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 | ||
53 | blks = 1; | |
2e9eeaa1 | 54 | first = sdp->sd_ldptrs; |
b3b94faa DT |
55 | |
56 | if (nstruct > first) { | |
2e9eeaa1 | 57 | second = sdp->sd_inptrs; |
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 | ||
68942870 | 73 | 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, | |
69511080 | 92 | struct gfs2_trans *tr) |
d6a079e8 DC |
93 | __releases(&sdp->sd_ail_lock) |
94 | __acquires(&sdp->sd_ail_lock) | |
ddacfaf7 | 95 | { |
5ac048bb | 96 | struct gfs2_glock *gl = NULL; |
4667a0ec | 97 | struct address_space *mapping; |
ddacfaf7 SW |
98 | struct gfs2_bufdata *bd, *s; |
99 | struct buffer_head *bh; | |
b1676cbb | 100 | int ret = 0; |
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)) { |
30fe70a8 BP |
108 | if (buffer_uptodate(bh)) { |
109 | list_move(&bd->bd_ail_st_list, | |
110 | &tr->tr_ail2_list); | |
111 | continue; | |
112 | } | |
036330c9 | 113 | if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) { |
4667a0ec | 114 | gfs2_io_error_bh(sdp, bh); |
69511080 | 115 | gfs2_withdraw_delayed(sdp); |
9e1a9ecd | 116 | } |
4667a0ec SW |
117 | } |
118 | ||
30fe70a8 BP |
119 | if (gfs2_withdrawn(sdp)) { |
120 | gfs2_remove_from_ail(bd); | |
121 | continue; | |
122 | } | |
4667a0ec SW |
123 | if (!buffer_dirty(bh)) |
124 | continue; | |
125 | if (gl == bd->bd_gl) | |
126 | continue; | |
127 | gl = bd->bd_gl; | |
16ca9412 | 128 | list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list); |
4667a0ec | 129 | mapping = bh->b_page->mapping; |
4f1de018 SW |
130 | if (!mapping) |
131 | continue; | |
4667a0ec | 132 | spin_unlock(&sdp->sd_ail_lock); |
b1676cbb | 133 | ret = generic_writepages(mapping, wbc); |
4667a0ec | 134 | spin_lock(&sdp->sd_ail_lock); |
4e79e3f0 BP |
135 | if (ret == -ENODATA) /* if a jdata write into a new hole */ |
136 | ret = 0; /* ignore it */ | |
b1676cbb | 137 | if (ret || wbc->nr_to_write <= 0) |
4667a0ec | 138 | break; |
b1676cbb | 139 | return -EBUSY; |
4667a0ec | 140 | } |
4f1de018 | 141 | |
b1676cbb | 142 | return ret; |
4667a0ec | 143 | } |
ddacfaf7 | 144 | |
9592ea80 BP |
145 | static void dump_ail_list(struct gfs2_sbd *sdp) |
146 | { | |
147 | struct gfs2_trans *tr; | |
148 | struct gfs2_bufdata *bd; | |
149 | struct buffer_head *bh; | |
150 | ||
9592ea80 BP |
151 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
152 | list_for_each_entry_reverse(bd, &tr->tr_ail1_list, | |
153 | bd_ail_st_list) { | |
154 | bh = bd->bd_bh; | |
155 | fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd, | |
156 | (unsigned long long)bd->bd_blkno, bh); | |
157 | if (!bh) { | |
158 | fs_err(sdp, "\n"); | |
159 | continue; | |
160 | } | |
161 | fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d " | |
162 | "map:%d new:%d ar:%d aw:%d delay:%d " | |
163 | "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n", | |
164 | (unsigned long long)bh->b_blocknr, | |
165 | buffer_uptodate(bh), buffer_dirty(bh), | |
166 | buffer_locked(bh), buffer_req(bh), | |
167 | buffer_mapped(bh), buffer_new(bh), | |
168 | buffer_async_read(bh), buffer_async_write(bh), | |
169 | buffer_delay(bh), buffer_write_io_error(bh), | |
170 | buffer_unwritten(bh), | |
171 | buffer_defer_completion(bh), | |
172 | buffer_pinned(bh), buffer_escaped(bh)); | |
173 | } | |
174 | } | |
175 | } | |
ddacfaf7 | 176 | |
4667a0ec SW |
177 | /** |
178 | * gfs2_ail1_flush - start writeback of some ail1 entries | |
179 | * @sdp: The super block | |
180 | * @wbc: The writeback control structure | |
181 | * | |
182 | * Writes back some ail1 entries, according to the limits in the | |
183 | * writeback control structure | |
184 | */ | |
ddacfaf7 | 185 | |
4667a0ec SW |
186 | void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) |
187 | { | |
188 | struct list_head *head = &sdp->sd_ail1_list; | |
16ca9412 | 189 | struct gfs2_trans *tr; |
885bceca | 190 | struct blk_plug plug; |
75b46c43 | 191 | int ret; |
9592ea80 | 192 | unsigned long flush_start = jiffies; |
ddacfaf7 | 193 | |
c83ae9ca | 194 | trace_gfs2_ail_flush(sdp, wbc, 1); |
885bceca | 195 | blk_start_plug(&plug); |
4667a0ec | 196 | spin_lock(&sdp->sd_ail_lock); |
4f1de018 | 197 | restart: |
75b46c43 | 198 | ret = 0; |
9592ea80 | 199 | if (time_after(jiffies, flush_start + (HZ * 600))) { |
d5dc3d96 BP |
200 | fs_err(sdp, "Error: In %s for ten minutes! t=%d\n", |
201 | __func__, current->journal_info ? 1 : 0); | |
9592ea80 BP |
202 | dump_ail_list(sdp); |
203 | goto out; | |
204 | } | |
16ca9412 | 205 | list_for_each_entry_reverse(tr, head, tr_list) { |
4667a0ec | 206 | if (wbc->nr_to_write <= 0) |
ddacfaf7 | 207 | break; |
b1676cbb BP |
208 | ret = gfs2_ail1_start_one(sdp, wbc, tr); |
209 | if (ret) { | |
210 | if (ret == -EBUSY) | |
211 | goto restart; | |
212 | break; | |
213 | } | |
4667a0ec | 214 | } |
9592ea80 | 215 | out: |
4667a0ec | 216 | spin_unlock(&sdp->sd_ail_lock); |
885bceca | 217 | blk_finish_plug(&plug); |
49003128 BP |
218 | if (ret) { |
219 | gfs2_lm(sdp, "gfs2_ail1_start_one (generic_writepages) " | |
220 | "returned: %d\n", ret); | |
badb55ec | 221 | gfs2_withdraw(sdp); |
49003128 | 222 | } |
c83ae9ca | 223 | trace_gfs2_ail_flush(sdp, wbc, 0); |
4667a0ec SW |
224 | } |
225 | ||
226 | /** | |
227 | * gfs2_ail1_start - start writeback of all ail1 entries | |
228 | * @sdp: The superblock | |
229 | */ | |
230 | ||
231 | static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
232 | { | |
233 | struct writeback_control wbc = { | |
234 | .sync_mode = WB_SYNC_NONE, | |
235 | .nr_to_write = LONG_MAX, | |
236 | .range_start = 0, | |
237 | .range_end = LLONG_MAX, | |
238 | }; | |
239 | ||
240 | return gfs2_ail1_flush(sdp, &wbc); | |
ddacfaf7 SW |
241 | } |
242 | ||
243 | /** | |
244 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
245 | * @sdp: the filesystem | |
5e4c7632 BP |
246 | * @tr: the transaction |
247 | * @max_revokes: If nonzero, issue revokes for the bd items for written buffers | |
ddacfaf7 | 248 | * |
36c78309 | 249 | * returns: the transaction's count of remaining active items |
ddacfaf7 SW |
250 | */ |
251 | ||
36c78309 | 252 | static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
5e4c7632 | 253 | int *max_revokes) |
ddacfaf7 SW |
254 | { |
255 | struct gfs2_bufdata *bd, *s; | |
256 | struct buffer_head *bh; | |
36c78309 | 257 | int active_count = 0; |
ddacfaf7 | 258 | |
16ca9412 | 259 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, |
ddacfaf7 SW |
260 | bd_ail_st_list) { |
261 | bh = bd->bd_bh; | |
16ca9412 | 262 | gfs2_assert(sdp, bd->bd_tr == tr); |
036330c9 BP |
263 | /* |
264 | * If another process flagged an io error, e.g. writing to the | |
265 | * journal, error all other bhs and move them off the ail1 to | |
266 | * prevent a tight loop when unmount tries to flush ail1, | |
267 | * regardless of whether they're still busy. If no outside | |
268 | * errors were found and the buffer is busy, move to the next. | |
269 | * If the ail buffer is not busy and caught an error, flag it | |
270 | * for others. | |
271 | */ | |
36c78309 BP |
272 | if (!sdp->sd_log_error && buffer_busy(bh)) { |
273 | active_count++; | |
4667a0ec | 274 | continue; |
36c78309 | 275 | } |
b524abcc | 276 | if (!buffer_uptodate(bh) && |
036330c9 | 277 | !cmpxchg(&sdp->sd_log_error, 0, -EIO)) { |
ddacfaf7 | 278 | gfs2_io_error_bh(sdp, bh); |
69511080 | 279 | gfs2_withdraw_delayed(sdp); |
9e1a9ecd | 280 | } |
5e4c7632 BP |
281 | /* |
282 | * If we have space for revokes and the bd is no longer on any | |
283 | * buf list, we can just add a revoke for it immediately and | |
284 | * avoid having to put it on the ail2 list, where it would need | |
285 | * to be revoked later. | |
286 | */ | |
287 | if (*max_revokes && list_empty(&bd->bd_list)) { | |
288 | gfs2_add_revoke(sdp, bd); | |
289 | (*max_revokes)--; | |
290 | continue; | |
291 | } | |
16ca9412 | 292 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
ddacfaf7 | 293 | } |
36c78309 | 294 | return active_count; |
ddacfaf7 SW |
295 | } |
296 | ||
4667a0ec SW |
297 | /** |
298 | * gfs2_ail1_empty - Try to empty the ail1 lists | |
299 | * @sdp: The superblock | |
5e4c7632 | 300 | * @max_revokes: If non-zero, add revokes where appropriate |
4667a0ec SW |
301 | * |
302 | * Tries to empty the ail1 lists, starting with the oldest first | |
303 | */ | |
b3b94faa | 304 | |
5e4c7632 | 305 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes) |
b3b94faa | 306 | { |
16ca9412 | 307 | struct gfs2_trans *tr, *s; |
5d054964 | 308 | int oldest_tr = 1; |
b3b94faa DT |
309 | int ret; |
310 | ||
d6a079e8 | 311 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 312 | list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) { |
36c78309 | 313 | if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr) |
16ca9412 | 314 | list_move(&tr->tr_list, &sdp->sd_ail2_list); |
4667a0ec | 315 | else |
5d054964 | 316 | oldest_tr = 0; |
b3b94faa | 317 | } |
b3b94faa | 318 | ret = list_empty(&sdp->sd_ail1_list); |
d6a079e8 | 319 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 320 | |
69511080 | 321 | if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) { |
badb55ec AG |
322 | gfs2_lm(sdp, "fatal: I/O error(s)\n"); |
323 | gfs2_withdraw(sdp); | |
324 | } | |
9e1a9ecd | 325 | |
b3b94faa DT |
326 | return ret; |
327 | } | |
328 | ||
26b06a69 SW |
329 | static void gfs2_ail1_wait(struct gfs2_sbd *sdp) |
330 | { | |
16ca9412 | 331 | struct gfs2_trans *tr; |
26b06a69 SW |
332 | struct gfs2_bufdata *bd; |
333 | struct buffer_head *bh; | |
334 | ||
335 | spin_lock(&sdp->sd_ail_lock); | |
16ca9412 BM |
336 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
337 | list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) { | |
26b06a69 SW |
338 | bh = bd->bd_bh; |
339 | if (!buffer_locked(bh)) | |
340 | continue; | |
341 | get_bh(bh); | |
342 | spin_unlock(&sdp->sd_ail_lock); | |
343 | wait_on_buffer(bh); | |
344 | brelse(bh); | |
345 | return; | |
346 | } | |
347 | } | |
348 | spin_unlock(&sdp->sd_ail_lock); | |
349 | } | |
ddacfaf7 SW |
350 | |
351 | /** | |
2ca0c2fb | 352 | * gfs2_ail_empty_tr - empty one of the ail lists for a transaction |
ddacfaf7 SW |
353 | */ |
354 | ||
2ca0c2fb BP |
355 | static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
356 | struct list_head *head) | |
ddacfaf7 | 357 | { |
ddacfaf7 SW |
358 | struct gfs2_bufdata *bd; |
359 | ||
360 | while (!list_empty(head)) { | |
2ca0c2fb BP |
361 | bd = list_first_entry(head, struct gfs2_bufdata, |
362 | bd_ail_st_list); | |
16ca9412 | 363 | gfs2_assert(sdp, bd->bd_tr == tr); |
f91a0d3e | 364 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
365 | } |
366 | } | |
367 | ||
b3b94faa DT |
368 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
369 | { | |
16ca9412 | 370 | struct gfs2_trans *tr, *safe; |
b3b94faa DT |
371 | unsigned int old_tail = sdp->sd_log_tail; |
372 | int wrap = (new_tail < old_tail); | |
373 | int a, b, rm; | |
374 | ||
d6a079e8 | 375 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 376 | |
16ca9412 BM |
377 | list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) { |
378 | a = (old_tail <= tr->tr_first); | |
379 | b = (tr->tr_first < new_tail); | |
b3b94faa DT |
380 | rm = (wrap) ? (a || b) : (a && b); |
381 | if (!rm) | |
382 | continue; | |
383 | ||
2ca0c2fb | 384 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); |
16ca9412 BM |
385 | list_del(&tr->tr_list); |
386 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list)); | |
387 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list)); | |
b839dada | 388 | gfs2_trans_free(sdp, tr); |
b3b94faa DT |
389 | } |
390 | ||
d6a079e8 | 391 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
392 | } |
393 | ||
24972557 BM |
394 | /** |
395 | * gfs2_log_release - Release a given number of log blocks | |
396 | * @sdp: The GFS2 superblock | |
397 | * @blks: The number of blocks | |
398 | * | |
399 | */ | |
400 | ||
401 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
402 | { | |
403 | ||
404 | atomic_add(blks, &sdp->sd_log_blks_free); | |
405 | trace_gfs2_log_blocks(sdp, blks); | |
406 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= | |
407 | sdp->sd_jdesc->jd_blocks); | |
408 | up_read(&sdp->sd_log_flush_lock); | |
409 | } | |
410 | ||
b3b94faa DT |
411 | /** |
412 | * gfs2_log_reserve - Make a log reservation | |
413 | * @sdp: The GFS2 superblock | |
414 | * @blks: The number of blocks to reserve | |
415 | * | |
89918647 | 416 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 417 | * due to the fact that there is a small number of header blocks |
b004157a SW |
418 | * associated with each log flush. The exact number can't be known until |
419 | * flush time, so we ensure that we have just enough free blocks at all | |
420 | * times to avoid running out during a log flush. | |
421 | * | |
5e687eac BM |
422 | * We no longer flush the log here, instead we wake up logd to do that |
423 | * for us. To avoid the thundering herd and to ensure that we deal fairly | |
424 | * with queued waiters, we use an exclusive wait. This means that when we | |
425 | * get woken with enough journal space to get our reservation, we need to | |
426 | * wake the next waiter on the list. | |
427 | * | |
b3b94faa DT |
428 | * Returns: errno |
429 | */ | |
430 | ||
431 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
432 | { | |
2e60d768 | 433 | int ret = 0; |
5d054964 | 434 | unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize); |
5e687eac BM |
435 | unsigned wanted = blks + reserved_blks; |
436 | DEFINE_WAIT(wait); | |
437 | int did_wait = 0; | |
438 | unsigned int free_blocks; | |
b3b94faa DT |
439 | |
440 | if (gfs2_assert_warn(sdp, blks) || | |
441 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
442 | return -EINVAL; | |
f07b3520 | 443 | atomic_add(blks, &sdp->sd_log_blks_needed); |
5e687eac BM |
444 | retry: |
445 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
446 | if (unlikely(free_blocks <= wanted)) { | |
447 | do { | |
448 | prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait, | |
449 | TASK_UNINTERRUPTIBLE); | |
450 | wake_up(&sdp->sd_logd_waitq); | |
451 | did_wait = 1; | |
452 | if (atomic_read(&sdp->sd_log_blks_free) <= wanted) | |
453 | io_schedule(); | |
454 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
455 | } while(free_blocks <= wanted); | |
456 | finish_wait(&sdp->sd_log_waitq, &wait); | |
b3b94faa | 457 | } |
2e60d768 | 458 | atomic_inc(&sdp->sd_reserving_log); |
5e687eac | 459 | if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks, |
2e60d768 BM |
460 | free_blocks - blks) != free_blocks) { |
461 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) | |
462 | wake_up(&sdp->sd_reserving_log_wait); | |
5e687eac | 463 | goto retry; |
2e60d768 | 464 | } |
f07b3520 | 465 | atomic_sub(blks, &sdp->sd_log_blks_needed); |
63997775 | 466 | trace_gfs2_log_blocks(sdp, -blks); |
5e687eac BM |
467 | |
468 | /* | |
469 | * If we waited, then so might others, wake them up _after_ we get | |
470 | * our share of the log. | |
471 | */ | |
472 | if (unlikely(did_wait)) | |
473 | wake_up(&sdp->sd_log_waitq); | |
484adff8 SW |
474 | |
475 | down_read(&sdp->sd_log_flush_lock); | |
24972557 BM |
476 | if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { |
477 | gfs2_log_release(sdp, blks); | |
2e60d768 | 478 | ret = -EROFS; |
24972557 | 479 | } |
2e60d768 BM |
480 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) |
481 | wake_up(&sdp->sd_reserving_log_wait); | |
482 | return ret; | |
b3b94faa DT |
483 | } |
484 | ||
b3b94faa DT |
485 | /** |
486 | * log_distance - Compute distance between two journal blocks | |
487 | * @sdp: The GFS2 superblock | |
488 | * @newer: The most recent journal block of the pair | |
489 | * @older: The older journal block of the pair | |
490 | * | |
491 | * Compute the distance (in the journal direction) between two | |
492 | * blocks in the journal | |
493 | * | |
494 | * Returns: the distance in blocks | |
495 | */ | |
496 | ||
faa31ce8 | 497 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
498 | unsigned int older) |
499 | { | |
500 | int dist; | |
501 | ||
502 | dist = newer - older; | |
503 | if (dist < 0) | |
504 | dist += sdp->sd_jdesc->jd_blocks; | |
505 | ||
506 | return dist; | |
507 | } | |
508 | ||
2332c443 RP |
509 | /** |
510 | * calc_reserved - Calculate the number of blocks to reserve when | |
511 | * refunding a transaction's unused buffers. | |
512 | * @sdp: The GFS2 superblock | |
513 | * | |
514 | * This is complex. We need to reserve room for all our currently used | |
515 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
516 | * all our journaled data buffers for journaled files (e.g. files in the | |
517 | * meta_fs like rindex, or files for which chattr +j was done.) | |
518 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
519 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
520 | * | |
521 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
522 | * type gets its own log header, for which we need to reserve a block. | |
523 | * In fact, each type has the potential for needing more than one header | |
524 | * in cases where we have more buffers than will fit on a journal page. | |
525 | * Metadata journal entries take up half the space of journaled buffer entries. | |
526 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
527 | * databuf_limit (251) before they cause a wrap around. | |
528 | * | |
529 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
530 | * overall header for the lot. | |
531 | * | |
532 | * Returns: the number of blocks reserved | |
533 | */ | |
534 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
535 | { | |
536 | unsigned int reserved = 0; | |
022ef4fe SW |
537 | unsigned int mbuf; |
538 | unsigned int dbuf; | |
539 | struct gfs2_trans *tr = sdp->sd_log_tr; | |
2332c443 | 540 | |
022ef4fe SW |
541 | if (tr) { |
542 | mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
543 | dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; | |
544 | reserved = mbuf + dbuf; | |
545 | /* Account for header blocks */ | |
546 | reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp)); | |
547 | reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp)); | |
548 | } | |
2332c443 | 549 | |
5d439758 AG |
550 | if (sdp->sd_log_committed_revoke > 0) |
551 | reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke); | |
2332c443 RP |
552 | /* One for the overall header */ |
553 | if (reserved) | |
554 | reserved++; | |
555 | return reserved; | |
556 | } | |
557 | ||
b3b94faa DT |
558 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
559 | { | |
16ca9412 | 560 | struct gfs2_trans *tr; |
b3b94faa DT |
561 | unsigned int tail; |
562 | ||
d6a079e8 | 563 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 564 | |
faa31ce8 | 565 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 566 | tail = sdp->sd_log_head; |
faa31ce8 | 567 | } else { |
969183bc | 568 | tr = list_last_entry(&sdp->sd_ail1_list, struct gfs2_trans, |
16ca9412 BM |
569 | tr_list); |
570 | tail = tr->tr_first; | |
b3b94faa DT |
571 | } |
572 | ||
d6a079e8 | 573 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
574 | |
575 | return tail; | |
576 | } | |
577 | ||
2332c443 | 578 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
579 | { |
580 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
581 | ||
582 | ail2_empty(sdp, new_tail); | |
583 | ||
fd041f0b | 584 | atomic_add(dist, &sdp->sd_log_blks_free); |
63997775 | 585 | trace_gfs2_log_blocks(sdp, dist); |
5e687eac BM |
586 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
587 | sdp->sd_jdesc->jd_blocks); | |
b3b94faa DT |
588 | |
589 | sdp->sd_log_tail = new_tail; | |
590 | } | |
591 | ||
b3b94faa | 592 | |
9ff78289 | 593 | void log_flush_wait(struct gfs2_sbd *sdp) |
b3b94faa | 594 | { |
16615be1 SW |
595 | DEFINE_WAIT(wait); |
596 | ||
597 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
598 | do { | |
599 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
600 | TASK_UNINTERRUPTIBLE); | |
601 | if (atomic_read(&sdp->sd_log_in_flight)) | |
602 | io_schedule(); | |
603 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
604 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa | 605 | } |
b3b94faa DT |
606 | } |
607 | ||
45138990 | 608 | static int ip_cmp(void *priv, struct list_head *a, struct list_head *b) |
4a36d08d | 609 | { |
45138990 | 610 | struct gfs2_inode *ipa, *ipb; |
4a36d08d | 611 | |
45138990 SW |
612 | ipa = list_entry(a, struct gfs2_inode, i_ordered); |
613 | ipb = list_entry(b, struct gfs2_inode, i_ordered); | |
4a36d08d | 614 | |
45138990 | 615 | if (ipa->i_no_addr < ipb->i_no_addr) |
4a36d08d | 616 | return -1; |
45138990 | 617 | if (ipa->i_no_addr > ipb->i_no_addr) |
4a36d08d BP |
618 | return 1; |
619 | return 0; | |
620 | } | |
621 | ||
7542486b BP |
622 | static void __ordered_del_inode(struct gfs2_inode *ip) |
623 | { | |
624 | if (!list_empty(&ip->i_ordered)) | |
625 | list_del_init(&ip->i_ordered); | |
626 | } | |
627 | ||
d7b616e2 SW |
628 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
629 | { | |
45138990 | 630 | struct gfs2_inode *ip; |
d7b616e2 SW |
631 | LIST_HEAD(written); |
632 | ||
45138990 | 633 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
634 | list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp); |
635 | while (!list_empty(&sdp->sd_log_ordered)) { | |
969183bc | 636 | ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered); |
1f23bc78 | 637 | if (ip->i_inode.i_mapping->nrpages == 0) { |
7542486b | 638 | __ordered_del_inode(ip); |
d7b616e2 | 639 | continue; |
1f23bc78 AD |
640 | } |
641 | list_move(&ip->i_ordered, &written); | |
45138990 SW |
642 | spin_unlock(&sdp->sd_ordered_lock); |
643 | filemap_fdatawrite(ip->i_inode.i_mapping); | |
644 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 645 | } |
a5b1d3fc | 646 | list_splice(&written, &sdp->sd_log_ordered); |
45138990 | 647 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
648 | } |
649 | ||
650 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
651 | { | |
45138990 | 652 | struct gfs2_inode *ip; |
d7b616e2 | 653 | |
45138990 | 654 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc | 655 | while (!list_empty(&sdp->sd_log_ordered)) { |
969183bc | 656 | ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered); |
7542486b | 657 | __ordered_del_inode(ip); |
45138990 | 658 | if (ip->i_inode.i_mapping->nrpages == 0) |
d7b616e2 | 659 | continue; |
45138990 SW |
660 | spin_unlock(&sdp->sd_ordered_lock); |
661 | filemap_fdatawait(ip->i_inode.i_mapping); | |
662 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 663 | } |
45138990 SW |
664 | spin_unlock(&sdp->sd_ordered_lock); |
665 | } | |
666 | ||
667 | void gfs2_ordered_del_inode(struct gfs2_inode *ip) | |
668 | { | |
669 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
670 | ||
671 | spin_lock(&sdp->sd_ordered_lock); | |
7542486b | 672 | __ordered_del_inode(ip); |
45138990 | 673 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
674 | } |
675 | ||
5d054964 BM |
676 | void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) |
677 | { | |
678 | struct buffer_head *bh = bd->bd_bh; | |
679 | struct gfs2_glock *gl = bd->bd_gl; | |
680 | ||
f4e2f5e1 AG |
681 | sdp->sd_log_num_revoke++; |
682 | if (atomic_inc_return(&gl->gl_revokes) == 1) | |
683 | gfs2_glock_hold(gl); | |
5d054964 BM |
684 | bh->b_private = NULL; |
685 | bd->bd_blkno = bh->b_blocknr; | |
9290a9a7 BP |
686 | gfs2_remove_from_ail(bd); /* drops ref on bh */ |
687 | bd->bd_bh = NULL; | |
5d054964 | 688 | set_bit(GLF_LFLUSH, &gl->gl_flags); |
a5b1d3fc | 689 | list_add(&bd->bd_list, &sdp->sd_log_revokes); |
5d054964 BM |
690 | } |
691 | ||
fe5e7ba1 BP |
692 | void gfs2_glock_remove_revoke(struct gfs2_glock *gl) |
693 | { | |
694 | if (atomic_dec_return(&gl->gl_revokes) == 0) { | |
695 | clear_bit(GLF_LFLUSH, &gl->gl_flags); | |
696 | gfs2_glock_queue_put(gl); | |
697 | } | |
698 | } | |
699 | ||
5e4c7632 BP |
700 | /** |
701 | * gfs2_write_revokes - Add as many revokes to the system transaction as we can | |
702 | * @sdp: The GFS2 superblock | |
703 | * | |
704 | * Our usual strategy is to defer writing revokes as much as we can in the hope | |
705 | * that we'll eventually overwrite the journal, which will make those revokes | |
706 | * go away. This changes when we flush the log: at that point, there will | |
707 | * likely be some left-over space in the last revoke block of that transaction. | |
708 | * We can fill that space with additional revokes for blocks that have already | |
709 | * been written back. This will basically come at no cost now, and will save | |
710 | * us from having to keep track of those blocks on the AIL2 list later. | |
711 | */ | |
5d054964 BM |
712 | void gfs2_write_revokes(struct gfs2_sbd *sdp) |
713 | { | |
5e4c7632 | 714 | /* number of revokes we still have room for */ |
5d054964 BM |
715 | int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64); |
716 | ||
5e4c7632 | 717 | gfs2_log_lock(sdp); |
5d054964 BM |
718 | while (sdp->sd_log_num_revoke > max_revokes) |
719 | max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64); | |
720 | max_revokes -= sdp->sd_log_num_revoke; | |
721 | if (!sdp->sd_log_num_revoke) { | |
722 | atomic_dec(&sdp->sd_log_blks_free); | |
723 | /* If no blocks have been reserved, we need to also | |
724 | * reserve a block for the header */ | |
77650bdb | 725 | if (!sdp->sd_log_blks_reserved) { |
5d054964 | 726 | atomic_dec(&sdp->sd_log_blks_free); |
77650bdb BP |
727 | trace_gfs2_log_blocks(sdp, -2); |
728 | } else { | |
729 | trace_gfs2_log_blocks(sdp, -1); | |
730 | } | |
5d054964 | 731 | } |
5e4c7632 | 732 | gfs2_ail1_empty(sdp, max_revokes); |
5d054964 BM |
733 | gfs2_log_unlock(sdp); |
734 | ||
735 | if (!sdp->sd_log_num_revoke) { | |
736 | atomic_inc(&sdp->sd_log_blks_free); | |
77650bdb | 737 | if (!sdp->sd_log_blks_reserved) { |
5d054964 | 738 | atomic_inc(&sdp->sd_log_blks_free); |
77650bdb BP |
739 | trace_gfs2_log_blocks(sdp, 2); |
740 | } else { | |
741 | trace_gfs2_log_blocks(sdp, 1); | |
742 | } | |
5d054964 BM |
743 | } |
744 | } | |
745 | ||
34cc1781 | 746 | /** |
7c70b896 | 747 | * gfs2_write_log_header - Write a journal log header buffer at lblock |
34cc1781 | 748 | * @sdp: The GFS2 superblock |
c1696fb8 | 749 | * @jd: journal descriptor of the journal to which we are writing |
588bff95 BP |
750 | * @seq: sequence number |
751 | * @tail: tail of the log | |
7c70b896 | 752 | * @lblock: value for lh_blkno (block number relative to start of journal) |
c1696fb8 | 753 | * @flags: log header flags GFS2_LOG_HEAD_* |
588bff95 | 754 | * @op_flags: flags to pass to the bio |
34cc1781 SW |
755 | * |
756 | * Returns: the initialized log buffer descriptor | |
757 | */ | |
758 | ||
c1696fb8 | 759 | void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, |
7c70b896 BP |
760 | u64 seq, u32 tail, u32 lblock, u32 flags, |
761 | int op_flags) | |
34cc1781 | 762 | { |
34cc1781 | 763 | struct gfs2_log_header *lh; |
c1696fb8 | 764 | u32 hash, crc; |
ade48088 | 765 | struct page *page; |
c1696fb8 BP |
766 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
767 | struct timespec64 tv; | |
768 | struct super_block *sb = sdp->sd_vfs; | |
7c70b896 | 769 | u64 dblock; |
588bff95 | 770 | |
ade48088 BP |
771 | if (gfs2_withdrawn(sdp)) |
772 | goto out; | |
773 | ||
774 | page = mempool_alloc(gfs2_page_pool, GFP_NOIO); | |
e8c92ed7 SW |
775 | lh = page_address(page); |
776 | clear_page(lh); | |
34cc1781 | 777 | |
34cc1781 SW |
778 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); |
779 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); | |
780 | lh->lh_header.__pad0 = cpu_to_be64(0); | |
781 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
782 | lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); | |
588bff95 | 783 | lh->lh_sequence = cpu_to_be64(seq); |
34cc1781 SW |
784 | lh->lh_flags = cpu_to_be32(flags); |
785 | lh->lh_tail = cpu_to_be32(tail); | |
7c70b896 | 786 | lh->lh_blkno = cpu_to_be32(lblock); |
c1696fb8 | 787 | hash = ~crc32(~0, lh, LH_V1_SIZE); |
34cc1781 SW |
788 | lh->lh_hash = cpu_to_be32(hash); |
789 | ||
ee9c7f9a | 790 | ktime_get_coarse_real_ts64(&tv); |
c1696fb8 BP |
791 | lh->lh_nsec = cpu_to_be32(tv.tv_nsec); |
792 | lh->lh_sec = cpu_to_be64(tv.tv_sec); | |
7c70b896 | 793 | if (!list_empty(&jd->extent_list)) |
19ebc050 | 794 | dblock = gfs2_log_bmap(jd, lblock); |
7c70b896 BP |
795 | else { |
796 | int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock); | |
797 | if (gfs2_assert_withdraw(sdp, ret == 0)) | |
798 | return; | |
799 | } | |
800 | lh->lh_addr = cpu_to_be64(dblock); | |
c1696fb8 BP |
801 | lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr); |
802 | ||
803 | /* We may only write local statfs, quota, etc., when writing to our | |
804 | own journal. The values are left 0 when recovering a journal | |
805 | different from our own. */ | |
806 | if (!(flags & GFS2_LOG_HEAD_RECOVERY)) { | |
807 | lh->lh_statfs_addr = | |
808 | cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr); | |
809 | lh->lh_quota_addr = | |
810 | cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr); | |
811 | ||
812 | spin_lock(&sdp->sd_statfs_spin); | |
813 | lh->lh_local_total = cpu_to_be64(l_sc->sc_total); | |
814 | lh->lh_local_free = cpu_to_be64(l_sc->sc_free); | |
815 | lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes); | |
816 | spin_unlock(&sdp->sd_statfs_spin); | |
817 | } | |
818 | ||
819 | BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE); | |
820 | ||
821 | crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4, | |
822 | sb->s_blocksize - LH_V1_SIZE - 4); | |
823 | lh->lh_crc = cpu_to_be32(crc); | |
824 | ||
7c70b896 | 825 | gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); |
f4686c26 | 826 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags); |
ade48088 | 827 | out: |
588bff95 BP |
828 | log_flush_wait(sdp); |
829 | } | |
830 | ||
831 | /** | |
832 | * log_write_header - Get and initialize a journal header buffer | |
833 | * @sdp: The GFS2 superblock | |
c1696fb8 | 834 | * @flags: The log header flags, including log header origin |
588bff95 BP |
835 | * |
836 | * Returns: the initialized log buffer descriptor | |
837 | */ | |
838 | ||
839 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags) | |
840 | { | |
841 | unsigned int tail; | |
842 | int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC; | |
843 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); | |
844 | ||
845 | gfs2_assert_withdraw(sdp, (state != SFS_FROZEN)); | |
846 | tail = current_tail(sdp); | |
847 | ||
34cc1781 SW |
848 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { |
849 | gfs2_ordered_wait(sdp); | |
850 | log_flush_wait(sdp); | |
70fd7614 | 851 | op_flags = REQ_SYNC | REQ_META | REQ_PRIO; |
34cc1781 | 852 | } |
e8c92ed7 | 853 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); |
c1696fb8 | 854 | gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail, |
7c70b896 | 855 | sdp->sd_log_flush_head, flags, op_flags); |
19ebc050 | 856 | gfs2_log_incr_head(sdp); |
34cc1781 SW |
857 | |
858 | if (sdp->sd_log_tail != tail) | |
859 | log_pull_tail(sdp, tail); | |
34cc1781 SW |
860 | } |
861 | ||
2ca0c2fb BP |
862 | /** |
863 | * ail_drain - drain the ail lists after a withdraw | |
864 | * @sdp: Pointer to GFS2 superblock | |
865 | */ | |
866 | static void ail_drain(struct gfs2_sbd *sdp) | |
867 | { | |
868 | struct gfs2_trans *tr; | |
869 | ||
870 | spin_lock(&sdp->sd_ail_lock); | |
871 | /* | |
872 | * For transactions on the sd_ail1_list we need to drain both the | |
873 | * ail1 and ail2 lists. That's because function gfs2_ail1_start_one | |
874 | * (temporarily) moves items from its tr_ail1 list to tr_ail2 list | |
875 | * before revokes are sent for that block. Items on the sd_ail2_list | |
876 | * should have already gotten beyond that point, so no need. | |
877 | */ | |
878 | while (!list_empty(&sdp->sd_ail1_list)) { | |
879 | tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans, | |
880 | tr_list); | |
881 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list); | |
882 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); | |
883 | list_del(&tr->tr_list); | |
b839dada | 884 | gfs2_trans_free(sdp, tr); |
2ca0c2fb BP |
885 | } |
886 | while (!list_empty(&sdp->sd_ail2_list)) { | |
887 | tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans, | |
888 | tr_list); | |
889 | gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list); | |
890 | list_del(&tr->tr_list); | |
b839dada | 891 | gfs2_trans_free(sdp, tr); |
2ca0c2fb BP |
892 | } |
893 | spin_unlock(&sdp->sd_ail_lock); | |
894 | } | |
895 | ||
d5dc3d96 BP |
896 | /** |
897 | * empty_ail1_list - try to start IO and empty the ail1 list | |
898 | * @sdp: Pointer to GFS2 superblock | |
899 | */ | |
900 | static void empty_ail1_list(struct gfs2_sbd *sdp) | |
901 | { | |
902 | unsigned long start = jiffies; | |
903 | ||
904 | for (;;) { | |
905 | if (time_after(jiffies, start + (HZ * 600))) { | |
906 | fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n", | |
907 | __func__, current->journal_info ? 1 : 0); | |
908 | dump_ail_list(sdp); | |
909 | return; | |
910 | } | |
911 | gfs2_ail1_start(sdp); | |
912 | gfs2_ail1_wait(sdp); | |
913 | if (gfs2_ail1_empty(sdp, 0)) | |
914 | return; | |
915 | } | |
916 | } | |
917 | ||
462582b9 | 918 | /** |
521031fa | 919 | * trans_drain - drain the buf and databuf queue for a failed transaction |
462582b9 BP |
920 | * @tr: the transaction to drain |
921 | * | |
922 | * When this is called, we're taking an error exit for a log write that failed | |
923 | * but since we bypassed the after_commit functions, we need to remove the | |
924 | * items from the buf and databuf queue. | |
925 | */ | |
926 | static void trans_drain(struct gfs2_trans *tr) | |
927 | { | |
928 | struct gfs2_bufdata *bd; | |
929 | struct list_head *head; | |
930 | ||
931 | if (!tr) | |
932 | return; | |
933 | ||
934 | head = &tr->tr_buf; | |
935 | while (!list_empty(head)) { | |
936 | bd = list_first_entry(head, struct gfs2_bufdata, bd_list); | |
937 | list_del_init(&bd->bd_list); | |
938 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
939 | } | |
940 | head = &tr->tr_databuf; | |
941 | while (!list_empty(head)) { | |
942 | bd = list_first_entry(head, struct gfs2_bufdata, bd_list); | |
943 | list_del_init(&bd->bd_list); | |
944 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
945 | } | |
946 | } | |
947 | ||
b3b94faa | 948 | /** |
b09e593d | 949 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
950 | * @sdp: the filesystem |
951 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
805c0907 | 952 | * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags |
b3b94faa DT |
953 | * |
954 | */ | |
955 | ||
c1696fb8 | 956 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags) |
b3b94faa | 957 | { |
2ca0c2fb | 958 | struct gfs2_trans *tr = NULL; |
2e60d768 | 959 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); |
b3b94faa | 960 | |
484adff8 | 961 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 962 | |
2ca0c2fb BP |
963 | /* |
964 | * Do this check while holding the log_flush_lock to prevent new | |
965 | * buffers from being added to the ail via gfs2_pin() | |
966 | */ | |
967 | if (gfs2_withdrawn(sdp)) | |
968 | goto out; | |
969 | ||
2bcd610d | 970 | /* Log might have been flushed while we waited for the flush lock */ |
5a61ae14 AG |
971 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) |
972 | goto out; | |
805c0907 | 973 | trace_gfs2_log_flush(sdp, 1, flags); |
f55ab26a | 974 | |
c1696fb8 | 975 | if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN) |
400ac52e BM |
976 | clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
977 | ||
b1ab1e44 | 978 | sdp->sd_log_flush_head = sdp->sd_log_head; |
16ca9412 BM |
979 | tr = sdp->sd_log_tr; |
980 | if (tr) { | |
981 | sdp->sd_log_tr = NULL; | |
b1ab1e44 | 982 | tr->tr_first = sdp->sd_log_flush_head; |
2e60d768 | 983 | if (unlikely (state == SFS_FROZEN)) |
ca399c96 BP |
984 | if (gfs2_assert_withdraw_delayed(sdp, |
985 | !tr->tr_num_buf_new && !tr->tr_num_databuf_new)) | |
5a61ae14 | 986 | goto out_withdraw; |
16ca9412 | 987 | } |
b3b94faa | 988 | |
2e60d768 | 989 | if (unlikely(state == SFS_FROZEN)) |
ca399c96 | 990 | if (gfs2_assert_withdraw_delayed(sdp, !sdp->sd_log_num_revoke)) |
5a61ae14 | 991 | goto out_withdraw; |
ca399c96 BP |
992 | if (gfs2_assert_withdraw_delayed(sdp, |
993 | sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke)) | |
5a61ae14 | 994 | goto out_withdraw; |
b3b94faa | 995 | |
d7b616e2 | 996 | gfs2_ordered_write(sdp); |
2ca0c2fb | 997 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 998 | goto out_withdraw; |
d69a3c65 | 999 | lops_before_commit(sdp, tr); |
2ca0c2fb | 1000 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1001 | goto out_withdraw; |
f4686c26 | 1002 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE); |
2ca0c2fb | 1003 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1004 | goto out_withdraw; |
d7b616e2 | 1005 | |
34cc1781 | 1006 | if (sdp->sd_log_head != sdp->sd_log_flush_head) { |
428fd95d | 1007 | log_flush_wait(sdp); |
c1696fb8 | 1008 | log_write_header(sdp, flags); |
34cc1781 | 1009 | } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
fd041f0b | 1010 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 1011 | trace_gfs2_log_blocks(sdp, -1); |
c1696fb8 | 1012 | log_write_header(sdp, flags); |
2332c443 | 1013 | } |
2ca0c2fb | 1014 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1015 | goto out_withdraw; |
16ca9412 | 1016 | lops_after_commit(sdp, tr); |
b09e593d | 1017 | |
fe1a698f SW |
1018 | gfs2_log_lock(sdp); |
1019 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 | 1020 | sdp->sd_log_blks_reserved = 0; |
5d439758 | 1021 | sdp->sd_log_committed_revoke = 0; |
b3b94faa | 1022 | |
d6a079e8 | 1023 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 BM |
1024 | if (tr && !list_empty(&tr->tr_ail1_list)) { |
1025 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
1026 | tr = NULL; | |
b3b94faa | 1027 | } |
d6a079e8 | 1028 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 1029 | gfs2_log_unlock(sdp); |
24972557 | 1030 | |
c1696fb8 | 1031 | if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) { |
24972557 | 1032 | if (!sdp->sd_log_idle) { |
d5dc3d96 | 1033 | empty_ail1_list(sdp); |
30fe70a8 | 1034 | if (gfs2_withdrawn(sdp)) |
5a61ae14 | 1035 | goto out_withdraw; |
24972557 BM |
1036 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
1037 | trace_gfs2_log_blocks(sdp, -1); | |
c1696fb8 | 1038 | log_write_header(sdp, flags); |
24972557 BM |
1039 | sdp->sd_log_head = sdp->sd_log_flush_head; |
1040 | } | |
c1696fb8 BP |
1041 | if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
1042 | GFS2_LOG_HEAD_FLUSH_FREEZE)) | |
24972557 | 1043 | gfs2_log_shutdown(sdp); |
c1696fb8 | 1044 | if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE) |
2e60d768 | 1045 | atomic_set(&sdp->sd_freeze_state, SFS_FROZEN); |
24972557 BM |
1046 | } |
1047 | ||
5a61ae14 | 1048 | out_end: |
805c0907 | 1049 | trace_gfs2_log_flush(sdp, 0, flags); |
5a61ae14 | 1050 | out: |
484adff8 | 1051 | up_write(&sdp->sd_log_flush_lock); |
b839dada | 1052 | gfs2_trans_free(sdp, tr); |
5a61ae14 AG |
1053 | if (gfs2_withdrawing(sdp)) |
1054 | gfs2_withdraw(sdp); | |
1055 | return; | |
1056 | ||
1057 | out_withdraw: | |
1058 | trans_drain(tr); | |
1059 | /** | |
1060 | * If the tr_list is empty, we're withdrawing during a log | |
1061 | * flush that targets a transaction, but the transaction was | |
1062 | * never queued onto any of the ail lists. Here we add it to | |
1063 | * ail1 just so that ail_drain() will find and free it. | |
1064 | */ | |
1065 | spin_lock(&sdp->sd_ail_lock); | |
1066 | if (tr && list_empty(&tr->tr_list)) | |
1067 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
1068 | spin_unlock(&sdp->sd_ail_lock); | |
1069 | ail_drain(sdp); /* frees all transactions */ | |
1070 | tr = NULL; | |
1071 | goto out_end; | |
b3b94faa DT |
1072 | } |
1073 | ||
d69a3c65 SW |
1074 | /** |
1075 | * gfs2_merge_trans - Merge a new transaction into a cached transaction | |
1076 | * @old: Original transaction to be expanded | |
1077 | * @new: New transaction to be merged | |
1078 | */ | |
1079 | ||
83d060ca | 1080 | static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new) |
d69a3c65 | 1081 | { |
83d060ca BP |
1082 | struct gfs2_trans *old = sdp->sd_log_tr; |
1083 | ||
9862ca05 | 1084 | WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags)); |
d69a3c65 SW |
1085 | |
1086 | old->tr_num_buf_new += new->tr_num_buf_new; | |
1087 | old->tr_num_databuf_new += new->tr_num_databuf_new; | |
1088 | old->tr_num_buf_rm += new->tr_num_buf_rm; | |
1089 | old->tr_num_databuf_rm += new->tr_num_databuf_rm; | |
1090 | old->tr_num_revoke += new->tr_num_revoke; | |
a31b4ec5 | 1091 | old->tr_num_revoke_rm += new->tr_num_revoke_rm; |
d69a3c65 SW |
1092 | |
1093 | list_splice_tail_init(&new->tr_databuf, &old->tr_databuf); | |
1094 | list_splice_tail_init(&new->tr_buf, &old->tr_buf); | |
83d060ca BP |
1095 | |
1096 | spin_lock(&sdp->sd_ail_lock); | |
1097 | list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list); | |
1098 | list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list); | |
1099 | spin_unlock(&sdp->sd_ail_lock); | |
d69a3c65 SW |
1100 | } |
1101 | ||
b3b94faa DT |
1102 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
1103 | { | |
2332c443 | 1104 | unsigned int reserved; |
ac39aadd | 1105 | unsigned int unused; |
022ef4fe | 1106 | unsigned int maxres; |
b3b94faa DT |
1107 | |
1108 | gfs2_log_lock(sdp); | |
1109 | ||
022ef4fe | 1110 | if (sdp->sd_log_tr) { |
83d060ca | 1111 | gfs2_merge_trans(sdp, tr); |
022ef4fe | 1112 | } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) { |
9862ca05 | 1113 | gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags)); |
022ef4fe | 1114 | sdp->sd_log_tr = tr; |
9862ca05 | 1115 | set_bit(TR_ATTACHED, &tr->tr_flags); |
022ef4fe SW |
1116 | } |
1117 | ||
a31b4ec5 | 1118 | sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
2332c443 | 1119 | reserved = calc_reserved(sdp); |
022ef4fe SW |
1120 | maxres = sdp->sd_log_blks_reserved + tr->tr_reserved; |
1121 | gfs2_assert_withdraw(sdp, maxres >= reserved); | |
1122 | unused = maxres - reserved; | |
ac39aadd | 1123 | atomic_add(unused, &sdp->sd_log_blks_free); |
63997775 | 1124 | trace_gfs2_log_blocks(sdp, unused); |
fd041f0b | 1125 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
2332c443 | 1126 | sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
1127 | sdp->sd_log_blks_reserved = reserved; |
1128 | ||
1129 | gfs2_log_unlock(sdp); | |
1130 | } | |
1131 | ||
1132 | /** | |
1133 | * gfs2_log_commit - Commit a transaction to the log | |
1134 | * @sdp: the filesystem | |
1135 | * @tr: the transaction | |
1136 | * | |
5e687eac BM |
1137 | * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 |
1138 | * or the total number of used blocks (pinned blocks plus AIL blocks) | |
1139 | * is greater than thresh2. | |
1140 | * | |
b57bc0fb | 1141 | * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of |
5e687eac BM |
1142 | * journal size. |
1143 | * | |
b3b94faa DT |
1144 | * Returns: errno |
1145 | */ | |
1146 | ||
1147 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
1148 | { | |
1149 | log_refund(sdp, tr); | |
b3b94faa | 1150 | |
5e687eac BM |
1151 | if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || |
1152 | ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > | |
1153 | atomic_read(&sdp->sd_log_thresh2))) | |
1154 | wake_up(&sdp->sd_logd_waitq); | |
b3b94faa DT |
1155 | } |
1156 | ||
1157 | /** | |
1158 | * gfs2_log_shutdown - write a shutdown header into a journal | |
1159 | * @sdp: the filesystem | |
1160 | * | |
1161 | */ | |
1162 | ||
feed98a8 | 1163 | static void gfs2_log_shutdown(struct gfs2_sbd *sdp) |
b3b94faa | 1164 | { |
b3b94faa | 1165 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa | 1166 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); |
b3b94faa DT |
1167 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); |
1168 | ||
1169 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
b3b94faa | 1170 | |
805c0907 | 1171 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN); |
b3b94faa | 1172 | |
a74604be SW |
1173 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
1174 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
1175 | |
1176 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa | 1177 | sdp->sd_log_tail = sdp->sd_log_head; |
a25311c8 SW |
1178 | } |
1179 | ||
5e687eac BM |
1180 | static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) |
1181 | { | |
f07b3520 BP |
1182 | return (atomic_read(&sdp->sd_log_pinned) + |
1183 | atomic_read(&sdp->sd_log_blks_needed) >= | |
1184 | atomic_read(&sdp->sd_log_thresh1)); | |
5e687eac BM |
1185 | } |
1186 | ||
1187 | static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) | |
1188 | { | |
1189 | unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); | |
b066a4ee AD |
1190 | |
1191 | if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags)) | |
1192 | return 1; | |
1193 | ||
f07b3520 BP |
1194 | return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >= |
1195 | atomic_read(&sdp->sd_log_thresh2); | |
5e687eac | 1196 | } |
ec69b188 SW |
1197 | |
1198 | /** | |
1199 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
1200 | * @sdp: Pointer to GFS2 superblock | |
1201 | * | |
1202 | * Also, periodically check to make sure that we're using the most recent | |
1203 | * journal index. | |
1204 | */ | |
1205 | ||
1206 | int gfs2_logd(void *data) | |
1207 | { | |
1208 | struct gfs2_sbd *sdp = data; | |
5e687eac BM |
1209 | unsigned long t = 1; |
1210 | DEFINE_WAIT(wait); | |
b63f5e84 | 1211 | bool did_flush; |
ec69b188 SW |
1212 | |
1213 | while (!kthread_should_stop()) { | |
ec69b188 | 1214 | |
d22f69a0 BP |
1215 | if (gfs2_withdrawn(sdp)) { |
1216 | msleep_interruptible(HZ); | |
1217 | continue; | |
1218 | } | |
942b0cdd BP |
1219 | /* Check for errors writing to the journal */ |
1220 | if (sdp->sd_log_error) { | |
badb55ec AG |
1221 | gfs2_lm(sdp, |
1222 | "GFS2: fsid=%s: error %d: " | |
1223 | "withdrawing the file system to " | |
1224 | "prevent further damage.\n", | |
1225 | sdp->sd_fsname, sdp->sd_log_error); | |
1226 | gfs2_withdraw(sdp); | |
d22f69a0 | 1227 | continue; |
942b0cdd BP |
1228 | } |
1229 | ||
b63f5e84 | 1230 | did_flush = false; |
5e687eac | 1231 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { |
5e4c7632 | 1232 | gfs2_ail1_empty(sdp, 0); |
805c0907 BP |
1233 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1234 | GFS2_LFC_LOGD_JFLUSH_REQD); | |
b63f5e84 | 1235 | did_flush = true; |
5e687eac | 1236 | } |
ec69b188 | 1237 | |
5e687eac BM |
1238 | if (gfs2_ail_flush_reqd(sdp)) { |
1239 | gfs2_ail1_start(sdp); | |
26b06a69 | 1240 | gfs2_ail1_wait(sdp); |
5e4c7632 | 1241 | gfs2_ail1_empty(sdp, 0); |
805c0907 BP |
1242 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1243 | GFS2_LFC_LOGD_AIL_FLUSH_REQD); | |
b63f5e84 | 1244 | did_flush = true; |
ec69b188 SW |
1245 | } |
1246 | ||
b63f5e84 | 1247 | if (!gfs2_ail_flush_reqd(sdp) || did_flush) |
26b06a69 SW |
1248 | wake_up(&sdp->sd_log_waitq); |
1249 | ||
ec69b188 | 1250 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
a0acae0e TH |
1251 | |
1252 | try_to_freeze(); | |
5e687eac BM |
1253 | |
1254 | do { | |
1255 | prepare_to_wait(&sdp->sd_logd_waitq, &wait, | |
5f487490 | 1256 | TASK_INTERRUPTIBLE); |
5e687eac BM |
1257 | if (!gfs2_ail_flush_reqd(sdp) && |
1258 | !gfs2_jrnl_flush_reqd(sdp) && | |
1259 | !kthread_should_stop()) | |
1260 | t = schedule_timeout(t); | |
1261 | } while(t && !gfs2_ail_flush_reqd(sdp) && | |
1262 | !gfs2_jrnl_flush_reqd(sdp) && | |
1263 | !kthread_should_stop()); | |
1264 | finish_wait(&sdp->sd_logd_waitq, &wait); | |
ec69b188 SW |
1265 | } |
1266 | ||
1267 | return 0; | |
1268 | } | |
1269 |