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> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/highmem.h> | |
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" | |
39 | ||
40 | #include "buffer_head_io.h" | |
41 | ||
970e4936 JB |
42 | /* |
43 | * Bits on bh->b_state used by ocfs2. | |
44 | * | |
45 | * These MUST be after the JBD2 bits. Currently BH_Unshadow is the last | |
46 | * JBD2 bit. | |
47 | */ | |
48 | enum ocfs2_state_bits { | |
49 | BH_NeedsValidate = BH_Unshadow + 1, | |
50 | }; | |
51 | ||
52 | /* Expand the magic b_state functions */ | |
53 | BUFFER_FNS(NeedsValidate, needs_validate); | |
54 | ||
ccd979bd MF |
55 | int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh, |
56 | struct inode *inode) | |
57 | { | |
58 | int ret = 0; | |
59 | ||
60 | mlog_entry("(bh->b_blocknr = %llu, inode=%p)\n", | |
61 | (unsigned long long)bh->b_blocknr, inode); | |
62 | ||
63 | BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO); | |
64 | BUG_ON(buffer_jbd(bh)); | |
65 | ||
66 | /* No need to check for a soft readonly file system here. non | |
67 | * journalled writes are only ever done on system files which | |
68 | * can get modified during recovery even if read-only. */ | |
69 | if (ocfs2_is_hard_readonly(osb)) { | |
70 | ret = -EROFS; | |
71 | goto out; | |
72 | } | |
73 | ||
251b6ecc | 74 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
75 | |
76 | lock_buffer(bh); | |
77 | set_buffer_uptodate(bh); | |
78 | ||
79 | /* remove from dirty list before I/O. */ | |
80 | clear_buffer_dirty(bh); | |
81 | ||
da1e9098 | 82 | get_bh(bh); /* for end_buffer_write_sync() */ |
ccd979bd MF |
83 | bh->b_end_io = end_buffer_write_sync; |
84 | submit_bh(WRITE, bh); | |
85 | ||
86 | wait_on_buffer(bh); | |
87 | ||
88 | if (buffer_uptodate(bh)) { | |
89 | ocfs2_set_buffer_uptodate(inode, bh); | |
90 | } else { | |
91 | /* We don't need to remove the clustered uptodate | |
92 | * information for this bh as it's not marked locally | |
93 | * uptodate. */ | |
94 | ret = -EIO; | |
2fe5c1d7 | 95 | put_bh(bh); |
ccd979bd MF |
96 | } |
97 | ||
251b6ecc | 98 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
99 | out: |
100 | mlog_exit(ret); | |
101 | return ret; | |
102 | } | |
103 | ||
da1e9098 JB |
104 | int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block, |
105 | unsigned int nr, struct buffer_head *bhs[]) | |
106 | { | |
107 | int status = 0; | |
108 | unsigned int i; | |
109 | struct buffer_head *bh; | |
110 | ||
111 | if (!nr) { | |
112 | mlog(ML_BH_IO, "No buffers will be read!\n"); | |
113 | goto bail; | |
114 | } | |
115 | ||
116 | for (i = 0 ; i < nr ; i++) { | |
117 | if (bhs[i] == NULL) { | |
118 | bhs[i] = sb_getblk(osb->sb, block++); | |
119 | if (bhs[i] == NULL) { | |
120 | status = -EIO; | |
121 | mlog_errno(status); | |
122 | goto bail; | |
123 | } | |
124 | } | |
125 | bh = bhs[i]; | |
126 | ||
127 | if (buffer_jbd(bh)) { | |
d6b58f89 | 128 | mlog(ML_BH_IO, |
da1e9098 JB |
129 | "trying to sync read a jbd " |
130 | "managed bh (blocknr = %llu), skipping\n", | |
131 | (unsigned long long)bh->b_blocknr); | |
132 | continue; | |
133 | } | |
134 | ||
135 | if (buffer_dirty(bh)) { | |
136 | /* This should probably be a BUG, or | |
137 | * at least return an error. */ | |
138 | mlog(ML_ERROR, | |
139 | "trying to sync read a dirty " | |
140 | "buffer! (blocknr = %llu), skipping\n", | |
141 | (unsigned long long)bh->b_blocknr); | |
142 | continue; | |
143 | } | |
144 | ||
145 | lock_buffer(bh); | |
146 | if (buffer_jbd(bh)) { | |
147 | mlog(ML_ERROR, | |
148 | "block %llu had the JBD bit set " | |
149 | "while I was in lock_buffer!", | |
150 | (unsigned long long)bh->b_blocknr); | |
151 | BUG(); | |
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 | ||
31d33073 | 181 | int ocfs2_read_blocks(struct inode *inode, 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; | |
189 | ||
31d33073 JB |
190 | mlog_entry("(inode=%p, block=(%llu), nr=(%d), flags=%d)\n", |
191 | inode, (unsigned long long)block, nr, flags); | |
ccd979bd | 192 | |
31d33073 | 193 | BUG_ON(!inode); |
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) { | |
211 | mlog(ML_BH_IO, "No buffers will be read!\n"); | |
212 | status = 0; | |
213 | goto bail; | |
214 | } | |
215 | ||
31d33073 | 216 | mutex_lock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
217 | for (i = 0 ; i < nr ; i++) { |
218 | if (bhs[i] == NULL) { | |
31d33073 | 219 | bhs[i] = sb_getblk(inode->i_sb, block++); |
ccd979bd | 220 | if (bhs[i] == NULL) { |
31d33073 | 221 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd MF |
222 | status = -EIO; |
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 | ||
d4a8c93c | 254 | if (!ignore_cache && !ocfs2_buffer_uptodate(inode, bh)) { |
ccd979bd | 255 | mlog(ML_UPTODATE, |
b0697053 | 256 | "bh (%llu), inode %llu not uptodate\n", |
ccd979bd | 257 | (unsigned long long)bh->b_blocknr, |
b0697053 | 258 | (unsigned long long)OCFS2_I(inode)->ip_blkno); |
d4a8c93c JB |
259 | /* We're using ignore_cache here to say |
260 | * "go to disk" */ | |
ccd979bd MF |
261 | ignore_cache = 1; |
262 | } | |
263 | ||
ccd979bd | 264 | if (buffer_jbd(bh)) { |
d4a8c93c | 265 | if (ignore_cache) |
ccd979bd MF |
266 | mlog(ML_BH_IO, "trying to sync read a jbd " |
267 | "managed bh (blocknr = %llu)\n", | |
268 | (unsigned long long)bh->b_blocknr); | |
269 | continue; | |
270 | } | |
271 | ||
d4a8c93c | 272 | if (ignore_cache) { |
ccd979bd MF |
273 | if (buffer_dirty(bh)) { |
274 | /* This should probably be a BUG, or | |
275 | * at least return an error. */ | |
276 | mlog(ML_BH_IO, "asking me to sync read a dirty " | |
277 | "buffer! (blocknr = %llu)\n", | |
278 | (unsigned long long)bh->b_blocknr); | |
279 | continue; | |
280 | } | |
281 | ||
aa958874 MF |
282 | /* A read-ahead request was made - if the |
283 | * buffer is already under read-ahead from a | |
284 | * previously submitted request than we are | |
285 | * done here. */ | |
286 | if ((flags & OCFS2_BH_READAHEAD) | |
287 | && ocfs2_buffer_read_ahead(inode, bh)) | |
288 | continue; | |
289 | ||
ccd979bd MF |
290 | lock_buffer(bh); |
291 | if (buffer_jbd(bh)) { | |
292 | #ifdef CATCH_BH_JBD_RACES | |
293 | mlog(ML_ERROR, "block %llu had the JBD bit set " | |
294 | "while I was in lock_buffer!", | |
295 | (unsigned long long)bh->b_blocknr); | |
296 | BUG(); | |
297 | #else | |
298 | unlock_buffer(bh); | |
299 | continue; | |
300 | #endif | |
301 | } | |
aa958874 MF |
302 | |
303 | /* Re-check ocfs2_buffer_uptodate() as a | |
304 | * previously read-ahead buffer may have | |
305 | * completed I/O while we were waiting for the | |
306 | * buffer lock. */ | |
d4a8c93c | 307 | if (!(flags & OCFS2_BH_IGNORE_CACHE) |
aa958874 MF |
308 | && !(flags & OCFS2_BH_READAHEAD) |
309 | && ocfs2_buffer_uptodate(inode, bh)) { | |
310 | unlock_buffer(bh); | |
311 | continue; | |
312 | } | |
313 | ||
ccd979bd MF |
314 | clear_buffer_uptodate(bh); |
315 | get_bh(bh); /* for end_buffer_read_sync() */ | |
970e4936 JB |
316 | if (validate) |
317 | set_buffer_needs_validate(bh); | |
ccd979bd | 318 | bh->b_end_io = end_buffer_read_sync; |
aa958874 | 319 | submit_bh(READ, bh); |
ccd979bd MF |
320 | continue; |
321 | } | |
322 | } | |
323 | ||
324 | status = 0; | |
325 | ||
326 | for (i = (nr - 1); i >= 0; i--) { | |
327 | bh = bhs[i]; | |
328 | ||
aa958874 MF |
329 | if (!(flags & OCFS2_BH_READAHEAD)) { |
330 | /* We know this can't have changed as we hold the | |
331 | * inode sem. Avoid doing any work on the bh if the | |
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); | |
355 | status = validate(inode->i_sb, bh); | |
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. */ | |
31d33073 | 367 | ocfs2_set_buffer_uptodate(inode, bh); |
ccd979bd | 368 | } |
31d33073 | 369 | mutex_unlock(&OCFS2_I(inode)->ip_io_mutex); |
ccd979bd | 370 | |
aa958874 | 371 | mlog(ML_BH_IO, "block=(%llu), nr=(%d), cached=%s, flags=0x%x\n", |
b0697053 | 372 | (unsigned long long)block, nr, |
d4a8c93c JB |
373 | ((flags & OCFS2_BH_IGNORE_CACHE) || ignore_cache) ? "no" : "yes", |
374 | flags); | |
ccd979bd MF |
375 | |
376 | bail: | |
377 | ||
378 | mlog_exit(status); | |
379 | return status; | |
380 | } | |
d659072f TM |
381 | |
382 | /* Check whether the blkno is the super block or one of the backups. */ | |
383 | static void ocfs2_check_super_or_backup(struct super_block *sb, | |
384 | sector_t blkno) | |
385 | { | |
386 | int i; | |
387 | u64 backup_blkno; | |
388 | ||
389 | if (blkno == OCFS2_SUPER_BLOCK_BLKNO) | |
390 | return; | |
391 | ||
392 | for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) { | |
393 | backup_blkno = ocfs2_backup_super_blkno(sb, i); | |
394 | if (backup_blkno == blkno) | |
395 | return; | |
396 | } | |
397 | ||
398 | BUG(); | |
399 | } | |
400 | ||
401 | /* | |
402 | * Write super block and backups doesn't need to collaborate with journal, | |
403 | * so we don't need to lock ip_io_mutex and inode doesn't need to bea passed | |
404 | * into this function. | |
405 | */ | |
406 | int ocfs2_write_super_or_backup(struct ocfs2_super *osb, | |
407 | struct buffer_head *bh) | |
408 | { | |
409 | int ret = 0; | |
410 | ||
411 | mlog_entry_void(); | |
412 | ||
413 | BUG_ON(buffer_jbd(bh)); | |
414 | ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr); | |
415 | ||
416 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) { | |
417 | ret = -EROFS; | |
418 | goto out; | |
419 | } | |
420 | ||
421 | lock_buffer(bh); | |
422 | set_buffer_uptodate(bh); | |
423 | ||
424 | /* remove from dirty list before I/O. */ | |
425 | clear_buffer_dirty(bh); | |
426 | ||
427 | get_bh(bh); /* for end_buffer_write_sync() */ | |
428 | bh->b_end_io = end_buffer_write_sync; | |
429 | submit_bh(WRITE, bh); | |
430 | ||
431 | wait_on_buffer(bh); | |
432 | ||
433 | if (!buffer_uptodate(bh)) { | |
434 | ret = -EIO; | |
2fe5c1d7 | 435 | put_bh(bh); |
d659072f TM |
436 | } |
437 | ||
438 | out: | |
439 | mlog_exit(ret); | |
440 | return ret; | |
441 | } |