]>
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> |
4667a0ec | 21 | #include <linux/writeback.h> |
4a36d08d | 22 | #include <linux/list_sort.h> |
b3b94faa DT |
23 | |
24 | #include "gfs2.h" | |
5c676f6d | 25 | #include "incore.h" |
b3b94faa DT |
26 | #include "bmap.h" |
27 | #include "glock.h" | |
28 | #include "log.h" | |
29 | #include "lops.h" | |
30 | #include "meta_io.h" | |
5c676f6d | 31 | #include "util.h" |
71b86f56 | 32 | #include "dir.h" |
63997775 | 33 | #include "trace_gfs2.h" |
b3b94faa | 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 | * | |
c618e87a | 70 | * The ail lock _must_ be held when calling this function |
1e1a3d03 SW |
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 | |
4667a0ec SW |
86 | * @wbc: The writeback control structure |
87 | * @ai: The ail structure | |
ddacfaf7 SW |
88 | * |
89 | */ | |
90 | ||
4f1de018 SW |
91 | static int gfs2_ail1_start_one(struct gfs2_sbd *sdp, |
92 | struct writeback_control *wbc, | |
93 | struct gfs2_ail *ai) | |
d6a079e8 DC |
94 | __releases(&sdp->sd_ail_lock) |
95 | __acquires(&sdp->sd_ail_lock) | |
ddacfaf7 | 96 | { |
5ac048bb | 97 | struct gfs2_glock *gl = NULL; |
4667a0ec | 98 | struct address_space *mapping; |
ddacfaf7 SW |
99 | struct gfs2_bufdata *bd, *s; |
100 | struct buffer_head *bh; | |
ddacfaf7 | 101 | |
4667a0ec SW |
102 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, bd_ail_st_list) { |
103 | bh = bd->bd_bh; | |
ddacfaf7 | 104 | |
4667a0ec | 105 | gfs2_assert(sdp, bd->bd_ail == ai); |
ddacfaf7 | 106 | |
4667a0ec SW |
107 | if (!buffer_busy(bh)) { |
108 | if (!buffer_uptodate(bh)) | |
109 | gfs2_io_error_bh(sdp, bh); | |
110 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); | |
111 | continue; | |
112 | } | |
113 | ||
114 | if (!buffer_dirty(bh)) | |
115 | continue; | |
116 | if (gl == bd->bd_gl) | |
117 | continue; | |
118 | gl = bd->bd_gl; | |
119 | list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list); | |
120 | mapping = bh->b_page->mapping; | |
4f1de018 SW |
121 | if (!mapping) |
122 | continue; | |
4667a0ec SW |
123 | spin_unlock(&sdp->sd_ail_lock); |
124 | generic_writepages(mapping, wbc); | |
125 | spin_lock(&sdp->sd_ail_lock); | |
126 | if (wbc->nr_to_write <= 0) | |
127 | break; | |
4f1de018 | 128 | return 1; |
4667a0ec | 129 | } |
4f1de018 SW |
130 | |
131 | return 0; | |
4667a0ec | 132 | } |
ddacfaf7 | 133 | |
ddacfaf7 | 134 | |
4667a0ec SW |
135 | /** |
136 | * gfs2_ail1_flush - start writeback of some ail1 entries | |
137 | * @sdp: The super block | |
138 | * @wbc: The writeback control structure | |
139 | * | |
140 | * Writes back some ail1 entries, according to the limits in the | |
141 | * writeback control structure | |
142 | */ | |
ddacfaf7 | 143 | |
4667a0ec SW |
144 | void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) |
145 | { | |
146 | struct list_head *head = &sdp->sd_ail1_list; | |
147 | struct gfs2_ail *ai; | |
ddacfaf7 | 148 | |
c83ae9ca | 149 | trace_gfs2_ail_flush(sdp, wbc, 1); |
4667a0ec | 150 | spin_lock(&sdp->sd_ail_lock); |
4f1de018 | 151 | restart: |
4667a0ec SW |
152 | list_for_each_entry_reverse(ai, head, ai_list) { |
153 | if (wbc->nr_to_write <= 0) | |
ddacfaf7 | 154 | break; |
4f1de018 SW |
155 | if (gfs2_ail1_start_one(sdp, wbc, ai)) |
156 | goto restart; | |
4667a0ec SW |
157 | } |
158 | spin_unlock(&sdp->sd_ail_lock); | |
c83ae9ca | 159 | trace_gfs2_ail_flush(sdp, wbc, 0); |
4667a0ec SW |
160 | } |
161 | ||
162 | /** | |
163 | * gfs2_ail1_start - start writeback of all ail1 entries | |
164 | * @sdp: The superblock | |
165 | */ | |
166 | ||
167 | static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
168 | { | |
169 | struct writeback_control wbc = { | |
170 | .sync_mode = WB_SYNC_NONE, | |
171 | .nr_to_write = LONG_MAX, | |
172 | .range_start = 0, | |
173 | .range_end = LLONG_MAX, | |
174 | }; | |
175 | ||
176 | return gfs2_ail1_flush(sdp, &wbc); | |
ddacfaf7 SW |
177 | } |
178 | ||
179 | /** | |
180 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
181 | * @sdp: the filesystem | |
182 | * @ai: the AIL entry | |
183 | * | |
184 | */ | |
185 | ||
4667a0ec | 186 | static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) |
ddacfaf7 SW |
187 | { |
188 | struct gfs2_bufdata *bd, *s; | |
189 | struct buffer_head *bh; | |
190 | ||
191 | list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list, | |
192 | bd_ail_st_list) { | |
193 | bh = bd->bd_bh; | |
ddacfaf7 | 194 | gfs2_assert(sdp, bd->bd_ail == ai); |
4667a0ec SW |
195 | if (buffer_busy(bh)) |
196 | continue; | |
ddacfaf7 SW |
197 | if (!buffer_uptodate(bh)) |
198 | gfs2_io_error_bh(sdp, bh); | |
ddacfaf7 SW |
199 | list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list); |
200 | } | |
201 | ||
ddacfaf7 SW |
202 | } |
203 | ||
4667a0ec SW |
204 | /** |
205 | * gfs2_ail1_empty - Try to empty the ail1 lists | |
206 | * @sdp: The superblock | |
207 | * | |
208 | * Tries to empty the ail1 lists, starting with the oldest first | |
209 | */ | |
b3b94faa | 210 | |
4667a0ec | 211 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp) |
b3b94faa DT |
212 | { |
213 | struct gfs2_ail *ai, *s; | |
214 | int ret; | |
215 | ||
d6a079e8 | 216 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 217 | list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) { |
4667a0ec SW |
218 | gfs2_ail1_empty_one(sdp, ai); |
219 | if (list_empty(&ai->ai_ail1_list)) | |
b3b94faa | 220 | list_move(&ai->ai_list, &sdp->sd_ail2_list); |
4667a0ec | 221 | else |
b3b94faa DT |
222 | break; |
223 | } | |
b3b94faa | 224 | ret = list_empty(&sdp->sd_ail1_list); |
d6a079e8 | 225 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
226 | |
227 | return ret; | |
228 | } | |
229 | ||
26b06a69 SW |
230 | static void gfs2_ail1_wait(struct gfs2_sbd *sdp) |
231 | { | |
232 | struct gfs2_ail *ai; | |
233 | struct gfs2_bufdata *bd; | |
234 | struct buffer_head *bh; | |
235 | ||
236 | spin_lock(&sdp->sd_ail_lock); | |
237 | list_for_each_entry_reverse(ai, &sdp->sd_ail1_list, ai_list) { | |
238 | list_for_each_entry(bd, &ai->ai_ail1_list, bd_ail_st_list) { | |
239 | bh = bd->bd_bh; | |
240 | if (!buffer_locked(bh)) | |
241 | continue; | |
242 | get_bh(bh); | |
243 | spin_unlock(&sdp->sd_ail_lock); | |
244 | wait_on_buffer(bh); | |
245 | brelse(bh); | |
246 | return; | |
247 | } | |
248 | } | |
249 | spin_unlock(&sdp->sd_ail_lock); | |
250 | } | |
ddacfaf7 SW |
251 | |
252 | /** | |
253 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
254 | * @sdp: the filesystem | |
255 | * @ai: the AIL entry | |
256 | * | |
257 | */ | |
258 | ||
259 | static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai) | |
260 | { | |
261 | struct list_head *head = &ai->ai_ail2_list; | |
262 | struct gfs2_bufdata *bd; | |
263 | ||
264 | while (!list_empty(head)) { | |
265 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
266 | bd_ail_st_list); | |
267 | gfs2_assert(sdp, bd->bd_ail == ai); | |
f91a0d3e | 268 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
269 | } |
270 | } | |
271 | ||
b3b94faa DT |
272 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
273 | { | |
274 | struct gfs2_ail *ai, *safe; | |
275 | unsigned int old_tail = sdp->sd_log_tail; | |
276 | int wrap = (new_tail < old_tail); | |
277 | int a, b, rm; | |
278 | ||
d6a079e8 | 279 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa DT |
280 | |
281 | list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) { | |
282 | a = (old_tail <= ai->ai_first); | |
283 | b = (ai->ai_first < new_tail); | |
284 | rm = (wrap) ? (a || b) : (a && b); | |
285 | if (!rm) | |
286 | continue; | |
287 | ||
288 | gfs2_ail2_empty_one(sdp, ai); | |
289 | list_del(&ai->ai_list); | |
290 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list)); | |
291 | gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list)); | |
292 | kfree(ai); | |
293 | } | |
294 | ||
d6a079e8 | 295 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
296 | } |
297 | ||
298 | /** | |
299 | * gfs2_log_reserve - Make a log reservation | |
300 | * @sdp: The GFS2 superblock | |
301 | * @blks: The number of blocks to reserve | |
302 | * | |
89918647 | 303 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 304 | * due to the fact that there is a small number of header blocks |
b004157a SW |
305 | * associated with each log flush. The exact number can't be known until |
306 | * flush time, so we ensure that we have just enough free blocks at all | |
307 | * times to avoid running out during a log flush. | |
308 | * | |
5e687eac BM |
309 | * We no longer flush the log here, instead we wake up logd to do that |
310 | * for us. To avoid the thundering herd and to ensure that we deal fairly | |
311 | * with queued waiters, we use an exclusive wait. This means that when we | |
312 | * get woken with enough journal space to get our reservation, we need to | |
313 | * wake the next waiter on the list. | |
314 | * | |
b3b94faa DT |
315 | * Returns: errno |
316 | */ | |
317 | ||
318 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
319 | { | |
89918647 | 320 | unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize); |
5e687eac BM |
321 | unsigned wanted = blks + reserved_blks; |
322 | DEFINE_WAIT(wait); | |
323 | int did_wait = 0; | |
324 | unsigned int free_blocks; | |
b3b94faa DT |
325 | |
326 | if (gfs2_assert_warn(sdp, blks) || | |
327 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
328 | return -EINVAL; | |
5e687eac BM |
329 | retry: |
330 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
331 | if (unlikely(free_blocks <= wanted)) { | |
332 | do { | |
333 | prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait, | |
334 | TASK_UNINTERRUPTIBLE); | |
335 | wake_up(&sdp->sd_logd_waitq); | |
336 | did_wait = 1; | |
337 | if (atomic_read(&sdp->sd_log_blks_free) <= wanted) | |
338 | io_schedule(); | |
339 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
340 | } while(free_blocks <= wanted); | |
341 | finish_wait(&sdp->sd_log_waitq, &wait); | |
b3b94faa | 342 | } |
5e687eac BM |
343 | if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks, |
344 | free_blocks - blks) != free_blocks) | |
345 | goto retry; | |
63997775 | 346 | trace_gfs2_log_blocks(sdp, -blks); |
5e687eac BM |
347 | |
348 | /* | |
349 | * If we waited, then so might others, wake them up _after_ we get | |
350 | * our share of the log. | |
351 | */ | |
352 | if (unlikely(did_wait)) | |
353 | wake_up(&sdp->sd_log_waitq); | |
484adff8 SW |
354 | |
355 | down_read(&sdp->sd_log_flush_lock); | |
b3b94faa DT |
356 | |
357 | return 0; | |
358 | } | |
359 | ||
47ac5537 | 360 | u64 gfs2_log_bmap(struct gfs2_sbd *sdp, unsigned int lbn) |
b3b94faa | 361 | { |
da6dd40d BP |
362 | struct gfs2_journal_extent *je; |
363 | ||
364 | list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) { | |
365 | if (lbn >= je->lblock && lbn < je->lblock + je->blocks) | |
ff91cc9b | 366 | return je->dblock + lbn - je->lblock; |
da6dd40d BP |
367 | } |
368 | ||
369 | return -1; | |
b3b94faa DT |
370 | } |
371 | ||
372 | /** | |
373 | * log_distance - Compute distance between two journal blocks | |
374 | * @sdp: The GFS2 superblock | |
375 | * @newer: The most recent journal block of the pair | |
376 | * @older: The older journal block of the pair | |
377 | * | |
378 | * Compute the distance (in the journal direction) between two | |
379 | * blocks in the journal | |
380 | * | |
381 | * Returns: the distance in blocks | |
382 | */ | |
383 | ||
faa31ce8 | 384 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
385 | unsigned int older) |
386 | { | |
387 | int dist; | |
388 | ||
389 | dist = newer - older; | |
390 | if (dist < 0) | |
391 | dist += sdp->sd_jdesc->jd_blocks; | |
392 | ||
393 | return dist; | |
394 | } | |
395 | ||
2332c443 RP |
396 | /** |
397 | * calc_reserved - Calculate the number of blocks to reserve when | |
398 | * refunding a transaction's unused buffers. | |
399 | * @sdp: The GFS2 superblock | |
400 | * | |
401 | * This is complex. We need to reserve room for all our currently used | |
402 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
403 | * all our journaled data buffers for journaled files (e.g. files in the | |
404 | * meta_fs like rindex, or files for which chattr +j was done.) | |
405 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
406 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
407 | * | |
408 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
409 | * type gets its own log header, for which we need to reserve a block. | |
410 | * In fact, each type has the potential for needing more than one header | |
411 | * in cases where we have more buffers than will fit on a journal page. | |
412 | * Metadata journal entries take up half the space of journaled buffer entries. | |
413 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
414 | * databuf_limit (251) before they cause a wrap around. | |
415 | * | |
416 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
417 | * overall header for the lot. | |
418 | * | |
419 | * Returns: the number of blocks reserved | |
420 | */ | |
421 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
422 | { | |
423 | unsigned int reserved = 0; | |
424 | unsigned int mbuf_limit, metabufhdrs_needed; | |
425 | unsigned int dbuf_limit, databufhdrs_needed; | |
426 | unsigned int revokes = 0; | |
427 | ||
428 | mbuf_limit = buf_limit(sdp); | |
429 | metabufhdrs_needed = (sdp->sd_log_commited_buf + | |
430 | (mbuf_limit - 1)) / mbuf_limit; | |
431 | dbuf_limit = databuf_limit(sdp); | |
432 | databufhdrs_needed = (sdp->sd_log_commited_databuf + | |
433 | (dbuf_limit - 1)) / dbuf_limit; | |
434 | ||
2e95e3f6 | 435 | if (sdp->sd_log_commited_revoke > 0) |
2332c443 RP |
436 | revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke, |
437 | sizeof(u64)); | |
438 | ||
439 | reserved = sdp->sd_log_commited_buf + metabufhdrs_needed + | |
440 | sdp->sd_log_commited_databuf + databufhdrs_needed + | |
441 | revokes; | |
442 | /* One for the overall header */ | |
443 | if (reserved) | |
444 | reserved++; | |
445 | return reserved; | |
446 | } | |
447 | ||
b3b94faa DT |
448 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
449 | { | |
450 | struct gfs2_ail *ai; | |
451 | unsigned int tail; | |
452 | ||
d6a079e8 | 453 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 454 | |
faa31ce8 | 455 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 456 | tail = sdp->sd_log_head; |
faa31ce8 SW |
457 | } else { |
458 | ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list); | |
b3b94faa DT |
459 | tail = ai->ai_first; |
460 | } | |
461 | ||
d6a079e8 | 462 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
463 | |
464 | return tail; | |
465 | } | |
466 | ||
16615be1 | 467 | void gfs2_log_incr_head(struct gfs2_sbd *sdp) |
b3b94faa | 468 | { |
47ac5537 SW |
469 | BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) && |
470 | (sdp->sd_log_flush_head != sdp->sd_log_head)); | |
b3b94faa DT |
471 | |
472 | if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) { | |
473 | sdp->sd_log_flush_head = 0; | |
474 | sdp->sd_log_flush_wrapped = 1; | |
475 | } | |
476 | } | |
477 | ||
2332c443 | 478 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
479 | { |
480 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
481 | ||
482 | ail2_empty(sdp, new_tail); | |
483 | ||
fd041f0b | 484 | atomic_add(dist, &sdp->sd_log_blks_free); |
63997775 | 485 | trace_gfs2_log_blocks(sdp, dist); |
5e687eac BM |
486 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
487 | sdp->sd_jdesc->jd_blocks); | |
b3b94faa DT |
488 | |
489 | sdp->sd_log_tail = new_tail; | |
490 | } | |
491 | ||
b3b94faa | 492 | |
34cc1781 | 493 | static void log_flush_wait(struct gfs2_sbd *sdp) |
b3b94faa | 494 | { |
16615be1 SW |
495 | DEFINE_WAIT(wait); |
496 | ||
497 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
498 | do { | |
499 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
500 | TASK_UNINTERRUPTIBLE); | |
501 | if (atomic_read(&sdp->sd_log_in_flight)) | |
502 | io_schedule(); | |
503 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
504 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa | 505 | } |
b3b94faa DT |
506 | } |
507 | ||
08728f2d | 508 | static int bd_cmp(void *priv, struct list_head *a, struct list_head *b) |
4a36d08d BP |
509 | { |
510 | struct gfs2_bufdata *bda, *bdb; | |
511 | ||
512 | bda = list_entry(a, struct gfs2_bufdata, bd_le.le_list); | |
513 | bdb = list_entry(b, struct gfs2_bufdata, bd_le.le_list); | |
514 | ||
515 | if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr) | |
516 | return -1; | |
517 | if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr) | |
518 | return 1; | |
519 | return 0; | |
520 | } | |
521 | ||
d7b616e2 SW |
522 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
523 | { | |
524 | struct gfs2_bufdata *bd; | |
525 | struct buffer_head *bh; | |
526 | LIST_HEAD(written); | |
527 | ||
528 | gfs2_log_lock(sdp); | |
4a36d08d | 529 | list_sort(NULL, &sdp->sd_log_le_ordered, &bd_cmp); |
d7b616e2 SW |
530 | while (!list_empty(&sdp->sd_log_le_ordered)) { |
531 | bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list); | |
532 | list_move(&bd->bd_le.le_list, &written); | |
533 | bh = bd->bd_bh; | |
534 | if (!buffer_dirty(bh)) | |
535 | continue; | |
536 | get_bh(bh); | |
537 | gfs2_log_unlock(sdp); | |
538 | lock_buffer(bh); | |
b8e7cbb6 | 539 | if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) { |
d7b616e2 | 540 | bh->b_end_io = end_buffer_write_sync; |
721a9602 | 541 | submit_bh(WRITE_SYNC, bh); |
d7b616e2 SW |
542 | } else { |
543 | unlock_buffer(bh); | |
544 | brelse(bh); | |
545 | } | |
546 | gfs2_log_lock(sdp); | |
547 | } | |
548 | list_splice(&written, &sdp->sd_log_le_ordered); | |
549 | gfs2_log_unlock(sdp); | |
550 | } | |
551 | ||
552 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
553 | { | |
554 | struct gfs2_bufdata *bd; | |
555 | struct buffer_head *bh; | |
556 | ||
557 | gfs2_log_lock(sdp); | |
558 | while (!list_empty(&sdp->sd_log_le_ordered)) { | |
559 | bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list); | |
560 | bh = bd->bd_bh; | |
561 | if (buffer_locked(bh)) { | |
562 | get_bh(bh); | |
563 | gfs2_log_unlock(sdp); | |
564 | wait_on_buffer(bh); | |
565 | brelse(bh); | |
566 | gfs2_log_lock(sdp); | |
567 | continue; | |
568 | } | |
569 | list_del_init(&bd->bd_le.le_list); | |
570 | } | |
571 | gfs2_log_unlock(sdp); | |
572 | } | |
573 | ||
34cc1781 SW |
574 | /** |
575 | * log_write_header - Get and initialize a journal header buffer | |
576 | * @sdp: The GFS2 superblock | |
577 | * | |
578 | * Returns: the initialized log buffer descriptor | |
579 | */ | |
580 | ||
fdb76a42 | 581 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags) |
34cc1781 SW |
582 | { |
583 | u64 blkno = gfs2_log_bmap(sdp, sdp->sd_log_flush_head); | |
584 | struct buffer_head *bh; | |
585 | struct gfs2_log_header *lh; | |
586 | unsigned int tail; | |
587 | u32 hash; | |
588 | ||
589 | bh = sb_getblk(sdp->sd_vfs, blkno); | |
590 | lock_buffer(bh); | |
591 | memset(bh->b_data, 0, bh->b_size); | |
592 | set_buffer_uptodate(bh); | |
593 | clear_buffer_dirty(bh); | |
594 | ||
595 | gfs2_ail1_empty(sdp); | |
596 | tail = current_tail(sdp); | |
597 | ||
598 | lh = (struct gfs2_log_header *)bh->b_data; | |
599 | memset(lh, 0, sizeof(struct gfs2_log_header)); | |
600 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); | |
601 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); | |
602 | lh->lh_header.__pad0 = cpu_to_be64(0); | |
603 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
604 | lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); | |
605 | lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++); | |
606 | lh->lh_flags = cpu_to_be32(flags); | |
607 | lh->lh_tail = cpu_to_be32(tail); | |
608 | lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head); | |
609 | hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header)); | |
610 | lh->lh_hash = cpu_to_be32(hash); | |
611 | ||
612 | bh->b_end_io = end_buffer_write_sync; | |
613 | get_bh(bh); | |
614 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { | |
615 | gfs2_ordered_wait(sdp); | |
616 | log_flush_wait(sdp); | |
617 | submit_bh(WRITE_SYNC | REQ_META | REQ_PRIO, bh); | |
618 | } else { | |
619 | submit_bh(WRITE_FLUSH_FUA | REQ_META, bh); | |
620 | } | |
621 | wait_on_buffer(bh); | |
622 | ||
623 | if (!buffer_uptodate(bh)) | |
624 | gfs2_io_error_bh(sdp, bh); | |
625 | brelse(bh); | |
626 | ||
627 | if (sdp->sd_log_tail != tail) | |
628 | log_pull_tail(sdp, tail); | |
34cc1781 SW |
629 | |
630 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); | |
631 | gfs2_log_incr_head(sdp); | |
632 | } | |
633 | ||
b3b94faa | 634 | /** |
b09e593d | 635 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
636 | * @sdp: the filesystem |
637 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
638 | * | |
639 | */ | |
640 | ||
ed4878e8 | 641 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl) |
b3b94faa DT |
642 | { |
643 | struct gfs2_ail *ai; | |
644 | ||
484adff8 | 645 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 646 | |
2bcd610d SW |
647 | /* Log might have been flushed while we waited for the flush lock */ |
648 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) { | |
649 | up_write(&sdp->sd_log_flush_lock); | |
650 | return; | |
f55ab26a | 651 | } |
63997775 | 652 | trace_gfs2_log_flush(sdp, 1); |
f55ab26a | 653 | |
b09e593d SW |
654 | ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL); |
655 | INIT_LIST_HEAD(&ai->ai_ail1_list); | |
656 | INIT_LIST_HEAD(&ai->ai_ail2_list); | |
b3b94faa | 657 | |
16615be1 SW |
658 | if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) { |
659 | printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf, | |
660 | sdp->sd_log_commited_buf); | |
661 | gfs2_assert_withdraw(sdp, 0); | |
662 | } | |
663 | if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) { | |
664 | printk(KERN_INFO "GFS2: log databuf %u %u\n", | |
665 | sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf); | |
666 | gfs2_assert_withdraw(sdp, 0); | |
667 | } | |
b3b94faa DT |
668 | gfs2_assert_withdraw(sdp, |
669 | sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke); | |
670 | ||
b3b94faa DT |
671 | sdp->sd_log_flush_head = sdp->sd_log_head; |
672 | sdp->sd_log_flush_wrapped = 0; | |
673 | ai->ai_first = sdp->sd_log_flush_head; | |
674 | ||
d7b616e2 | 675 | gfs2_ordered_write(sdp); |
b3b94faa | 676 | lops_before_commit(sdp); |
d7b616e2 | 677 | |
34cc1781 | 678 | if (sdp->sd_log_head != sdp->sd_log_flush_head) { |
fdb76a42 | 679 | log_write_header(sdp, 0); |
34cc1781 | 680 | } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
2332c443 | 681 | gfs2_log_lock(sdp); |
fd041f0b | 682 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 683 | trace_gfs2_log_blocks(sdp, -1); |
2332c443 | 684 | gfs2_log_unlock(sdp); |
fdb76a42 | 685 | log_write_header(sdp, 0); |
2332c443 | 686 | } |
b3b94faa | 687 | lops_after_commit(sdp, ai); |
b09e593d | 688 | |
fe1a698f SW |
689 | gfs2_log_lock(sdp); |
690 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 SW |
691 | sdp->sd_log_blks_reserved = 0; |
692 | sdp->sd_log_commited_buf = 0; | |
2332c443 | 693 | sdp->sd_log_commited_databuf = 0; |
faa31ce8 | 694 | sdp->sd_log_commited_revoke = 0; |
b3b94faa | 695 | |
d6a079e8 | 696 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa DT |
697 | if (!list_empty(&ai->ai_ail1_list)) { |
698 | list_add(&ai->ai_list, &sdp->sd_ail1_list); | |
699 | ai = NULL; | |
700 | } | |
d6a079e8 | 701 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 702 | gfs2_log_unlock(sdp); |
63997775 | 703 | trace_gfs2_log_flush(sdp, 0); |
484adff8 | 704 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
705 | |
706 | kfree(ai); | |
707 | } | |
708 | ||
709 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
710 | { | |
2332c443 | 711 | unsigned int reserved; |
ac39aadd | 712 | unsigned int unused; |
b3b94faa DT |
713 | |
714 | gfs2_log_lock(sdp); | |
715 | ||
716 | sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
2332c443 RP |
717 | sdp->sd_log_commited_databuf += tr->tr_num_databuf_new - |
718 | tr->tr_num_databuf_rm; | |
719 | gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) || | |
720 | (((int)sdp->sd_log_commited_databuf) >= 0)); | |
b3b94faa | 721 | sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
2332c443 | 722 | reserved = calc_reserved(sdp); |
62be1f71 | 723 | gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved); |
ac39aadd | 724 | unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved; |
ac39aadd | 725 | atomic_add(unused, &sdp->sd_log_blks_free); |
63997775 | 726 | trace_gfs2_log_blocks(sdp, unused); |
fd041f0b | 727 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
2332c443 | 728 | sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
729 | sdp->sd_log_blks_reserved = reserved; |
730 | ||
731 | gfs2_log_unlock(sdp); | |
732 | } | |
733 | ||
d0109bfa BP |
734 | static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
735 | { | |
736 | struct list_head *head = &tr->tr_list_buf; | |
737 | struct gfs2_bufdata *bd; | |
738 | ||
739 | gfs2_log_lock(sdp); | |
740 | while (!list_empty(head)) { | |
741 | bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr); | |
742 | list_del_init(&bd->bd_list_tr); | |
743 | tr->tr_num_buf--; | |
744 | } | |
745 | gfs2_log_unlock(sdp); | |
746 | gfs2_assert_warn(sdp, !tr->tr_num_buf); | |
747 | } | |
748 | ||
b3b94faa DT |
749 | /** |
750 | * gfs2_log_commit - Commit a transaction to the log | |
751 | * @sdp: the filesystem | |
752 | * @tr: the transaction | |
753 | * | |
5e687eac BM |
754 | * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 |
755 | * or the total number of used blocks (pinned blocks plus AIL blocks) | |
756 | * is greater than thresh2. | |
757 | * | |
758 | * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of | |
759 | * journal size. | |
760 | * | |
b3b94faa DT |
761 | * Returns: errno |
762 | */ | |
763 | ||
764 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
765 | { | |
766 | log_refund(sdp, tr); | |
d0109bfa | 767 | buf_lo_incore_commit(sdp, tr); |
b3b94faa | 768 | |
484adff8 | 769 | up_read(&sdp->sd_log_flush_lock); |
b3b94faa | 770 | |
5e687eac BM |
771 | if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || |
772 | ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > | |
773 | atomic_read(&sdp->sd_log_thresh2))) | |
774 | wake_up(&sdp->sd_logd_waitq); | |
b3b94faa DT |
775 | } |
776 | ||
777 | /** | |
778 | * gfs2_log_shutdown - write a shutdown header into a journal | |
779 | * @sdp: the filesystem | |
780 | * | |
781 | */ | |
782 | ||
783 | void gfs2_log_shutdown(struct gfs2_sbd *sdp) | |
784 | { | |
484adff8 | 785 | down_write(&sdp->sd_log_flush_lock); |
b3b94faa | 786 | |
b3b94faa | 787 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa DT |
788 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf); |
789 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
790 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg); | |
791 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf); | |
792 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); | |
793 | ||
794 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
795 | sdp->sd_log_flush_wrapped = 0; | |
796 | ||
fdb76a42 | 797 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT); |
b3b94faa | 798 | |
fd041f0b | 799 | gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks); |
a74604be SW |
800 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
801 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
802 | |
803 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa DT |
804 | sdp->sd_log_tail = sdp->sd_log_head; |
805 | ||
484adff8 | 806 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa DT |
807 | } |
808 | ||
a25311c8 SW |
809 | |
810 | /** | |
811 | * gfs2_meta_syncfs - sync all the buffers in a filesystem | |
812 | * @sdp: the filesystem | |
813 | * | |
814 | */ | |
815 | ||
816 | void gfs2_meta_syncfs(struct gfs2_sbd *sdp) | |
817 | { | |
818 | gfs2_log_flush(sdp, NULL); | |
819 | for (;;) { | |
5e687eac | 820 | gfs2_ail1_start(sdp); |
26b06a69 | 821 | gfs2_ail1_wait(sdp); |
4667a0ec | 822 | if (gfs2_ail1_empty(sdp)) |
a25311c8 | 823 | break; |
a25311c8 | 824 | } |
380f7c65 | 825 | gfs2_log_flush(sdp, NULL); |
a25311c8 SW |
826 | } |
827 | ||
5e687eac BM |
828 | static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) |
829 | { | |
830 | return (atomic_read(&sdp->sd_log_pinned) >= atomic_read(&sdp->sd_log_thresh1)); | |
831 | } | |
832 | ||
833 | static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) | |
834 | { | |
835 | unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); | |
836 | return used_blocks >= atomic_read(&sdp->sd_log_thresh2); | |
837 | } | |
ec69b188 SW |
838 | |
839 | /** | |
840 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
841 | * @sdp: Pointer to GFS2 superblock | |
842 | * | |
843 | * Also, periodically check to make sure that we're using the most recent | |
844 | * journal index. | |
845 | */ | |
846 | ||
847 | int gfs2_logd(void *data) | |
848 | { | |
849 | struct gfs2_sbd *sdp = data; | |
5e687eac BM |
850 | unsigned long t = 1; |
851 | DEFINE_WAIT(wait); | |
852 | unsigned preflush; | |
ec69b188 SW |
853 | |
854 | while (!kthread_should_stop()) { | |
ec69b188 | 855 | |
5e687eac BM |
856 | preflush = atomic_read(&sdp->sd_log_pinned); |
857 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { | |
4667a0ec | 858 | gfs2_ail1_empty(sdp); |
5e687eac | 859 | gfs2_log_flush(sdp, NULL); |
5e687eac | 860 | } |
ec69b188 | 861 | |
5e687eac BM |
862 | if (gfs2_ail_flush_reqd(sdp)) { |
863 | gfs2_ail1_start(sdp); | |
26b06a69 | 864 | gfs2_ail1_wait(sdp); |
4667a0ec | 865 | gfs2_ail1_empty(sdp); |
ec69b188 | 866 | gfs2_log_flush(sdp, NULL); |
ec69b188 SW |
867 | } |
868 | ||
26b06a69 SW |
869 | if (!gfs2_ail_flush_reqd(sdp)) |
870 | wake_up(&sdp->sd_log_waitq); | |
871 | ||
ec69b188 | 872 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
a0acae0e TH |
873 | |
874 | try_to_freeze(); | |
5e687eac BM |
875 | |
876 | do { | |
877 | prepare_to_wait(&sdp->sd_logd_waitq, &wait, | |
5f487490 | 878 | TASK_INTERRUPTIBLE); |
5e687eac BM |
879 | if (!gfs2_ail_flush_reqd(sdp) && |
880 | !gfs2_jrnl_flush_reqd(sdp) && | |
881 | !kthread_should_stop()) | |
882 | t = schedule_timeout(t); | |
883 | } while(t && !gfs2_ail_flush_reqd(sdp) && | |
884 | !gfs2_jrnl_flush_reqd(sdp) && | |
885 | !kthread_should_stop()); | |
886 | finish_wait(&sdp->sd_logd_waitq, &wait); | |
ec69b188 SW |
887 | } |
888 | ||
889 | return 0; | |
890 | } | |
891 |