Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
da6dd40d | 3 | * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
5c676f6d | 15 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 16 | #include <linux/crc32.h> |
a25311c8 | 17 | #include <linux/delay.h> |
ec69b188 SW |
18 | #include <linux/kthread.h> |
19 | #include <linux/freezer.h> | |
254db57f | 20 | #include <linux/bio.h> |
b3b94faa DT |
21 | |
22 | #include "gfs2.h" | |
5c676f6d | 23 | #include "incore.h" |
b3b94faa DT |
24 | #include "bmap.h" |
25 | #include "glock.h" | |
26 | #include "log.h" | |
27 | #include "lops.h" | |
28 | #include "meta_io.h" | |
5c676f6d | 29 | #include "util.h" |
71b86f56 | 30 | #include "dir.h" |
63997775 | 31 | #include "trace_gfs2.h" |
b3b94faa DT |
32 | |
33 | #define PULL 1 | |
34 | ||
b3b94faa DT |
35 | /** |
36 | * gfs2_struct2blk - compute stuff | |
37 | * @sdp: the filesystem | |
38 | * @nstruct: the number of structures | |
39 | * @ssize: the size of the structures | |
40 | * | |
41 | * Compute the number of log descriptor blocks needed to hold a certain number | |
42 | * of structures of a certain size. | |
43 | * | |
44 | * Returns: the number of blocks needed (minimum is always 1) | |
45 | */ | |
46 | ||
47 | unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct, | |
48 | unsigned int ssize) | |
49 | { | |
50 | unsigned int blks; | |
51 | unsigned int first, second; | |
52 | ||
53 | blks = 1; | |
faa31ce8 | 54 | first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize; |
b3b94faa DT |
55 | |
56 | if (nstruct > first) { | |
568f4c96 SW |
57 | second = (sdp->sd_sb.sb_bsize - |
58 | sizeof(struct gfs2_meta_header)) / ssize; | |
5c676f6d | 59 | blks += DIV_ROUND_UP(nstruct - first, second); |
b3b94faa DT |
60 | } |
61 | ||
62 | return blks; | |
63 | } | |
64 | ||
1e1a3d03 SW |
65 | /** |
66 | * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters | |
67 | * @mapping: The associated mapping (maybe NULL) | |
68 | * @bd: The gfs2_bufdata to remove | |
69 | * | |
70 | * The log lock _must_ be held when calling this function | |
71 | * | |
72 | */ | |
73 | ||
f91a0d3e | 74 | void gfs2_remove_from_ail(struct gfs2_bufdata *bd) |
1e1a3d03 SW |
75 | { |
76 | bd->bd_ail = NULL; | |
1ad38c43 SW |
77 | list_del_init(&bd->bd_ail_st_list); |
78 | list_del_init(&bd->bd_ail_gl_list); | |
1e1a3d03 | 79 | atomic_dec(&bd->bd_gl->gl_ail_count); |
1e1a3d03 SW |
80 | brelse(bd->bd_bh); |
81 | } | |
82 | ||
ddacfaf7 SW |
83 | /** |
84 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
85 | * @sdp: the filesystem | |
86 | * @tr: the part of the AIL | |
87 | * | |
88 | */ | |
89 | ||
90 | static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
2d81afb8 HH |
91 | __releases(&sdp->sd_log_lock) |
92 | __acquires(&sdp->sd_log_lock) | |
ddacfaf7 SW |
93 | { |
94 | struct gfs2_bufdata *bd, *s; | |
95 | struct buffer_head *bh; | |
96 | int retry; | |
97 | ||
ddacfaf7 SW |
98 | do { |
99 | retry = 0; | |
100 | ||
101 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
102 | bd_ail_st_list) { | |
103 | bh = bd->bd_bh; | |
104 | ||
105 | gfs2_assert(sdp, bd->bd_ail == ai); | |
106 | ||
107 | if (!buffer_busy(bh)) { | |
16615be1 | 108 | if (!buffer_uptodate(bh)) |
ddacfaf7 | 109 | gfs2_io_error_bh(sdp, bh); |
ddacfaf7 SW |
110 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); |
111 | continue; | |
112 | } | |
113 | ||
114 | if (!buffer_dirty(bh)) | |
115 | continue; | |
116 | ||
117 | list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
118 | ||
16615be1 | 119 | get_bh(bh); |
ddacfaf7 | 120 | gfs2_log_unlock(sdp); |
16615be1 SW |
121 | lock_buffer(bh); |
122 | if (test_clear_buffer_dirty(bh)) { | |
123 | bh->b_end_io = end_buffer_write_sync; | |
c969f58c | 124 | submit_bh(WRITE_SYNC_PLUG, bh); |
16615be1 SW |
125 | } else { |
126 | unlock_buffer(bh); | |
127 | brelse(bh); | |
128 | } | |
ddacfaf7 SW |
129 | gfs2_log_lock(sdp); |
130 | ||
131 | retry = 1; | |
132 | break; | |
133 | } | |
134 | } while (retry); | |
135 | } | |
136 | ||
137 | /** | |
138 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
139 | * @sdp: the filesystem | |
140 | * @ai: the AIL entry | |
141 | * | |
142 | */ | |
143 | ||
144 | static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags) | |
145 | { | |
146 | struct gfs2_bufdata *bd, *s; | |
147 | struct buffer_head *bh; | |
148 | ||
149 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
150 | bd_ail_st_list) { | |
151 | bh = bd->bd_bh; | |
152 | ||
153 | gfs2_assert(sdp, bd->bd_ail == ai); | |
154 | ||
155 | if (buffer_busy(bh)) { | |
156 | if (flags & DIO_ALL) | |
157 | continue; | |
158 | else | |
159 | break; | |
160 | } | |
161 | ||
162 | if (!buffer_uptodate(bh)) | |
163 | gfs2_io_error_bh(sdp, bh); | |
164 | ||
165 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
166 | } | |
167 | ||
168 | return list_empty(&ai->ai_ail1_list); | |
169 | } | |
170 | ||
a25311c8 | 171 | static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags) |
b3b94faa | 172 | { |
693ddeab | 173 | struct list_head *head; |
cd915493 | 174 | u64 sync_gen; |
74669416 SW |
175 | struct list_head *first; |
176 | struct gfs2_ail *first_ai, *ai, *tmp; | |
177 | int done = 0; | |
b3b94faa DT |
178 | |
179 | gfs2_log_lock(sdp); | |
693ddeab | 180 | head = &sdp->sd_ail1_list; |
b3b94faa DT |
181 | if (list_empty(head)) { |
182 | gfs2_log_unlock(sdp); | |
183 | return; | |
184 | } | |
185 | sync_gen = sdp->sd_ail_sync_gen++; | |
186 | ||
187 | first = head->prev; | |
188 | first_ai = list_entry(first, struct gfs2_ail, ai_list); | |
189 | first_ai->ai_sync_gen = sync_gen; | |
74669416 | 190 | gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */ |
b3b94faa DT |
191 | |
192 | if (flags & DIO_ALL) | |
193 | first = NULL; | |
194 | ||
74669416 | 195 | while(!done) { |
484adff8 SW |
196 | if (first && (head->prev != first || |
197 | gfs2_ail1_empty_one(sdp, first_ai, 0))) | |
b3b94faa DT |
198 | break; |
199 | ||
74669416 SW |
200 | done = 1; |
201 | list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) { | |
b3b94faa DT |
202 | if (ai->ai_sync_gen >= sync_gen) |
203 | continue; | |
204 | ai->ai_sync_gen = sync_gen; | |
74669416 SW |
205 | gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */ |
206 | done = 0; | |
b3b94faa DT |
207 | break; |
208 | } | |
b3b94faa DT |
209 | } |
210 | ||
211 | gfs2_log_unlock(sdp); | |
212 | } | |
213 | ||
ec69b188 | 214 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags) |
b3b94faa DT |
215 | { |
216 | struct gfs2_ail *ai, *s; | |
217 | int ret; | |
218 | ||
219 | gfs2_log_lock(sdp); | |
220 | ||
221 | list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) { | |
222 | if (gfs2_ail1_empty_one(sdp, ai, flags)) | |
223 | list_move(&ai->ai_list, &sdp->sd_ail2_list); | |
224 | else if (!(flags & DIO_ALL)) | |
225 | break; | |
226 | } | |
227 | ||
228 | ret = list_empty(&sdp->sd_ail1_list); | |
229 | ||
230 | gfs2_log_unlock(sdp); | |
231 | ||
232 | return ret; | |
233 | } | |
234 | ||
ddacfaf7 SW |
235 | |
236 | /** | |
237 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
238 | * @sdp: the filesystem | |
239 | * @ai: the AIL entry | |
240 | * | |
241 | */ | |
242 | ||
243 | static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
244 | { | |
245 | struct list_head *head = &ai->ai_ail2_list; | |
246 | struct gfs2_bufdata *bd; | |
247 | ||
248 | while (!list_empty(head)) { | |
249 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
250 | bd_ail_st_list); | |
251 | gfs2_assert(sdp, bd->bd_ail == ai); | |
f91a0d3e | 252 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
253 | } |
254 | } | |
255 | ||
b3b94faa DT |
256 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
257 | { | |
258 | struct gfs2_ail *ai, *safe; | |
259 | unsigned int old_tail = sdp->sd_log_tail; | |
260 | int wrap = (new_tail < old_tail); | |
261 | int a, b, rm; | |
262 | ||
263 | gfs2_log_lock(sdp); | |
264 | ||
265 | list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) { | |
266 | a = (old_tail <= ai->ai_first); | |
267 | b = (ai->ai_first < new_tail); | |
268 | rm = (wrap) ? (a || b) : (a && b); | |
269 | if (!rm) | |
270 | continue; | |
271 | ||
272 | gfs2_ail2_empty_one(sdp, ai); | |
273 | list_del(&ai->ai_list); | |
274 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list)); | |
275 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list)); | |
276 | kfree(ai); | |
277 | } | |
278 | ||
279 | gfs2_log_unlock(sdp); | |
280 | } | |
281 | ||
282 | /** | |
283 | * gfs2_log_reserve - Make a log reservation | |
284 | * @sdp: The GFS2 superblock | |
285 | * @blks: The number of blocks to reserve | |
286 | * | |
89918647 | 287 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 288 | * due to the fact that there is a small number of header blocks |
b004157a SW |
289 | * associated with each log flush. The exact number can't be known until |
290 | * flush time, so we ensure that we have just enough free blocks at all | |
291 | * times to avoid running out during a log flush. | |
292 | * | |
b3b94faa DT |
293 | * Returns: errno |
294 | */ | |
295 | ||
296 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
297 | { | |
b3b94faa | 298 | unsigned int try = 0; |
89918647 | 299 | unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize); |
b3b94faa DT |
300 | |
301 | if (gfs2_assert_warn(sdp, blks) || | |
302 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
303 | return -EINVAL; | |
304 | ||
71b86f56 | 305 | mutex_lock(&sdp->sd_log_reserve_mutex); |
484adff8 | 306 | gfs2_log_lock(sdp); |
fd041f0b | 307 | while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) { |
b3b94faa | 308 | gfs2_log_unlock(sdp); |
b3b94faa | 309 | gfs2_ail1_empty(sdp, 0); |
b09e593d | 310 | gfs2_log_flush(sdp, NULL); |
b3b94faa DT |
311 | |
312 | if (try++) | |
313 | gfs2_ail1_start(sdp, 0); | |
484adff8 | 314 | gfs2_log_lock(sdp); |
b3b94faa | 315 | } |
fd041f0b | 316 | atomic_sub(blks, &sdp->sd_log_blks_free); |
63997775 | 317 | trace_gfs2_log_blocks(sdp, -blks); |
484adff8 SW |
318 | gfs2_log_unlock(sdp); |
319 | mutex_unlock(&sdp->sd_log_reserve_mutex); | |
320 | ||
321 | down_read(&sdp->sd_log_flush_lock); | |
b3b94faa DT |
322 | |
323 | return 0; | |
324 | } | |
325 | ||
326 | /** | |
327 | * gfs2_log_release - Release a given number of log blocks | |
328 | * @sdp: The GFS2 superblock | |
329 | * @blks: The number of blocks | |
330 | * | |
331 | */ | |
332 | ||
333 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
334 | { | |
b3b94faa DT |
335 | |
336 | gfs2_log_lock(sdp); | |
fd041f0b | 337 | atomic_add(blks, &sdp->sd_log_blks_free); |
63997775 | 338 | trace_gfs2_log_blocks(sdp, blks); |
b3b94faa | 339 | gfs2_assert_withdraw(sdp, |
fd041f0b | 340 | atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks); |
b3b94faa | 341 | gfs2_log_unlock(sdp); |
ed386507 | 342 | up_read(&sdp->sd_log_flush_lock); |
b3b94faa DT |
343 | } |
344 | ||
cd915493 | 345 | static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn) |
b3b94faa | 346 | { |
da6dd40d BP |
347 | struct gfs2_journal_extent *je; |
348 | ||
349 | list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) { | |
350 | if (lbn >= je->lblock && lbn < je->lblock + je->blocks) | |
ff91cc9b | 351 | return je->dblock + lbn - je->lblock; |
da6dd40d BP |
352 | } |
353 | ||
354 | return -1; | |
b3b94faa DT |
355 | } |
356 | ||
357 | /** | |
358 | * log_distance - Compute distance between two journal blocks | |
359 | * @sdp: The GFS2 superblock | |
360 | * @newer: The most recent journal block of the pair | |
361 | * @older: The older journal block of the pair | |
362 | * | |
363 | * Compute the distance (in the journal direction) between two | |
364 | * blocks in the journal | |
365 | * | |
366 | * Returns: the distance in blocks | |
367 | */ | |
368 | ||
faa31ce8 | 369 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
370 | unsigned int older) |
371 | { | |
372 | int dist; | |
373 | ||
374 | dist = newer - older; | |
375 | if (dist < 0) | |
376 | dist += sdp->sd_jdesc->jd_blocks; | |
377 | ||
378 | return dist; | |
379 | } | |
380 | ||
2332c443 RP |
381 | /** |
382 | * calc_reserved - Calculate the number of blocks to reserve when | |
383 | * refunding a transaction's unused buffers. | |
384 | * @sdp: The GFS2 superblock | |
385 | * | |
386 | * This is complex. We need to reserve room for all our currently used | |
387 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
388 | * all our journaled data buffers for journaled files (e.g. files in the | |
389 | * meta_fs like rindex, or files for which chattr +j was done.) | |
390 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
391 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
392 | * | |
393 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
394 | * type gets its own log header, for which we need to reserve a block. | |
395 | * In fact, each type has the potential for needing more than one header | |
396 | * in cases where we have more buffers than will fit on a journal page. | |
397 | * Metadata journal entries take up half the space of journaled buffer entries. | |
398 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
399 | * databuf_limit (251) before they cause a wrap around. | |
400 | * | |
401 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
402 | * overall header for the lot. | |
403 | * | |
404 | * Returns: the number of blocks reserved | |
405 | */ | |
406 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
407 | { | |
408 | unsigned int reserved = 0; | |
409 | unsigned int mbuf_limit, metabufhdrs_needed; | |
410 | unsigned int dbuf_limit, databufhdrs_needed; | |
411 | unsigned int revokes = 0; | |
412 | ||
413 | mbuf_limit = buf_limit(sdp); | |
414 | metabufhdrs_needed = (sdp->sd_log_commited_buf + | |
415 | (mbuf_limit - 1)) / mbuf_limit; | |
416 | dbuf_limit = databuf_limit(sdp); | |
417 | databufhdrs_needed = (sdp->sd_log_commited_databuf + | |
418 | (dbuf_limit - 1)) / dbuf_limit; | |
419 | ||
420 | if (sdp->sd_log_commited_revoke) | |
421 | revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke, | |
422 | sizeof(u64)); | |
423 | ||
424 | reserved = sdp->sd_log_commited_buf + metabufhdrs_needed + | |
425 | sdp->sd_log_commited_databuf + databufhdrs_needed + | |
426 | revokes; | |
427 | /* One for the overall header */ | |
428 | if (reserved) | |
429 | reserved++; | |
430 | return reserved; | |
431 | } | |
432 | ||
b3b94faa DT |
433 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
434 | { | |
435 | struct gfs2_ail *ai; | |
436 | unsigned int tail; | |
437 | ||
438 | gfs2_log_lock(sdp); | |
439 | ||
faa31ce8 | 440 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 441 | tail = sdp->sd_log_head; |
faa31ce8 SW |
442 | } else { |
443 | ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list); | |
b3b94faa DT |
444 | tail = ai->ai_first; |
445 | } | |
446 | ||
447 | gfs2_log_unlock(sdp); | |
448 | ||
449 | return tail; | |
450 | } | |
451 | ||
16615be1 | 452 | void gfs2_log_incr_head(struct gfs2_sbd *sdp) |
b3b94faa DT |
453 | { |
454 | if (sdp->sd_log_flush_head == sdp->sd_log_tail) | |
16615be1 | 455 | BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head); |
b3b94faa DT |
456 | |
457 | if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) { | |
458 | sdp->sd_log_flush_head = 0; | |
459 | sdp->sd_log_flush_wrapped = 1; | |
460 | } | |
461 | } | |
462 | ||
16615be1 SW |
463 | /** |
464 | * gfs2_log_write_endio - End of I/O for a log buffer | |
465 | * @bh: The buffer head | |
466 | * @uptodate: I/O Status | |
467 | * | |
468 | */ | |
469 | ||
470 | static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate) | |
471 | { | |
472 | struct gfs2_sbd *sdp = bh->b_private; | |
473 | bh->b_private = NULL; | |
474 | ||
475 | end_buffer_write_sync(bh, uptodate); | |
476 | if (atomic_dec_and_test(&sdp->sd_log_in_flight)) | |
477 | wake_up(&sdp->sd_log_flush_wait); | |
478 | } | |
479 | ||
b3b94faa DT |
480 | /** |
481 | * gfs2_log_get_buf - Get and initialize a buffer to use for log control data | |
482 | * @sdp: The GFS2 superblock | |
483 | * | |
484 | * Returns: the buffer_head | |
485 | */ | |
486 | ||
487 | struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp) | |
488 | { | |
cd915493 | 489 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
490 | struct buffer_head *bh; |
491 | ||
16615be1 | 492 | bh = sb_getblk(sdp->sd_vfs, blkno); |
b3b94faa DT |
493 | lock_buffer(bh); |
494 | memset(bh->b_data, 0, bh->b_size); | |
495 | set_buffer_uptodate(bh); | |
496 | clear_buffer_dirty(bh); | |
16615be1 SW |
497 | gfs2_log_incr_head(sdp); |
498 | atomic_inc(&sdp->sd_log_in_flight); | |
499 | bh->b_private = sdp; | |
500 | bh->b_end_io = gfs2_log_write_endio; | |
b3b94faa DT |
501 | |
502 | return bh; | |
503 | } | |
504 | ||
16615be1 SW |
505 | /** |
506 | * gfs2_fake_write_endio - | |
507 | * @bh: The buffer head | |
508 | * @uptodate: The I/O Status | |
509 | * | |
510 | */ | |
511 | ||
512 | static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate) | |
513 | { | |
514 | struct buffer_head *real_bh = bh->b_private; | |
5a60c532 SW |
515 | struct gfs2_bufdata *bd = real_bh->b_private; |
516 | struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd; | |
16615be1 SW |
517 | |
518 | end_buffer_write_sync(bh, uptodate); | |
519 | free_buffer_head(bh); | |
520 | unlock_buffer(real_bh); | |
521 | brelse(real_bh); | |
522 | if (atomic_dec_and_test(&sdp->sd_log_in_flight)) | |
523 | wake_up(&sdp->sd_log_flush_wait); | |
524 | } | |
525 | ||
b3b94faa DT |
526 | /** |
527 | * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log | |
528 | * @sdp: the filesystem | |
529 | * @data: the data the buffer_head should point to | |
530 | * | |
531 | * Returns: the log buffer descriptor | |
532 | */ | |
533 | ||
534 | struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp, | |
535 | struct buffer_head *real) | |
536 | { | |
cd915493 | 537 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
538 | struct buffer_head *bh; |
539 | ||
16615be1 | 540 | bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL); |
b3b94faa | 541 | atomic_set(&bh->b_count, 1); |
16615be1 | 542 | bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock); |
18ec7d5c | 543 | set_bh_page(bh, real->b_page, bh_offset(real)); |
b3b94faa DT |
544 | bh->b_blocknr = blkno; |
545 | bh->b_size = sdp->sd_sb.sb_bsize; | |
546 | bh->b_bdev = sdp->sd_vfs->s_bdev; | |
16615be1 SW |
547 | bh->b_private = real; |
548 | bh->b_end_io = gfs2_fake_write_endio; | |
b3b94faa | 549 | |
16615be1 SW |
550 | gfs2_log_incr_head(sdp); |
551 | atomic_inc(&sdp->sd_log_in_flight); | |
b3b94faa DT |
552 | |
553 | return bh; | |
554 | } | |
555 | ||
2332c443 | 556 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
557 | { |
558 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
559 | ||
560 | ail2_empty(sdp, new_tail); | |
561 | ||
562 | gfs2_log_lock(sdp); | |
fd041f0b | 563 | atomic_add(dist, &sdp->sd_log_blks_free); |
63997775 | 564 | trace_gfs2_log_blocks(sdp, dist); |
fd041f0b | 565 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
566 | gfs2_log_unlock(sdp); |
567 | ||
568 | sdp->sd_log_tail = new_tail; | |
569 | } | |
570 | ||
571 | /** | |
572 | * log_write_header - Get and initialize a journal header buffer | |
573 | * @sdp: The GFS2 superblock | |
574 | * | |
575 | * Returns: the initialized log buffer descriptor | |
576 | */ | |
577 | ||
cd915493 | 578 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull) |
b3b94faa | 579 | { |
cd915493 | 580 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
581 | struct buffer_head *bh; |
582 | struct gfs2_log_header *lh; | |
583 | unsigned int tail; | |
cd915493 | 584 | u32 hash; |
b3b94faa | 585 | |
b3b94faa DT |
586 | bh = sb_getblk(sdp->sd_vfs, blkno); |
587 | lock_buffer(bh); | |
588 | memset(bh->b_data, 0, bh->b_size); | |
589 | set_buffer_uptodate(bh); | |
590 | clear_buffer_dirty(bh); | |
b3b94faa DT |
591 | |
592 | gfs2_ail1_empty(sdp, 0); | |
593 | tail = current_tail(sdp); | |
594 | ||
595 | lh = (struct gfs2_log_header *)bh->b_data; | |
596 | memset(lh, 0, sizeof(struct gfs2_log_header)); | |
597 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
598 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); |
599 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
e0f2bf78 SW |
600 | lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++); |
601 | lh->lh_flags = cpu_to_be32(flags); | |
602 | lh->lh_tail = cpu_to_be32(tail); | |
603 | lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head); | |
b3b94faa DT |
604 | hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header)); |
605 | lh->lh_hash = cpu_to_be32(hash); | |
606 | ||
254db57f SW |
607 | bh->b_end_io = end_buffer_write_sync; |
608 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) | |
609 | goto skip_barrier; | |
610 | get_bh(bh); | |
c969f58c | 611 | submit_bh(WRITE_SYNC | (1 << BIO_RW_BARRIER) | (1 << BIO_RW_META), bh); |
254db57f SW |
612 | wait_on_buffer(bh); |
613 | if (buffer_eopnotsupp(bh)) { | |
614 | clear_buffer_eopnotsupp(bh); | |
615 | set_buffer_uptodate(bh); | |
616 | set_bit(SDF_NOBARRIERS, &sdp->sd_flags); | |
617 | lock_buffer(bh); | |
618 | skip_barrier: | |
619 | get_bh(bh); | |
620 | submit_bh(WRITE_SYNC | (1 << BIO_RW_META), bh); | |
621 | wait_on_buffer(bh); | |
622 | } | |
623 | if (!buffer_uptodate(bh)) | |
b3b94faa DT |
624 | gfs2_io_error_bh(sdp, bh); |
625 | brelse(bh); | |
626 | ||
627 | if (sdp->sd_log_tail != tail) | |
2332c443 | 628 | log_pull_tail(sdp, tail); |
b3b94faa DT |
629 | else |
630 | gfs2_assert_withdraw(sdp, !pull); | |
631 | ||
632 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); | |
16615be1 | 633 | gfs2_log_incr_head(sdp); |
b3b94faa DT |
634 | } |
635 | ||
636 | static void log_flush_commit(struct gfs2_sbd *sdp) | |
637 | { | |
16615be1 SW |
638 | DEFINE_WAIT(wait); |
639 | ||
640 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
641 | do { | |
642 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
643 | TASK_UNINTERRUPTIBLE); | |
644 | if (atomic_read(&sdp->sd_log_in_flight)) | |
645 | io_schedule(); | |
646 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
647 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa DT |
648 | } |
649 | ||
16615be1 | 650 | log_write_header(sdp, 0, 0); |
b3b94faa DT |
651 | } |
652 | ||
d7b616e2 SW |
653 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
654 | { | |
655 | struct gfs2_bufdata *bd; | |
656 | struct buffer_head *bh; | |
657 | LIST_HEAD(written); | |
658 | ||
659 | gfs2_log_lock(sdp); | |
660 | while (!list_empty(&sdp->sd_log_le_ordered)) { | |
661 | bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list); | |
662 | list_move(&bd->bd_le.le_list, &written); | |
663 | bh = bd->bd_bh; | |
664 | if (!buffer_dirty(bh)) | |
665 | continue; | |
666 | get_bh(bh); | |
667 | gfs2_log_unlock(sdp); | |
668 | lock_buffer(bh); | |
b8e7cbb6 | 669 | if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) { |
d7b616e2 | 670 | bh->b_end_io = end_buffer_write_sync; |
c969f58c | 671 | submit_bh(WRITE_SYNC_PLUG, bh); |
d7b616e2 SW |
672 | } else { |
673 | unlock_buffer(bh); | |
674 | brelse(bh); | |
675 | } | |
676 | gfs2_log_lock(sdp); | |
677 | } | |
678 | list_splice(&written, &sdp->sd_log_le_ordered); | |
679 | gfs2_log_unlock(sdp); | |
680 | } | |
681 | ||
682 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
683 | { | |
684 | struct gfs2_bufdata *bd; | |
685 | struct buffer_head *bh; | |
686 | ||
687 | gfs2_log_lock(sdp); | |
688 | while (!list_empty(&sdp->sd_log_le_ordered)) { | |
689 | bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list); | |
690 | bh = bd->bd_bh; | |
691 | if (buffer_locked(bh)) { | |
692 | get_bh(bh); | |
693 | gfs2_log_unlock(sdp); | |
694 | wait_on_buffer(bh); | |
695 | brelse(bh); | |
696 | gfs2_log_lock(sdp); | |
697 | continue; | |
698 | } | |
699 | list_del_init(&bd->bd_le.le_list); | |
700 | } | |
701 | gfs2_log_unlock(sdp); | |
702 | } | |
703 | ||
b3b94faa | 704 | /** |
b09e593d | 705 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
706 | * @sdp: the filesystem |
707 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
708 | * | |
709 | */ | |
710 | ||
2bcd610d | 711 | void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl) |
b3b94faa DT |
712 | { |
713 | struct gfs2_ail *ai; | |
714 | ||
484adff8 | 715 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 716 | |
2bcd610d SW |
717 | /* Log might have been flushed while we waited for the flush lock */ |
718 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) { | |
719 | up_write(&sdp->sd_log_flush_lock); | |
720 | return; | |
f55ab26a | 721 | } |
63997775 | 722 | trace_gfs2_log_flush(sdp, 1); |
f55ab26a | 723 | |
b09e593d SW |
724 | ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL); |
725 | INIT_LIST_HEAD(&ai->ai_ail1_list); | |
726 | INIT_LIST_HEAD(&ai->ai_ail2_list); | |
b3b94faa | 727 | |
16615be1 SW |
728 | if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) { |
729 | printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf, | |
730 | sdp->sd_log_commited_buf); | |
731 | gfs2_assert_withdraw(sdp, 0); | |
732 | } | |
733 | if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) { | |
734 | printk(KERN_INFO "GFS2: log databuf %u %u\n", | |
735 | sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf); | |
736 | gfs2_assert_withdraw(sdp, 0); | |
737 | } | |
b3b94faa DT |
738 | gfs2_assert_withdraw(sdp, |
739 | sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke); | |
740 | ||
b3b94faa DT |
741 | sdp->sd_log_flush_head = sdp->sd_log_head; |
742 | sdp->sd_log_flush_wrapped = 0; | |
743 | ai->ai_first = sdp->sd_log_flush_head; | |
744 | ||
d7b616e2 | 745 | gfs2_ordered_write(sdp); |
b3b94faa | 746 | lops_before_commit(sdp); |
d7b616e2 SW |
747 | gfs2_ordered_wait(sdp); |
748 | ||
16615be1 | 749 | if (sdp->sd_log_head != sdp->sd_log_flush_head) |
b3b94faa | 750 | log_flush_commit(sdp); |
2332c443 RP |
751 | else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
752 | gfs2_log_lock(sdp); | |
fd041f0b | 753 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 754 | trace_gfs2_log_blocks(sdp, -1); |
2332c443 | 755 | gfs2_log_unlock(sdp); |
b3b94faa | 756 | log_write_header(sdp, 0, PULL); |
2332c443 | 757 | } |
b3b94faa | 758 | lops_after_commit(sdp, ai); |
b09e593d | 759 | |
fe1a698f SW |
760 | gfs2_log_lock(sdp); |
761 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 SW |
762 | sdp->sd_log_blks_reserved = 0; |
763 | sdp->sd_log_commited_buf = 0; | |
2332c443 | 764 | sdp->sd_log_commited_databuf = 0; |
faa31ce8 | 765 | sdp->sd_log_commited_revoke = 0; |
b3b94faa | 766 | |
b3b94faa DT |
767 | if (!list_empty(&ai->ai_ail1_list)) { |
768 | list_add(&ai->ai_list, &sdp->sd_ail1_list); | |
769 | ai = NULL; | |
770 | } | |
771 | gfs2_log_unlock(sdp); | |
63997775 | 772 | trace_gfs2_log_flush(sdp, 0); |
484adff8 | 773 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
774 | |
775 | kfree(ai); | |
776 | } | |
777 | ||
778 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
779 | { | |
2332c443 | 780 | unsigned int reserved; |
ac39aadd | 781 | unsigned int unused; |
b3b94faa DT |
782 | |
783 | gfs2_log_lock(sdp); | |
784 | ||
785 | sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
2332c443 RP |
786 | sdp->sd_log_commited_databuf += tr->tr_num_databuf_new - |
787 | tr->tr_num_databuf_rm; | |
788 | gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) || | |
789 | (((int)sdp->sd_log_commited_databuf) >= 0)); | |
b3b94faa DT |
790 | sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
791 | gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0); | |
2332c443 | 792 | reserved = calc_reserved(sdp); |
62be1f71 | 793 | gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved); |
ac39aadd | 794 | unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved; |
ac39aadd | 795 | atomic_add(unused, &sdp->sd_log_blks_free); |
63997775 | 796 | trace_gfs2_log_blocks(sdp, unused); |
fd041f0b | 797 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
2332c443 | 798 | sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
799 | sdp->sd_log_blks_reserved = reserved; |
800 | ||
801 | gfs2_log_unlock(sdp); | |
802 | } | |
803 | ||
d0109bfa BP |
804 | static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
805 | { | |
806 | struct list_head *head = &tr->tr_list_buf; | |
807 | struct gfs2_bufdata *bd; | |
808 | ||
809 | gfs2_log_lock(sdp); | |
810 | while (!list_empty(head)) { | |
811 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr); | |
812 | list_del_init(&bd->bd_list_tr); | |
813 | tr->tr_num_buf--; | |
814 | } | |
815 | gfs2_log_unlock(sdp); | |
816 | gfs2_assert_warn(sdp, !tr->tr_num_buf); | |
817 | } | |
818 | ||
b3b94faa DT |
819 | /** |
820 | * gfs2_log_commit - Commit a transaction to the log | |
821 | * @sdp: the filesystem | |
822 | * @tr: the transaction | |
823 | * | |
824 | * Returns: errno | |
825 | */ | |
826 | ||
827 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
828 | { | |
829 | log_refund(sdp, tr); | |
d0109bfa | 830 | buf_lo_incore_commit(sdp, tr); |
b3b94faa | 831 | |
484adff8 | 832 | up_read(&sdp->sd_log_flush_lock); |
b3b94faa | 833 | |
b3b94faa | 834 | gfs2_log_lock(sdp); |
b004157a SW |
835 | if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) |
836 | wake_up_process(sdp->sd_logd_process); | |
837 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
838 | } |
839 | ||
840 | /** | |
841 | * gfs2_log_shutdown - write a shutdown header into a journal | |
842 | * @sdp: the filesystem | |
843 | * | |
844 | */ | |
845 | ||
846 | void gfs2_log_shutdown(struct gfs2_sbd *sdp) | |
847 | { | |
484adff8 | 848 | down_write(&sdp->sd_log_flush_lock); |
b3b94faa | 849 | |
b3b94faa | 850 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa DT |
851 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf); |
852 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
853 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg); | |
854 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf); | |
855 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); | |
856 | ||
857 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
858 | sdp->sd_log_flush_wrapped = 0; | |
859 | ||
2332c443 RP |
860 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, |
861 | (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL); | |
b3b94faa | 862 | |
fd041f0b | 863 | gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks); |
a74604be SW |
864 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
865 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
866 | |
867 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa DT |
868 | sdp->sd_log_tail = sdp->sd_log_head; |
869 | ||
484adff8 | 870 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
871 | } |
872 | ||
a25311c8 SW |
873 | |
874 | /** | |
875 | * gfs2_meta_syncfs - sync all the buffers in a filesystem | |
876 | * @sdp: the filesystem | |
877 | * | |
878 | */ | |
879 | ||
880 | void gfs2_meta_syncfs(struct gfs2_sbd *sdp) | |
881 | { | |
882 | gfs2_log_flush(sdp, NULL); | |
883 | for (;;) { | |
884 | gfs2_ail1_start(sdp, DIO_ALL); | |
885 | if (gfs2_ail1_empty(sdp, DIO_ALL)) | |
886 | break; | |
887 | msleep(10); | |
888 | } | |
889 | } | |
890 | ||
ec69b188 SW |
891 | |
892 | /** | |
893 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
894 | * @sdp: Pointer to GFS2 superblock | |
895 | * | |
896 | * Also, periodically check to make sure that we're using the most recent | |
897 | * journal index. | |
898 | */ | |
899 | ||
900 | int gfs2_logd(void *data) | |
901 | { | |
902 | struct gfs2_sbd *sdp = data; | |
ec69b188 SW |
903 | unsigned long t; |
904 | int need_flush; | |
905 | ||
906 | while (!kthread_should_stop()) { | |
907 | /* Advance the log tail */ | |
908 | ||
909 | t = sdp->sd_log_flush_time + | |
910 | gfs2_tune_get(sdp, gt_log_flush_secs) * HZ; | |
911 | ||
912 | gfs2_ail1_empty(sdp, DIO_ALL); | |
913 | gfs2_log_lock(sdp); | |
914 | need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks); | |
915 | gfs2_log_unlock(sdp); | |
916 | if (need_flush || time_after_eq(jiffies, t)) { | |
917 | gfs2_log_flush(sdp, NULL); | |
918 | sdp->sd_log_flush_time = jiffies; | |
919 | } | |
920 | ||
ec69b188 SW |
921 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
922 | if (freezing(current)) | |
923 | refrigerator(); | |
924 | schedule_timeout_interruptible(t); | |
925 | } | |
926 | ||
927 | return 0; | |
928 | } | |
929 |