]>
Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
da6dd40d | 4 | * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
7 | #include <linux/sched.h> | |
8 | #include <linux/slab.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/completion.h> | |
11 | #include <linux/buffer_head.h> | |
5c676f6d | 12 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 13 | #include <linux/crc32.h> |
c1696fb8 | 14 | #include <linux/crc32c.h> |
a25311c8 | 15 | #include <linux/delay.h> |
ec69b188 SW |
16 | #include <linux/kthread.h> |
17 | #include <linux/freezer.h> | |
254db57f | 18 | #include <linux/bio.h> |
885bceca | 19 | #include <linux/blkdev.h> |
4667a0ec | 20 | #include <linux/writeback.h> |
4a36d08d | 21 | #include <linux/list_sort.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "bmap.h" |
26 | #include "glock.h" | |
27 | #include "log.h" | |
28 | #include "lops.h" | |
29 | #include "meta_io.h" | |
5c676f6d | 30 | #include "util.h" |
71b86f56 | 31 | #include "dir.h" |
63997775 | 32 | #include "trace_gfs2.h" |
b3b94faa | 33 | |
b3b94faa DT |
34 | /** |
35 | * gfs2_struct2blk - compute stuff | |
36 | * @sdp: the filesystem | |
37 | * @nstruct: the number of structures | |
38 | * @ssize: the size of the structures | |
39 | * | |
40 | * Compute the number of log descriptor blocks needed to hold a certain number | |
41 | * of structures of a certain size. | |
42 | * | |
43 | * Returns: the number of blocks needed (minimum is always 1) | |
44 | */ | |
45 | ||
46 | unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct, | |
47 | unsigned int ssize) | |
48 | { | |
49 | unsigned int blks; | |
50 | unsigned int first, second; | |
51 | ||
52 | blks = 1; | |
faa31ce8 | 53 | first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize; |
b3b94faa DT |
54 | |
55 | if (nstruct > first) { | |
568f4c96 SW |
56 | second = (sdp->sd_sb.sb_bsize - |
57 | sizeof(struct gfs2_meta_header)) / ssize; | |
5c676f6d | 58 | blks += DIV_ROUND_UP(nstruct - first, second); |
b3b94faa DT |
59 | } |
60 | ||
61 | return blks; | |
62 | } | |
63 | ||
1e1a3d03 SW |
64 | /** |
65 | * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters | |
66 | * @mapping: The associated mapping (maybe NULL) | |
67 | * @bd: The gfs2_bufdata to remove | |
68 | * | |
c618e87a | 69 | * The ail lock _must_ be held when calling this function |
1e1a3d03 SW |
70 | * |
71 | */ | |
72 | ||
9bc980cd | 73 | static void gfs2_remove_from_ail(struct gfs2_bufdata *bd) |
1e1a3d03 | 74 | { |
16ca9412 | 75 | bd->bd_tr = NULL; |
1ad38c43 SW |
76 | list_del_init(&bd->bd_ail_st_list); |
77 | list_del_init(&bd->bd_ail_gl_list); | |
1e1a3d03 | 78 | atomic_dec(&bd->bd_gl->gl_ail_count); |
1e1a3d03 SW |
79 | brelse(bd->bd_bh); |
80 | } | |
81 | ||
ddacfaf7 SW |
82 | /** |
83 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
84 | * @sdp: the filesystem | |
4667a0ec SW |
85 | * @wbc: The writeback control structure |
86 | * @ai: The ail structure | |
ddacfaf7 SW |
87 | * |
88 | */ | |
89 | ||
4f1de018 SW |
90 | static int gfs2_ail1_start_one(struct gfs2_sbd *sdp, |
91 | struct writeback_control *wbc, | |
9e1a9ecd AG |
92 | struct gfs2_trans *tr, |
93 | bool *withdraw) | |
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 | |
16ca9412 | 102 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) { |
4667a0ec | 103 | bh = bd->bd_bh; |
ddacfaf7 | 104 | |
16ca9412 | 105 | gfs2_assert(sdp, bd->bd_tr == tr); |
ddacfaf7 | 106 | |
4667a0ec | 107 | if (!buffer_busy(bh)) { |
b524abcc BP |
108 | if (!buffer_uptodate(bh) && |
109 | !test_and_set_bit(SDF_AIL1_IO_ERROR, | |
110 | &sdp->sd_flags)) { | |
4667a0ec | 111 | gfs2_io_error_bh(sdp, bh); |
9e1a9ecd AG |
112 | *withdraw = true; |
113 | } | |
16ca9412 | 114 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
4667a0ec SW |
115 | continue; |
116 | } | |
117 | ||
118 | if (!buffer_dirty(bh)) | |
119 | continue; | |
120 | if (gl == bd->bd_gl) | |
121 | continue; | |
122 | gl = bd->bd_gl; | |
16ca9412 | 123 | list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list); |
4667a0ec | 124 | mapping = bh->b_page->mapping; |
4f1de018 SW |
125 | if (!mapping) |
126 | continue; | |
4667a0ec SW |
127 | spin_unlock(&sdp->sd_ail_lock); |
128 | generic_writepages(mapping, wbc); | |
129 | spin_lock(&sdp->sd_ail_lock); | |
130 | if (wbc->nr_to_write <= 0) | |
131 | break; | |
4f1de018 | 132 | return 1; |
4667a0ec | 133 | } |
4f1de018 SW |
134 | |
135 | return 0; | |
4667a0ec | 136 | } |
ddacfaf7 | 137 | |
ddacfaf7 | 138 | |
4667a0ec SW |
139 | /** |
140 | * gfs2_ail1_flush - start writeback of some ail1 entries | |
141 | * @sdp: The super block | |
142 | * @wbc: The writeback control structure | |
143 | * | |
144 | * Writes back some ail1 entries, according to the limits in the | |
145 | * writeback control structure | |
146 | */ | |
ddacfaf7 | 147 | |
4667a0ec SW |
148 | void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) |
149 | { | |
150 | struct list_head *head = &sdp->sd_ail1_list; | |
16ca9412 | 151 | struct gfs2_trans *tr; |
885bceca | 152 | struct blk_plug plug; |
9e1a9ecd | 153 | bool withdraw = false; |
ddacfaf7 | 154 | |
c83ae9ca | 155 | trace_gfs2_ail_flush(sdp, wbc, 1); |
885bceca | 156 | blk_start_plug(&plug); |
4667a0ec | 157 | spin_lock(&sdp->sd_ail_lock); |
4f1de018 | 158 | restart: |
16ca9412 | 159 | list_for_each_entry_reverse(tr, head, tr_list) { |
4667a0ec | 160 | if (wbc->nr_to_write <= 0) |
ddacfaf7 | 161 | break; |
9e1a9ecd | 162 | if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw)) |
4f1de018 | 163 | goto restart; |
4667a0ec SW |
164 | } |
165 | spin_unlock(&sdp->sd_ail_lock); | |
885bceca | 166 | blk_finish_plug(&plug); |
9e1a9ecd AG |
167 | if (withdraw) |
168 | gfs2_lm_withdraw(sdp, NULL); | |
c83ae9ca | 169 | trace_gfs2_ail_flush(sdp, wbc, 0); |
4667a0ec SW |
170 | } |
171 | ||
172 | /** | |
173 | * gfs2_ail1_start - start writeback of all ail1 entries | |
174 | * @sdp: The superblock | |
175 | */ | |
176 | ||
177 | static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
178 | { | |
179 | struct writeback_control wbc = { | |
180 | .sync_mode = WB_SYNC_NONE, | |
181 | .nr_to_write = LONG_MAX, | |
182 | .range_start = 0, | |
183 | .range_end = LLONG_MAX, | |
184 | }; | |
185 | ||
186 | return gfs2_ail1_flush(sdp, &wbc); | |
ddacfaf7 SW |
187 | } |
188 | ||
189 | /** | |
190 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
191 | * @sdp: the filesystem | |
192 | * @ai: the AIL entry | |
193 | * | |
194 | */ | |
195 | ||
9e1a9ecd AG |
196 | static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
197 | bool *withdraw) | |
ddacfaf7 SW |
198 | { |
199 | struct gfs2_bufdata *bd, *s; | |
200 | struct buffer_head *bh; | |
201 | ||
16ca9412 | 202 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, |
ddacfaf7 SW |
203 | bd_ail_st_list) { |
204 | bh = bd->bd_bh; | |
16ca9412 | 205 | gfs2_assert(sdp, bd->bd_tr == tr); |
4667a0ec SW |
206 | if (buffer_busy(bh)) |
207 | continue; | |
b524abcc BP |
208 | if (!buffer_uptodate(bh) && |
209 | !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) { | |
ddacfaf7 | 210 | gfs2_io_error_bh(sdp, bh); |
9e1a9ecd AG |
211 | *withdraw = true; |
212 | } | |
16ca9412 | 213 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
ddacfaf7 | 214 | } |
ddacfaf7 SW |
215 | } |
216 | ||
4667a0ec SW |
217 | /** |
218 | * gfs2_ail1_empty - Try to empty the ail1 lists | |
219 | * @sdp: The superblock | |
220 | * | |
221 | * Tries to empty the ail1 lists, starting with the oldest first | |
222 | */ | |
b3b94faa | 223 | |
4667a0ec | 224 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp) |
b3b94faa | 225 | { |
16ca9412 | 226 | struct gfs2_trans *tr, *s; |
5d054964 | 227 | int oldest_tr = 1; |
b3b94faa | 228 | int ret; |
9e1a9ecd | 229 | bool withdraw = false; |
b3b94faa | 230 | |
d6a079e8 | 231 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 232 | list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) { |
9e1a9ecd | 233 | gfs2_ail1_empty_one(sdp, tr, &withdraw); |
5d054964 | 234 | if (list_empty(&tr->tr_ail1_list) && oldest_tr) |
16ca9412 | 235 | list_move(&tr->tr_list, &sdp->sd_ail2_list); |
4667a0ec | 236 | else |
5d054964 | 237 | oldest_tr = 0; |
b3b94faa | 238 | } |
b3b94faa | 239 | ret = list_empty(&sdp->sd_ail1_list); |
d6a079e8 | 240 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 241 | |
9e1a9ecd AG |
242 | if (withdraw) |
243 | gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n"); | |
244 | ||
b3b94faa DT |
245 | return ret; |
246 | } | |
247 | ||
26b06a69 SW |
248 | static void gfs2_ail1_wait(struct gfs2_sbd *sdp) |
249 | { | |
16ca9412 | 250 | struct gfs2_trans *tr; |
26b06a69 SW |
251 | struct gfs2_bufdata *bd; |
252 | struct buffer_head *bh; | |
253 | ||
254 | spin_lock(&sdp->sd_ail_lock); | |
16ca9412 BM |
255 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
256 | list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) { | |
26b06a69 SW |
257 | bh = bd->bd_bh; |
258 | if (!buffer_locked(bh)) | |
259 | continue; | |
260 | get_bh(bh); | |
261 | spin_unlock(&sdp->sd_ail_lock); | |
262 | wait_on_buffer(bh); | |
263 | brelse(bh); | |
264 | return; | |
265 | } | |
266 | } | |
267 | spin_unlock(&sdp->sd_ail_lock); | |
268 | } | |
ddacfaf7 SW |
269 | |
270 | /** | |
271 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
272 | * @sdp: the filesystem | |
273 | * @ai: the AIL entry | |
274 | * | |
275 | */ | |
276 | ||
16ca9412 | 277 | static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
ddacfaf7 | 278 | { |
16ca9412 | 279 | struct list_head *head = &tr->tr_ail2_list; |
ddacfaf7 SW |
280 | struct gfs2_bufdata *bd; |
281 | ||
282 | while (!list_empty(head)) { | |
283 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
284 | bd_ail_st_list); | |
16ca9412 | 285 | gfs2_assert(sdp, bd->bd_tr == tr); |
f91a0d3e | 286 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
287 | } |
288 | } | |
289 | ||
b3b94faa DT |
290 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
291 | { | |
16ca9412 | 292 | struct gfs2_trans *tr, *safe; |
b3b94faa DT |
293 | unsigned int old_tail = sdp->sd_log_tail; |
294 | int wrap = (new_tail < old_tail); | |
295 | int a, b, rm; | |
296 | ||
d6a079e8 | 297 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 298 | |
16ca9412 BM |
299 | list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) { |
300 | a = (old_tail <= tr->tr_first); | |
301 | b = (tr->tr_first < new_tail); | |
b3b94faa DT |
302 | rm = (wrap) ? (a || b) : (a && b); |
303 | if (!rm) | |
304 | continue; | |
305 | ||
16ca9412 BM |
306 | gfs2_ail2_empty_one(sdp, tr); |
307 | list_del(&tr->tr_list); | |
308 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list)); | |
309 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list)); | |
310 | kfree(tr); | |
b3b94faa DT |
311 | } |
312 | ||
d6a079e8 | 313 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
314 | } |
315 | ||
24972557 BM |
316 | /** |
317 | * gfs2_log_release - Release a given number of log blocks | |
318 | * @sdp: The GFS2 superblock | |
319 | * @blks: The number of blocks | |
320 | * | |
321 | */ | |
322 | ||
323 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
324 | { | |
325 | ||
326 | atomic_add(blks, &sdp->sd_log_blks_free); | |
327 | trace_gfs2_log_blocks(sdp, blks); | |
328 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= | |
329 | sdp->sd_jdesc->jd_blocks); | |
330 | up_read(&sdp->sd_log_flush_lock); | |
331 | } | |
332 | ||
b3b94faa DT |
333 | /** |
334 | * gfs2_log_reserve - Make a log reservation | |
335 | * @sdp: The GFS2 superblock | |
336 | * @blks: The number of blocks to reserve | |
337 | * | |
89918647 | 338 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 339 | * due to the fact that there is a small number of header blocks |
b004157a SW |
340 | * associated with each log flush. The exact number can't be known until |
341 | * flush time, so we ensure that we have just enough free blocks at all | |
342 | * times to avoid running out during a log flush. | |
343 | * | |
5e687eac BM |
344 | * We no longer flush the log here, instead we wake up logd to do that |
345 | * for us. To avoid the thundering herd and to ensure that we deal fairly | |
346 | * with queued waiters, we use an exclusive wait. This means that when we | |
347 | * get woken with enough journal space to get our reservation, we need to | |
348 | * wake the next waiter on the list. | |
349 | * | |
b3b94faa DT |
350 | * Returns: errno |
351 | */ | |
352 | ||
353 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
354 | { | |
2e60d768 | 355 | int ret = 0; |
5d054964 | 356 | unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize); |
5e687eac BM |
357 | unsigned wanted = blks + reserved_blks; |
358 | DEFINE_WAIT(wait); | |
359 | int did_wait = 0; | |
360 | unsigned int free_blocks; | |
b3b94faa DT |
361 | |
362 | if (gfs2_assert_warn(sdp, blks) || | |
363 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
364 | return -EINVAL; | |
f07b3520 | 365 | atomic_add(blks, &sdp->sd_log_blks_needed); |
5e687eac BM |
366 | retry: |
367 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
368 | if (unlikely(free_blocks <= wanted)) { | |
369 | do { | |
370 | prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait, | |
371 | TASK_UNINTERRUPTIBLE); | |
372 | wake_up(&sdp->sd_logd_waitq); | |
373 | did_wait = 1; | |
374 | if (atomic_read(&sdp->sd_log_blks_free) <= wanted) | |
375 | io_schedule(); | |
376 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
377 | } while(free_blocks <= wanted); | |
378 | finish_wait(&sdp->sd_log_waitq, &wait); | |
b3b94faa | 379 | } |
2e60d768 | 380 | atomic_inc(&sdp->sd_reserving_log); |
5e687eac | 381 | if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks, |
2e60d768 BM |
382 | free_blocks - blks) != free_blocks) { |
383 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) | |
384 | wake_up(&sdp->sd_reserving_log_wait); | |
5e687eac | 385 | goto retry; |
2e60d768 | 386 | } |
f07b3520 | 387 | atomic_sub(blks, &sdp->sd_log_blks_needed); |
63997775 | 388 | trace_gfs2_log_blocks(sdp, -blks); |
5e687eac BM |
389 | |
390 | /* | |
391 | * If we waited, then so might others, wake them up _after_ we get | |
392 | * our share of the log. | |
393 | */ | |
394 | if (unlikely(did_wait)) | |
395 | wake_up(&sdp->sd_log_waitq); | |
484adff8 SW |
396 | |
397 | down_read(&sdp->sd_log_flush_lock); | |
24972557 BM |
398 | if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { |
399 | gfs2_log_release(sdp, blks); | |
2e60d768 | 400 | ret = -EROFS; |
24972557 | 401 | } |
2e60d768 BM |
402 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) |
403 | wake_up(&sdp->sd_reserving_log_wait); | |
404 | return ret; | |
b3b94faa DT |
405 | } |
406 | ||
b3b94faa DT |
407 | /** |
408 | * log_distance - Compute distance between two journal blocks | |
409 | * @sdp: The GFS2 superblock | |
410 | * @newer: The most recent journal block of the pair | |
411 | * @older: The older journal block of the pair | |
412 | * | |
413 | * Compute the distance (in the journal direction) between two | |
414 | * blocks in the journal | |
415 | * | |
416 | * Returns: the distance in blocks | |
417 | */ | |
418 | ||
faa31ce8 | 419 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
420 | unsigned int older) |
421 | { | |
422 | int dist; | |
423 | ||
424 | dist = newer - older; | |
425 | if (dist < 0) | |
426 | dist += sdp->sd_jdesc->jd_blocks; | |
427 | ||
428 | return dist; | |
429 | } | |
430 | ||
2332c443 RP |
431 | /** |
432 | * calc_reserved - Calculate the number of blocks to reserve when | |
433 | * refunding a transaction's unused buffers. | |
434 | * @sdp: The GFS2 superblock | |
435 | * | |
436 | * This is complex. We need to reserve room for all our currently used | |
437 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
438 | * all our journaled data buffers for journaled files (e.g. files in the | |
439 | * meta_fs like rindex, or files for which chattr +j was done.) | |
440 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
441 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
442 | * | |
443 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
444 | * type gets its own log header, for which we need to reserve a block. | |
445 | * In fact, each type has the potential for needing more than one header | |
446 | * in cases where we have more buffers than will fit on a journal page. | |
447 | * Metadata journal entries take up half the space of journaled buffer entries. | |
448 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
449 | * databuf_limit (251) before they cause a wrap around. | |
450 | * | |
451 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
452 | * overall header for the lot. | |
453 | * | |
454 | * Returns: the number of blocks reserved | |
455 | */ | |
456 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
457 | { | |
458 | unsigned int reserved = 0; | |
022ef4fe SW |
459 | unsigned int mbuf; |
460 | unsigned int dbuf; | |
461 | struct gfs2_trans *tr = sdp->sd_log_tr; | |
2332c443 | 462 | |
022ef4fe SW |
463 | if (tr) { |
464 | mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
465 | dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; | |
466 | reserved = mbuf + dbuf; | |
467 | /* Account for header blocks */ | |
468 | reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp)); | |
469 | reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp)); | |
470 | } | |
2332c443 | 471 | |
2e95e3f6 | 472 | if (sdp->sd_log_commited_revoke > 0) |
022ef4fe | 473 | reserved += gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke, |
2332c443 | 474 | sizeof(u64)); |
2332c443 RP |
475 | /* One for the overall header */ |
476 | if (reserved) | |
477 | reserved++; | |
478 | return reserved; | |
479 | } | |
480 | ||
b3b94faa DT |
481 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
482 | { | |
16ca9412 | 483 | struct gfs2_trans *tr; |
b3b94faa DT |
484 | unsigned int tail; |
485 | ||
d6a079e8 | 486 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 487 | |
faa31ce8 | 488 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 489 | tail = sdp->sd_log_head; |
faa31ce8 | 490 | } else { |
16ca9412 BM |
491 | tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans, |
492 | tr_list); | |
493 | tail = tr->tr_first; | |
b3b94faa DT |
494 | } |
495 | ||
d6a079e8 | 496 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
497 | |
498 | return tail; | |
499 | } | |
500 | ||
2332c443 | 501 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
502 | { |
503 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
504 | ||
505 | ail2_empty(sdp, new_tail); | |
506 | ||
fd041f0b | 507 | atomic_add(dist, &sdp->sd_log_blks_free); |
63997775 | 508 | trace_gfs2_log_blocks(sdp, dist); |
5e687eac BM |
509 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
510 | sdp->sd_jdesc->jd_blocks); | |
b3b94faa DT |
511 | |
512 | sdp->sd_log_tail = new_tail; | |
513 | } | |
514 | ||
b3b94faa | 515 | |
34cc1781 | 516 | static void log_flush_wait(struct gfs2_sbd *sdp) |
b3b94faa | 517 | { |
16615be1 SW |
518 | DEFINE_WAIT(wait); |
519 | ||
520 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
521 | do { | |
522 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
523 | TASK_UNINTERRUPTIBLE); | |
524 | if (atomic_read(&sdp->sd_log_in_flight)) | |
525 | io_schedule(); | |
526 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
527 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa | 528 | } |
b3b94faa DT |
529 | } |
530 | ||
45138990 | 531 | static int ip_cmp(void *priv, struct list_head *a, struct list_head *b) |
4a36d08d | 532 | { |
45138990 | 533 | struct gfs2_inode *ipa, *ipb; |
4a36d08d | 534 | |
45138990 SW |
535 | ipa = list_entry(a, struct gfs2_inode, i_ordered); |
536 | ipb = list_entry(b, struct gfs2_inode, i_ordered); | |
4a36d08d | 537 | |
45138990 | 538 | if (ipa->i_no_addr < ipb->i_no_addr) |
4a36d08d | 539 | return -1; |
45138990 | 540 | if (ipa->i_no_addr > ipb->i_no_addr) |
4a36d08d BP |
541 | return 1; |
542 | return 0; | |
543 | } | |
544 | ||
d7b616e2 SW |
545 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
546 | { | |
45138990 | 547 | struct gfs2_inode *ip; |
d7b616e2 SW |
548 | LIST_HEAD(written); |
549 | ||
45138990 | 550 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
551 | list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp); |
552 | while (!list_empty(&sdp->sd_log_ordered)) { | |
553 | ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); | |
1f23bc78 AD |
554 | if (ip->i_inode.i_mapping->nrpages == 0) { |
555 | test_and_clear_bit(GIF_ORDERED, &ip->i_flags); | |
556 | list_del(&ip->i_ordered); | |
d7b616e2 | 557 | continue; |
1f23bc78 AD |
558 | } |
559 | list_move(&ip->i_ordered, &written); | |
45138990 SW |
560 | spin_unlock(&sdp->sd_ordered_lock); |
561 | filemap_fdatawrite(ip->i_inode.i_mapping); | |
562 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 563 | } |
a5b1d3fc | 564 | list_splice(&written, &sdp->sd_log_ordered); |
45138990 | 565 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
566 | } |
567 | ||
568 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
569 | { | |
45138990 | 570 | struct gfs2_inode *ip; |
d7b616e2 | 571 | |
45138990 | 572 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
573 | while (!list_empty(&sdp->sd_log_ordered)) { |
574 | ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); | |
45138990 SW |
575 | list_del(&ip->i_ordered); |
576 | WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags)); | |
577 | if (ip->i_inode.i_mapping->nrpages == 0) | |
d7b616e2 | 578 | continue; |
45138990 SW |
579 | spin_unlock(&sdp->sd_ordered_lock); |
580 | filemap_fdatawait(ip->i_inode.i_mapping); | |
581 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 582 | } |
45138990 SW |
583 | spin_unlock(&sdp->sd_ordered_lock); |
584 | } | |
585 | ||
586 | void gfs2_ordered_del_inode(struct gfs2_inode *ip) | |
587 | { | |
588 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
589 | ||
590 | spin_lock(&sdp->sd_ordered_lock); | |
591 | if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags)) | |
592 | list_del(&ip->i_ordered); | |
593 | spin_unlock(&sdp->sd_ordered_lock); | |
d7b616e2 SW |
594 | } |
595 | ||
5d054964 BM |
596 | void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) |
597 | { | |
598 | struct buffer_head *bh = bd->bd_bh; | |
599 | struct gfs2_glock *gl = bd->bd_gl; | |
600 | ||
5d054964 BM |
601 | bh->b_private = NULL; |
602 | bd->bd_blkno = bh->b_blocknr; | |
9290a9a7 BP |
603 | gfs2_remove_from_ail(bd); /* drops ref on bh */ |
604 | bd->bd_bh = NULL; | |
5d054964 | 605 | sdp->sd_log_num_revoke++; |
638803d4 | 606 | if (atomic_inc_return(&gl->gl_revokes) == 1) |
9287c645 | 607 | gfs2_glock_hold(gl); |
5d054964 | 608 | set_bit(GLF_LFLUSH, &gl->gl_flags); |
a5b1d3fc | 609 | list_add(&bd->bd_list, &sdp->sd_log_revokes); |
5d054964 BM |
610 | } |
611 | ||
612 | void gfs2_write_revokes(struct gfs2_sbd *sdp) | |
613 | { | |
614 | struct gfs2_trans *tr; | |
615 | struct gfs2_bufdata *bd, *tmp; | |
616 | int have_revokes = 0; | |
617 | int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64); | |
618 | ||
619 | gfs2_ail1_empty(sdp); | |
620 | spin_lock(&sdp->sd_ail_lock); | |
c9e58fb2 | 621 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
5d054964 BM |
622 | list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) { |
623 | if (list_empty(&bd->bd_list)) { | |
624 | have_revokes = 1; | |
625 | goto done; | |
626 | } | |
627 | } | |
628 | } | |
629 | done: | |
630 | spin_unlock(&sdp->sd_ail_lock); | |
631 | if (have_revokes == 0) | |
632 | return; | |
633 | while (sdp->sd_log_num_revoke > max_revokes) | |
634 | max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64); | |
635 | max_revokes -= sdp->sd_log_num_revoke; | |
636 | if (!sdp->sd_log_num_revoke) { | |
637 | atomic_dec(&sdp->sd_log_blks_free); | |
638 | /* If no blocks have been reserved, we need to also | |
639 | * reserve a block for the header */ | |
640 | if (!sdp->sd_log_blks_reserved) | |
641 | atomic_dec(&sdp->sd_log_blks_free); | |
642 | } | |
643 | gfs2_log_lock(sdp); | |
644 | spin_lock(&sdp->sd_ail_lock); | |
c9e58fb2 | 645 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
5d054964 BM |
646 | list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) { |
647 | if (max_revokes == 0) | |
648 | goto out_of_blocks; | |
649 | if (!list_empty(&bd->bd_list)) | |
650 | continue; | |
651 | gfs2_add_revoke(sdp, bd); | |
652 | max_revokes--; | |
653 | } | |
654 | } | |
655 | out_of_blocks: | |
656 | spin_unlock(&sdp->sd_ail_lock); | |
657 | gfs2_log_unlock(sdp); | |
658 | ||
659 | if (!sdp->sd_log_num_revoke) { | |
660 | atomic_inc(&sdp->sd_log_blks_free); | |
661 | if (!sdp->sd_log_blks_reserved) | |
662 | atomic_inc(&sdp->sd_log_blks_free); | |
663 | } | |
664 | } | |
665 | ||
34cc1781 | 666 | /** |
7c70b896 | 667 | * gfs2_write_log_header - Write a journal log header buffer at lblock |
34cc1781 | 668 | * @sdp: The GFS2 superblock |
c1696fb8 | 669 | * @jd: journal descriptor of the journal to which we are writing |
588bff95 BP |
670 | * @seq: sequence number |
671 | * @tail: tail of the log | |
7c70b896 | 672 | * @lblock: value for lh_blkno (block number relative to start of journal) |
c1696fb8 | 673 | * @flags: log header flags GFS2_LOG_HEAD_* |
588bff95 | 674 | * @op_flags: flags to pass to the bio |
34cc1781 SW |
675 | * |
676 | * Returns: the initialized log buffer descriptor | |
677 | */ | |
678 | ||
c1696fb8 | 679 | void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, |
7c70b896 BP |
680 | u64 seq, u32 tail, u32 lblock, u32 flags, |
681 | int op_flags) | |
34cc1781 | 682 | { |
34cc1781 | 683 | struct gfs2_log_header *lh; |
c1696fb8 | 684 | u32 hash, crc; |
e8c92ed7 | 685 | struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO); |
c1696fb8 BP |
686 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
687 | struct timespec64 tv; | |
688 | struct super_block *sb = sdp->sd_vfs; | |
7c70b896 | 689 | u64 dblock; |
588bff95 | 690 | |
e8c92ed7 SW |
691 | lh = page_address(page); |
692 | clear_page(lh); | |
34cc1781 | 693 | |
34cc1781 SW |
694 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); |
695 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); | |
696 | lh->lh_header.__pad0 = cpu_to_be64(0); | |
697 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
698 | lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); | |
588bff95 | 699 | lh->lh_sequence = cpu_to_be64(seq); |
34cc1781 SW |
700 | lh->lh_flags = cpu_to_be32(flags); |
701 | lh->lh_tail = cpu_to_be32(tail); | |
7c70b896 | 702 | lh->lh_blkno = cpu_to_be32(lblock); |
c1696fb8 | 703 | hash = ~crc32(~0, lh, LH_V1_SIZE); |
34cc1781 SW |
704 | lh->lh_hash = cpu_to_be32(hash); |
705 | ||
ee9c7f9a | 706 | ktime_get_coarse_real_ts64(&tv); |
c1696fb8 BP |
707 | lh->lh_nsec = cpu_to_be32(tv.tv_nsec); |
708 | lh->lh_sec = cpu_to_be64(tv.tv_sec); | |
7c70b896 BP |
709 | if (!list_empty(&jd->extent_list)) |
710 | dblock = gfs2_log_bmap(sdp); | |
711 | else { | |
712 | int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock); | |
713 | if (gfs2_assert_withdraw(sdp, ret == 0)) | |
714 | return; | |
715 | } | |
716 | lh->lh_addr = cpu_to_be64(dblock); | |
c1696fb8 BP |
717 | lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr); |
718 | ||
719 | /* We may only write local statfs, quota, etc., when writing to our | |
720 | own journal. The values are left 0 when recovering a journal | |
721 | different from our own. */ | |
722 | if (!(flags & GFS2_LOG_HEAD_RECOVERY)) { | |
723 | lh->lh_statfs_addr = | |
724 | cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr); | |
725 | lh->lh_quota_addr = | |
726 | cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr); | |
727 | ||
728 | spin_lock(&sdp->sd_statfs_spin); | |
729 | lh->lh_local_total = cpu_to_be64(l_sc->sc_total); | |
730 | lh->lh_local_free = cpu_to_be64(l_sc->sc_free); | |
731 | lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes); | |
732 | spin_unlock(&sdp->sd_statfs_spin); | |
733 | } | |
734 | ||
735 | BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE); | |
736 | ||
737 | crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4, | |
738 | sb->s_blocksize - LH_V1_SIZE - 4); | |
739 | lh->lh_crc = cpu_to_be32(crc); | |
740 | ||
7c70b896 | 741 | gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); |
f4686c26 | 742 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags); |
588bff95 BP |
743 | log_flush_wait(sdp); |
744 | } | |
745 | ||
746 | /** | |
747 | * log_write_header - Get and initialize a journal header buffer | |
748 | * @sdp: The GFS2 superblock | |
c1696fb8 | 749 | * @flags: The log header flags, including log header origin |
588bff95 BP |
750 | * |
751 | * Returns: the initialized log buffer descriptor | |
752 | */ | |
753 | ||
754 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags) | |
755 | { | |
756 | unsigned int tail; | |
757 | int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC; | |
758 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); | |
759 | ||
760 | gfs2_assert_withdraw(sdp, (state != SFS_FROZEN)); | |
761 | tail = current_tail(sdp); | |
762 | ||
34cc1781 SW |
763 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { |
764 | gfs2_ordered_wait(sdp); | |
765 | log_flush_wait(sdp); | |
70fd7614 | 766 | op_flags = REQ_SYNC | REQ_META | REQ_PRIO; |
34cc1781 | 767 | } |
e8c92ed7 | 768 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); |
c1696fb8 | 769 | gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail, |
7c70b896 | 770 | sdp->sd_log_flush_head, flags, op_flags); |
34cc1781 SW |
771 | |
772 | if (sdp->sd_log_tail != tail) | |
773 | log_pull_tail(sdp, tail); | |
34cc1781 SW |
774 | } |
775 | ||
b3b94faa | 776 | /** |
b09e593d | 777 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
778 | * @sdp: the filesystem |
779 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
805c0907 | 780 | * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags |
b3b94faa DT |
781 | * |
782 | */ | |
783 | ||
c1696fb8 | 784 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags) |
b3b94faa | 785 | { |
16ca9412 | 786 | struct gfs2_trans *tr; |
2e60d768 | 787 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); |
b3b94faa | 788 | |
484adff8 | 789 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 790 | |
2bcd610d SW |
791 | /* Log might have been flushed while we waited for the flush lock */ |
792 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) { | |
793 | up_write(&sdp->sd_log_flush_lock); | |
794 | return; | |
f55ab26a | 795 | } |
805c0907 | 796 | trace_gfs2_log_flush(sdp, 1, flags); |
f55ab26a | 797 | |
c1696fb8 | 798 | if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN) |
400ac52e BM |
799 | clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
800 | ||
b1ab1e44 | 801 | sdp->sd_log_flush_head = sdp->sd_log_head; |
16ca9412 BM |
802 | tr = sdp->sd_log_tr; |
803 | if (tr) { | |
804 | sdp->sd_log_tr = NULL; | |
805 | INIT_LIST_HEAD(&tr->tr_ail1_list); | |
806 | INIT_LIST_HEAD(&tr->tr_ail2_list); | |
b1ab1e44 | 807 | tr->tr_first = sdp->sd_log_flush_head; |
2e60d768 BM |
808 | if (unlikely (state == SFS_FROZEN)) |
809 | gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new); | |
16ca9412 | 810 | } |
b3b94faa | 811 | |
2e60d768 BM |
812 | if (unlikely(state == SFS_FROZEN)) |
813 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
b3b94faa DT |
814 | gfs2_assert_withdraw(sdp, |
815 | sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke); | |
816 | ||
d7b616e2 | 817 | gfs2_ordered_write(sdp); |
d69a3c65 | 818 | lops_before_commit(sdp, tr); |
f4686c26 | 819 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE); |
d7b616e2 | 820 | |
34cc1781 | 821 | if (sdp->sd_log_head != sdp->sd_log_flush_head) { |
428fd95d | 822 | log_flush_wait(sdp); |
c1696fb8 | 823 | log_write_header(sdp, flags); |
34cc1781 | 824 | } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
fd041f0b | 825 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 826 | trace_gfs2_log_blocks(sdp, -1); |
c1696fb8 | 827 | log_write_header(sdp, flags); |
2332c443 | 828 | } |
16ca9412 | 829 | lops_after_commit(sdp, tr); |
b09e593d | 830 | |
fe1a698f SW |
831 | gfs2_log_lock(sdp); |
832 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 | 833 | sdp->sd_log_blks_reserved = 0; |
faa31ce8 | 834 | sdp->sd_log_commited_revoke = 0; |
b3b94faa | 835 | |
d6a079e8 | 836 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 BM |
837 | if (tr && !list_empty(&tr->tr_ail1_list)) { |
838 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
839 | tr = NULL; | |
b3b94faa | 840 | } |
d6a079e8 | 841 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 842 | gfs2_log_unlock(sdp); |
24972557 | 843 | |
c1696fb8 | 844 | if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) { |
24972557 BM |
845 | if (!sdp->sd_log_idle) { |
846 | for (;;) { | |
847 | gfs2_ail1_start(sdp); | |
848 | gfs2_ail1_wait(sdp); | |
849 | if (gfs2_ail1_empty(sdp)) | |
850 | break; | |
851 | } | |
852 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ | |
853 | trace_gfs2_log_blocks(sdp, -1); | |
c1696fb8 | 854 | log_write_header(sdp, flags); |
24972557 BM |
855 | sdp->sd_log_head = sdp->sd_log_flush_head; |
856 | } | |
c1696fb8 BP |
857 | if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
858 | GFS2_LOG_HEAD_FLUSH_FREEZE)) | |
24972557 | 859 | gfs2_log_shutdown(sdp); |
c1696fb8 | 860 | if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE) |
2e60d768 | 861 | atomic_set(&sdp->sd_freeze_state, SFS_FROZEN); |
24972557 BM |
862 | } |
863 | ||
805c0907 | 864 | trace_gfs2_log_flush(sdp, 0, flags); |
484adff8 | 865 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa | 866 | |
16ca9412 | 867 | kfree(tr); |
b3b94faa DT |
868 | } |
869 | ||
d69a3c65 SW |
870 | /** |
871 | * gfs2_merge_trans - Merge a new transaction into a cached transaction | |
872 | * @old: Original transaction to be expanded | |
873 | * @new: New transaction to be merged | |
874 | */ | |
875 | ||
876 | static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new) | |
877 | { | |
9862ca05 | 878 | WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags)); |
d69a3c65 SW |
879 | |
880 | old->tr_num_buf_new += new->tr_num_buf_new; | |
881 | old->tr_num_databuf_new += new->tr_num_databuf_new; | |
882 | old->tr_num_buf_rm += new->tr_num_buf_rm; | |
883 | old->tr_num_databuf_rm += new->tr_num_databuf_rm; | |
884 | old->tr_num_revoke += new->tr_num_revoke; | |
885 | old->tr_num_revoke_rm += new->tr_num_revoke_rm; | |
886 | ||
887 | list_splice_tail_init(&new->tr_databuf, &old->tr_databuf); | |
888 | list_splice_tail_init(&new->tr_buf, &old->tr_buf); | |
889 | } | |
890 | ||
b3b94faa DT |
891 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
892 | { | |
2332c443 | 893 | unsigned int reserved; |
ac39aadd | 894 | unsigned int unused; |
022ef4fe | 895 | unsigned int maxres; |
b3b94faa DT |
896 | |
897 | gfs2_log_lock(sdp); | |
898 | ||
022ef4fe SW |
899 | if (sdp->sd_log_tr) { |
900 | gfs2_merge_trans(sdp->sd_log_tr, tr); | |
901 | } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) { | |
9862ca05 | 902 | gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags)); |
022ef4fe | 903 | sdp->sd_log_tr = tr; |
9862ca05 | 904 | set_bit(TR_ATTACHED, &tr->tr_flags); |
022ef4fe SW |
905 | } |
906 | ||
b3b94faa | 907 | sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
2332c443 | 908 | reserved = calc_reserved(sdp); |
022ef4fe SW |
909 | maxres = sdp->sd_log_blks_reserved + tr->tr_reserved; |
910 | gfs2_assert_withdraw(sdp, maxres >= reserved); | |
911 | unused = maxres - reserved; | |
ac39aadd | 912 | atomic_add(unused, &sdp->sd_log_blks_free); |
63997775 | 913 | trace_gfs2_log_blocks(sdp, unused); |
fd041f0b | 914 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
2332c443 | 915 | sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
916 | sdp->sd_log_blks_reserved = reserved; |
917 | ||
918 | gfs2_log_unlock(sdp); | |
919 | } | |
920 | ||
921 | /** | |
922 | * gfs2_log_commit - Commit a transaction to the log | |
923 | * @sdp: the filesystem | |
924 | * @tr: the transaction | |
925 | * | |
5e687eac BM |
926 | * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 |
927 | * or the total number of used blocks (pinned blocks plus AIL blocks) | |
928 | * is greater than thresh2. | |
929 | * | |
930 | * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of | |
931 | * journal size. | |
932 | * | |
b3b94faa DT |
933 | * Returns: errno |
934 | */ | |
935 | ||
936 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
937 | { | |
938 | log_refund(sdp, tr); | |
b3b94faa | 939 | |
5e687eac BM |
940 | if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || |
941 | ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > | |
942 | atomic_read(&sdp->sd_log_thresh2))) | |
943 | wake_up(&sdp->sd_logd_waitq); | |
b3b94faa DT |
944 | } |
945 | ||
946 | /** | |
947 | * gfs2_log_shutdown - write a shutdown header into a journal | |
948 | * @sdp: the filesystem | |
949 | * | |
950 | */ | |
951 | ||
952 | void gfs2_log_shutdown(struct gfs2_sbd *sdp) | |
953 | { | |
b3b94faa | 954 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa | 955 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); |
b3b94faa DT |
956 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); |
957 | ||
958 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
b3b94faa | 959 | |
805c0907 | 960 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN); |
b3b94faa | 961 | |
a74604be SW |
962 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
963 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
964 | |
965 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa | 966 | sdp->sd_log_tail = sdp->sd_log_head; |
a25311c8 SW |
967 | } |
968 | ||
5e687eac BM |
969 | static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) |
970 | { | |
f07b3520 BP |
971 | return (atomic_read(&sdp->sd_log_pinned) + |
972 | atomic_read(&sdp->sd_log_blks_needed) >= | |
973 | atomic_read(&sdp->sd_log_thresh1)); | |
5e687eac BM |
974 | } |
975 | ||
976 | static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) | |
977 | { | |
978 | unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); | |
b066a4ee AD |
979 | |
980 | if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags)) | |
981 | return 1; | |
982 | ||
f07b3520 BP |
983 | return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >= |
984 | atomic_read(&sdp->sd_log_thresh2); | |
5e687eac | 985 | } |
ec69b188 SW |
986 | |
987 | /** | |
988 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
989 | * @sdp: Pointer to GFS2 superblock | |
990 | * | |
991 | * Also, periodically check to make sure that we're using the most recent | |
992 | * journal index. | |
993 | */ | |
994 | ||
995 | int gfs2_logd(void *data) | |
996 | { | |
997 | struct gfs2_sbd *sdp = data; | |
5e687eac BM |
998 | unsigned long t = 1; |
999 | DEFINE_WAIT(wait); | |
b63f5e84 | 1000 | bool did_flush; |
ec69b188 SW |
1001 | |
1002 | while (!kthread_should_stop()) { | |
ec69b188 | 1003 | |
942b0cdd BP |
1004 | /* Check for errors writing to the journal */ |
1005 | if (sdp->sd_log_error) { | |
1006 | gfs2_lm_withdraw(sdp, | |
1007 | "GFS2: fsid=%s: error %d: " | |
1008 | "withdrawing the file system to " | |
1009 | "prevent further damage.\n", | |
1010 | sdp->sd_fsname, sdp->sd_log_error); | |
1011 | } | |
1012 | ||
b63f5e84 | 1013 | did_flush = false; |
5e687eac | 1014 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { |
4667a0ec | 1015 | gfs2_ail1_empty(sdp); |
805c0907 BP |
1016 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1017 | GFS2_LFC_LOGD_JFLUSH_REQD); | |
b63f5e84 | 1018 | did_flush = true; |
5e687eac | 1019 | } |
ec69b188 | 1020 | |
5e687eac BM |
1021 | if (gfs2_ail_flush_reqd(sdp)) { |
1022 | gfs2_ail1_start(sdp); | |
26b06a69 | 1023 | gfs2_ail1_wait(sdp); |
4667a0ec | 1024 | gfs2_ail1_empty(sdp); |
805c0907 BP |
1025 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1026 | GFS2_LFC_LOGD_AIL_FLUSH_REQD); | |
b63f5e84 | 1027 | did_flush = true; |
ec69b188 SW |
1028 | } |
1029 | ||
b63f5e84 | 1030 | if (!gfs2_ail_flush_reqd(sdp) || did_flush) |
26b06a69 SW |
1031 | wake_up(&sdp->sd_log_waitq); |
1032 | ||
ec69b188 | 1033 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
a0acae0e TH |
1034 | |
1035 | try_to_freeze(); | |
5e687eac BM |
1036 | |
1037 | do { | |
1038 | prepare_to_wait(&sdp->sd_logd_waitq, &wait, | |
5f487490 | 1039 | TASK_INTERRUPTIBLE); |
5e687eac BM |
1040 | if (!gfs2_ail_flush_reqd(sdp) && |
1041 | !gfs2_jrnl_flush_reqd(sdp) && | |
1042 | !kthread_should_stop()) | |
1043 | t = schedule_timeout(t); | |
1044 | } while(t && !gfs2_ail_flush_reqd(sdp) && | |
1045 | !gfs2_jrnl_flush_reqd(sdp) && | |
1046 | !kthread_should_stop()); | |
1047 | finish_wait(&sdp->sd_logd_waitq, &wait); | |
ec69b188 SW |
1048 | } |
1049 | ||
1050 | return 0; | |
1051 | } | |
1052 |