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