]>
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" |
b839dada | 33 | #include "trans.h" |
b3b94faa | 34 | |
feed98a8 BP |
35 | static void gfs2_log_shutdown(struct gfs2_sbd *sdp); |
36 | ||
b3b94faa DT |
37 | /** |
38 | * gfs2_struct2blk - compute stuff | |
39 | * @sdp: the filesystem | |
40 | * @nstruct: the number of structures | |
b3b94faa DT |
41 | * |
42 | * Compute the number of log descriptor blocks needed to hold a certain number | |
43 | * of structures of a certain size. | |
44 | * | |
45 | * Returns: the number of blocks needed (minimum is always 1) | |
46 | */ | |
47 | ||
2e9eeaa1 | 48 | unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct) |
b3b94faa DT |
49 | { |
50 | unsigned int blks; | |
51 | unsigned int first, second; | |
52 | ||
6188e877 | 53 | /* The initial struct gfs2_log_descriptor block */ |
b3b94faa | 54 | blks = 1; |
2e9eeaa1 | 55 | first = sdp->sd_ldptrs; |
b3b94faa DT |
56 | |
57 | if (nstruct > first) { | |
6188e877 | 58 | /* Subsequent struct gfs2_meta_header blocks */ |
2e9eeaa1 | 59 | second = sdp->sd_inptrs; |
5c676f6d | 60 | blks += DIV_ROUND_UP(nstruct - first, second); |
b3b94faa DT |
61 | } |
62 | ||
63 | return blks; | |
64 | } | |
65 | ||
1e1a3d03 SW |
66 | /** |
67 | * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters | |
68 | * @mapping: The associated mapping (maybe NULL) | |
69 | * @bd: The gfs2_bufdata to remove | |
70 | * | |
c618e87a | 71 | * The ail lock _must_ be held when calling this function |
1e1a3d03 SW |
72 | * |
73 | */ | |
74 | ||
68942870 | 75 | void gfs2_remove_from_ail(struct gfs2_bufdata *bd) |
1e1a3d03 | 76 | { |
16ca9412 | 77 | bd->bd_tr = NULL; |
1ad38c43 SW |
78 | list_del_init(&bd->bd_ail_st_list); |
79 | list_del_init(&bd->bd_ail_gl_list); | |
1e1a3d03 | 80 | atomic_dec(&bd->bd_gl->gl_ail_count); |
1e1a3d03 SW |
81 | brelse(bd->bd_bh); |
82 | } | |
83 | ||
ddacfaf7 SW |
84 | /** |
85 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
86 | * @sdp: the filesystem | |
4667a0ec SW |
87 | * @wbc: The writeback control structure |
88 | * @ai: The ail structure | |
ddacfaf7 SW |
89 | * |
90 | */ | |
91 | ||
4f1de018 SW |
92 | static int gfs2_ail1_start_one(struct gfs2_sbd *sdp, |
93 | struct writeback_control *wbc, | |
69511080 | 94 | struct gfs2_trans *tr) |
d6a079e8 DC |
95 | __releases(&sdp->sd_ail_lock) |
96 | __acquires(&sdp->sd_ail_lock) | |
ddacfaf7 | 97 | { |
5ac048bb | 98 | struct gfs2_glock *gl = NULL; |
4667a0ec | 99 | struct address_space *mapping; |
ddacfaf7 SW |
100 | struct gfs2_bufdata *bd, *s; |
101 | struct buffer_head *bh; | |
b1676cbb | 102 | int ret = 0; |
ddacfaf7 | 103 | |
16ca9412 | 104 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) { |
4667a0ec | 105 | bh = bd->bd_bh; |
ddacfaf7 | 106 | |
16ca9412 | 107 | gfs2_assert(sdp, bd->bd_tr == tr); |
ddacfaf7 | 108 | |
4667a0ec | 109 | if (!buffer_busy(bh)) { |
30fe70a8 BP |
110 | if (buffer_uptodate(bh)) { |
111 | list_move(&bd->bd_ail_st_list, | |
112 | &tr->tr_ail2_list); | |
113 | continue; | |
114 | } | |
036330c9 | 115 | if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) { |
4667a0ec | 116 | gfs2_io_error_bh(sdp, bh); |
69511080 | 117 | gfs2_withdraw_delayed(sdp); |
9e1a9ecd | 118 | } |
4667a0ec SW |
119 | } |
120 | ||
30fe70a8 BP |
121 | if (gfs2_withdrawn(sdp)) { |
122 | gfs2_remove_from_ail(bd); | |
123 | continue; | |
124 | } | |
4667a0ec SW |
125 | if (!buffer_dirty(bh)) |
126 | continue; | |
127 | if (gl == bd->bd_gl) | |
128 | continue; | |
129 | gl = bd->bd_gl; | |
16ca9412 | 130 | list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list); |
4667a0ec | 131 | mapping = bh->b_page->mapping; |
4f1de018 SW |
132 | if (!mapping) |
133 | continue; | |
4667a0ec | 134 | spin_unlock(&sdp->sd_ail_lock); |
b1676cbb | 135 | ret = generic_writepages(mapping, wbc); |
4667a0ec | 136 | spin_lock(&sdp->sd_ail_lock); |
4e79e3f0 BP |
137 | if (ret == -ENODATA) /* if a jdata write into a new hole */ |
138 | ret = 0; /* ignore it */ | |
b1676cbb | 139 | if (ret || wbc->nr_to_write <= 0) |
4667a0ec | 140 | break; |
b1676cbb | 141 | return -EBUSY; |
4667a0ec | 142 | } |
4f1de018 | 143 | |
b1676cbb | 144 | return ret; |
4667a0ec | 145 | } |
ddacfaf7 | 146 | |
9592ea80 BP |
147 | static void dump_ail_list(struct gfs2_sbd *sdp) |
148 | { | |
149 | struct gfs2_trans *tr; | |
150 | struct gfs2_bufdata *bd; | |
151 | struct buffer_head *bh; | |
152 | ||
9592ea80 BP |
153 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
154 | list_for_each_entry_reverse(bd, &tr->tr_ail1_list, | |
155 | bd_ail_st_list) { | |
156 | bh = bd->bd_bh; | |
157 | fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd, | |
158 | (unsigned long long)bd->bd_blkno, bh); | |
159 | if (!bh) { | |
160 | fs_err(sdp, "\n"); | |
161 | continue; | |
162 | } | |
163 | fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d " | |
164 | "map:%d new:%d ar:%d aw:%d delay:%d " | |
165 | "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n", | |
166 | (unsigned long long)bh->b_blocknr, | |
167 | buffer_uptodate(bh), buffer_dirty(bh), | |
168 | buffer_locked(bh), buffer_req(bh), | |
169 | buffer_mapped(bh), buffer_new(bh), | |
170 | buffer_async_read(bh), buffer_async_write(bh), | |
171 | buffer_delay(bh), buffer_write_io_error(bh), | |
172 | buffer_unwritten(bh), | |
173 | buffer_defer_completion(bh), | |
174 | buffer_pinned(bh), buffer_escaped(bh)); | |
175 | } | |
176 | } | |
177 | } | |
ddacfaf7 | 178 | |
4667a0ec SW |
179 | /** |
180 | * gfs2_ail1_flush - start writeback of some ail1 entries | |
181 | * @sdp: The super block | |
182 | * @wbc: The writeback control structure | |
183 | * | |
184 | * Writes back some ail1 entries, according to the limits in the | |
185 | * writeback control structure | |
186 | */ | |
ddacfaf7 | 187 | |
4667a0ec SW |
188 | void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) |
189 | { | |
190 | struct list_head *head = &sdp->sd_ail1_list; | |
16ca9412 | 191 | struct gfs2_trans *tr; |
885bceca | 192 | struct blk_plug plug; |
75b46c43 | 193 | int ret; |
9592ea80 | 194 | unsigned long flush_start = jiffies; |
ddacfaf7 | 195 | |
c83ae9ca | 196 | trace_gfs2_ail_flush(sdp, wbc, 1); |
885bceca | 197 | blk_start_plug(&plug); |
4667a0ec | 198 | spin_lock(&sdp->sd_ail_lock); |
4f1de018 | 199 | restart: |
75b46c43 | 200 | ret = 0; |
9592ea80 | 201 | if (time_after(jiffies, flush_start + (HZ * 600))) { |
d5dc3d96 BP |
202 | fs_err(sdp, "Error: In %s for ten minutes! t=%d\n", |
203 | __func__, current->journal_info ? 1 : 0); | |
9592ea80 BP |
204 | dump_ail_list(sdp); |
205 | goto out; | |
206 | } | |
16ca9412 | 207 | list_for_each_entry_reverse(tr, head, tr_list) { |
4667a0ec | 208 | if (wbc->nr_to_write <= 0) |
ddacfaf7 | 209 | break; |
b1676cbb BP |
210 | ret = gfs2_ail1_start_one(sdp, wbc, tr); |
211 | if (ret) { | |
212 | if (ret == -EBUSY) | |
213 | goto restart; | |
214 | break; | |
215 | } | |
4667a0ec | 216 | } |
9592ea80 | 217 | out: |
4667a0ec | 218 | spin_unlock(&sdp->sd_ail_lock); |
885bceca | 219 | blk_finish_plug(&plug); |
49003128 BP |
220 | if (ret) { |
221 | gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) " | |
222 | "returned: %d\n", ret); | |
badb55ec | 223 | gfs2_withdraw(sdp); |
49003128 | 224 | } |
c83ae9ca | 225 | trace_gfs2_ail_flush(sdp, wbc, 0); |
4667a0ec SW |
226 | } |
227 | ||
228 | /** | |
229 | * gfs2_ail1_start - start writeback of all ail1 entries | |
230 | * @sdp: The superblock | |
231 | */ | |
232 | ||
233 | static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
234 | { | |
235 | struct writeback_control wbc = { | |
236 | .sync_mode = WB_SYNC_NONE, | |
237 | .nr_to_write = LONG_MAX, | |
238 | .range_start = 0, | |
239 | .range_end = LLONG_MAX, | |
240 | }; | |
241 | ||
242 | return gfs2_ail1_flush(sdp, &wbc); | |
ddacfaf7 SW |
243 | } |
244 | ||
5cb738b5 AG |
245 | static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp) |
246 | { | |
247 | unsigned int new_flush_tail = sdp->sd_log_head; | |
248 | struct gfs2_trans *tr; | |
249 | ||
250 | if (!list_empty(&sdp->sd_ail1_list)) { | |
251 | tr = list_last_entry(&sdp->sd_ail1_list, | |
252 | struct gfs2_trans, tr_list); | |
253 | new_flush_tail = tr->tr_first; | |
254 | } | |
255 | sdp->sd_log_flush_tail = new_flush_tail; | |
256 | } | |
257 | ||
258 | static void gfs2_log_update_head(struct gfs2_sbd *sdp) | |
259 | { | |
260 | unsigned int new_head = sdp->sd_log_flush_head; | |
261 | ||
262 | if (sdp->sd_log_flush_tail == sdp->sd_log_head) | |
263 | sdp->sd_log_flush_tail = new_head; | |
264 | sdp->sd_log_head = new_head; | |
265 | } | |
266 | ||
ddacfaf7 SW |
267 | /** |
268 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
269 | * @sdp: the filesystem | |
5e4c7632 BP |
270 | * @tr: the transaction |
271 | * @max_revokes: If nonzero, issue revokes for the bd items for written buffers | |
ddacfaf7 | 272 | * |
36c78309 | 273 | * returns: the transaction's count of remaining active items |
ddacfaf7 SW |
274 | */ |
275 | ||
36c78309 | 276 | static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
5e4c7632 | 277 | int *max_revokes) |
ddacfaf7 SW |
278 | { |
279 | struct gfs2_bufdata *bd, *s; | |
280 | struct buffer_head *bh; | |
36c78309 | 281 | int active_count = 0; |
ddacfaf7 | 282 | |
16ca9412 | 283 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, |
ddacfaf7 SW |
284 | bd_ail_st_list) { |
285 | bh = bd->bd_bh; | |
16ca9412 | 286 | gfs2_assert(sdp, bd->bd_tr == tr); |
036330c9 BP |
287 | /* |
288 | * If another process flagged an io error, e.g. writing to the | |
289 | * journal, error all other bhs and move them off the ail1 to | |
290 | * prevent a tight loop when unmount tries to flush ail1, | |
291 | * regardless of whether they're still busy. If no outside | |
292 | * errors were found and the buffer is busy, move to the next. | |
293 | * If the ail buffer is not busy and caught an error, flag it | |
294 | * for others. | |
295 | */ | |
36c78309 BP |
296 | if (!sdp->sd_log_error && buffer_busy(bh)) { |
297 | active_count++; | |
4667a0ec | 298 | continue; |
36c78309 | 299 | } |
b524abcc | 300 | if (!buffer_uptodate(bh) && |
036330c9 | 301 | !cmpxchg(&sdp->sd_log_error, 0, -EIO)) { |
ddacfaf7 | 302 | gfs2_io_error_bh(sdp, bh); |
69511080 | 303 | gfs2_withdraw_delayed(sdp); |
9e1a9ecd | 304 | } |
5e4c7632 BP |
305 | /* |
306 | * If we have space for revokes and the bd is no longer on any | |
307 | * buf list, we can just add a revoke for it immediately and | |
308 | * avoid having to put it on the ail2 list, where it would need | |
309 | * to be revoked later. | |
310 | */ | |
311 | if (*max_revokes && list_empty(&bd->bd_list)) { | |
312 | gfs2_add_revoke(sdp, bd); | |
313 | (*max_revokes)--; | |
314 | continue; | |
315 | } | |
16ca9412 | 316 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
ddacfaf7 | 317 | } |
36c78309 | 318 | return active_count; |
ddacfaf7 SW |
319 | } |
320 | ||
4667a0ec SW |
321 | /** |
322 | * gfs2_ail1_empty - Try to empty the ail1 lists | |
323 | * @sdp: The superblock | |
5e4c7632 | 324 | * @max_revokes: If non-zero, add revokes where appropriate |
4667a0ec SW |
325 | * |
326 | * Tries to empty the ail1 lists, starting with the oldest first | |
327 | */ | |
b3b94faa | 328 | |
5e4c7632 | 329 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes) |
b3b94faa | 330 | { |
16ca9412 | 331 | struct gfs2_trans *tr, *s; |
5d054964 | 332 | int oldest_tr = 1; |
b3b94faa DT |
333 | int ret; |
334 | ||
d6a079e8 | 335 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 336 | list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) { |
36c78309 | 337 | if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr) |
16ca9412 | 338 | list_move(&tr->tr_list, &sdp->sd_ail2_list); |
4667a0ec | 339 | else |
5d054964 | 340 | oldest_tr = 0; |
b3b94faa | 341 | } |
5cb738b5 | 342 | gfs2_log_update_flush_tail(sdp); |
b3b94faa | 343 | ret = list_empty(&sdp->sd_ail1_list); |
d6a079e8 | 344 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 345 | |
69511080 | 346 | if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) { |
badb55ec AG |
347 | gfs2_lm(sdp, "fatal: I/O error(s)\n"); |
348 | gfs2_withdraw(sdp); | |
349 | } | |
9e1a9ecd | 350 | |
b3b94faa DT |
351 | return ret; |
352 | } | |
353 | ||
26b06a69 SW |
354 | static void gfs2_ail1_wait(struct gfs2_sbd *sdp) |
355 | { | |
16ca9412 | 356 | struct gfs2_trans *tr; |
26b06a69 SW |
357 | struct gfs2_bufdata *bd; |
358 | struct buffer_head *bh; | |
359 | ||
360 | spin_lock(&sdp->sd_ail_lock); | |
16ca9412 BM |
361 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
362 | list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) { | |
26b06a69 SW |
363 | bh = bd->bd_bh; |
364 | if (!buffer_locked(bh)) | |
365 | continue; | |
366 | get_bh(bh); | |
367 | spin_unlock(&sdp->sd_ail_lock); | |
368 | wait_on_buffer(bh); | |
369 | brelse(bh); | |
370 | return; | |
371 | } | |
372 | } | |
373 | spin_unlock(&sdp->sd_ail_lock); | |
374 | } | |
ddacfaf7 SW |
375 | |
376 | /** | |
2ca0c2fb | 377 | * gfs2_ail_empty_tr - empty one of the ail lists for a transaction |
ddacfaf7 SW |
378 | */ |
379 | ||
2ca0c2fb BP |
380 | static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
381 | struct list_head *head) | |
ddacfaf7 | 382 | { |
ddacfaf7 SW |
383 | struct gfs2_bufdata *bd; |
384 | ||
385 | while (!list_empty(head)) { | |
2ca0c2fb BP |
386 | bd = list_first_entry(head, struct gfs2_bufdata, |
387 | bd_ail_st_list); | |
16ca9412 | 388 | gfs2_assert(sdp, bd->bd_tr == tr); |
f91a0d3e | 389 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
390 | } |
391 | } | |
392 | ||
6e80674a AG |
393 | static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
394 | { | |
395 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); | |
396 | list_del(&tr->tr_list); | |
397 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list)); | |
398 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list)); | |
399 | gfs2_trans_free(sdp, tr); | |
400 | } | |
401 | ||
b3b94faa DT |
402 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
403 | { | |
6e80674a | 404 | struct list_head *ail2_list = &sdp->sd_ail2_list; |
b3b94faa | 405 | unsigned int old_tail = sdp->sd_log_tail; |
6e80674a | 406 | struct gfs2_trans *tr, *safe; |
b3b94faa | 407 | |
d6a079e8 | 408 | spin_lock(&sdp->sd_ail_lock); |
6e80674a AG |
409 | if (old_tail <= new_tail) { |
410 | list_for_each_entry_safe(tr, safe, ail2_list, tr_list) { | |
411 | if (old_tail <= tr->tr_first && tr->tr_first < new_tail) | |
412 | __ail2_empty(sdp, tr); | |
413 | } | |
414 | } else { | |
415 | list_for_each_entry_safe(tr, safe, ail2_list, tr_list) { | |
416 | if (old_tail <= tr->tr_first || tr->tr_first < new_tail) | |
417 | __ail2_empty(sdp, tr); | |
418 | } | |
b3b94faa | 419 | } |
d6a079e8 | 420 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
421 | } |
422 | ||
f3708fb5 AG |
423 | /** |
424 | * gfs2_log_is_empty - Check if the log is empty | |
425 | * @sdp: The GFS2 superblock | |
426 | */ | |
427 | ||
428 | bool gfs2_log_is_empty(struct gfs2_sbd *sdp) { | |
429 | return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks; | |
430 | } | |
431 | ||
24972557 BM |
432 | /** |
433 | * gfs2_log_release - Release a given number of log blocks | |
434 | * @sdp: The GFS2 superblock | |
435 | * @blks: The number of blocks | |
436 | * | |
437 | */ | |
438 | ||
439 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
440 | { | |
24972557 BM |
441 | atomic_add(blks, &sdp->sd_log_blks_free); |
442 | trace_gfs2_log_blocks(sdp, blks); | |
443 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= | |
444 | sdp->sd_jdesc->jd_blocks); | |
5ae8fff8 AG |
445 | if (atomic_read(&sdp->sd_log_blks_needed)) |
446 | wake_up(&sdp->sd_log_waitq); | |
24972557 BM |
447 | } |
448 | ||
b3b94faa DT |
449 | /** |
450 | * gfs2_log_reserve - Make a log reservation | |
451 | * @sdp: The GFS2 superblock | |
452 | * @blks: The number of blocks to reserve | |
453 | * | |
89918647 | 454 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 455 | * due to the fact that there is a small number of header blocks |
b004157a SW |
456 | * associated with each log flush. The exact number can't be known until |
457 | * flush time, so we ensure that we have just enough free blocks at all | |
458 | * times to avoid running out during a log flush. | |
459 | * | |
5e687eac BM |
460 | * We no longer flush the log here, instead we wake up logd to do that |
461 | * for us. To avoid the thundering herd and to ensure that we deal fairly | |
462 | * with queued waiters, we use an exclusive wait. This means that when we | |
463 | * get woken with enough journal space to get our reservation, we need to | |
464 | * wake the next waiter on the list. | |
b3b94faa DT |
465 | */ |
466 | ||
c1eba1b0 | 467 | void gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) |
b3b94faa | 468 | { |
5d054964 | 469 | unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize); |
5e687eac | 470 | unsigned wanted = blks + reserved_blks; |
5e687eac | 471 | unsigned int free_blocks; |
b3b94faa | 472 | |
5e687eac | 473 | free_blocks = atomic_read(&sdp->sd_log_blks_free); |
5ae8fff8 AG |
474 | while (free_blocks >= wanted) { |
475 | if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks, | |
476 | free_blocks - blks)) | |
477 | return; | |
478 | } | |
479 | ||
480 | atomic_add(blks, &sdp->sd_log_blks_needed); | |
481 | for (;;) { | |
482 | if (current != sdp->sd_logd_process) | |
5e687eac | 483 | wake_up(&sdp->sd_logd_waitq); |
5ae8fff8 AG |
484 | io_wait_event(sdp->sd_log_waitq, |
485 | (free_blocks = atomic_read(&sdp->sd_log_blks_free), | |
486 | free_blocks >= wanted)); | |
487 | do { | |
488 | if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, | |
489 | &free_blocks, | |
490 | free_blocks - blks)) | |
491 | goto reserved; | |
492 | } while (free_blocks >= wanted); | |
b3b94faa | 493 | } |
5e687eac | 494 | |
5ae8fff8 AG |
495 | reserved: |
496 | trace_gfs2_log_blocks(sdp, -blks); | |
497 | if (atomic_sub_return(blks, &sdp->sd_log_blks_needed)) | |
5e687eac | 498 | wake_up(&sdp->sd_log_waitq); |
b3b94faa DT |
499 | } |
500 | ||
b3b94faa DT |
501 | /** |
502 | * log_distance - Compute distance between two journal blocks | |
503 | * @sdp: The GFS2 superblock | |
504 | * @newer: The most recent journal block of the pair | |
505 | * @older: The older journal block of the pair | |
506 | * | |
507 | * Compute the distance (in the journal direction) between two | |
508 | * blocks in the journal | |
509 | * | |
510 | * Returns: the distance in blocks | |
511 | */ | |
512 | ||
faa31ce8 | 513 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
514 | unsigned int older) |
515 | { | |
516 | int dist; | |
517 | ||
518 | dist = newer - older; | |
519 | if (dist < 0) | |
520 | dist += sdp->sd_jdesc->jd_blocks; | |
521 | ||
522 | return dist; | |
523 | } | |
524 | ||
2332c443 | 525 | /** |
6188e877 | 526 | * calc_reserved - Calculate the number of blocks to keep reserved |
2332c443 RP |
527 | * @sdp: The GFS2 superblock |
528 | * | |
529 | * This is complex. We need to reserve room for all our currently used | |
6188e877 AG |
530 | * metadata blocks (e.g. normal file I/O rewriting file time stamps) and |
531 | * all our journaled data blocks for journaled files (e.g. files in the | |
2332c443 | 532 | * meta_fs like rindex, or files for which chattr +j was done.) |
6188e877 | 533 | * If we don't reserve enough space, corruption will follow. |
2332c443 | 534 | * |
6188e877 AG |
535 | * We can have metadata blocks and jdata blocks in the same journal. Each |
536 | * type gets its own log descriptor, for which we need to reserve a block. | |
537 | * In fact, each type has the potential for needing more than one log descriptor | |
538 | * in cases where we have more blocks than will fit in a log descriptor. | |
2332c443 | 539 | * Metadata journal entries take up half the space of journaled buffer entries. |
2332c443 RP |
540 | * |
541 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
542 | * overall header for the lot. | |
543 | * | |
544 | * Returns: the number of blocks reserved | |
545 | */ | |
546 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
547 | { | |
548 | unsigned int reserved = 0; | |
022ef4fe SW |
549 | unsigned int mbuf; |
550 | unsigned int dbuf; | |
551 | struct gfs2_trans *tr = sdp->sd_log_tr; | |
2332c443 | 552 | |
022ef4fe SW |
553 | if (tr) { |
554 | mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
555 | dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; | |
556 | reserved = mbuf + dbuf; | |
557 | /* Account for header blocks */ | |
558 | reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp)); | |
559 | reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp)); | |
560 | } | |
2332c443 | 561 | |
5d439758 AG |
562 | if (sdp->sd_log_committed_revoke > 0) |
563 | reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke); | |
2332c443 RP |
564 | /* One for the overall header */ |
565 | if (reserved) | |
566 | reserved++; | |
567 | return reserved; | |
568 | } | |
569 | ||
5cb738b5 | 570 | static void log_pull_tail(struct gfs2_sbd *sdp) |
b3b94faa | 571 | { |
5cb738b5 AG |
572 | unsigned int new_tail = sdp->sd_log_flush_tail; |
573 | unsigned int dist; | |
b3b94faa | 574 | |
5cb738b5 AG |
575 | if (new_tail == sdp->sd_log_tail) |
576 | return; | |
577 | dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
b3b94faa | 578 | ail2_empty(sdp, new_tail); |
c1eba1b0 | 579 | gfs2_log_release(sdp, dist); |
b3b94faa DT |
580 | sdp->sd_log_tail = new_tail; |
581 | } | |
582 | ||
b3b94faa | 583 | |
9ff78289 | 584 | void log_flush_wait(struct gfs2_sbd *sdp) |
b3b94faa | 585 | { |
16615be1 SW |
586 | DEFINE_WAIT(wait); |
587 | ||
588 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
589 | do { | |
590 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
591 | TASK_UNINTERRUPTIBLE); | |
592 | if (atomic_read(&sdp->sd_log_in_flight)) | |
593 | io_schedule(); | |
594 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
595 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa | 596 | } |
b3b94faa DT |
597 | } |
598 | ||
45138990 | 599 | static int ip_cmp(void *priv, struct list_head *a, struct list_head *b) |
4a36d08d | 600 | { |
45138990 | 601 | struct gfs2_inode *ipa, *ipb; |
4a36d08d | 602 | |
45138990 SW |
603 | ipa = list_entry(a, struct gfs2_inode, i_ordered); |
604 | ipb = list_entry(b, struct gfs2_inode, i_ordered); | |
4a36d08d | 605 | |
45138990 | 606 | if (ipa->i_no_addr < ipb->i_no_addr) |
4a36d08d | 607 | return -1; |
45138990 | 608 | if (ipa->i_no_addr > ipb->i_no_addr) |
4a36d08d BP |
609 | return 1; |
610 | return 0; | |
611 | } | |
612 | ||
7542486b BP |
613 | static void __ordered_del_inode(struct gfs2_inode *ip) |
614 | { | |
615 | if (!list_empty(&ip->i_ordered)) | |
616 | list_del_init(&ip->i_ordered); | |
617 | } | |
618 | ||
d7b616e2 SW |
619 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
620 | { | |
45138990 | 621 | struct gfs2_inode *ip; |
d7b616e2 SW |
622 | LIST_HEAD(written); |
623 | ||
45138990 | 624 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
625 | list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp); |
626 | while (!list_empty(&sdp->sd_log_ordered)) { | |
969183bc | 627 | ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered); |
1f23bc78 | 628 | if (ip->i_inode.i_mapping->nrpages == 0) { |
7542486b | 629 | __ordered_del_inode(ip); |
d7b616e2 | 630 | continue; |
1f23bc78 AD |
631 | } |
632 | list_move(&ip->i_ordered, &written); | |
45138990 SW |
633 | spin_unlock(&sdp->sd_ordered_lock); |
634 | filemap_fdatawrite(ip->i_inode.i_mapping); | |
635 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 636 | } |
a5b1d3fc | 637 | list_splice(&written, &sdp->sd_log_ordered); |
45138990 | 638 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
639 | } |
640 | ||
641 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
642 | { | |
45138990 | 643 | struct gfs2_inode *ip; |
d7b616e2 | 644 | |
45138990 | 645 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc | 646 | while (!list_empty(&sdp->sd_log_ordered)) { |
969183bc | 647 | ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered); |
7542486b | 648 | __ordered_del_inode(ip); |
45138990 | 649 | if (ip->i_inode.i_mapping->nrpages == 0) |
d7b616e2 | 650 | continue; |
45138990 SW |
651 | spin_unlock(&sdp->sd_ordered_lock); |
652 | filemap_fdatawait(ip->i_inode.i_mapping); | |
653 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 654 | } |
45138990 SW |
655 | spin_unlock(&sdp->sd_ordered_lock); |
656 | } | |
657 | ||
658 | void gfs2_ordered_del_inode(struct gfs2_inode *ip) | |
659 | { | |
660 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
661 | ||
662 | spin_lock(&sdp->sd_ordered_lock); | |
7542486b | 663 | __ordered_del_inode(ip); |
45138990 | 664 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
665 | } |
666 | ||
5d054964 BM |
667 | void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) |
668 | { | |
669 | struct buffer_head *bh = bd->bd_bh; | |
670 | struct gfs2_glock *gl = bd->bd_gl; | |
671 | ||
f4e2f5e1 AG |
672 | sdp->sd_log_num_revoke++; |
673 | if (atomic_inc_return(&gl->gl_revokes) == 1) | |
674 | gfs2_glock_hold(gl); | |
5d054964 BM |
675 | bh->b_private = NULL; |
676 | bd->bd_blkno = bh->b_blocknr; | |
9290a9a7 BP |
677 | gfs2_remove_from_ail(bd); /* drops ref on bh */ |
678 | bd->bd_bh = NULL; | |
5d054964 | 679 | set_bit(GLF_LFLUSH, &gl->gl_flags); |
a5b1d3fc | 680 | list_add(&bd->bd_list, &sdp->sd_log_revokes); |
5d054964 BM |
681 | } |
682 | ||
fe5e7ba1 BP |
683 | void gfs2_glock_remove_revoke(struct gfs2_glock *gl) |
684 | { | |
685 | if (atomic_dec_return(&gl->gl_revokes) == 0) { | |
686 | clear_bit(GLF_LFLUSH, &gl->gl_flags); | |
687 | gfs2_glock_queue_put(gl); | |
688 | } | |
689 | } | |
690 | ||
5e4c7632 | 691 | /** |
e7501bf8 | 692 | * gfs2_flush_revokes - Add as many revokes to the system transaction as we can |
5e4c7632 BP |
693 | * @sdp: The GFS2 superblock |
694 | * | |
695 | * Our usual strategy is to defer writing revokes as much as we can in the hope | |
696 | * that we'll eventually overwrite the journal, which will make those revokes | |
697 | * go away. This changes when we flush the log: at that point, there will | |
698 | * likely be some left-over space in the last revoke block of that transaction. | |
699 | * We can fill that space with additional revokes for blocks that have already | |
700 | * been written back. This will basically come at no cost now, and will save | |
701 | * us from having to keep track of those blocks on the AIL2 list later. | |
702 | */ | |
e7501bf8 | 703 | void gfs2_flush_revokes(struct gfs2_sbd *sdp) |
5d054964 | 704 | { |
5e4c7632 | 705 | /* number of revokes we still have room for */ |
5a4e9c60 | 706 | unsigned int max_revokes; |
5d054964 | 707 | |
5e4c7632 | 708 | gfs2_log_lock(sdp); |
5a4e9c60 AG |
709 | max_revokes = sdp->sd_ldptrs; |
710 | if (sdp->sd_log_num_revoke > sdp->sd_ldptrs) | |
711 | max_revokes += roundup(sdp->sd_log_num_revoke - sdp->sd_ldptrs, | |
712 | sdp->sd_inptrs); | |
5d054964 BM |
713 | max_revokes -= sdp->sd_log_num_revoke; |
714 | if (!sdp->sd_log_num_revoke) { | |
715 | atomic_dec(&sdp->sd_log_blks_free); | |
716 | /* If no blocks have been reserved, we need to also | |
717 | * reserve a block for the header */ | |
77650bdb | 718 | if (!sdp->sd_log_blks_reserved) { |
5d054964 | 719 | atomic_dec(&sdp->sd_log_blks_free); |
77650bdb BP |
720 | trace_gfs2_log_blocks(sdp, -2); |
721 | } else { | |
722 | trace_gfs2_log_blocks(sdp, -1); | |
723 | } | |
5d054964 | 724 | } |
5e4c7632 | 725 | gfs2_ail1_empty(sdp, max_revokes); |
5d054964 BM |
726 | gfs2_log_unlock(sdp); |
727 | ||
728 | if (!sdp->sd_log_num_revoke) { | |
729 | atomic_inc(&sdp->sd_log_blks_free); | |
77650bdb | 730 | if (!sdp->sd_log_blks_reserved) { |
5d054964 | 731 | atomic_inc(&sdp->sd_log_blks_free); |
77650bdb BP |
732 | trace_gfs2_log_blocks(sdp, 2); |
733 | } else { | |
734 | trace_gfs2_log_blocks(sdp, 1); | |
735 | } | |
5d054964 BM |
736 | } |
737 | } | |
738 | ||
34cc1781 | 739 | /** |
7c70b896 | 740 | * gfs2_write_log_header - Write a journal log header buffer at lblock |
34cc1781 | 741 | * @sdp: The GFS2 superblock |
c1696fb8 | 742 | * @jd: journal descriptor of the journal to which we are writing |
588bff95 BP |
743 | * @seq: sequence number |
744 | * @tail: tail of the log | |
7c70b896 | 745 | * @lblock: value for lh_blkno (block number relative to start of journal) |
c1696fb8 | 746 | * @flags: log header flags GFS2_LOG_HEAD_* |
588bff95 | 747 | * @op_flags: flags to pass to the bio |
34cc1781 SW |
748 | * |
749 | * Returns: the initialized log buffer descriptor | |
750 | */ | |
751 | ||
c1696fb8 | 752 | void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, |
7c70b896 BP |
753 | u64 seq, u32 tail, u32 lblock, u32 flags, |
754 | int op_flags) | |
34cc1781 | 755 | { |
34cc1781 | 756 | struct gfs2_log_header *lh; |
c1696fb8 | 757 | u32 hash, crc; |
ade48088 | 758 | struct page *page; |
c1696fb8 BP |
759 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
760 | struct timespec64 tv; | |
761 | struct super_block *sb = sdp->sd_vfs; | |
7c70b896 | 762 | u64 dblock; |
588bff95 | 763 | |
ade48088 | 764 | if (gfs2_withdrawn(sdp)) |
4a3d049d | 765 | return; |
ade48088 BP |
766 | |
767 | page = mempool_alloc(gfs2_page_pool, GFP_NOIO); | |
e8c92ed7 SW |
768 | lh = page_address(page); |
769 | clear_page(lh); | |
34cc1781 | 770 | |
34cc1781 SW |
771 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); |
772 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); | |
773 | lh->lh_header.__pad0 = cpu_to_be64(0); | |
774 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
775 | lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); | |
588bff95 | 776 | lh->lh_sequence = cpu_to_be64(seq); |
34cc1781 SW |
777 | lh->lh_flags = cpu_to_be32(flags); |
778 | lh->lh_tail = cpu_to_be32(tail); | |
7c70b896 | 779 | lh->lh_blkno = cpu_to_be32(lblock); |
c1696fb8 | 780 | hash = ~crc32(~0, lh, LH_V1_SIZE); |
34cc1781 SW |
781 | lh->lh_hash = cpu_to_be32(hash); |
782 | ||
ee9c7f9a | 783 | ktime_get_coarse_real_ts64(&tv); |
c1696fb8 BP |
784 | lh->lh_nsec = cpu_to_be32(tv.tv_nsec); |
785 | lh->lh_sec = cpu_to_be64(tv.tv_sec); | |
7c70b896 | 786 | if (!list_empty(&jd->extent_list)) |
19ebc050 | 787 | dblock = gfs2_log_bmap(jd, lblock); |
7c70b896 BP |
788 | else { |
789 | int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock); | |
790 | if (gfs2_assert_withdraw(sdp, ret == 0)) | |
791 | return; | |
792 | } | |
793 | lh->lh_addr = cpu_to_be64(dblock); | |
c1696fb8 BP |
794 | lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr); |
795 | ||
796 | /* We may only write local statfs, quota, etc., when writing to our | |
797 | own journal. The values are left 0 when recovering a journal | |
798 | different from our own. */ | |
799 | if (!(flags & GFS2_LOG_HEAD_RECOVERY)) { | |
800 | lh->lh_statfs_addr = | |
801 | cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr); | |
802 | lh->lh_quota_addr = | |
803 | cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr); | |
804 | ||
805 | spin_lock(&sdp->sd_statfs_spin); | |
806 | lh->lh_local_total = cpu_to_be64(l_sc->sc_total); | |
807 | lh->lh_local_free = cpu_to_be64(l_sc->sc_free); | |
808 | lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes); | |
809 | spin_unlock(&sdp->sd_statfs_spin); | |
810 | } | |
811 | ||
812 | BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE); | |
813 | ||
814 | crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4, | |
815 | sb->s_blocksize - LH_V1_SIZE - 4); | |
816 | lh->lh_crc = cpu_to_be32(crc); | |
817 | ||
7c70b896 | 818 | gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); |
f4686c26 | 819 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags); |
588bff95 BP |
820 | } |
821 | ||
822 | /** | |
823 | * log_write_header - Get and initialize a journal header buffer | |
824 | * @sdp: The GFS2 superblock | |
c1696fb8 | 825 | * @flags: The log header flags, including log header origin |
588bff95 BP |
826 | * |
827 | * Returns: the initialized log buffer descriptor | |
828 | */ | |
829 | ||
830 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags) | |
831 | { | |
588bff95 BP |
832 | int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC; |
833 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); | |
834 | ||
835 | gfs2_assert_withdraw(sdp, (state != SFS_FROZEN)); | |
588bff95 | 836 | |
34cc1781 SW |
837 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { |
838 | gfs2_ordered_wait(sdp); | |
839 | log_flush_wait(sdp); | |
70fd7614 | 840 | op_flags = REQ_SYNC | REQ_META | REQ_PRIO; |
34cc1781 | 841 | } |
5cb738b5 AG |
842 | sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head); |
843 | gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, | |
844 | sdp->sd_log_flush_tail, sdp->sd_log_flush_head, | |
845 | flags, op_flags); | |
19ebc050 | 846 | gfs2_log_incr_head(sdp); |
4a3d049d | 847 | log_flush_wait(sdp); |
5cb738b5 | 848 | log_pull_tail(sdp); |
34cc1781 SW |
849 | } |
850 | ||
2ca0c2fb BP |
851 | /** |
852 | * ail_drain - drain the ail lists after a withdraw | |
853 | * @sdp: Pointer to GFS2 superblock | |
854 | */ | |
855 | static void ail_drain(struct gfs2_sbd *sdp) | |
856 | { | |
857 | struct gfs2_trans *tr; | |
858 | ||
859 | spin_lock(&sdp->sd_ail_lock); | |
860 | /* | |
861 | * For transactions on the sd_ail1_list we need to drain both the | |
862 | * ail1 and ail2 lists. That's because function gfs2_ail1_start_one | |
863 | * (temporarily) moves items from its tr_ail1 list to tr_ail2 list | |
864 | * before revokes are sent for that block. Items on the sd_ail2_list | |
865 | * should have already gotten beyond that point, so no need. | |
866 | */ | |
867 | while (!list_empty(&sdp->sd_ail1_list)) { | |
868 | tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans, | |
869 | tr_list); | |
870 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list); | |
871 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); | |
872 | list_del(&tr->tr_list); | |
b839dada | 873 | gfs2_trans_free(sdp, tr); |
2ca0c2fb BP |
874 | } |
875 | while (!list_empty(&sdp->sd_ail2_list)) { | |
876 | tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans, | |
877 | tr_list); | |
878 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); | |
879 | list_del(&tr->tr_list); | |
b839dada | 880 | gfs2_trans_free(sdp, tr); |
2ca0c2fb BP |
881 | } |
882 | spin_unlock(&sdp->sd_ail_lock); | |
883 | } | |
884 | ||
d5dc3d96 BP |
885 | /** |
886 | * empty_ail1_list - try to start IO and empty the ail1 list | |
887 | * @sdp: Pointer to GFS2 superblock | |
888 | */ | |
889 | static void empty_ail1_list(struct gfs2_sbd *sdp) | |
890 | { | |
891 | unsigned long start = jiffies; | |
892 | ||
893 | for (;;) { | |
894 | if (time_after(jiffies, start + (HZ * 600))) { | |
895 | fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n", | |
896 | __func__, current->journal_info ? 1 : 0); | |
897 | dump_ail_list(sdp); | |
898 | return; | |
899 | } | |
900 | gfs2_ail1_start(sdp); | |
901 | gfs2_ail1_wait(sdp); | |
902 | if (gfs2_ail1_empty(sdp, 0)) | |
903 | return; | |
904 | } | |
905 | } | |
906 | ||
462582b9 | 907 | /** |
521031fa | 908 | * trans_drain - drain the buf and databuf queue for a failed transaction |
462582b9 BP |
909 | * @tr: the transaction to drain |
910 | * | |
911 | * When this is called, we're taking an error exit for a log write that failed | |
912 | * but since we bypassed the after_commit functions, we need to remove the | |
913 | * items from the buf and databuf queue. | |
914 | */ | |
915 | static void trans_drain(struct gfs2_trans *tr) | |
916 | { | |
917 | struct gfs2_bufdata *bd; | |
918 | struct list_head *head; | |
919 | ||
920 | if (!tr) | |
921 | return; | |
922 | ||
923 | head = &tr->tr_buf; | |
924 | while (!list_empty(head)) { | |
925 | bd = list_first_entry(head, struct gfs2_bufdata, bd_list); | |
926 | list_del_init(&bd->bd_list); | |
927 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
928 | } | |
929 | head = &tr->tr_databuf; | |
930 | while (!list_empty(head)) { | |
931 | bd = list_first_entry(head, struct gfs2_bufdata, bd_list); | |
932 | list_del_init(&bd->bd_list); | |
933 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
934 | } | |
935 | } | |
936 | ||
b3b94faa | 937 | /** |
b09e593d | 938 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
939 | * @sdp: the filesystem |
940 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
805c0907 | 941 | * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags |
b3b94faa DT |
942 | * |
943 | */ | |
944 | ||
c1696fb8 | 945 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags) |
b3b94faa | 946 | { |
2ca0c2fb | 947 | struct gfs2_trans *tr = NULL; |
2e60d768 | 948 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); |
b3b94faa | 949 | |
484adff8 | 950 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 951 | |
2ca0c2fb BP |
952 | /* |
953 | * Do this check while holding the log_flush_lock to prevent new | |
954 | * buffers from being added to the ail via gfs2_pin() | |
955 | */ | |
956 | if (gfs2_withdrawn(sdp)) | |
957 | goto out; | |
958 | ||
2bcd610d | 959 | /* Log might have been flushed while we waited for the flush lock */ |
5a61ae14 AG |
960 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) |
961 | goto out; | |
805c0907 | 962 | trace_gfs2_log_flush(sdp, 1, flags); |
f55ab26a | 963 | |
c1696fb8 | 964 | if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN) |
400ac52e BM |
965 | clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
966 | ||
b1ab1e44 | 967 | sdp->sd_log_flush_head = sdp->sd_log_head; |
16ca9412 BM |
968 | tr = sdp->sd_log_tr; |
969 | if (tr) { | |
970 | sdp->sd_log_tr = NULL; | |
b1ab1e44 | 971 | tr->tr_first = sdp->sd_log_flush_head; |
2e60d768 | 972 | if (unlikely (state == SFS_FROZEN)) |
ca399c96 BP |
973 | if (gfs2_assert_withdraw_delayed(sdp, |
974 | !tr->tr_num_buf_new && !tr->tr_num_databuf_new)) | |
5a61ae14 | 975 | goto out_withdraw; |
16ca9412 | 976 | } |
b3b94faa | 977 | |
2e60d768 | 978 | if (unlikely(state == SFS_FROZEN)) |
ca399c96 | 979 | if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke)) |
5a61ae14 | 980 | goto out_withdraw; |
ca399c96 BP |
981 | if (gfs2_assert_withdraw_delayed(sdp, |
982 | sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke)) | |
5a61ae14 | 983 | goto out_withdraw; |
b3b94faa | 984 | |
d7b616e2 | 985 | gfs2_ordered_write(sdp); |
2ca0c2fb | 986 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 987 | goto out_withdraw; |
d69a3c65 | 988 | lops_before_commit(sdp, tr); |
2ca0c2fb | 989 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 990 | goto out_withdraw; |
f4686c26 | 991 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE); |
2ca0c2fb | 992 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 993 | goto out_withdraw; |
d7b616e2 | 994 | |
34cc1781 | 995 | if (sdp->sd_log_head != sdp->sd_log_flush_head) { |
428fd95d | 996 | log_flush_wait(sdp); |
c1696fb8 | 997 | log_write_header(sdp, flags); |
5cb738b5 | 998 | } else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) { |
fd041f0b | 999 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 1000 | trace_gfs2_log_blocks(sdp, -1); |
c1696fb8 | 1001 | log_write_header(sdp, flags); |
2332c443 | 1002 | } |
2ca0c2fb | 1003 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1004 | goto out_withdraw; |
16ca9412 | 1005 | lops_after_commit(sdp, tr); |
b09e593d | 1006 | |
fe1a698f | 1007 | gfs2_log_lock(sdp); |
5cb738b5 | 1008 | gfs2_log_update_head(sdp); |
faa31ce8 | 1009 | sdp->sd_log_blks_reserved = 0; |
5d439758 | 1010 | sdp->sd_log_committed_revoke = 0; |
b3b94faa | 1011 | |
d6a079e8 | 1012 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 BM |
1013 | if (tr && !list_empty(&tr->tr_ail1_list)) { |
1014 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
1015 | tr = NULL; | |
b3b94faa | 1016 | } |
d6a079e8 | 1017 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 1018 | gfs2_log_unlock(sdp); |
24972557 | 1019 | |
c1696fb8 | 1020 | if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) { |
24972557 | 1021 | if (!sdp->sd_log_idle) { |
d5dc3d96 | 1022 | empty_ail1_list(sdp); |
30fe70a8 | 1023 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1024 | goto out_withdraw; |
24972557 BM |
1025 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
1026 | trace_gfs2_log_blocks(sdp, -1); | |
c1696fb8 | 1027 | log_write_header(sdp, flags); |
5cb738b5 | 1028 | gfs2_log_update_head(sdp); |
24972557 | 1029 | } |
c1696fb8 BP |
1030 | if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
1031 | GFS2_LOG_HEAD_FLUSH_FREEZE)) | |
24972557 | 1032 | gfs2_log_shutdown(sdp); |
c1696fb8 | 1033 | if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE) |
2e60d768 | 1034 | atomic_set(&sdp->sd_freeze_state, SFS_FROZEN); |
24972557 BM |
1035 | } |
1036 | ||
5a61ae14 | 1037 | out_end: |
805c0907 | 1038 | trace_gfs2_log_flush(sdp, 0, flags); |
5a61ae14 | 1039 | out: |
484adff8 | 1040 | up_write(&sdp->sd_log_flush_lock); |
b839dada | 1041 | gfs2_trans_free(sdp, tr); |
5a61ae14 AG |
1042 | if (gfs2_withdrawing(sdp)) |
1043 | gfs2_withdraw(sdp); | |
1044 | return; | |
1045 | ||
1046 | out_withdraw: | |
1047 | trans_drain(tr); | |
1048 | /** | |
1049 | * If the tr_list is empty, we're withdrawing during a log | |
1050 | * flush that targets a transaction, but the transaction was | |
1051 | * never queued onto any of the ail lists. Here we add it to | |
1052 | * ail1 just so that ail_drain() will find and free it. | |
1053 | */ | |
1054 | spin_lock(&sdp->sd_ail_lock); | |
1055 | if (tr && list_empty(&tr->tr_list)) | |
1056 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
1057 | spin_unlock(&sdp->sd_ail_lock); | |
1058 | ail_drain(sdp); /* frees all transactions */ | |
1059 | tr = NULL; | |
1060 | goto out_end; | |
b3b94faa DT |
1061 | } |
1062 | ||
d69a3c65 SW |
1063 | /** |
1064 | * gfs2_merge_trans - Merge a new transaction into a cached transaction | |
1065 | * @old: Original transaction to be expanded | |
1066 | * @new: New transaction to be merged | |
1067 | */ | |
1068 | ||
83d060ca | 1069 | static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new) |
d69a3c65 | 1070 | { |
83d060ca BP |
1071 | struct gfs2_trans *old = sdp->sd_log_tr; |
1072 | ||
9862ca05 | 1073 | WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags)); |
d69a3c65 SW |
1074 | |
1075 | old->tr_num_buf_new += new->tr_num_buf_new; | |
1076 | old->tr_num_databuf_new += new->tr_num_databuf_new; | |
1077 | old->tr_num_buf_rm += new->tr_num_buf_rm; | |
1078 | old->tr_num_databuf_rm += new->tr_num_databuf_rm; | |
1079 | old->tr_num_revoke += new->tr_num_revoke; | |
a31b4ec5 | 1080 | old->tr_num_revoke_rm += new->tr_num_revoke_rm; |
d69a3c65 SW |
1081 | |
1082 | list_splice_tail_init(&new->tr_databuf, &old->tr_databuf); | |
1083 | list_splice_tail_init(&new->tr_buf, &old->tr_buf); | |
83d060ca BP |
1084 | |
1085 | spin_lock(&sdp->sd_ail_lock); | |
1086 | list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list); | |
1087 | list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list); | |
1088 | spin_unlock(&sdp->sd_ail_lock); | |
d69a3c65 SW |
1089 | } |
1090 | ||
b3b94faa DT |
1091 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
1092 | { | |
2332c443 | 1093 | unsigned int reserved; |
ac39aadd | 1094 | unsigned int unused; |
022ef4fe | 1095 | unsigned int maxres; |
b3b94faa DT |
1096 | |
1097 | gfs2_log_lock(sdp); | |
1098 | ||
022ef4fe | 1099 | if (sdp->sd_log_tr) { |
83d060ca | 1100 | gfs2_merge_trans(sdp, tr); |
022ef4fe | 1101 | } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) { |
c968f578 | 1102 | gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags)); |
022ef4fe | 1103 | sdp->sd_log_tr = tr; |
9862ca05 | 1104 | set_bit(TR_ATTACHED, &tr->tr_flags); |
022ef4fe SW |
1105 | } |
1106 | ||
a31b4ec5 | 1107 | sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
2332c443 | 1108 | reserved = calc_reserved(sdp); |
022ef4fe SW |
1109 | maxres = sdp->sd_log_blks_reserved + tr->tr_reserved; |
1110 | gfs2_assert_withdraw(sdp, maxres >= reserved); | |
1111 | unused = maxres - reserved; | |
5ae8fff8 AG |
1112 | if (unused) |
1113 | gfs2_log_release(sdp, unused); | |
b3b94faa DT |
1114 | sdp->sd_log_blks_reserved = reserved; |
1115 | ||
1116 | gfs2_log_unlock(sdp); | |
1117 | } | |
1118 | ||
1119 | /** | |
1120 | * gfs2_log_commit - Commit a transaction to the log | |
1121 | * @sdp: the filesystem | |
1122 | * @tr: the transaction | |
1123 | * | |
5e687eac BM |
1124 | * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 |
1125 | * or the total number of used blocks (pinned blocks plus AIL blocks) | |
1126 | * is greater than thresh2. | |
1127 | * | |
b57bc0fb | 1128 | * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of |
5e687eac BM |
1129 | * journal size. |
1130 | * | |
b3b94faa DT |
1131 | * Returns: errno |
1132 | */ | |
1133 | ||
1134 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
1135 | { | |
1136 | log_refund(sdp, tr); | |
b3b94faa | 1137 | |
5e687eac BM |
1138 | if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || |
1139 | ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > | |
1140 | atomic_read(&sdp->sd_log_thresh2))) | |
1141 | wake_up(&sdp->sd_logd_waitq); | |
b3b94faa DT |
1142 | } |
1143 | ||
1144 | /** | |
1145 | * gfs2_log_shutdown - write a shutdown header into a journal | |
1146 | * @sdp: the filesystem | |
1147 | * | |
1148 | */ | |
1149 | ||
feed98a8 | 1150 | static void gfs2_log_shutdown(struct gfs2_sbd *sdp) |
b3b94faa | 1151 | { |
b3b94faa | 1152 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa | 1153 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); |
b3b94faa DT |
1154 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); |
1155 | ||
1156 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
b3b94faa | 1157 | |
805c0907 | 1158 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN); |
b3b94faa | 1159 | |
a74604be SW |
1160 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
1161 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa | 1162 | |
5cb738b5 | 1163 | gfs2_log_update_head(sdp); |
b3b94faa | 1164 | sdp->sd_log_tail = sdp->sd_log_head; |
a25311c8 SW |
1165 | } |
1166 | ||
5e687eac BM |
1167 | static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) |
1168 | { | |
f07b3520 BP |
1169 | return (atomic_read(&sdp->sd_log_pinned) + |
1170 | atomic_read(&sdp->sd_log_blks_needed) >= | |
1171 | atomic_read(&sdp->sd_log_thresh1)); | |
5e687eac BM |
1172 | } |
1173 | ||
1174 | static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) | |
1175 | { | |
1176 | unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); | |
b066a4ee AD |
1177 | |
1178 | if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags)) | |
1179 | return 1; | |
1180 | ||
f07b3520 BP |
1181 | return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >= |
1182 | atomic_read(&sdp->sd_log_thresh2); | |
5e687eac | 1183 | } |
ec69b188 SW |
1184 | |
1185 | /** | |
1186 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
1187 | * @sdp: Pointer to GFS2 superblock | |
1188 | * | |
1189 | * Also, periodically check to make sure that we're using the most recent | |
1190 | * journal index. | |
1191 | */ | |
1192 | ||
1193 | int gfs2_logd(void *data) | |
1194 | { | |
1195 | struct gfs2_sbd *sdp = data; | |
5e687eac BM |
1196 | unsigned long t = 1; |
1197 | DEFINE_WAIT(wait); | |
ec69b188 SW |
1198 | |
1199 | while (!kthread_should_stop()) { | |
ec69b188 | 1200 | |
d22f69a0 BP |
1201 | if (gfs2_withdrawn(sdp)) { |
1202 | msleep_interruptible(HZ); | |
1203 | continue; | |
1204 | } | |
942b0cdd BP |
1205 | /* Check for errors writing to the journal */ |
1206 | if (sdp->sd_log_error) { | |
badb55ec AG |
1207 | gfs2_lm(sdp, |
1208 | "GFS2: fsid=%s: error %d: " | |
1209 | "withdrawing the file system to " | |
1210 | "prevent further damage.\n", | |
1211 | sdp->sd_fsname, sdp->sd_log_error); | |
1212 | gfs2_withdraw(sdp); | |
d22f69a0 | 1213 | continue; |
942b0cdd BP |
1214 | } |
1215 | ||
5e687eac | 1216 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { |
5e4c7632 | 1217 | gfs2_ail1_empty(sdp, 0); |
805c0907 BP |
1218 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1219 | GFS2_LFC_LOGD_JFLUSH_REQD); | |
5e687eac | 1220 | } |
ec69b188 | 1221 | |
5e687eac BM |
1222 | if (gfs2_ail_flush_reqd(sdp)) { |
1223 | gfs2_ail1_start(sdp); | |
26b06a69 | 1224 | gfs2_ail1_wait(sdp); |
5e4c7632 | 1225 | gfs2_ail1_empty(sdp, 0); |
805c0907 BP |
1226 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1227 | GFS2_LFC_LOGD_AIL_FLUSH_REQD); | |
ec69b188 SW |
1228 | } |
1229 | ||
ec69b188 | 1230 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
a0acae0e TH |
1231 | |
1232 | try_to_freeze(); | |
5e687eac BM |
1233 | |
1234 | do { | |
1235 | prepare_to_wait(&sdp->sd_logd_waitq, &wait, | |
5f487490 | 1236 | TASK_INTERRUPTIBLE); |
5e687eac BM |
1237 | if (!gfs2_ail_flush_reqd(sdp) && |
1238 | !gfs2_jrnl_flush_reqd(sdp) && | |
1239 | !kthread_should_stop()) | |
1240 | t = schedule_timeout(t); | |
1241 | } while(t && !gfs2_ail_flush_reqd(sdp) && | |
1242 | !gfs2_jrnl_flush_reqd(sdp) && | |
1243 | !kthread_should_stop()); | |
1244 | finish_wait(&sdp->sd_logd_waitq, &wait); | |
ec69b188 SW |
1245 | } |
1246 | ||
1247 | return 0; | |
1248 | } | |
1249 |