++); | |
117 | if (bhs[i] == NULL) { | |
7391a294 | 118 | status = -ENOMEM; |
da1e9098 JB |
119 | mlog_errno(status); |
120 | goto bail; | |
121 | } | |
122 | } | |
123 | bh = bhs[i]; | |
124 | ||
125 | if (buffer_jbd(bh)) { | |
15057e98 TM |
126 | trace_ocfs2_read_blocks_sync_jbd( |
127 | (unsigned long long)bh->b_blocknr); | |
da1e9098 JB |
128 | continue; |
129 | } | |
130 | ||
131 | if (buffer_dirty(bh)) { | |
132 | /* This should probably be a BUG, or | |
133 | * at least return an error. */ | |
134 | mlog(ML_ERROR, | |
135 | "trying to sync read a dirty " | |
136 | "buffer! (blocknr = %llu), skipping\n", | |
137 | (unsigned long long)bh->b_blocknr); | |
138 | continue; | |
139 | } | |
140 | ||
141 | lock_buffer(bh); | |
142 | if (buffer_jbd(bh)) { | |
7186ee06 | 143 | #ifdef CATCH_BH_JBD_RACES |
da1e9098 JB |
144 | mlog(ML_ERROR, |
145 | "block %llu had the JBD bit set " | |
146 | "while I was in lock_buffer!", | |
147 | (unsigned long long)bh->b_blocknr); | |
148 | BUG(); | |
7186ee06 GH |
149 | #else |
150 | unlock_buffer(bh); | |
151 | continue; | |
152 | #endif | |
da1e9098 JB |
153 | } |
154 | ||
155 | clear_buffer_uptodate(bh); | |
156 | get_bh(bh); /* for end_buffer_read_sync() */ | |
157 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 158 | submit_bh(REQ_OP_READ, 0, bh); |
da1e9098 JB |
159 | } |
160 | ||
161 | for (i = nr; i > 0; i--) { | |
162 | bh = bhs[i - 1]; | |
163 | ||
d6b58f89 MF |
164 | /* No need to wait on the buffer if it's managed by JBD. */ |
165 | if (!buffer_jbd(bh)) | |
166 | wait_on_buffer(bh); | |
da1e9098 | 167 | |
da1e9098 JB |
168 | if (!buffer_uptodate(bh)) { |
169 | /* Status won't be cleared from here on out, | |
170 | * so we can safely record this and loop back | |
171 | * to cleanup the other buffers. */ | |
172 | status = -EIO; | |
173 | put_bh(bh); | |
174 | bhs[i - 1] = NULL; | |
175 | } | |
176 | } | |
177 | ||
178 | bail: | |
179 | return status; | |
180 | } | |
181 | ||
8cb471e8 | 182 | int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr, |
970e4936 JB |
183 | struct buffer_head *bhs[], int flags, |
184 | int (*validate)(struct super_block *sb, | |
185 | struct buffer_head *bh)) | |
ccd979bd MF |
186 | { |
187 | int status = 0; | |
ccd979bd MF |
188 | int i, ignore_cache = 0; |
189 | struct buffer_head *bh; | |
8cb471e8 | 190 | struct super_block *sb = ocfs2_metadata_cache_get_super(ci); |
ccd979bd | 191 | |
15057e98 | 192 | trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags); |
ccd979bd | 193 | |
8cb471e8 | 194 | BUG_ON(!ci); |
d4a8c93c JB |
195 | BUG_ON((flags & OCFS2_BH_READAHEAD) && |
196 | (flags & OCFS2_BH_IGNORE_CACHE)); | |
aa958874 | 197 | |
31d33073 | 198 | if (bhs == NULL) { |
ccd979bd MF |
199 | status = -EINVAL; |
200 | mlog_errno(status); | |
201 | goto bail; | |
202 | } | |
203 | ||
204 | if (nr < 0) { | |
205 | mlog(ML_ERROR, "asked to read %d blocks!\n", nr); | |
206 | status = -EINVAL; | |
207 | mlog_errno(status); | |
208 | goto bail; | |
209 | } | |
210 | ||
211 | if (nr == 0) { | |
ccd979bd MF |
212 | status = 0; |
213 | goto bail; | |
214 | } | |
215 | ||
8cb471e8 | 216 | ocfs2_metadata_cache_io_lock(ci); |
ccd979bd MF |
217 | for (i = 0 ; i < nr ; i++) { |
218 | if (bhs[i] == NULL) { | |
8cb471e8 | 219 | bhs[i] = sb_getblk(sb, block++); |
ccd979bd | 220 | if (bhs[i] == NULL) { |
8cb471e8 | 221 | ocfs2_metadata_cache_io_unlock(ci); |
7391a294 | 222 | status = -ENOMEM; |
ccd979bd MF |
223 | mlog_errno(status); |
224 | goto bail; | |
225 | } | |
226 | } | |
227 | bh = bhs[i]; | |
d4a8c93c | 228 | ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE); |
ccd979bd | 229 | |
aa958874 MF |
230 | /* There are three read-ahead cases here which we need to |
231 | * be concerned with. All three assume a buffer has | |
232 | * previously been submitted with OCFS2_BH_READAHEAD | |
233 | * and it hasn't yet completed I/O. | |
234 | * | |
235 | * 1) The current request is sync to disk. This rarely | |
236 | * happens these days, and never when performance | |
237 | * matters - the code can just wait on the buffer | |
238 | * lock and re-submit. | |
239 | * | |
240 | * 2) The current request is cached, but not | |
241 | * readahead. ocfs2_buffer_uptodate() will return | |
242 | * false anyway, so we'll wind up waiting on the | |
243 | * buffer lock to do I/O. We re-check the request | |
244 | * with after getting the lock to avoid a re-submit. | |
245 | * | |
246 | * 3) The current request is readahead (and so must | |
247 | * also be a caching one). We short circuit if the | |
248 | * buffer is locked (under I/O) and if it's in the | |
249 | * uptodate cache. The re-check from #2 catches the | |
250 | * case that the previous read-ahead completes just | |
251 | * before our is-it-in-flight check. | |
252 | */ | |
253 | ||
8cb471e8 | 254 | if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) { |
d701485a | 255 | trace_ocfs2_read_blocks_from_disk( |
ccd979bd | 256 | (unsigned long long)bh->b_blocknr, |
8cb471e8 | 257 | (unsigned long long)ocfs2_metadata_cache_owner(ci)); |
d4a8c93c JB |
258 | /* We're using ignore_cache here to say |
259 | * "go to disk" */ | |
ccd979bd MF |
260 | ignore_cache = 1; |
261 | } | |
262 | ||
15057e98 TM |
263 | trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr, |
264 | ignore_cache, buffer_jbd(bh), buffer_dirty(bh)); | |
265 | ||
ccd979bd | 266 | if (buffer_jbd(bh)) { |
ccd979bd MF |
267 | continue; |
268 | } | |
269 | ||
d4a8c93c | 270 | if (ignore_cache) { |
ccd979bd MF |
271 | if (buffer_dirty(bh)) { |
272 | /* This should probably be a BUG, or | |
273 | * at least return an error. */ | |
ccd979bd MF |
274 | continue; |
275 | } | |
276 | ||
aa958874 MF |
277 | /* A read-ahead request was made - if the |
278 | * buffer is already under read-ahead from a | |
279 | * previously submitted request than we are | |
280 | * done here. */ | |
281 | if ((flags & OCFS2_BH_READAHEAD) | |
8cb471e8 | 282 | && ocfs2_buffer_read_ahead(ci, bh)) |
aa958874 MF |
283 | continue; |
284 | ||
ccd979bd MF |
285 | lock_buffer(bh); |
286 | if (buffer_jbd(bh)) { | |
287 | #ifdef CATCH_BH_JBD_RACES | |
288 | mlog(ML_ERROR, "block %llu had the JBD bit set " | |
289 | "while I was in lock_buffer!", | |
290 | (unsigned long long)bh->b_blocknr); | |
291 | BUG(); | |
292 | #else | |
293 | unlock_buffer(bh); | |
294 | continue; | |
295 | #endif | |
296 | } | |
aa958874 MF |
297 | |
298 | /* Re-check ocfs2_buffer_uptodate() as a | |
299 | * previously read-ahead buffer may have | |
300 | * completed I/O while we were waiting for the | |
301 | * buffer lock. */ | |
d4a8c93c | 302 | if (!(flags & OCFS2_BH_IGNORE_CACHE) |
aa958874 | 303 | && !(flags & OCFS2_BH_READAHEAD) |
8cb471e8 | 304 | && ocfs2_buffer_uptodate(ci, bh)) { |
aa958874 MF |
305 | unlock_buffer(bh); |
306 | continue; | |
307 | } | |
308 | ||
ccd979bd MF |
309 | clear_buffer_uptodate(bh); |
310 | get_bh(bh); /* for end_buffer_read_sync() */ | |
970e4936 JB |
311 | if (validate) |
312 | set_buffer_needs_validate(bh); | |
ccd979bd | 313 | bh->b_end_io = end_buffer_read_sync; |
2a222ca9 | 314 | submit_bh(REQ_OP_READ, 0, bh); |
ccd979bd MF |
315 | continue; |
316 | } | |
317 | } | |
318 | ||
319 | status = 0; | |
320 | ||
321 | for (i = (nr - 1); i >= 0; i--) { | |
322 | bh = bhs[i]; | |
323 | ||
aa958874 | 324 | if (!(flags & OCFS2_BH_READAHEAD)) { |
34237681 GR |
325 | if (status) { |
326 | /* Clear the rest of the buffers on error */ | |
327 | put_bh(bh); | |
328 | bhs[i] = NULL; | |
329 | continue; | |
330 | } | |
aa958874 | 331 | /* We know this can't have changed as we hold the |
8cb471e8 | 332 | * owner sem. Avoid doing any work on the bh if the |
aa958874 MF |
333 | * journal has it. */ |
334 | if (!buffer_jbd(bh)) | |
335 | wait_on_buffer(bh); | |
336 | ||
337 | if (!buffer_uptodate(bh)) { | |
338 | /* Status won't be cleared from here on out, | |
339 | * so we can safely record this and loop back | |
340 | * to cleanup the other buffers. Don't need to | |
341 | * remove the clustered uptodate information | |
342 | * for this bh as it's not marked locally | |
343 | * uptodate. */ | |
344 | status = -EIO; | |
234b69e3 | 345 | clear_buffer_needs_validate(bh); |
2fe5c1d7 | 346 | put_bh(bh); |
aa958874 MF |
347 | bhs[i] = NULL; |
348 | continue; | |
349 | } | |
970e4936 JB |
350 | |
351 | if (buffer_needs_validate(bh)) { | |
352 | /* We never set NeedsValidate if the | |
353 | * buffer was held by the journal, so | |
354 | * that better not have changed */ | |
355 | BUG_ON(buffer_jbd(bh)); | |
356 | clear_buffer_needs_validate(bh); | |
8cb471e8 | 357 | status = validate(sb, bh); |
970e4936 JB |
358 | if (status) { |
359 | put_bh(bh); | |
360 | bhs[i] = NULL; | |
361 | continue; | |
362 | } | |
363 | } | |
ccd979bd MF |
364 | } |
365 | ||
aa958874 MF |
366 | /* Always set the buffer in the cache, even if it was |
367 | * a forced read, or read-ahead which hasn't yet | |
368 | * completed. */ | |
8cb471e8 | 369 | ocfs2_set_buffer_uptodate(ci, bh); |
ccd979bd | 370 | } |
8cb471e8 | 371 | ocfs2_metadata_cache_io_unlock(ci); |
ccd979bd | 372 | |
15057e98 TM |
373 | trace_ocfs2_read_blocks_end((unsigned long long)block, nr, |
374 | flags, ignore_cache); | |
ccd979bd MF |
375 | |
376 | bail: | |
377 | ||
ccd979bd MF |
378 | return status; |
379 | } | |
d659072f TM |
380 | |
381 | /* Check whether the blkno is the super block or one of the backups. */ | |
382 | static void ocfs2_check_super_or_backup(struct super_block *sb, | |
383 | sector_t blkno) | |
384 | { | |
385 | int i; | |
386 | u64 backup_blkno; | |
387 | ||
388 | if (blkno == OCFS2_SUPER_BLOCK_BLKNO) | |
389 | return; | |
390 | ||
391 | for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { | |
392 | backup_blkno = ocfs2_backup_super_blkno(sb, i); | |
393 | if (backup_blkno == blkno) | |
394 | return; | |
395 | } | |
396 | ||
397 | BUG(); | |
398 | } | |
399 | ||
400 | /* | |
401 | * Write super block and backups doesn't need to collaborate with journal, | |
8cb471e8 | 402 | * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed |
d659072f TM |
403 | * into this function. |
404 | */ | |
405 | int ocfs2_write_super_or_backup(struct ocfs2_super *osb, | |
406 | struct buffer_head *bh) | |
407 | { | |
408 | int ret = 0; | |
a42ab8e1 | 409 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
d659072f | 410 | |
d659072f TM |
411 | BUG_ON(buffer_jbd(bh)); |
412 | ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr); | |
413 | ||
414 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) { | |
415 | ret = -EROFS; | |
c1e8d35e | 416 | mlog_errno(ret); |
d659072f TM |
417 | goto out; |
418 | } | |
419 | ||
420 | lock_buffer(bh); | |
421 | set_buffer_uptodate(bh); | |
422 | ||
423 | /* remove from dirty list before I/O. */ | |
424 | clear_buffer_dirty(bh); | |
425 | ||
426 | get_bh(bh); /* for end_buffer_write_sync() */ | |
427 | bh->b_end_io = end_buffer_write_sync; | |
a42ab8e1 | 428 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check); |
2a222ca9 | 429 | submit_bh(REQ_OP_WRITE, 0, bh); |
d659072f TM |
430 | |
431 | wait_on_buffer(bh); | |
432 | ||
433 | if (!buffer_uptodate(bh)) { | |
434 | ret = -EIO; | |
c1e8d35e | 435 | mlog_errno(ret); |
d659072f TM |
436 | } |
437 | ||
438 | out: | |
d659072f TM |
439 | return ret; |
440 | } |