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)) { | |
142 | mlog(ML_ERROR, | |
143 | "block %llu had the JBD bit set " | |
144 | "while I was in lock_buffer!", | |
145 | (unsigned long long)bh->b_blocknr); | |
146 | BUG(); | |
147 | } | |
148 | ||
149 | clear_buffer_uptodate(bh); | |
150 | get_bh(bh); /* for end_buffer_read_sync() */ | |
151 | bh->b_end_io = end_buffer_read_sync; | |
152 | submit_bh(READ, bh); | |
153 | } | |
154 | ||
155 | for (i = nr; i > 0; i--) { | |
156 | bh = bhs[i - 1]; | |
157 | ||
d6b58f89 MF |
158 | /* No need to wait on the buffer if it's managed by JBD. */ |
159 | if (!buffer_jbd(bh)) | |
160 | wait_on_buffer(bh); | |
da1e9098 | 161 | |
da1e9098 JB |
162 | if (!buffer_uptodate(bh)) { |
163 | /* Status won't be cleared from here on out, | |
164 | * so we can safely record this and loop back | |
165 | * to cleanup the other buffers. */ | |
166 | status = -EIO; | |
167 | put_bh(bh); | |
168 | bhs[i - 1] = NULL; | |
169 | } | |
170 | } | |
171 | ||
172 | bail: | |
173 | return status; | |
174 | } | |
175 | ||
8cb471e8 | 176 | int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr, |
970e4936 JB |
177 | struct buffer_head *bhs[], int flags, |
178 | int (*validate)(struct super_block *sb, | |
179 | struct buffer_head *bh)) | |
ccd979bd MF |
180 | { |
181 | int status = 0; | |
ccd979bd MF |
182 | int i, ignore_cache = 0; |
183 | struct buffer_head *bh; | |
8cb471e8 | 184 | struct super_block *sb = ocfs2_metadata_cache_get_super(ci); |
ccd979bd | 185 | |
15057e98 | 186 | trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags); |
ccd979bd | 187 | |
8cb471e8 | 188 | BUG_ON(!ci); |
d4a8c93c JB |
189 | BUG_ON((flags & OCFS2_BH_READAHEAD) && |
190 | (flags & OCFS2_BH_IGNORE_CACHE)); | |
aa958874 | 191 | |
31d33073 | 192 | if (bhs == NULL) { |
ccd979bd MF |
193 | status = -EINVAL; |
194 | mlog_errno(status); | |
195 | goto bail; | |
196 | } | |
197 | ||
198 | if (nr < 0) { | |
199 | mlog(ML_ERROR, "asked to read %d blocks!\n", nr); | |
200 | status = -EINVAL; | |
201 | mlog_errno(status); | |
202 | goto bail; | |
203 | } | |
204 | ||
205 | if (nr == 0) { | |
ccd979bd MF |
206 | status = 0; |
207 | goto bail; | |
208 | } | |
209 | ||
8cb471e8 | 210 | ocfs2_metadata_cache_io_lock(ci); |
ccd979bd MF |
211 | for (i = 0 ; i < nr ; i++) { |
212 | if (bhs[i] == NULL) { | |
8cb471e8 | 213 | bhs[i] = sb_getblk(sb, block++); |
ccd979bd | 214 | if (bhs[i] == NULL) { |
8cb471e8 | 215 | ocfs2_metadata_cache_io_unlock(ci); |
7391a294 | 216 | status = -ENOMEM; |
ccd979bd MF |
217 | mlog_errno(status); |
218 | goto bail; | |
219 | } | |
220 | } | |
221 | bh = bhs[i]; | |
d4a8c93c | 222 | ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE); |
ccd979bd | 223 | |
aa958874 MF |
224 | /* There are three read-ahead cases here which we need to |
225 | * be concerned with. All three assume a buffer has | |
226 | * previously been submitted with OCFS2_BH_READAHEAD | |
227 | * and it hasn't yet completed I/O. | |
228 | * | |
229 | * 1) The current request is sync to disk. This rarely | |
230 | * happens these days, and never when performance | |
231 | * matters - the code can just wait on the buffer | |
232 | * lock and re-submit. | |
233 | * | |
234 | * 2) The current request is cached, but not | |
235 | * readahead. ocfs2_buffer_uptodate() will return | |
236 | * false anyway, so we'll wind up waiting on the | |
237 | * buffer lock to do I/O. We re-check the request | |
238 | * with after getting the lock to avoid a re-submit. | |
239 | * | |
240 | * 3) The current request is readahead (and so must | |
241 | * also be a caching one). We short circuit if the | |
242 | * buffer is locked (under I/O) and if it's in the | |
243 | * uptodate cache. The re-check from #2 catches the | |
244 | * case that the previous read-ahead completes just | |
245 | * before our is-it-in-flight check. | |
246 | */ | |
247 | ||
8cb471e8 | 248 | if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) { |
d701485a | 249 | trace_ocfs2_read_blocks_from_disk( |
ccd979bd | 250 | (unsigned long long)bh->b_blocknr, |
8cb471e8 | 251 | (unsigned long long)ocfs2_metadata_cache_owner(ci)); |
d4a8c93c JB |
252 | /* We're using ignore_cache here to say |
253 | * "go to disk" */ | |
ccd979bd MF |
254 | ignore_cache = 1; |
255 | } | |
256 | ||
15057e98 TM |
257 | trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr, |
258 | ignore_cache, buffer_jbd(bh), buffer_dirty(bh)); | |
259 | ||
ccd979bd | 260 | if (buffer_jbd(bh)) { |
ccd979bd MF |
261 | continue; |
262 | } | |
263 | ||
d4a8c93c | 264 | if (ignore_cache) { |
ccd979bd MF |
265 | if (buffer_dirty(bh)) { |
266 | /* This should probably be a BUG, or | |
267 | * at least return an error. */ | |
ccd979bd MF |
268 | continue; |
269 | } | |
270 | ||
aa958874 MF |
271 | /* A read-ahead request was made - if the |
272 | * buffer is already under read-ahead from a | |
273 | * previously submitted request than we are | |
274 | * done here. */ | |
275 | if ((flags & OCFS2_BH_READAHEAD) | |
8cb471e8 | 276 | && ocfs2_buffer_read_ahead(ci, bh)) |
aa958874 MF |
277 | continue; |
278 | ||
ccd979bd MF |
279 | lock_buffer(bh); |
280 | if (buffer_jbd(bh)) { | |
281 | #ifdef CATCH_BH_JBD_RACES | |
282 | mlog(ML_ERROR, "block %llu had the JBD bit set " | |
283 | "while I was in lock_buffer!", | |
284 | (unsigned long long)bh->b_blocknr); | |
285 | BUG(); | |
286 | #else | |
287 | unlock_buffer(bh); | |
288 | continue; | |
289 | #endif | |
290 | } | |
aa958874 MF |
291 | |
292 | /* Re-check ocfs2_buffer_uptodate() as a | |
293 | * previously read-ahead buffer may have | |
294 | * completed I/O while we were waiting for the | |
295 | * buffer lock. */ | |
d4a8c93c | 296 | if (!(flags & OCFS2_BH_IGNORE_CACHE) |
aa958874 | 297 | && !(flags & OCFS2_BH_READAHEAD) |
8cb471e8 | 298 | && ocfs2_buffer_uptodate(ci, bh)) { |
aa958874 MF |
299 | unlock_buffer(bh); |
300 | continue; | |
301 | } | |
302 | ||
ccd979bd MF |
303 | clear_buffer_uptodate(bh); |
304 | get_bh(bh); /* for end_buffer_read_sync() */ | |
970e4936 JB |
305 | if (validate) |
306 | set_buffer_needs_validate(bh); | |
ccd979bd | 307 | bh->b_end_io = end_buffer_read_sync; |
aa958874 | 308 | submit_bh(READ, bh); |
ccd979bd MF |
309 | continue; |
310 | } | |
311 | } | |
312 | ||
313 | status = 0; | |
314 | ||
315 | for (i = (nr - 1); i >= 0; i--) { | |
316 | bh = bhs[i]; | |
317 | ||
aa958874 | 318 | if (!(flags & OCFS2_BH_READAHEAD)) { |
34237681 GR |
319 | if (status) { |
320 | /* Clear the rest of the buffers on error */ | |
321 | put_bh(bh); | |
322 | bhs[i] = NULL; | |
323 | continue; | |
324 | } | |
aa958874 | 325 | /* We know this can't have changed as we hold the |
8cb471e8 | 326 | * owner sem. Avoid doing any work on the bh if the |
aa958874 MF |
327 | * journal has it. */ |
328 | if (!buffer_jbd(bh)) | |
329 | wait_on_buffer(bh); | |
330 | ||
331 | if (!buffer_uptodate(bh)) { | |
332 | /* Status won't be cleared from here on out, | |
333 | * so we can safely record this and loop back | |
334 | * to cleanup the other buffers. Don't need to | |
335 | * remove the clustered uptodate information | |
336 | * for this bh as it's not marked locally | |
337 | * uptodate. */ | |
338 | status = -EIO; | |
2fe5c1d7 | 339 | put_bh(bh); |
aa958874 MF |
340 | bhs[i] = NULL; |
341 | continue; | |
342 | } | |
970e4936 JB |
343 | |
344 | if (buffer_needs_validate(bh)) { | |
345 | /* We never set NeedsValidate if the | |
346 | * buffer was held by the journal, so | |
347 | * that better not have changed */ | |
348 | BUG_ON(buffer_jbd(bh)); | |
349 | clear_buffer_needs_validate(bh); | |
8cb471e8 | 350 | status = validate(sb, bh); |
970e4936 JB |
351 | if (status) { |
352 | put_bh(bh); | |
353 | bhs[i] = NULL; | |
354 | continue; | |
355 | } | |
356 | } | |
ccd979bd MF |
357 | } |
358 | ||
aa958874 MF |
359 | /* Always set the buffer in the cache, even if it was |
360 | * a forced read, or read-ahead which hasn't yet | |
361 | * completed. */ | |
8cb471e8 | 362 | ocfs2_set_buffer_uptodate(ci, bh); |
ccd979bd | 363 | } |
8cb471e8 | 364 | ocfs2_metadata_cache_io_unlock(ci); |
ccd979bd | 365 | |
15057e98 TM |
366 | trace_ocfs2_read_blocks_end((unsigned long long)block, nr, |
367 | flags, ignore_cache); | |
ccd979bd MF |
368 | |
369 | bail: | |
370 | ||
ccd979bd MF |
371 | return status; |
372 | } | |
d659072f TM |
373 | |
374 | /* Check whether the blkno is the super block or one of the backups. */ | |
375 | static void ocfs2_check_super_or_backup(struct super_block *sb, | |
376 | sector_t blkno) | |
377 | { | |
378 | int i; | |
379 | u64 backup_blkno; | |
380 | ||
381 | if (blkno == OCFS2_SUPER_BLOCK_BLKNO) | |
382 | return; | |
383 | ||
384 | for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { | |
385 | backup_blkno = ocfs2_backup_super_blkno(sb, i); | |
386 | if (backup_blkno == blkno) | |
387 | return; | |
388 | } | |
389 | ||
390 | BUG(); | |
391 | } | |
392 | ||
393 | /* | |
394 | * Write super block and backups doesn't need to collaborate with journal, | |
8cb471e8 | 395 | * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed |
d659072f TM |
396 | * into this function. |
397 | */ | |
398 | int ocfs2_write_super_or_backup(struct ocfs2_super *osb, | |
399 | struct buffer_head *bh) | |
400 | { | |
401 | int ret = 0; | |
a42ab8e1 | 402 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data; |
d659072f | 403 | |
d659072f TM |
404 | BUG_ON(buffer_jbd(bh)); |
405 | ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr); | |
406 | ||
407 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) { | |
408 | ret = -EROFS; | |
c1e8d35e | 409 | mlog_errno(ret); |
d659072f TM |
410 | goto out; |
411 | } | |
412 | ||
413 | lock_buffer(bh); | |
414 | set_buffer_uptodate(bh); | |
415 | ||
416 | /* remove from dirty list before I/O. */ | |
417 | clear_buffer_dirty(bh); | |
418 | ||
419 | get_bh(bh); /* for end_buffer_write_sync() */ | |
420 | bh->b_end_io = end_buffer_write_sync; | |
a42ab8e1 | 421 | ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check); |
d659072f TM |
422 | submit_bh(WRITE, bh); |
423 | ||
424 | wait_on_buffer(bh); | |
425 | ||
426 | if (!buffer_uptodate(bh)) { | |
427 | ret = -EIO; | |
c1e8d35e | 428 | mlog_errno(ret); |
d659072f TM |
429 | } |
430 | ||
431 | out: | |
d659072f TM |
432 | return ret; |
433 | } |