]>
Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
da6dd40d | 4 | * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
7 | #include <linux/sched.h> | |
8 | #include <linux/slab.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/completion.h> | |
11 | #include <linux/buffer_head.h> | |
5c676f6d | 12 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 13 | #include <linux/crc32.h> |
c1696fb8 | 14 | #include <linux/crc32c.h> |
a25311c8 | 15 | #include <linux/delay.h> |
ec69b188 SW |
16 | #include <linux/kthread.h> |
17 | #include <linux/freezer.h> | |
254db57f | 18 | #include <linux/bio.h> |
885bceca | 19 | #include <linux/blkdev.h> |
4667a0ec | 20 | #include <linux/writeback.h> |
4a36d08d | 21 | #include <linux/list_sort.h> |
b3b94faa DT |
22 | |
23 | #include "gfs2.h" | |
5c676f6d | 24 | #include "incore.h" |
b3b94faa DT |
25 | #include "bmap.h" |
26 | #include "glock.h" | |
27 | #include "log.h" | |
28 | #include "lops.h" | |
29 | #include "meta_io.h" | |
5c676f6d | 30 | #include "util.h" |
71b86f56 | 31 | #include "dir.h" |
63997775 | 32 | #include "trace_gfs2.h" |
b3b94faa | 33 | |
feed98a8 BP |
34 | static void gfs2_log_shutdown(struct gfs2_sbd *sdp); |
35 | ||
b3b94faa DT |
36 | /** |
37 | * gfs2_struct2blk - compute stuff | |
38 | * @sdp: the filesystem | |
39 | * @nstruct: the number of structures | |
b3b94faa DT |
40 | * |
41 | * Compute the number of log descriptor blocks needed to hold a certain number | |
42 | * of structures of a certain size. | |
43 | * | |
44 | * Returns: the number of blocks needed (minimum is always 1) | |
45 | */ | |
46 | ||
2e9eeaa1 | 47 | unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct) |
b3b94faa DT |
48 | { |
49 | unsigned int blks; | |
50 | unsigned int first, second; | |
51 | ||
52 | blks = 1; | |
2e9eeaa1 | 53 | first = sdp->sd_ldptrs; |
b3b94faa DT |
54 | |
55 | if (nstruct > first) { | |
2e9eeaa1 | 56 | second = sdp->sd_inptrs; |
5c676f6d | 57 | blks += DIV_ROUND_UP(nstruct - first, second); |
b3b94faa DT |
58 | } |
59 | ||
60 | return blks; | |
61 | } | |
62 | ||
1e1a3d03 SW |
63 | /** |
64 | * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters | |
65 | * @mapping: The associated mapping (maybe NULL) | |
66 | * @bd: The gfs2_bufdata to remove | |
67 | * | |
c618e87a | 68 | * The ail lock _must_ be held when calling this function |
1e1a3d03 SW |
69 | * |
70 | */ | |
71 | ||
9bc980cd | 72 | static void gfs2_remove_from_ail(struct gfs2_bufdata *bd) |
1e1a3d03 | 73 | { |
16ca9412 | 74 | bd->bd_tr = NULL; |
1ad38c43 SW |
75 | list_del_init(&bd->bd_ail_st_list); |
76 | list_del_init(&bd->bd_ail_gl_list); | |
1e1a3d03 | 77 | atomic_dec(&bd->bd_gl->gl_ail_count); |
1e1a3d03 SW |
78 | brelse(bd->bd_bh); |
79 | } | |
80 | ||
ddacfaf7 SW |
81 | /** |
82 | * gfs2_ail1_start_one - Start I/O on a part of the AIL | |
83 | * @sdp: the filesystem | |
4667a0ec SW |
84 | * @wbc: The writeback control structure |
85 | * @ai: The ail structure | |
ddacfaf7 SW |
86 | * |
87 | */ | |
88 | ||
4f1de018 SW |
89 | static int gfs2_ail1_start_one(struct gfs2_sbd *sdp, |
90 | struct writeback_control *wbc, | |
9e1a9ecd AG |
91 | struct gfs2_trans *tr, |
92 | bool *withdraw) | |
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; | |
ddacfaf7 | 100 | |
16ca9412 | 101 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) { |
4667a0ec | 102 | bh = bd->bd_bh; |
ddacfaf7 | 103 | |
16ca9412 | 104 | gfs2_assert(sdp, bd->bd_tr == tr); |
ddacfaf7 | 105 | |
4667a0ec | 106 | if (!buffer_busy(bh)) { |
b524abcc BP |
107 | if (!buffer_uptodate(bh) && |
108 | !test_and_set_bit(SDF_AIL1_IO_ERROR, | |
109 | &sdp->sd_flags)) { | |
4667a0ec | 110 | gfs2_io_error_bh(sdp, bh); |
9e1a9ecd AG |
111 | *withdraw = true; |
112 | } | |
16ca9412 | 113 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
4667a0ec SW |
114 | continue; |
115 | } | |
116 | ||
117 | if (!buffer_dirty(bh)) | |
118 | continue; | |
119 | if (gl == bd->bd_gl) | |
120 | continue; | |
121 | gl = bd->bd_gl; | |
16ca9412 | 122 | list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list); |
4667a0ec | 123 | mapping = bh->b_page->mapping; |
4f1de018 SW |
124 | if (!mapping) |
125 | continue; | |
4667a0ec SW |
126 | spin_unlock(&sdp->sd_ail_lock); |
127 | generic_writepages(mapping, wbc); | |
128 | spin_lock(&sdp->sd_ail_lock); | |
129 | if (wbc->nr_to_write <= 0) | |
130 | break; | |
4f1de018 | 131 | return 1; |
4667a0ec | 132 | } |
4f1de018 SW |
133 | |
134 | return 0; | |
4667a0ec | 135 | } |
ddacfaf7 | 136 | |
ddacfaf7 | 137 | |
4667a0ec SW |
138 | /** |
139 | * gfs2_ail1_flush - start writeback of some ail1 entries | |
140 | * @sdp: The super block | |
141 | * @wbc: The writeback control structure | |
142 | * | |
143 | * Writes back some ail1 entries, according to the limits in the | |
144 | * writeback control structure | |
145 | */ | |
ddacfaf7 | 146 | |
4667a0ec SW |
147 | void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc) |
148 | { | |
149 | struct list_head *head = &sdp->sd_ail1_list; | |
16ca9412 | 150 | struct gfs2_trans *tr; |
885bceca | 151 | struct blk_plug plug; |
9e1a9ecd | 152 | bool withdraw = false; |
ddacfaf7 | 153 | |
c83ae9ca | 154 | trace_gfs2_ail_flush(sdp, wbc, 1); |
885bceca | 155 | blk_start_plug(&plug); |
4667a0ec | 156 | spin_lock(&sdp->sd_ail_lock); |
4f1de018 | 157 | restart: |
16ca9412 | 158 | list_for_each_entry_reverse(tr, head, tr_list) { |
4667a0ec | 159 | if (wbc->nr_to_write <= 0) |
ddacfaf7 | 160 | break; |
f155f5e0 BP |
161 | if (gfs2_ail1_start_one(sdp, wbc, tr, &withdraw) && |
162 | !gfs2_withdrawn(sdp)) | |
4f1de018 | 163 | goto restart; |
4667a0ec SW |
164 | } |
165 | spin_unlock(&sdp->sd_ail_lock); | |
885bceca | 166 | blk_finish_plug(&plug); |
9e1a9ecd AG |
167 | if (withdraw) |
168 | gfs2_lm_withdraw(sdp, NULL); | |
c83ae9ca | 169 | trace_gfs2_ail_flush(sdp, wbc, 0); |
4667a0ec SW |
170 | } |
171 | ||
172 | /** | |
173 | * gfs2_ail1_start - start writeback of all ail1 entries | |
174 | * @sdp: The superblock | |
175 | */ | |
176 | ||
177 | static void gfs2_ail1_start(struct gfs2_sbd *sdp) | |
178 | { | |
179 | struct writeback_control wbc = { | |
180 | .sync_mode = WB_SYNC_NONE, | |
181 | .nr_to_write = LONG_MAX, | |
182 | .range_start = 0, | |
183 | .range_end = LLONG_MAX, | |
184 | }; | |
185 | ||
186 | return gfs2_ail1_flush(sdp, &wbc); | |
ddacfaf7 SW |
187 | } |
188 | ||
189 | /** | |
190 | * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced | |
191 | * @sdp: the filesystem | |
192 | * @ai: the AIL entry | |
193 | * | |
194 | */ | |
195 | ||
9e1a9ecd AG |
196 | static void gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr, |
197 | bool *withdraw) | |
ddacfaf7 SW |
198 | { |
199 | struct gfs2_bufdata *bd, *s; | |
200 | struct buffer_head *bh; | |
201 | ||
16ca9412 | 202 | list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, |
ddacfaf7 SW |
203 | bd_ail_st_list) { |
204 | bh = bd->bd_bh; | |
16ca9412 | 205 | gfs2_assert(sdp, bd->bd_tr == tr); |
4667a0ec SW |
206 | if (buffer_busy(bh)) |
207 | continue; | |
b524abcc BP |
208 | if (!buffer_uptodate(bh) && |
209 | !test_and_set_bit(SDF_AIL1_IO_ERROR, &sdp->sd_flags)) { | |
ddacfaf7 | 210 | gfs2_io_error_bh(sdp, bh); |
9e1a9ecd AG |
211 | *withdraw = true; |
212 | } | |
16ca9412 | 213 | list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list); |
ddacfaf7 | 214 | } |
ddacfaf7 SW |
215 | } |
216 | ||
4667a0ec SW |
217 | /** |
218 | * gfs2_ail1_empty - Try to empty the ail1 lists | |
219 | * @sdp: The superblock | |
220 | * | |
221 | * Tries to empty the ail1 lists, starting with the oldest first | |
222 | */ | |
b3b94faa | 223 | |
4667a0ec | 224 | static int gfs2_ail1_empty(struct gfs2_sbd *sdp) |
b3b94faa | 225 | { |
16ca9412 | 226 | struct gfs2_trans *tr, *s; |
5d054964 | 227 | int oldest_tr = 1; |
b3b94faa | 228 | int ret; |
9e1a9ecd | 229 | bool withdraw = false; |
b3b94faa | 230 | |
d6a079e8 | 231 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 | 232 | list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) { |
9e1a9ecd | 233 | gfs2_ail1_empty_one(sdp, tr, &withdraw); |
5d054964 | 234 | if (list_empty(&tr->tr_ail1_list) && oldest_tr) |
16ca9412 | 235 | list_move(&tr->tr_list, &sdp->sd_ail2_list); |
4667a0ec | 236 | else |
5d054964 | 237 | oldest_tr = 0; |
b3b94faa | 238 | } |
b3b94faa | 239 | ret = list_empty(&sdp->sd_ail1_list); |
d6a079e8 | 240 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 241 | |
9e1a9ecd AG |
242 | if (withdraw) |
243 | gfs2_lm_withdraw(sdp, "fatal: I/O error(s)\n"); | |
244 | ||
b3b94faa DT |
245 | return ret; |
246 | } | |
247 | ||
26b06a69 SW |
248 | static void gfs2_ail1_wait(struct gfs2_sbd *sdp) |
249 | { | |
16ca9412 | 250 | struct gfs2_trans *tr; |
26b06a69 SW |
251 | struct gfs2_bufdata *bd; |
252 | struct buffer_head *bh; | |
253 | ||
254 | spin_lock(&sdp->sd_ail_lock); | |
16ca9412 BM |
255 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
256 | list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) { | |
26b06a69 SW |
257 | bh = bd->bd_bh; |
258 | if (!buffer_locked(bh)) | |
259 | continue; | |
260 | get_bh(bh); | |
261 | spin_unlock(&sdp->sd_ail_lock); | |
262 | wait_on_buffer(bh); | |
263 | brelse(bh); | |
264 | return; | |
265 | } | |
266 | } | |
267 | spin_unlock(&sdp->sd_ail_lock); | |
268 | } | |
ddacfaf7 SW |
269 | |
270 | /** | |
271 | * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced | |
272 | * @sdp: the filesystem | |
273 | * @ai: the AIL entry | |
274 | * | |
275 | */ | |
276 | ||
16ca9412 | 277 | static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
ddacfaf7 | 278 | { |
16ca9412 | 279 | struct list_head *head = &tr->tr_ail2_list; |
ddacfaf7 SW |
280 | struct gfs2_bufdata *bd; |
281 | ||
282 | while (!list_empty(head)) { | |
283 | bd = list_entry(head->prev, struct gfs2_bufdata, | |
284 | bd_ail_st_list); | |
16ca9412 | 285 | gfs2_assert(sdp, bd->bd_tr == tr); |
f91a0d3e | 286 | gfs2_remove_from_ail(bd); |
ddacfaf7 SW |
287 | } |
288 | } | |
289 | ||
b3b94faa DT |
290 | static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail) |
291 | { | |
16ca9412 | 292 | struct gfs2_trans *tr, *safe; |
b3b94faa DT |
293 | unsigned int old_tail = sdp->sd_log_tail; |
294 | int wrap = (new_tail < old_tail); | |
295 | int a, b, rm; | |
296 | ||
d6a079e8 | 297 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 298 | |
16ca9412 BM |
299 | list_for_each_entry_safe(tr, safe, &sdp->sd_ail2_list, tr_list) { |
300 | a = (old_tail <= tr->tr_first); | |
301 | b = (tr->tr_first < new_tail); | |
b3b94faa DT |
302 | rm = (wrap) ? (a || b) : (a && b); |
303 | if (!rm) | |
304 | continue; | |
305 | ||
16ca9412 BM |
306 | gfs2_ail2_empty_one(sdp, tr); |
307 | list_del(&tr->tr_list); | |
308 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list)); | |
309 | gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list)); | |
310 | kfree(tr); | |
b3b94faa DT |
311 | } |
312 | ||
d6a079e8 | 313 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
314 | } |
315 | ||
24972557 BM |
316 | /** |
317 | * gfs2_log_release - Release a given number of log blocks | |
318 | * @sdp: The GFS2 superblock | |
319 | * @blks: The number of blocks | |
320 | * | |
321 | */ | |
322 | ||
323 | void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks) | |
324 | { | |
325 | ||
326 | atomic_add(blks, &sdp->sd_log_blks_free); | |
327 | trace_gfs2_log_blocks(sdp, blks); | |
328 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= | |
329 | sdp->sd_jdesc->jd_blocks); | |
330 | up_read(&sdp->sd_log_flush_lock); | |
331 | } | |
332 | ||
b3b94faa DT |
333 | /** |
334 | * gfs2_log_reserve - Make a log reservation | |
335 | * @sdp: The GFS2 superblock | |
336 | * @blks: The number of blocks to reserve | |
337 | * | |
89918647 | 338 | * Note that we never give out the last few blocks of the journal. Thats |
2332c443 | 339 | * due to the fact that there is a small number of header blocks |
b004157a SW |
340 | * associated with each log flush. The exact number can't be known until |
341 | * flush time, so we ensure that we have just enough free blocks at all | |
342 | * times to avoid running out during a log flush. | |
343 | * | |
5e687eac BM |
344 | * We no longer flush the log here, instead we wake up logd to do that |
345 | * for us. To avoid the thundering herd and to ensure that we deal fairly | |
346 | * with queued waiters, we use an exclusive wait. This means that when we | |
347 | * get woken with enough journal space to get our reservation, we need to | |
348 | * wake the next waiter on the list. | |
349 | * | |
b3b94faa DT |
350 | * Returns: errno |
351 | */ | |
352 | ||
353 | int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks) | |
354 | { | |
2e60d768 | 355 | int ret = 0; |
5d054964 | 356 | unsigned reserved_blks = 7 * (4096 / sdp->sd_vfs->s_blocksize); |
5e687eac BM |
357 | unsigned wanted = blks + reserved_blks; |
358 | DEFINE_WAIT(wait); | |
359 | int did_wait = 0; | |
360 | unsigned int free_blocks; | |
b3b94faa DT |
361 | |
362 | if (gfs2_assert_warn(sdp, blks) || | |
363 | gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks)) | |
364 | return -EINVAL; | |
f07b3520 | 365 | atomic_add(blks, &sdp->sd_log_blks_needed); |
5e687eac BM |
366 | retry: |
367 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
368 | if (unlikely(free_blocks <= wanted)) { | |
369 | do { | |
370 | prepare_to_wait_exclusive(&sdp->sd_log_waitq, &wait, | |
371 | TASK_UNINTERRUPTIBLE); | |
372 | wake_up(&sdp->sd_logd_waitq); | |
373 | did_wait = 1; | |
374 | if (atomic_read(&sdp->sd_log_blks_free) <= wanted) | |
375 | io_schedule(); | |
376 | free_blocks = atomic_read(&sdp->sd_log_blks_free); | |
377 | } while(free_blocks <= wanted); | |
378 | finish_wait(&sdp->sd_log_waitq, &wait); | |
b3b94faa | 379 | } |
2e60d768 | 380 | atomic_inc(&sdp->sd_reserving_log); |
5e687eac | 381 | if (atomic_cmpxchg(&sdp->sd_log_blks_free, free_blocks, |
2e60d768 BM |
382 | free_blocks - blks) != free_blocks) { |
383 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) | |
384 | wake_up(&sdp->sd_reserving_log_wait); | |
5e687eac | 385 | goto retry; |
2e60d768 | 386 | } |
f07b3520 | 387 | atomic_sub(blks, &sdp->sd_log_blks_needed); |
63997775 | 388 | trace_gfs2_log_blocks(sdp, -blks); |
5e687eac BM |
389 | |
390 | /* | |
391 | * If we waited, then so might others, wake them up _after_ we get | |
392 | * our share of the log. | |
393 | */ | |
394 | if (unlikely(did_wait)) | |
395 | wake_up(&sdp->sd_log_waitq); | |
484adff8 SW |
396 | |
397 | down_read(&sdp->sd_log_flush_lock); | |
24972557 BM |
398 | if (unlikely(!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))) { |
399 | gfs2_log_release(sdp, blks); | |
2e60d768 | 400 | ret = -EROFS; |
24972557 | 401 | } |
2e60d768 BM |
402 | if (atomic_dec_and_test(&sdp->sd_reserving_log)) |
403 | wake_up(&sdp->sd_reserving_log_wait); | |
404 | return ret; | |
b3b94faa DT |
405 | } |
406 | ||
b3b94faa DT |
407 | /** |
408 | * log_distance - Compute distance between two journal blocks | |
409 | * @sdp: The GFS2 superblock | |
410 | * @newer: The most recent journal block of the pair | |
411 | * @older: The older journal block of the pair | |
412 | * | |
413 | * Compute the distance (in the journal direction) between two | |
414 | * blocks in the journal | |
415 | * | |
416 | * Returns: the distance in blocks | |
417 | */ | |
418 | ||
faa31ce8 | 419 | static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer, |
b3b94faa DT |
420 | unsigned int older) |
421 | { | |
422 | int dist; | |
423 | ||
424 | dist = newer - older; | |
425 | if (dist < 0) | |
426 | dist += sdp->sd_jdesc->jd_blocks; | |
427 | ||
428 | return dist; | |
429 | } | |
430 | ||
2332c443 RP |
431 | /** |
432 | * calc_reserved - Calculate the number of blocks to reserve when | |
433 | * refunding a transaction's unused buffers. | |
434 | * @sdp: The GFS2 superblock | |
435 | * | |
436 | * This is complex. We need to reserve room for all our currently used | |
437 | * metadata buffers (e.g. normal file I/O rewriting file time stamps) and | |
438 | * all our journaled data buffers for journaled files (e.g. files in the | |
439 | * meta_fs like rindex, or files for which chattr +j was done.) | |
440 | * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush | |
441 | * will count it as free space (sd_log_blks_free) and corruption will follow. | |
442 | * | |
443 | * We can have metadata bufs and jdata bufs in the same journal. So each | |
444 | * type gets its own log header, for which we need to reserve a block. | |
445 | * In fact, each type has the potential for needing more than one header | |
446 | * in cases where we have more buffers than will fit on a journal page. | |
447 | * Metadata journal entries take up half the space of journaled buffer entries. | |
448 | * Thus, metadata entries have buf_limit (502) and journaled buffers have | |
449 | * databuf_limit (251) before they cause a wrap around. | |
450 | * | |
451 | * Also, we need to reserve blocks for revoke journal entries and one for an | |
452 | * overall header for the lot. | |
453 | * | |
454 | * Returns: the number of blocks reserved | |
455 | */ | |
456 | static unsigned int calc_reserved(struct gfs2_sbd *sdp) | |
457 | { | |
458 | unsigned int reserved = 0; | |
022ef4fe SW |
459 | unsigned int mbuf; |
460 | unsigned int dbuf; | |
461 | struct gfs2_trans *tr = sdp->sd_log_tr; | |
2332c443 | 462 | |
022ef4fe SW |
463 | if (tr) { |
464 | mbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm; | |
465 | dbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm; | |
466 | reserved = mbuf + dbuf; | |
467 | /* Account for header blocks */ | |
468 | reserved += DIV_ROUND_UP(mbuf, buf_limit(sdp)); | |
469 | reserved += DIV_ROUND_UP(dbuf, databuf_limit(sdp)); | |
470 | } | |
2332c443 | 471 | |
5d439758 AG |
472 | if (sdp->sd_log_committed_revoke > 0) |
473 | reserved += gfs2_struct2blk(sdp, sdp->sd_log_committed_revoke); | |
2332c443 RP |
474 | /* One for the overall header */ |
475 | if (reserved) | |
476 | reserved++; | |
477 | return reserved; | |
478 | } | |
479 | ||
b3b94faa DT |
480 | static unsigned int current_tail(struct gfs2_sbd *sdp) |
481 | { | |
16ca9412 | 482 | struct gfs2_trans *tr; |
b3b94faa DT |
483 | unsigned int tail; |
484 | ||
d6a079e8 | 485 | spin_lock(&sdp->sd_ail_lock); |
b3b94faa | 486 | |
faa31ce8 | 487 | if (list_empty(&sdp->sd_ail1_list)) { |
b3b94faa | 488 | tail = sdp->sd_log_head; |
faa31ce8 | 489 | } else { |
16ca9412 BM |
490 | tr = list_entry(sdp->sd_ail1_list.prev, struct gfs2_trans, |
491 | tr_list); | |
492 | tail = tr->tr_first; | |
b3b94faa DT |
493 | } |
494 | ||
d6a079e8 | 495 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa DT |
496 | |
497 | return tail; | |
498 | } | |
499 | ||
2332c443 | 500 | static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail) |
b3b94faa DT |
501 | { |
502 | unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail); | |
503 | ||
504 | ail2_empty(sdp, new_tail); | |
505 | ||
fd041f0b | 506 | atomic_add(dist, &sdp->sd_log_blks_free); |
63997775 | 507 | trace_gfs2_log_blocks(sdp, dist); |
5e687eac BM |
508 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
509 | sdp->sd_jdesc->jd_blocks); | |
b3b94faa DT |
510 | |
511 | sdp->sd_log_tail = new_tail; | |
512 | } | |
513 | ||
b3b94faa | 514 | |
34cc1781 | 515 | static void log_flush_wait(struct gfs2_sbd *sdp) |
b3b94faa | 516 | { |
16615be1 SW |
517 | DEFINE_WAIT(wait); |
518 | ||
519 | if (atomic_read(&sdp->sd_log_in_flight)) { | |
520 | do { | |
521 | prepare_to_wait(&sdp->sd_log_flush_wait, &wait, | |
522 | TASK_UNINTERRUPTIBLE); | |
523 | if (atomic_read(&sdp->sd_log_in_flight)) | |
524 | io_schedule(); | |
525 | } while(atomic_read(&sdp->sd_log_in_flight)); | |
526 | finish_wait(&sdp->sd_log_flush_wait, &wait); | |
b3b94faa | 527 | } |
b3b94faa DT |
528 | } |
529 | ||
45138990 | 530 | static int ip_cmp(void *priv, struct list_head *a, struct list_head *b) |
4a36d08d | 531 | { |
45138990 | 532 | struct gfs2_inode *ipa, *ipb; |
4a36d08d | 533 | |
45138990 SW |
534 | ipa = list_entry(a, struct gfs2_inode, i_ordered); |
535 | ipb = list_entry(b, struct gfs2_inode, i_ordered); | |
4a36d08d | 536 | |
45138990 | 537 | if (ipa->i_no_addr < ipb->i_no_addr) |
4a36d08d | 538 | return -1; |
45138990 | 539 | if (ipa->i_no_addr > ipb->i_no_addr) |
4a36d08d BP |
540 | return 1; |
541 | return 0; | |
542 | } | |
543 | ||
d7b616e2 SW |
544 | static void gfs2_ordered_write(struct gfs2_sbd *sdp) |
545 | { | |
45138990 | 546 | struct gfs2_inode *ip; |
d7b616e2 SW |
547 | LIST_HEAD(written); |
548 | ||
45138990 | 549 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
550 | list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp); |
551 | while (!list_empty(&sdp->sd_log_ordered)) { | |
552 | ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); | |
1f23bc78 AD |
553 | if (ip->i_inode.i_mapping->nrpages == 0) { |
554 | test_and_clear_bit(GIF_ORDERED, &ip->i_flags); | |
555 | list_del(&ip->i_ordered); | |
d7b616e2 | 556 | continue; |
1f23bc78 AD |
557 | } |
558 | list_move(&ip->i_ordered, &written); | |
45138990 SW |
559 | spin_unlock(&sdp->sd_ordered_lock); |
560 | filemap_fdatawrite(ip->i_inode.i_mapping); | |
561 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 562 | } |
a5b1d3fc | 563 | list_splice(&written, &sdp->sd_log_ordered); |
45138990 | 564 | spin_unlock(&sdp->sd_ordered_lock); |
d7b616e2 SW |
565 | } |
566 | ||
567 | static void gfs2_ordered_wait(struct gfs2_sbd *sdp) | |
568 | { | |
45138990 | 569 | struct gfs2_inode *ip; |
d7b616e2 | 570 | |
45138990 | 571 | spin_lock(&sdp->sd_ordered_lock); |
a5b1d3fc AG |
572 | while (!list_empty(&sdp->sd_log_ordered)) { |
573 | ip = list_entry(sdp->sd_log_ordered.next, struct gfs2_inode, i_ordered); | |
45138990 SW |
574 | list_del(&ip->i_ordered); |
575 | WARN_ON(!test_and_clear_bit(GIF_ORDERED, &ip->i_flags)); | |
576 | if (ip->i_inode.i_mapping->nrpages == 0) | |
d7b616e2 | 577 | continue; |
45138990 SW |
578 | spin_unlock(&sdp->sd_ordered_lock); |
579 | filemap_fdatawait(ip->i_inode.i_mapping); | |
580 | spin_lock(&sdp->sd_ordered_lock); | |
d7b616e2 | 581 | } |
45138990 SW |
582 | spin_unlock(&sdp->sd_ordered_lock); |
583 | } | |
584 | ||
585 | void gfs2_ordered_del_inode(struct gfs2_inode *ip) | |
586 | { | |
587 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); | |
588 | ||
589 | spin_lock(&sdp->sd_ordered_lock); | |
590 | if (test_and_clear_bit(GIF_ORDERED, &ip->i_flags)) | |
591 | list_del(&ip->i_ordered); | |
592 | spin_unlock(&sdp->sd_ordered_lock); | |
d7b616e2 SW |
593 | } |
594 | ||
5d054964 BM |
595 | void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd) |
596 | { | |
597 | struct buffer_head *bh = bd->bd_bh; | |
598 | struct gfs2_glock *gl = bd->bd_gl; | |
599 | ||
5d054964 BM |
600 | bh->b_private = NULL; |
601 | bd->bd_blkno = bh->b_blocknr; | |
9290a9a7 BP |
602 | gfs2_remove_from_ail(bd); /* drops ref on bh */ |
603 | bd->bd_bh = NULL; | |
5d054964 | 604 | sdp->sd_log_num_revoke++; |
638803d4 | 605 | if (atomic_inc_return(&gl->gl_revokes) == 1) |
9287c645 | 606 | gfs2_glock_hold(gl); |
5d054964 | 607 | set_bit(GLF_LFLUSH, &gl->gl_flags); |
a5b1d3fc | 608 | list_add(&bd->bd_list, &sdp->sd_log_revokes); |
5d054964 BM |
609 | } |
610 | ||
fe5e7ba1 BP |
611 | void gfs2_glock_remove_revoke(struct gfs2_glock *gl) |
612 | { | |
613 | if (atomic_dec_return(&gl->gl_revokes) == 0) { | |
614 | clear_bit(GLF_LFLUSH, &gl->gl_flags); | |
615 | gfs2_glock_queue_put(gl); | |
616 | } | |
617 | } | |
618 | ||
5d054964 BM |
619 | void gfs2_write_revokes(struct gfs2_sbd *sdp) |
620 | { | |
621 | struct gfs2_trans *tr; | |
622 | struct gfs2_bufdata *bd, *tmp; | |
623 | int have_revokes = 0; | |
624 | int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64); | |
625 | ||
626 | gfs2_ail1_empty(sdp); | |
627 | spin_lock(&sdp->sd_ail_lock); | |
c9e58fb2 | 628 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
5d054964 BM |
629 | list_for_each_entry(bd, &tr->tr_ail2_list, bd_ail_st_list) { |
630 | if (list_empty(&bd->bd_list)) { | |
631 | have_revokes = 1; | |
632 | goto done; | |
633 | } | |
634 | } | |
635 | } | |
636 | done: | |
637 | spin_unlock(&sdp->sd_ail_lock); | |
638 | if (have_revokes == 0) | |
639 | return; | |
640 | while (sdp->sd_log_num_revoke > max_revokes) | |
641 | max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64); | |
642 | max_revokes -= sdp->sd_log_num_revoke; | |
643 | if (!sdp->sd_log_num_revoke) { | |
644 | atomic_dec(&sdp->sd_log_blks_free); | |
645 | /* If no blocks have been reserved, we need to also | |
646 | * reserve a block for the header */ | |
647 | if (!sdp->sd_log_blks_reserved) | |
648 | atomic_dec(&sdp->sd_log_blks_free); | |
649 | } | |
650 | gfs2_log_lock(sdp); | |
651 | spin_lock(&sdp->sd_ail_lock); | |
c9e58fb2 | 652 | list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) { |
5d054964 BM |
653 | list_for_each_entry_safe(bd, tmp, &tr->tr_ail2_list, bd_ail_st_list) { |
654 | if (max_revokes == 0) | |
655 | goto out_of_blocks; | |
656 | if (!list_empty(&bd->bd_list)) | |
657 | continue; | |
658 | gfs2_add_revoke(sdp, bd); | |
659 | max_revokes--; | |
660 | } | |
661 | } | |
662 | out_of_blocks: | |
663 | spin_unlock(&sdp->sd_ail_lock); | |
664 | gfs2_log_unlock(sdp); | |
665 | ||
666 | if (!sdp->sd_log_num_revoke) { | |
667 | atomic_inc(&sdp->sd_log_blks_free); | |
668 | if (!sdp->sd_log_blks_reserved) | |
669 | atomic_inc(&sdp->sd_log_blks_free); | |
670 | } | |
671 | } | |
672 | ||
34cc1781 | 673 | /** |
7c70b896 | 674 | * gfs2_write_log_header - Write a journal log header buffer at lblock |
34cc1781 | 675 | * @sdp: The GFS2 superblock |
c1696fb8 | 676 | * @jd: journal descriptor of the journal to which we are writing |
588bff95 BP |
677 | * @seq: sequence number |
678 | * @tail: tail of the log | |
7c70b896 | 679 | * @lblock: value for lh_blkno (block number relative to start of journal) |
c1696fb8 | 680 | * @flags: log header flags GFS2_LOG_HEAD_* |
588bff95 | 681 | * @op_flags: flags to pass to the bio |
34cc1781 SW |
682 | * |
683 | * Returns: the initialized log buffer descriptor | |
684 | */ | |
685 | ||
c1696fb8 | 686 | void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd, |
7c70b896 BP |
687 | u64 seq, u32 tail, u32 lblock, u32 flags, |
688 | int op_flags) | |
34cc1781 | 689 | { |
34cc1781 | 690 | struct gfs2_log_header *lh; |
c1696fb8 | 691 | u32 hash, crc; |
ade48088 | 692 | struct page *page; |
c1696fb8 BP |
693 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; |
694 | struct timespec64 tv; | |
695 | struct super_block *sb = sdp->sd_vfs; | |
7c70b896 | 696 | u64 dblock; |
588bff95 | 697 | |
ade48088 BP |
698 | if (gfs2_withdrawn(sdp)) |
699 | goto out; | |
700 | ||
701 | page = mempool_alloc(gfs2_page_pool, GFP_NOIO); | |
e8c92ed7 SW |
702 | lh = page_address(page); |
703 | clear_page(lh); | |
34cc1781 | 704 | |
34cc1781 SW |
705 | lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC); |
706 | lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH); | |
707 | lh->lh_header.__pad0 = cpu_to_be64(0); | |
708 | lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH); | |
709 | lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid); | |
588bff95 | 710 | lh->lh_sequence = cpu_to_be64(seq); |
34cc1781 SW |
711 | lh->lh_flags = cpu_to_be32(flags); |
712 | lh->lh_tail = cpu_to_be32(tail); | |
7c70b896 | 713 | lh->lh_blkno = cpu_to_be32(lblock); |
c1696fb8 | 714 | hash = ~crc32(~0, lh, LH_V1_SIZE); |
34cc1781 SW |
715 | lh->lh_hash = cpu_to_be32(hash); |
716 | ||
ee9c7f9a | 717 | ktime_get_coarse_real_ts64(&tv); |
c1696fb8 BP |
718 | lh->lh_nsec = cpu_to_be32(tv.tv_nsec); |
719 | lh->lh_sec = cpu_to_be64(tv.tv_sec); | |
7c70b896 | 720 | if (!list_empty(&jd->extent_list)) |
19ebc050 | 721 | dblock = gfs2_log_bmap(jd, lblock); |
7c70b896 BP |
722 | else { |
723 | int ret = gfs2_lblk_to_dblk(jd->jd_inode, lblock, &dblock); | |
724 | if (gfs2_assert_withdraw(sdp, ret == 0)) | |
725 | return; | |
726 | } | |
727 | lh->lh_addr = cpu_to_be64(dblock); | |
c1696fb8 BP |
728 | lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr); |
729 | ||
730 | /* We may only write local statfs, quota, etc., when writing to our | |
731 | own journal. The values are left 0 when recovering a journal | |
732 | different from our own. */ | |
733 | if (!(flags & GFS2_LOG_HEAD_RECOVERY)) { | |
734 | lh->lh_statfs_addr = | |
735 | cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr); | |
736 | lh->lh_quota_addr = | |
737 | cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr); | |
738 | ||
739 | spin_lock(&sdp->sd_statfs_spin); | |
740 | lh->lh_local_total = cpu_to_be64(l_sc->sc_total); | |
741 | lh->lh_local_free = cpu_to_be64(l_sc->sc_free); | |
742 | lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes); | |
743 | spin_unlock(&sdp->sd_statfs_spin); | |
744 | } | |
745 | ||
746 | BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE); | |
747 | ||
748 | crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4, | |
749 | sb->s_blocksize - LH_V1_SIZE - 4); | |
750 | lh->lh_crc = cpu_to_be32(crc); | |
751 | ||
7c70b896 | 752 | gfs2_log_write(sdp, page, sb->s_blocksize, 0, dblock); |
f4686c26 | 753 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE | op_flags); |
ade48088 | 754 | out: |
588bff95 BP |
755 | log_flush_wait(sdp); |
756 | } | |
757 | ||
758 | /** | |
759 | * log_write_header - Get and initialize a journal header buffer | |
760 | * @sdp: The GFS2 superblock | |
c1696fb8 | 761 | * @flags: The log header flags, including log header origin |
588bff95 BP |
762 | * |
763 | * Returns: the initialized log buffer descriptor | |
764 | */ | |
765 | ||
766 | static void log_write_header(struct gfs2_sbd *sdp, u32 flags) | |
767 | { | |
768 | unsigned int tail; | |
769 | int op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC; | |
770 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); | |
771 | ||
772 | gfs2_assert_withdraw(sdp, (state != SFS_FROZEN)); | |
773 | tail = current_tail(sdp); | |
774 | ||
34cc1781 SW |
775 | if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) { |
776 | gfs2_ordered_wait(sdp); | |
777 | log_flush_wait(sdp); | |
70fd7614 | 778 | op_flags = REQ_SYNC | REQ_META | REQ_PRIO; |
34cc1781 | 779 | } |
e8c92ed7 | 780 | sdp->sd_log_idle = (tail == sdp->sd_log_flush_head); |
c1696fb8 | 781 | gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++, tail, |
7c70b896 | 782 | sdp->sd_log_flush_head, flags, op_flags); |
19ebc050 | 783 | gfs2_log_incr_head(sdp); |
34cc1781 SW |
784 | |
785 | if (sdp->sd_log_tail != tail) | |
786 | log_pull_tail(sdp, tail); | |
34cc1781 SW |
787 | } |
788 | ||
b3b94faa | 789 | /** |
b09e593d | 790 | * gfs2_log_flush - flush incore transaction(s) |
b3b94faa DT |
791 | * @sdp: the filesystem |
792 | * @gl: The glock structure to flush. If NULL, flush the whole incore log | |
805c0907 | 793 | * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags |
b3b94faa DT |
794 | * |
795 | */ | |
796 | ||
c1696fb8 | 797 | void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags) |
b3b94faa | 798 | { |
16ca9412 | 799 | struct gfs2_trans *tr; |
2e60d768 | 800 | enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state); |
b3b94faa | 801 | |
484adff8 | 802 | down_write(&sdp->sd_log_flush_lock); |
f55ab26a | 803 | |
2bcd610d SW |
804 | /* Log might have been flushed while we waited for the flush lock */ |
805 | if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) { | |
806 | up_write(&sdp->sd_log_flush_lock); | |
807 | return; | |
f55ab26a | 808 | } |
805c0907 | 809 | trace_gfs2_log_flush(sdp, 1, flags); |
f55ab26a | 810 | |
c1696fb8 | 811 | if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN) |
400ac52e BM |
812 | clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); |
813 | ||
b1ab1e44 | 814 | sdp->sd_log_flush_head = sdp->sd_log_head; |
16ca9412 BM |
815 | tr = sdp->sd_log_tr; |
816 | if (tr) { | |
817 | sdp->sd_log_tr = NULL; | |
818 | INIT_LIST_HEAD(&tr->tr_ail1_list); | |
819 | INIT_LIST_HEAD(&tr->tr_ail2_list); | |
b1ab1e44 | 820 | tr->tr_first = sdp->sd_log_flush_head; |
2e60d768 BM |
821 | if (unlikely (state == SFS_FROZEN)) |
822 | gfs2_assert_withdraw(sdp, !tr->tr_num_buf_new && !tr->tr_num_databuf_new); | |
16ca9412 | 823 | } |
b3b94faa | 824 | |
2e60d768 BM |
825 | if (unlikely(state == SFS_FROZEN)) |
826 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); | |
b3b94faa | 827 | gfs2_assert_withdraw(sdp, |
5d439758 | 828 | sdp->sd_log_num_revoke == sdp->sd_log_committed_revoke); |
b3b94faa | 829 | |
d7b616e2 | 830 | gfs2_ordered_write(sdp); |
d69a3c65 | 831 | lops_before_commit(sdp, tr); |
f4686c26 | 832 | gfs2_log_submit_bio(&sdp->sd_log_bio, REQ_OP_WRITE); |
d7b616e2 | 833 | |
34cc1781 | 834 | if (sdp->sd_log_head != sdp->sd_log_flush_head) { |
428fd95d | 835 | log_flush_wait(sdp); |
c1696fb8 | 836 | log_write_header(sdp, flags); |
34cc1781 | 837 | } else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){ |
fd041f0b | 838 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ |
63997775 | 839 | trace_gfs2_log_blocks(sdp, -1); |
c1696fb8 | 840 | log_write_header(sdp, flags); |
2332c443 | 841 | } |
16ca9412 | 842 | lops_after_commit(sdp, tr); |
b09e593d | 843 | |
fe1a698f SW |
844 | gfs2_log_lock(sdp); |
845 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
faa31ce8 | 846 | sdp->sd_log_blks_reserved = 0; |
5d439758 | 847 | sdp->sd_log_committed_revoke = 0; |
b3b94faa | 848 | |
d6a079e8 | 849 | spin_lock(&sdp->sd_ail_lock); |
16ca9412 BM |
850 | if (tr && !list_empty(&tr->tr_ail1_list)) { |
851 | list_add(&tr->tr_list, &sdp->sd_ail1_list); | |
852 | tr = NULL; | |
b3b94faa | 853 | } |
d6a079e8 | 854 | spin_unlock(&sdp->sd_ail_lock); |
b3b94faa | 855 | gfs2_log_unlock(sdp); |
24972557 | 856 | |
c1696fb8 | 857 | if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) { |
24972557 BM |
858 | if (!sdp->sd_log_idle) { |
859 | for (;;) { | |
860 | gfs2_ail1_start(sdp); | |
861 | gfs2_ail1_wait(sdp); | |
862 | if (gfs2_ail1_empty(sdp)) | |
863 | break; | |
864 | } | |
865 | atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */ | |
866 | trace_gfs2_log_blocks(sdp, -1); | |
c1696fb8 | 867 | log_write_header(sdp, flags); |
24972557 BM |
868 | sdp->sd_log_head = sdp->sd_log_flush_head; |
869 | } | |
c1696fb8 BP |
870 | if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN | |
871 | GFS2_LOG_HEAD_FLUSH_FREEZE)) | |
24972557 | 872 | gfs2_log_shutdown(sdp); |
c1696fb8 | 873 | if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE) |
2e60d768 | 874 | atomic_set(&sdp->sd_freeze_state, SFS_FROZEN); |
24972557 BM |
875 | } |
876 | ||
805c0907 | 877 | trace_gfs2_log_flush(sdp, 0, flags); |
484adff8 | 878 | up_write(&sdp->sd_log_flush_lock); |
b3b94faa | 879 | |
16ca9412 | 880 | kfree(tr); |
b3b94faa DT |
881 | } |
882 | ||
d69a3c65 SW |
883 | /** |
884 | * gfs2_merge_trans - Merge a new transaction into a cached transaction | |
885 | * @old: Original transaction to be expanded | |
886 | * @new: New transaction to be merged | |
887 | */ | |
888 | ||
889 | static void gfs2_merge_trans(struct gfs2_trans *old, struct gfs2_trans *new) | |
890 | { | |
9862ca05 | 891 | WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags)); |
d69a3c65 SW |
892 | |
893 | old->tr_num_buf_new += new->tr_num_buf_new; | |
894 | old->tr_num_databuf_new += new->tr_num_databuf_new; | |
895 | old->tr_num_buf_rm += new->tr_num_buf_rm; | |
896 | old->tr_num_databuf_rm += new->tr_num_databuf_rm; | |
897 | old->tr_num_revoke += new->tr_num_revoke; | |
a31b4ec5 | 898 | old->tr_num_revoke_rm += new->tr_num_revoke_rm; |
d69a3c65 SW |
899 | |
900 | list_splice_tail_init(&new->tr_databuf, &old->tr_databuf); | |
901 | list_splice_tail_init(&new->tr_buf, &old->tr_buf); | |
902 | } | |
903 | ||
b3b94faa DT |
904 | static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr) |
905 | { | |
2332c443 | 906 | unsigned int reserved; |
ac39aadd | 907 | unsigned int unused; |
022ef4fe | 908 | unsigned int maxres; |
b3b94faa DT |
909 | |
910 | gfs2_log_lock(sdp); | |
911 | ||
022ef4fe SW |
912 | if (sdp->sd_log_tr) { |
913 | gfs2_merge_trans(sdp->sd_log_tr, tr); | |
914 | } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) { | |
9862ca05 | 915 | gfs2_assert_withdraw(sdp, test_bit(TR_ALLOCED, &tr->tr_flags)); |
022ef4fe | 916 | sdp->sd_log_tr = tr; |
9862ca05 | 917 | set_bit(TR_ATTACHED, &tr->tr_flags); |
022ef4fe SW |
918 | } |
919 | ||
a31b4ec5 | 920 | sdp->sd_log_committed_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm; |
2332c443 | 921 | reserved = calc_reserved(sdp); |
022ef4fe SW |
922 | maxres = sdp->sd_log_blks_reserved + tr->tr_reserved; |
923 | gfs2_assert_withdraw(sdp, maxres >= reserved); | |
924 | unused = maxres - reserved; | |
ac39aadd | 925 | atomic_add(unused, &sdp->sd_log_blks_free); |
63997775 | 926 | trace_gfs2_log_blocks(sdp, unused); |
fd041f0b | 927 | gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= |
2332c443 | 928 | sdp->sd_jdesc->jd_blocks); |
b3b94faa DT |
929 | sdp->sd_log_blks_reserved = reserved; |
930 | ||
931 | gfs2_log_unlock(sdp); | |
932 | } | |
933 | ||
934 | /** | |
935 | * gfs2_log_commit - Commit a transaction to the log | |
936 | * @sdp: the filesystem | |
937 | * @tr: the transaction | |
938 | * | |
5e687eac BM |
939 | * We wake up gfs2_logd if the number of pinned blocks exceed thresh1 |
940 | * or the total number of used blocks (pinned blocks plus AIL blocks) | |
941 | * is greater than thresh2. | |
942 | * | |
943 | * At mount time thresh1 is 1/3rd of journal size, thresh2 is 2/3rd of | |
944 | * journal size. | |
945 | * | |
b3b94faa DT |
946 | * Returns: errno |
947 | */ | |
948 | ||
949 | void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr) | |
950 | { | |
951 | log_refund(sdp, tr); | |
b3b94faa | 952 | |
5e687eac BM |
953 | if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) || |
954 | ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) > | |
955 | atomic_read(&sdp->sd_log_thresh2))) | |
956 | wake_up(&sdp->sd_logd_waitq); | |
b3b94faa DT |
957 | } |
958 | ||
959 | /** | |
960 | * gfs2_log_shutdown - write a shutdown header into a journal | |
961 | * @sdp: the filesystem | |
962 | * | |
963 | */ | |
964 | ||
feed98a8 | 965 | static void gfs2_log_shutdown(struct gfs2_sbd *sdp) |
b3b94faa | 966 | { |
b3b94faa | 967 | gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved); |
b3b94faa | 968 | gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke); |
b3b94faa DT |
969 | gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list)); |
970 | ||
971 | sdp->sd_log_flush_head = sdp->sd_log_head; | |
b3b94faa | 972 | |
805c0907 | 973 | log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN); |
b3b94faa | 974 | |
a74604be SW |
975 | gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail); |
976 | gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list)); | |
b3b94faa DT |
977 | |
978 | sdp->sd_log_head = sdp->sd_log_flush_head; | |
b3b94faa | 979 | sdp->sd_log_tail = sdp->sd_log_head; |
a25311c8 SW |
980 | } |
981 | ||
5e687eac BM |
982 | static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp) |
983 | { | |
f07b3520 BP |
984 | return (atomic_read(&sdp->sd_log_pinned) + |
985 | atomic_read(&sdp->sd_log_blks_needed) >= | |
986 | atomic_read(&sdp->sd_log_thresh1)); | |
5e687eac BM |
987 | } |
988 | ||
989 | static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp) | |
990 | { | |
991 | unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free); | |
b066a4ee AD |
992 | |
993 | if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags)) | |
994 | return 1; | |
995 | ||
f07b3520 BP |
996 | return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >= |
997 | atomic_read(&sdp->sd_log_thresh2); | |
5e687eac | 998 | } |
ec69b188 SW |
999 | |
1000 | /** | |
1001 | * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks | |
1002 | * @sdp: Pointer to GFS2 superblock | |
1003 | * | |
1004 | * Also, periodically check to make sure that we're using the most recent | |
1005 | * journal index. | |
1006 | */ | |
1007 | ||
1008 | int gfs2_logd(void *data) | |
1009 | { | |
1010 | struct gfs2_sbd *sdp = data; | |
5e687eac BM |
1011 | unsigned long t = 1; |
1012 | DEFINE_WAIT(wait); | |
b63f5e84 | 1013 | bool did_flush; |
ec69b188 SW |
1014 | |
1015 | while (!kthread_should_stop()) { | |
ec69b188 | 1016 | |
942b0cdd BP |
1017 | /* Check for errors writing to the journal */ |
1018 | if (sdp->sd_log_error) { | |
1019 | gfs2_lm_withdraw(sdp, | |
1020 | "GFS2: fsid=%s: error %d: " | |
1021 | "withdrawing the file system to " | |
1022 | "prevent further damage.\n", | |
1023 | sdp->sd_fsname, sdp->sd_log_error); | |
1024 | } | |
1025 | ||
b63f5e84 | 1026 | did_flush = false; |
5e687eac | 1027 | if (gfs2_jrnl_flush_reqd(sdp) || t == 0) { |
4667a0ec | 1028 | gfs2_ail1_empty(sdp); |
805c0907 BP |
1029 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1030 | GFS2_LFC_LOGD_JFLUSH_REQD); | |
b63f5e84 | 1031 | did_flush = true; |
5e687eac | 1032 | } |
ec69b188 | 1033 | |
5e687eac BM |
1034 | if (gfs2_ail_flush_reqd(sdp)) { |
1035 | gfs2_ail1_start(sdp); | |
26b06a69 | 1036 | gfs2_ail1_wait(sdp); |
4667a0ec | 1037 | gfs2_ail1_empty(sdp); |
805c0907 BP |
1038 | gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL | |
1039 | GFS2_LFC_LOGD_AIL_FLUSH_REQD); | |
b63f5e84 | 1040 | did_flush = true; |
ec69b188 SW |
1041 | } |
1042 | ||
b63f5e84 | 1043 | if (!gfs2_ail_flush_reqd(sdp) || did_flush) |
26b06a69 SW |
1044 | wake_up(&sdp->sd_log_waitq); |
1045 | ||
ec69b188 | 1046 | t = gfs2_tune_get(sdp, gt_logd_secs) * HZ; |
a0acae0e TH |
1047 | |
1048 | try_to_freeze(); | |
5e687eac BM |
1049 | |
1050 | do { | |
1051 | prepare_to_wait(&sdp->sd_logd_waitq, &wait, | |
5f487490 | 1052 | TASK_INTERRUPTIBLE); |
5e687eac BM |
1053 | if (!gfs2_ail_flush_reqd(sdp) && |
1054 | !gfs2_jrnl_flush_reqd(sdp) && | |
1055 | !kthread_should_stop()) | |
1056 | t = schedule_timeout(t); | |
1057 | } while(t && !gfs2_ail_flush_reqd(sdp) && | |
1058 | !gfs2_jrnl_flush_reqd(sdp) && | |
1059 | !kthread_should_stop()); | |
1060 | finish_wait(&sdp->sd_logd_waitq, &wait); | |
ec69b188 SW |
1061 | } |
1062 | ||
1063 | return 0; | |
1064 | } | |
1065 |