]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 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> |
7d308590 | 17 | #include <linux/lm_interface.h> |
a25311c8 | 18 | #include <linux/delay.h> |
b3b94faa DT |
19 | |
20 | #include "gfs2.h" | |
5c676f6d | 21 | #include "incore.h" |
b3b94faa DT |
22 | #include "bmap.h" |
23 | #include "glock.h" | |
24 | #include "log.h" | |
25 | #include "lops.h" | |
26 | #include "meta_io.h" | |
5c676f6d | 27 | #include "util.h" |
71b86f56 | 28 | #include "dir.h" |
b3b94faa DT |
29 | |
30 | #define PULL 1 | |
31 | ||
b3b94faa DT |
32 | /** |
33 | * gfs2_struct2blk - compute stuff | |
34 | * @sdp: the filesystem | |
35 | * @nstruct: the number of structures | |
36 | * @ssize: the size of the structures | |
37 | * | |
38 | * Compute the number of log descriptor blocks needed to hold a certain number | |
39 | * of structures of a certain size. | |
40 | * | |
41 | * Returns: the number of blocks needed (minimum is always 1) | |
42 | */ | |
43 | ||
44 | unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct, | |
45 | unsigned int ssize) | |
46 | { | |
47 | unsigned int blks; | |
48 | unsigned int first, second; | |
49 | ||
50 | blks = 1; | |
faa31ce8 | 51 | first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize; |
b3b94faa DT |
52 | |
53 | if (nstruct > first) { | |
568f4c96 SW |
54 | second = (sdp->sd_sb.sb_bsize - |
55 | sizeof(struct gfs2_meta_header)) / ssize; | |
5c676f6d | 56 | blks += DIV_ROUND_UP(nstruct - first, second); |
b3b94faa DT |
57 | } |
58 | ||
59 | return blks; | |
60 | } | |
61 | ||
1e1a3d03 SW |
62 | /** |
63 | * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters | |
64 | * @mapping: The associated mapping (maybe NULL) | |
65 | * @bd: The gfs2_bufdata to remove | |
66 | * | |
67 | * The log lock _must_ be held when calling this function | |
68 | * | |
69 | */ | |
70 | ||
f91a0d3e | 71 | void gfs2_remove_from_ail(struct gfs2_bufdata *bd) |
1e1a3d03 SW |
72 | { |
73 | bd->bd_ail = NULL; | |
1ad38c43 SW |
74 | list_del_init(&bd->bd_ail_st_list); |
75 | list_del_init(&bd->bd_ail_gl_list); | |
1e1a3d03 | 76 | atomic_dec(&bd->bd_gl->gl_ail_count); |
1e1a3d03 SW |
77 | brelse(bd->bd_bh); |
78 | } | |
79 | ||
ddacfaf7 SW |
80 | /** |
81 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
82 | * @sdp: the filesystem | |
83 | * @tr: the part of the AIL | |
84 | * | |
85 | */ | |
86 | ||
87 | static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
88 | { | |
89 | struct gfs2_bufdata *bd, *s; | |
90 | struct buffer_head *bh; | |
91 | int retry; | |
92 | ||
93 | BUG_ON(!spin_is_locked(&sdp->sd_log_lock)); | |
94 | ||
95 | do { | |
96 | retry = 0; | |
97 | ||
98 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
99 | bd_ail_st_list) { | |
100 | bh = bd->bd_bh; | |
101 | ||
102 | gfs2_assert(sdp, bd->bd_ail == ai); | |
103 | ||
104 | if (!buffer_busy(bh)) { | |
16615be1 | 105 | if (!buffer_uptodate(bh)) |
ddacfaf7 | 106 | gfs2_io_error_bh(sdp, bh); |
ddacfaf7 SW |
107 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); |
108 | continue; | |
109 | } | |
110 | ||
111 | if (!buffer_dirty(bh)) | |
112 | continue; | |
113 | ||
114 | list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
115 | ||
16615be1 | 116 | get_bh(bh); |
ddacfaf7 | 117 | gfs2_log_unlock(sdp); |
16615be1 SW |
118 | lock_buffer(bh); |
119 | if (test_clear_buffer_dirty(bh)) { | |
120 | bh->b_end_io = end_buffer_write_sync; | |
121 | submit_bh(WRITE, bh); | |
122 | } else { | |
123 | unlock_buffer(bh); | |
124 | brelse(bh); | |
125 | } | |
ddacfaf7 SW |
126 | gfs2_log_lock(sdp); |
127 | ||
128 | retry = 1; | |
129 | break; | |
130 | } | |
131 | } while (retry); | |
132 | } | |
133 | ||
134 | /** | |
135 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
136 | * @sdp: the filesystem | |
137 | * @ai: the AIL entry | |
138 | * | |
139 | */ | |
140 | ||
141 | static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags) | |
142 | { | |
143 | struct gfs2_bufdata *bd, *s; | |
144 | struct buffer_head *bh; | |
145 | ||
146 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
147 | bd_ail_st_list) { | |
148 | bh = bd->bd_bh; | |
149 | ||
150 | gfs2_assert(sdp, bd->bd_ail == ai); | |
151 | ||
152 | if (buffer_busy(bh)) { | |
153 | if (flags & DIO_ALL) | |
154 | continue; | |
155 | else | |
156 | break; | |
157 | } | |
158 | ||
159 | if (!buffer_uptodate(bh)) | |
160 | gfs2_io_error_bh(sdp, bh); | |
161 | ||
162 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
163 | } | |
164 | ||
165 | return list_empty(&ai->ai_ail1_list); | |
166 | } | |
167 | ||
a25311c8 | 168 | static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags) |
b3b94faa | 169 | { |
693ddeab | 170 | struct list_head *head; |
cd915493 | 171 | u64 sync_gen; |
74669416 SW |
172 | struct list_head *first; |
173 | struct gfs2_ail *first_ai, *ai, *tmp; | |
174 | int done = 0; | |
b3b94faa DT |
175 | |
176 | gfs2_log_lock(sdp); | |
693ddeab | 177 | head = &sdp->sd_ail1_list; |
b3b94faa DT |
178 | if (list_empty(head)) { |
179 | gfs2_log_unlock(sdp); | |
180 | return; | |
181 | } | |
182 | sync_gen = sdp->sd_ail_sync_gen++; | |
183 | ||
184 | first = head->prev; | |
185 | first_ai = list_entry(first, struct gfs2_ail, ai_list); | |
186 | first_ai->ai_sync_gen = sync_gen; | |
74669416 | 187 | gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */ |
b3b94faa DT |
188 | |
189 | if (flags & DIO_ALL) | |
190 | first = NULL; | |
191 | ||
74669416 | 192 | while(!done) { |
484adff8 SW |
193 | if (first && (head->prev != first || |
194 | gfs2_ail1_empty_one(sdp, first_ai, 0))) | |
b3b94faa DT |
195 | break; |
196 | ||
74669416 SW |
197 | done = 1; |
198 | list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) { | |
b3b94faa DT |
199 | if (ai->ai_sync_gen >= sync_gen) |
200 | continue; | |
201 | ai->ai_sync_gen = sync_gen; | |
74669416 SW |
202 | gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */ |
203 | done = 0; | |
b3b94faa DT |
204 | break; |
205 | } | |
b3b94faa DT |
206 | } |
207 | ||
208 | gfs2_log_unlock(sdp); | |
209 | } | |
210 | ||
211 | int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags) | |
212 | { | |
213 | struct gfs2_ail *ai, *s; | |
214 | int ret; | |
215 | ||
216 | gfs2_log_lock(sdp); | |
217 | ||
218 | list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) { | |
219 | if (gfs2_ail1_empty_one(sdp, ai, flags)) | |
220 | list_move(&ai->ai_list, &sdp->sd_ail2_list); | |
221 | else if (!(flags & DIO_ALL)) | |
222 | break; | |
223 | } | |
224 | ||
225 | ret = list_empty(&sdp->sd_ail1_list); | |
226 | ||
227 | gfs2_log_unlock(sdp); | |
228 | ||
229 | return ret; | |
230 | } | |
231 | ||
ddacfaf7 SW |
232 | |
233 | /** | |
234 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
235 | * @sdp: the filesystem | |
236 | * @ai: the AIL entry | |
237 | * | |
238 | */ | |
239 | ||
240 | static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
241 | { | |
242 | struct list_head *head = &ai->ai_ail2_list; | |
243 | struct gfs2_bufdata *bd; | |
244 | ||
245 | while (!list_empty(head)) { | |
246 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
247 | bd_ail_st_list); | |
248 | gfs2_assert(sdp, bd->bd_ail == ai); | |
f91a0d3e | 249 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
250 | } |
251 | } | |
252 | ||
b3b94faa DT |
253 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
254 | { | |
255 | struct gfs2_ail *ai, *safe; | |
256 | unsigned int old_tail = sdp->sd_log_tail; | |
257 | int wrap = (new_tail < old_tail); | |
258 | int a, b, rm; | |
259 | ||
260 | gfs2_log_lock(sdp); | |
261 | ||
262 | list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) { | |
263 | a = (old_tail <= ai->ai_first); | |
264 | b = (ai->ai_first < new_tail); | |
265 | rm = (wrap) ? (a || b) : (a && b); | |
266 | if (!rm) | |
267 | continue; | |
268 | ||
269 | gfs2_ail2_empty_one(sdp, ai); | |
270 | list_del(&ai->ai_list); | |
271 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list)); | |
272 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list)); | |
273 | kfree(ai); | |
274 | } | |
275 | ||
276 | gfs2_log_unlock(sdp); | |
277 | } | |
278 | ||
279 | /** | |
280 | * gfs2_log_reserve - Make a log reservation | |
281 | * @sdp: The GFS2 superblock | |
282 | * @blks: The number of blocks to reserve | |
283 | * | |
89918647 | 284 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 285 | * due to the fact that there is a small number of header blocks |
b004157a SW |
286 | * associated with each log flush. The exact number can't be known until |
287 | * flush time, so we ensure that we have just enough free blocks at all | |
288 | * times to avoid running out during a log flush. | |
289 | * | |
b3b94faa DT |
290 | * Returns: errno |
291 | */ | |
292 | ||
293 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
294 | { | |
b3b94faa | 295 | unsigned int try = 0; |
89918647 | 296 | unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize); |
b3b94faa DT |
297 | |
298 | if (gfs2_assert_warn(sdp, blks) || | |
299 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
300 | return -EINVAL; | |
301 | ||
71b86f56 | 302 | mutex_lock(&sdp->sd_log_reserve_mutex); |
484adff8 | 303 | gfs2_log_lock(sdp); |
89918647 | 304 | while(sdp->sd_log_blks_free <= (blks + reserved_blks)) { |
b3b94faa | 305 | gfs2_log_unlock(sdp); |
b3b94faa | 306 | gfs2_ail1_empty(sdp, 0); |
b09e593d | 307 | gfs2_log_flush(sdp, NULL); |
b3b94faa DT |
308 | |
309 | if (try++) | |
310 | gfs2_ail1_start(sdp, 0); | |
484adff8 | 311 | gfs2_log_lock(sdp); |
b3b94faa | 312 | } |
484adff8 SW |
313 | sdp->sd_log_blks_free -= blks; |
314 | gfs2_log_unlock(sdp); | |
315 | mutex_unlock(&sdp->sd_log_reserve_mutex); | |
316 | ||
317 | down_read(&sdp->sd_log_flush_lock); | |
b3b94faa DT |
318 | |
319 | return 0; | |
320 | } | |
321 | ||
322 | /** | |
323 | * gfs2_log_release - Release a given number of log blocks | |
324 | * @sdp: The GFS2 superblock | |
325 | * @blks: The number of blocks | |
326 | * | |
327 | */ | |
328 | ||
329 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
330 | { | |
b3b94faa DT |
331 | |
332 | gfs2_log_lock(sdp); | |
333 | sdp->sd_log_blks_free += blks; | |
334 | gfs2_assert_withdraw(sdp, | |
335 | sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks); | |
336 | gfs2_log_unlock(sdp); | |
ed386507 | 337 | up_read(&sdp->sd_log_flush_lock); |
b3b94faa DT |
338 | } |
339 | ||
cd915493 | 340 | static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn) |
b3b94faa | 341 | { |
23591256 | 342 | struct inode *inode = sdp->sd_jdesc->jd_inode; |
b3b94faa | 343 | int error; |
23591256 | 344 | struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 }; |
b3b94faa | 345 | |
23591256 SW |
346 | bh_map.b_size = 1 << inode->i_blkbits; |
347 | error = gfs2_block_map(inode, lbn, 0, &bh_map); | |
7a6bbacb | 348 | if (error || !bh_map.b_blocknr) |
aed3255f RK |
349 | printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error, |
350 | (unsigned long long)bh_map.b_blocknr, lbn); | |
7a6bbacb | 351 | gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr); |
b3b94faa | 352 | |
7a6bbacb | 353 | return bh_map.b_blocknr; |
b3b94faa DT |
354 | } |
355 | ||
356 | /** | |
357 | * log_distance - Compute distance between two journal blocks | |
358 | * @sdp: The GFS2 superblock | |
359 | * @newer: The most recent journal block of the pair | |
360 | * @older: The older journal block of the pair | |
361 | * | |
362 | * Compute the distance (in the journal direction) between two | |
363 | * blocks in the journal | |
364 | * | |
365 | * Returns: the distance in blocks | |
366 | */ | |
367 | ||
faa31ce8 | 368 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
369 | unsigned int older) |
370 | { | |
371 | int dist; | |
372 | ||
373 | dist = newer - older; | |
374 | if (dist < 0) | |
375 | dist += sdp->sd_jdesc->jd_blocks; | |
376 | ||
377 | return dist; | |
378 | } | |
379 | ||
2332c443 RP |
380 | /** |
381 | * calc_reserved - Calculate the number of blocks to reserve when | |
382 | * refunding a transaction's unused buffers. | |
383 | * @sdp: The GFS2 superblock | |
384 | * | |
385 | * This is complex. We need to reserve room for all our currently used | |
386 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
387 | * all our journaled data buffers for journaled files (e.g. files in the | |
388 | * meta_fs like rindex, or files for which chattr +j was done.) | |
389 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
390 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
391 | * | |
392 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
393 | * type gets its own log header, for which we need to reserve a block. | |
394 | * In fact, each type has the potential for needing more than one header | |
395 | * in cases where we have more buffers than will fit on a journal page. | |
396 | * Metadata journal entries take up half the space of journaled buffer entries. | |
397 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
398 | * databuf_limit (251) before they cause a wrap around. | |
399 | * | |
400 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
401 | * overall header for the lot. | |
402 | * | |
403 | * Returns: the number of blocks reserved | |
404 | */ | |
405 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
406 | { | |
407 | unsigned int reserved = 0; | |
408 | unsigned int mbuf_limit, metabufhdrs_needed; | |
409 | unsigned int dbuf_limit, databufhdrs_needed; | |
410 | unsigned int revokes = 0; | |
411 | ||
412 | mbuf_limit = buf_limit(sdp); | |
413 | metabufhdrs_needed = (sdp->sd_log_commited_buf + | |
414 | (mbuf_limit - 1)) / mbuf_limit; | |
415 | dbuf_limit = databuf_limit(sdp); | |
416 | databufhdrs_needed = (sdp->sd_log_commited_databuf + | |
417 | (dbuf_limit - 1)) / dbuf_limit; | |
418 | ||
419 | if (sdp->sd_log_commited_revoke) | |
420 | revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke, | |
421 | sizeof(u64)); | |
422 | ||
423 | reserved = sdp->sd_log_commited_buf + metabufhdrs_needed + | |
424 | sdp->sd_log_commited_databuf + databufhdrs_needed + | |
425 | revokes; | |
426 | /* One for the overall header */ | |
427 | if (reserved) | |
428 | reserved++; | |
429 | return reserved; | |
430 | } | |
431 | ||
b3b94faa DT |
432 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
433 | { | |
434 | struct gfs2_ail *ai; | |
435 | unsigned int tail; | |
436 | ||
437 | gfs2_log_lock(sdp); | |
438 | ||
faa31ce8 | 439 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 440 | tail = sdp->sd_log_head; |
faa31ce8 SW |
441 | } else { |
442 | ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list); | |
b3b94faa DT |
443 | tail = ai->ai_first; |
444 | } | |
445 | ||
446 | gfs2_log_unlock(sdp); | |
447 | ||
448 | return tail; | |
449 | } | |
450 | ||
16615be1 | 451 | void gfs2_log_incr_head(struct gfs2_sbd *sdp) |
b3b94faa DT |
452 | { |
453 | if (sdp->sd_log_flush_head == sdp->sd_log_tail) | |
16615be1 | 454 | BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head); |
b3b94faa DT |
455 | |
456 | if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) { | |
457 | sdp->sd_log_flush_head = 0; | |
458 | sdp->sd_log_flush_wrapped = 1; | |
459 | } | |
460 | } | |
461 | ||
16615be1 SW |
462 | /** |
463 | * gfs2_log_write_endio - End of I/O for a log buffer | |
464 | * @bh: The buffer head | |
465 | * @uptodate: I/O Status | |
466 | * | |
467 | */ | |
468 | ||
469 | static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate) | |
470 | { | |
471 | struct gfs2_sbd *sdp = bh->b_private; | |
472 | bh->b_private = NULL; | |
473 | ||
474 | end_buffer_write_sync(bh, uptodate); | |
475 | if (atomic_dec_and_test(&sdp->sd_log_in_flight)) | |
476 | wake_up(&sdp->sd_log_flush_wait); | |
477 | } | |
478 | ||
b3b94faa DT |
479 | /** |
480 | * gfs2_log_get_buf - Get and initialize a buffer to use for log control data | |
481 | * @sdp: The GFS2 superblock | |
482 | * | |
483 | * Returns: the buffer_head | |
484 | */ | |
485 | ||
486 | struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp) | |
487 | { | |
cd915493 | 488 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
489 | struct buffer_head *bh; |
490 | ||
16615be1 | 491 | bh = sb_getblk(sdp->sd_vfs, blkno); |
b3b94faa DT |
492 | lock_buffer(bh); |
493 | memset(bh->b_data, 0, bh->b_size); | |
494 | set_buffer_uptodate(bh); | |
495 | clear_buffer_dirty(bh); | |
16615be1 SW |
496 | gfs2_log_incr_head(sdp); |
497 | atomic_inc(&sdp->sd_log_in_flight); | |
498 | bh->b_private = sdp; | |
499 | bh->b_end_io = gfs2_log_write_endio; | |
b3b94faa DT |
500 | |
501 | return bh; | |
502 | } | |
503 | ||
16615be1 SW |
504 | /** |
505 | * gfs2_fake_write_endio - | |
506 | * @bh: The buffer head | |
507 | * @uptodate: The I/O Status | |
508 | * | |
509 | */ | |
510 | ||
511 | static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate) | |
512 | { | |
513 | struct buffer_head *real_bh = bh->b_private; | |
5a60c532 SW |
514 | struct gfs2_bufdata *bd = real_bh->b_private; |
515 | struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd; | |
16615be1 SW |
516 | |
517 | end_buffer_write_sync(bh, uptodate); | |
518 | free_buffer_head(bh); | |
519 | unlock_buffer(real_bh); | |
520 | brelse(real_bh); | |
521 | if (atomic_dec_and_test(&sdp->sd_log_in_flight)) | |
522 | wake_up(&sdp->sd_log_flush_wait); | |
523 | } | |
524 | ||
b3b94faa DT |
525 | /** |
526 | * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log | |
527 | * @sdp: the filesystem | |
528 | * @data: the data the buffer_head should point to | |
529 | * | |
530 | * Returns: the log buffer descriptor | |
531 | */ | |
532 | ||
533 | struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp, | |
534 | struct buffer_head *real) | |
535 | { | |
cd915493 | 536 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
537 | struct buffer_head *bh; |
538 | ||
16615be1 | 539 | bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL); |
b3b94faa | 540 | atomic_set(&bh->b_count, 1); |
16615be1 | 541 | bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock); |
18ec7d5c | 542 | set_bh_page(bh, real->b_page, bh_offset(real)); |
b3b94faa DT |
543 | bh->b_blocknr = blkno; |
544 | bh->b_size = sdp->sd_sb.sb_bsize; | |
545 | bh->b_bdev = sdp->sd_vfs->s_bdev; | |
16615be1 SW |
546 | bh->b_private = real; |
547 | bh->b_end_io = gfs2_fake_write_endio; | |
b3b94faa | 548 | |
16615be1 SW |
549 | gfs2_log_incr_head(sdp); |
550 | atomic_inc(&sdp->sd_log_in_flight); | |
b3b94faa DT |
551 | |
552 | return bh; | |
553 | } | |
554 | ||
2332c443 | 555 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
556 | { |
557 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
558 | ||
559 | ail2_empty(sdp, new_tail); | |
560 | ||
561 | gfs2_log_lock(sdp); | |
2332c443 | 562 | sdp->sd_log_blks_free += dist; |
faa31ce8 | 563 | gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
564 | gfs2_log_unlock(sdp); |
565 | ||
566 | sdp->sd_log_tail = new_tail; | |
567 | } | |
568 | ||
569 | /** | |
570 | * log_write_header - Get and initialize a journal header buffer | |
571 | * @sdp: The GFS2 superblock | |
572 | * | |
573 | * Returns: the initialized log buffer descriptor | |
574 | */ | |
575 | ||
cd915493 | 576 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull) |
b3b94faa | 577 | { |
cd915493 | 578 | u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head); |
b3b94faa DT |
579 | struct buffer_head *bh; |
580 | struct gfs2_log_header *lh; | |
581 | unsigned int tail; | |
cd915493 | 582 | u32 hash; |
b3b94faa | 583 | |
b3b94faa DT |
584 | bh = sb_getblk(sdp->sd_vfs, blkno); |
585 | lock_buffer(bh); | |
586 | memset(bh->b_data, 0, bh->b_size); | |
587 | set_buffer_uptodate(bh); | |
588 | clear_buffer_dirty(bh); | |
589 | unlock_buffer(bh); | |
590 | ||
591 | gfs2_ail1_empty(sdp, 0); | |
592 | tail = current_tail(sdp); | |
593 | ||
594 | lh = (struct gfs2_log_header *)bh->b_data; | |
595 | memset(lh, 0, sizeof(struct gfs2_log_header)); | |
596 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
e3167ded SW |
597 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); |
598 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
e0f2bf78 SW |
599 | lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++); |
600 | lh->lh_flags = cpu_to_be32(flags); | |
601 | lh->lh_tail = cpu_to_be32(tail); | |
602 | lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head); | |
b3b94faa DT |
603 | hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header)); |
604 | lh->lh_hash = cpu_to_be32(hash); | |
605 | ||
606 | set_buffer_dirty(bh); | |
607 | if (sync_dirty_buffer(bh)) | |
608 | gfs2_io_error_bh(sdp, bh); | |
609 | brelse(bh); | |
610 | ||
611 | if (sdp->sd_log_tail != tail) | |
2332c443 | 612 | log_pull_tail(sdp, tail); |
b3b94faa DT |
613 | else |
614 | gfs2_assert_withdraw(sdp, !pull); | |
615 | ||
616 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); | |
16615be1 | 617 | gfs2_log_incr_head(sdp); |
b3b94faa DT |
618 | } |
619 | ||
620 | static void log_flush_commit(struct gfs2_sbd *sdp) | |
621 | { | |
16615be1 SW |
622 | DEFINE_WAIT(wait); |
623 | ||
624 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
625 | do { | |
626 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
627 | TASK_UNINTERRUPTIBLE); | |
628 | if (atomic_read(&sdp->sd_log_in_flight)) | |
629 | io_schedule(); | |
630 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
631 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa DT |
632 | } |
633 | ||
16615be1 | 634 | log_write_header(sdp, 0, 0); |
b3b94faa DT |
635 | } |
636 | ||
d7b616e2 SW |
637 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
638 | { | |
639 | struct gfs2_bufdata *bd; | |
640 | struct buffer_head *bh; | |
641 | LIST_HEAD(written); | |
642 | ||
643 | gfs2_log_lock(sdp); | |
644 | while (!list_empty(&sdp->sd_log_le_ordered)) { | |
645 | bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list); | |
646 | list_move(&bd->bd_le.le_list, &written); | |
647 | bh = bd->bd_bh; | |
648 | if (!buffer_dirty(bh)) | |
649 | continue; | |
650 | get_bh(bh); | |
651 | gfs2_log_unlock(sdp); | |
652 | lock_buffer(bh); | |
653 | if (test_clear_buffer_dirty(bh)) { | |
654 | bh->b_end_io = end_buffer_write_sync; | |
655 | submit_bh(WRITE, bh); | |
656 | } else { | |
657 | unlock_buffer(bh); | |
658 | brelse(bh); | |
659 | } | |
660 | gfs2_log_lock(sdp); | |
661 | } | |
662 | list_splice(&written, &sdp->sd_log_le_ordered); | |
663 | gfs2_log_unlock(sdp); | |
664 | } | |
665 | ||
666 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
667 | { | |
668 | struct gfs2_bufdata *bd; | |
669 | struct buffer_head *bh; | |
670 | ||
671 | gfs2_log_lock(sdp); | |
672 | while (!list_empty(&sdp->sd_log_le_ordered)) { | |
673 | bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list); | |
674 | bh = bd->bd_bh; | |
675 | if (buffer_locked(bh)) { | |
676 | get_bh(bh); | |
677 | gfs2_log_unlock(sdp); | |
678 | wait_on_buffer(bh); | |
679 | brelse(bh); | |
680 | gfs2_log_lock(sdp); | |
681 | continue; | |
682 | } | |
683 | list_del_init(&bd->bd_le.le_list); | |
684 | } | |
685 | gfs2_log_unlock(sdp); | |
686 | } | |
687 | ||
b3b94faa | 688 | /** |
b09e593d | 689 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
690 | * @sdp: the filesystem |
691 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
692 | * | |
693 | */ | |
694 | ||
b09e593d | 695 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl) |
b3b94faa DT |
696 | { |
697 | struct gfs2_ail *ai; | |
698 | ||
484adff8 | 699 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a SW |
700 | |
701 | if (gl) { | |
702 | gfs2_log_lock(sdp); | |
703 | if (list_empty(&gl->gl_le.le_list)) { | |
704 | gfs2_log_unlock(sdp); | |
484adff8 | 705 | up_write(&sdp->sd_log_flush_lock); |
f55ab26a SW |
706 | return; |
707 | } | |
708 | gfs2_log_unlock(sdp); | |
709 | } | |
710 | ||
b09e593d SW |
711 | ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL); |
712 | INIT_LIST_HEAD(&ai->ai_ail1_list); | |
713 | INIT_LIST_HEAD(&ai->ai_ail2_list); | |
b3b94faa | 714 | |
16615be1 SW |
715 | if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) { |
716 | printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf, | |
717 | sdp->sd_log_commited_buf); | |
718 | gfs2_assert_withdraw(sdp, 0); | |
719 | } | |
720 | if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) { | |
721 | printk(KERN_INFO "GFS2: log databuf %u %u\n", | |
722 | sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf); | |
723 | gfs2_assert_withdraw(sdp, 0); | |
724 | } | |
b3b94faa DT |
725 | gfs2_assert_withdraw(sdp, |
726 | sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke); | |
727 | ||
b3b94faa DT |
728 | sdp->sd_log_flush_head = sdp->sd_log_head; |
729 | sdp->sd_log_flush_wrapped = 0; | |
730 | ai->ai_first = sdp->sd_log_flush_head; | |
731 | ||
d7b616e2 | 732 | gfs2_ordered_write(sdp); |
b3b94faa | 733 | lops_before_commit(sdp); |
d7b616e2 SW |
734 | gfs2_ordered_wait(sdp); |
735 | ||
16615be1 | 736 | if (sdp->sd_log_head != sdp->sd_log_flush_head) |
b3b94faa | 737 | log_flush_commit(sdp); |
2332c443 RP |
738 | else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
739 | gfs2_log_lock(sdp); | |
740 | sdp->sd_log_blks_free--; /* Adjust for unreserved buffer */ | |
741 | gfs2_log_unlock(sdp); | |
b3b94faa | 742 | log_write_header(sdp, 0, PULL); |
2332c443 | 743 | } |
b3b94faa | 744 | lops_after_commit(sdp, ai); |
b09e593d | 745 | |
fe1a698f SW |
746 | gfs2_log_lock(sdp); |
747 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 SW |
748 | sdp->sd_log_blks_reserved = 0; |
749 | sdp->sd_log_commited_buf = 0; | |
2332c443 | 750 | sdp->sd_log_commited_databuf = 0; |
faa31ce8 | 751 | sdp->sd_log_commited_revoke = 0; |
b3b94faa | 752 | |
b3b94faa DT |
753 | if (!list_empty(&ai->ai_ail1_list)) { |
754 | list_add(&ai->ai_list, &sdp->sd_ail1_list); | |
755 | ai = NULL; | |
756 | } | |
757 | gfs2_log_unlock(sdp); | |
758 | ||
b3b94faa | 759 | sdp->sd_vfs->s_dirt = 0; |
484adff8 | 760 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
761 | |
762 | kfree(ai); | |
763 | } | |
764 | ||
765 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
766 | { | |
2332c443 | 767 | unsigned int reserved; |
b3b94faa DT |
768 | unsigned int old; |
769 | ||
770 | gfs2_log_lock(sdp); | |
771 | ||
772 | sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
2332c443 RP |
773 | sdp->sd_log_commited_databuf += tr->tr_num_databuf_new - |
774 | tr->tr_num_databuf_rm; | |
775 | gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) || | |
776 | (((int)sdp->sd_log_commited_databuf) >= 0)); | |
b3b94faa DT |
777 | sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
778 | gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0); | |
2332c443 | 779 | reserved = calc_reserved(sdp); |
b3b94faa DT |
780 | old = sdp->sd_log_blks_free; |
781 | sdp->sd_log_blks_free += tr->tr_reserved - | |
782 | (reserved - sdp->sd_log_blks_reserved); | |
783 | ||
b09e593d | 784 | gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old); |
2332c443 RP |
785 | gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= |
786 | sdp->sd_jdesc->jd_blocks); | |
b3b94faa DT |
787 | |
788 | sdp->sd_log_blks_reserved = reserved; | |
789 | ||
790 | gfs2_log_unlock(sdp); | |
791 | } | |
792 | ||
793 | /** | |
794 | * gfs2_log_commit - Commit a transaction to the log | |
795 | * @sdp: the filesystem | |
796 | * @tr: the transaction | |
797 | * | |
798 | * Returns: errno | |
799 | */ | |
800 | ||
801 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
802 | { | |
803 | log_refund(sdp, tr); | |
804 | lops_incore_commit(sdp, tr); | |
805 | ||
806 | sdp->sd_vfs->s_dirt = 1; | |
484adff8 | 807 | up_read(&sdp->sd_log_flush_lock); |
b3b94faa | 808 | |
b3b94faa | 809 | gfs2_log_lock(sdp); |
b004157a SW |
810 | if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks)) |
811 | wake_up_process(sdp->sd_logd_process); | |
812 | gfs2_log_unlock(sdp); | |
b3b94faa DT |
813 | } |
814 | ||
815 | /** | |
816 | * gfs2_log_shutdown - write a shutdown header into a journal | |
817 | * @sdp: the filesystem | |
818 | * | |
819 | */ | |
820 | ||
821 | void gfs2_log_shutdown(struct gfs2_sbd *sdp) | |
822 | { | |
484adff8 | 823 | down_write(&sdp->sd_log_flush_lock); |
b3b94faa | 824 | |
b3b94faa DT |
825 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
826 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl); | |
827 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf); | |
828 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
829 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg); | |
830 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf); | |
831 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); | |
832 | ||
833 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
834 | sdp->sd_log_flush_wrapped = 0; | |
835 | ||
2332c443 RP |
836 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT, |
837 | (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL); | |
b3b94faa | 838 | |
a74604be SW |
839 | gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks); |
840 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); | |
841 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
842 | |
843 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa DT |
844 | sdp->sd_log_tail = sdp->sd_log_head; |
845 | ||
484adff8 | 846 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
847 | } |
848 | ||
a25311c8 SW |
849 | |
850 | /** | |
851 | * gfs2_meta_syncfs - sync all the buffers in a filesystem | |
852 | * @sdp: the filesystem | |
853 | * | |
854 | */ | |
855 | ||
856 | void gfs2_meta_syncfs(struct gfs2_sbd *sdp) | |
857 | { | |
858 | gfs2_log_flush(sdp, NULL); | |
859 | for (;;) { | |
860 | gfs2_ail1_start(sdp, DIO_ALL); | |
861 | if (gfs2_ail1_empty(sdp, DIO_ALL)) | |
862 | break; | |
863 | msleep(10); | |
864 | } | |
865 | } | |
866 |