Commit | Line | Data |
---|---|---|
71e330b5 DC |
1 | /* |
2 | * Copyright (c) 2010 Red Hat, Inc. All Rights Reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it would be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
11 | * GNU General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License | |
14 | * along with this program; if not, write the Free Software Foundation, | |
15 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
16 | */ | |
17 | ||
18 | #include "xfs.h" | |
19 | #include "xfs_fs.h" | |
239880ef | 20 | #include "xfs_log_format.h" |
70a9883c | 21 | #include "xfs_shared.h" |
239880ef | 22 | #include "xfs_trans_resv.h" |
71e330b5 DC |
23 | #include "xfs_sb.h" |
24 | #include "xfs_ag.h" | |
71e330b5 DC |
25 | #include "xfs_mount.h" |
26 | #include "xfs_error.h" | |
27 | #include "xfs_alloc.h" | |
efc27b52 | 28 | #include "xfs_extent_busy.h" |
e84661aa | 29 | #include "xfs_discard.h" |
239880ef DC |
30 | #include "xfs_trans.h" |
31 | #include "xfs_trans_priv.h" | |
32 | #include "xfs_log.h" | |
33 | #include "xfs_log_priv.h" | |
71e330b5 | 34 | |
71e330b5 DC |
35 | /* |
36 | * Allocate a new ticket. Failing to get a new ticket makes it really hard to | |
37 | * recover, so we don't allow failure here. Also, we allocate in a context that | |
38 | * we don't want to be issuing transactions from, so we need to tell the | |
39 | * allocation code this as well. | |
40 | * | |
41 | * We don't reserve any space for the ticket - we are going to steal whatever | |
42 | * space we require from transactions as they commit. To ensure we reserve all | |
43 | * the space required, we need to set the current reservation of the ticket to | |
44 | * zero so that we know to steal the initial transaction overhead from the | |
45 | * first transaction commit. | |
46 | */ | |
47 | static struct xlog_ticket * | |
48 | xlog_cil_ticket_alloc( | |
f7bdf03a | 49 | struct xlog *log) |
71e330b5 DC |
50 | { |
51 | struct xlog_ticket *tic; | |
52 | ||
53 | tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0, | |
54 | KM_SLEEP|KM_NOFS); | |
55 | tic->t_trans_type = XFS_TRANS_CHECKPOINT; | |
56 | ||
57 | /* | |
58 | * set the current reservation to zero so we know to steal the basic | |
59 | * transaction overhead reservation from the first transaction commit. | |
60 | */ | |
61 | tic->t_curr_res = 0; | |
62 | return tic; | |
63 | } | |
64 | ||
65 | /* | |
66 | * After the first stage of log recovery is done, we know where the head and | |
67 | * tail of the log are. We need this log initialisation done before we can | |
68 | * initialise the first CIL checkpoint context. | |
69 | * | |
70 | * Here we allocate a log ticket to track space usage during a CIL push. This | |
71 | * ticket is passed to xlog_write() directly so that we don't slowly leak log | |
72 | * space by failing to account for space used by log headers and additional | |
73 | * region headers for split regions. | |
74 | */ | |
75 | void | |
76 | xlog_cil_init_post_recovery( | |
f7bdf03a | 77 | struct xlog *log) |
71e330b5 | 78 | { |
71e330b5 DC |
79 | log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log); |
80 | log->l_cilp->xc_ctx->sequence = 1; | |
71e330b5 DC |
81 | } |
82 | ||
991aaf65 DC |
83 | /* |
84 | * Prepare the log item for insertion into the CIL. Calculate the difference in | |
85 | * log space and vectors it will consume, and if it is a new item pin it as | |
86 | * well. | |
87 | */ | |
88 | STATIC void | |
89 | xfs_cil_prepare_item( | |
90 | struct xlog *log, | |
91 | struct xfs_log_vec *lv, | |
92 | struct xfs_log_vec *old_lv, | |
93 | int *diff_len, | |
94 | int *diff_iovecs) | |
95 | { | |
96 | /* Account for the new LV being passed in */ | |
97 | if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) { | |
110dc24a | 98 | *diff_len += lv->lv_bytes; |
991aaf65 DC |
99 | *diff_iovecs += lv->lv_niovecs; |
100 | } | |
101 | ||
102 | /* | |
103 | * If there is no old LV, this is the first time we've seen the item in | |
104 | * this CIL context and so we need to pin it. If we are replacing the | |
105 | * old_lv, then remove the space it accounts for and free it. | |
106 | */ | |
107 | if (!old_lv) | |
108 | lv->lv_item->li_ops->iop_pin(lv->lv_item); | |
109 | else if (old_lv != lv) { | |
110 | ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED); | |
111 | ||
110dc24a | 112 | *diff_len -= old_lv->lv_bytes; |
991aaf65 DC |
113 | *diff_iovecs -= old_lv->lv_niovecs; |
114 | kmem_free(old_lv); | |
115 | } | |
116 | ||
117 | /* attach new log vector to log item */ | |
118 | lv->lv_item->li_lv = lv; | |
119 | ||
120 | /* | |
121 | * If this is the first time the item is being committed to the | |
122 | * CIL, store the sequence number on the log item so we can | |
123 | * tell in future commits whether this is the first checkpoint | |
124 | * the item is being committed into. | |
125 | */ | |
126 | if (!lv->lv_item->li_seq) | |
127 | lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence; | |
128 | } | |
129 | ||
71e330b5 DC |
130 | /* |
131 | * Format log item into a flat buffers | |
132 | * | |
133 | * For delayed logging, we need to hold a formatted buffer containing all the | |
134 | * changes on the log item. This enables us to relog the item in memory and | |
135 | * write it out asynchronously without needing to relock the object that was | |
136 | * modified at the time it gets written into the iclog. | |
137 | * | |
138 | * This function builds a vector for the changes in each log item in the | |
139 | * transaction. It then works out the length of the buffer needed for each log | |
140 | * item, allocates them and formats the vector for the item into the buffer. | |
141 | * The buffer is then attached to the log item are then inserted into the | |
142 | * Committed Item List for tracking until the next checkpoint is written out. | |
143 | * | |
144 | * We don't set up region headers during this process; we simply copy the | |
145 | * regions into the flat buffer. We can do this because we still have to do a | |
146 | * formatting step to write the regions into the iclog buffer. Writing the | |
147 | * ophdrs during the iclog write means that we can support splitting large | |
148 | * regions across iclog boundares without needing a change in the format of the | |
149 | * item/region encapsulation. | |
150 | * | |
151 | * Hence what we need to do now is change the rewrite the vector array to point | |
152 | * to the copied region inside the buffer we just allocated. This allows us to | |
153 | * format the regions into the iclog as though they are being formatted | |
154 | * directly out of the objects themselves. | |
155 | */ | |
991aaf65 DC |
156 | static void |
157 | xlog_cil_insert_format_items( | |
158 | struct xlog *log, | |
159 | struct xfs_trans *tp, | |
160 | int *diff_len, | |
161 | int *diff_iovecs) | |
71e330b5 | 162 | { |
0244b960 | 163 | struct xfs_log_item_desc *lidp; |
71e330b5 | 164 | |
0244b960 CH |
165 | |
166 | /* Bail out if we didn't find a log item. */ | |
167 | if (list_empty(&tp->t_items)) { | |
168 | ASSERT(0); | |
991aaf65 | 169 | return; |
0244b960 CH |
170 | } |
171 | ||
172 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { | |
166d1368 | 173 | struct xfs_log_item *lip = lidp->lid_item; |
7492c5b4 | 174 | struct xfs_log_vec *lv; |
991aaf65 | 175 | struct xfs_log_vec *old_lv; |
7492c5b4 DC |
176 | int niovecs = 0; |
177 | int nbytes = 0; | |
178 | int buf_size; | |
fd63875c | 179 | bool ordered = false; |
71e330b5 | 180 | |
0244b960 CH |
181 | /* Skip items which aren't dirty in this transaction. */ |
182 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) | |
183 | continue; | |
184 | ||
166d1368 DC |
185 | /* get number of vecs and size of data to be stored */ |
186 | lip->li_ops->iop_size(lip, &niovecs, &nbytes); | |
187 | ||
0244b960 | 188 | /* Skip items that do not have any vectors for writing */ |
b3934213 | 189 | if (!niovecs) |
0244b960 CH |
190 | continue; |
191 | ||
fd63875c DC |
192 | /* |
193 | * Ordered items need to be tracked but we do not wish to write | |
194 | * them. We need a logvec to track the object, but we do not | |
195 | * need an iovec or buffer to be allocated for copying data. | |
196 | */ | |
197 | if (niovecs == XFS_LOG_VEC_ORDERED) { | |
198 | ordered = true; | |
199 | niovecs = 0; | |
7492c5b4 | 200 | nbytes = 0; |
fd63875c DC |
201 | } |
202 | ||
bde7cff6 CH |
203 | /* |
204 | * We 64-bit align the length of each iovec so that the start | |
205 | * of the next one is naturally aligned. We'll need to | |
3895e51f DC |
206 | * account for that slack space here. Then round nbytes up |
207 | * to 64-bit alignment so that the initial buffer alignment is | |
208 | * easy to calculate and verify. | |
bde7cff6 CH |
209 | */ |
210 | nbytes += niovecs * sizeof(uint64_t); | |
3895e51f | 211 | nbytes = round_up(nbytes, sizeof(uint64_t)); |
bde7cff6 | 212 | |
991aaf65 DC |
213 | /* grab the old item if it exists for reservation accounting */ |
214 | old_lv = lip->li_lv; | |
215 | ||
3895e51f DC |
216 | /* |
217 | * The data buffer needs to start 64-bit aligned, so round up | |
218 | * that space to ensure we can align it appropriately and not | |
219 | * overrun the buffer. | |
220 | */ | |
221 | buf_size = nbytes + | |
222 | round_up((sizeof(struct xfs_log_vec) + | |
223 | niovecs * sizeof(struct xfs_log_iovec)), | |
224 | sizeof(uint64_t)); | |
0244b960 | 225 | |
f5baac35 DC |
226 | /* compare to existing item size */ |
227 | if (lip->li_lv && buf_size <= lip->li_lv->lv_size) { | |
228 | /* same or smaller, optimise common overwrite case */ | |
229 | lv = lip->li_lv; | |
230 | lv->lv_next = NULL; | |
231 | ||
232 | if (ordered) | |
233 | goto insert; | |
234 | ||
991aaf65 DC |
235 | /* |
236 | * set the item up as though it is a new insertion so | |
237 | * that the space reservation accounting is correct. | |
238 | */ | |
239 | *diff_iovecs -= lv->lv_niovecs; | |
110dc24a | 240 | *diff_len -= lv->lv_bytes; |
9597df6b CH |
241 | } else { |
242 | /* allocate new data chunk */ | |
243 | lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS); | |
244 | lv->lv_item = lip; | |
245 | lv->lv_size = buf_size; | |
246 | if (ordered) { | |
247 | /* track as an ordered logvec */ | |
248 | ASSERT(lip->li_lv == NULL); | |
249 | lv->lv_buf_len = XFS_LOG_VEC_ORDERED; | |
250 | goto insert; | |
251 | } | |
252 | lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1]; | |
f5baac35 DC |
253 | } |
254 | ||
9597df6b | 255 | /* Ensure the lv is set up according to ->iop_size */ |
7492c5b4 | 256 | lv->lv_niovecs = niovecs; |
71e330b5 | 257 | |
7492c5b4 | 258 | /* The allocated data region lies beyond the iovec region */ |
bde7cff6 | 259 | lv->lv_buf_len = 0; |
110dc24a | 260 | lv->lv_bytes = 0; |
7492c5b4 | 261 | lv->lv_buf = (char *)lv + buf_size - nbytes; |
3895e51f DC |
262 | ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t))); |
263 | ||
bde7cff6 | 264 | lip->li_ops->iop_format(lip, lv); |
7492c5b4 | 265 | insert: |
f5baac35 | 266 | ASSERT(lv->lv_buf_len <= nbytes); |
991aaf65 | 267 | xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs); |
3b93c7aa | 268 | } |
d1583a38 DC |
269 | } |
270 | ||
271 | /* | |
272 | * Insert the log items into the CIL and calculate the difference in space | |
273 | * consumed by the item. Add the space to the checkpoint ticket and calculate | |
274 | * if the change requires additional log metadata. If it does, take that space | |
42b2aa86 | 275 | * as well. Remove the amount of space we added to the checkpoint ticket from |
d1583a38 DC |
276 | * the current transaction ticket so that the accounting works out correctly. |
277 | */ | |
3b93c7aa DC |
278 | static void |
279 | xlog_cil_insert_items( | |
f7bdf03a | 280 | struct xlog *log, |
991aaf65 | 281 | struct xfs_trans *tp) |
3b93c7aa | 282 | { |
d1583a38 DC |
283 | struct xfs_cil *cil = log->l_cilp; |
284 | struct xfs_cil_ctx *ctx = cil->xc_ctx; | |
991aaf65 | 285 | struct xfs_log_item_desc *lidp; |
d1583a38 DC |
286 | int len = 0; |
287 | int diff_iovecs = 0; | |
288 | int iclog_space; | |
3b93c7aa | 289 | |
991aaf65 | 290 | ASSERT(tp); |
d1583a38 DC |
291 | |
292 | /* | |
d1583a38 DC |
293 | * We can do this safely because the context can't checkpoint until we |
294 | * are done so it doesn't matter exactly how we update the CIL. | |
295 | */ | |
991aaf65 DC |
296 | xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs); |
297 | ||
298 | /* | |
299 | * Now (re-)position everything modified at the tail of the CIL. | |
300 | * We do this here so we only need to take the CIL lock once during | |
301 | * the transaction commit. | |
302 | */ | |
d1583a38 | 303 | spin_lock(&cil->xc_cil_lock); |
991aaf65 DC |
304 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { |
305 | struct xfs_log_item *lip = lidp->lid_item; | |
d1583a38 | 306 | |
991aaf65 DC |
307 | /* Skip items which aren't dirty in this transaction. */ |
308 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) | |
309 | continue; | |
fd63875c | 310 | |
991aaf65 | 311 | list_move_tail(&lip->li_cil, &cil->xc_cil); |
fd63875c | 312 | } |
d1583a38 | 313 | |
fd63875c DC |
314 | /* account for space used by new iovec headers */ |
315 | len += diff_iovecs * sizeof(xlog_op_header_t); | |
d1583a38 DC |
316 | ctx->nvecs += diff_iovecs; |
317 | ||
991aaf65 DC |
318 | /* attach the transaction to the CIL if it has any busy extents */ |
319 | if (!list_empty(&tp->t_busy)) | |
320 | list_splice_init(&tp->t_busy, &ctx->busy_extents); | |
321 | ||
d1583a38 DC |
322 | /* |
323 | * Now transfer enough transaction reservation to the context ticket | |
324 | * for the checkpoint. The context ticket is special - the unit | |
325 | * reservation has to grow as well as the current reservation as we | |
326 | * steal from tickets so we can correctly determine the space used | |
327 | * during the transaction commit. | |
328 | */ | |
329 | if (ctx->ticket->t_curr_res == 0) { | |
d1583a38 | 330 | ctx->ticket->t_curr_res = ctx->ticket->t_unit_res; |
991aaf65 | 331 | tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res; |
d1583a38 DC |
332 | } |
333 | ||
334 | /* do we need space for more log record headers? */ | |
335 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
336 | if (len > 0 && (ctx->space_used / iclog_space != | |
337 | (ctx->space_used + len) / iclog_space)) { | |
338 | int hdrs; | |
339 | ||
340 | hdrs = (len + iclog_space - 1) / iclog_space; | |
341 | /* need to take into account split region headers, too */ | |
342 | hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header); | |
343 | ctx->ticket->t_unit_res += hdrs; | |
344 | ctx->ticket->t_curr_res += hdrs; | |
991aaf65 DC |
345 | tp->t_ticket->t_curr_res -= hdrs; |
346 | ASSERT(tp->t_ticket->t_curr_res >= len); | |
d1583a38 | 347 | } |
991aaf65 | 348 | tp->t_ticket->t_curr_res -= len; |
d1583a38 DC |
349 | ctx->space_used += len; |
350 | ||
351 | spin_unlock(&cil->xc_cil_lock); | |
71e330b5 DC |
352 | } |
353 | ||
354 | static void | |
355 | xlog_cil_free_logvec( | |
356 | struct xfs_log_vec *log_vector) | |
357 | { | |
358 | struct xfs_log_vec *lv; | |
359 | ||
360 | for (lv = log_vector; lv; ) { | |
361 | struct xfs_log_vec *next = lv->lv_next; | |
71e330b5 DC |
362 | kmem_free(lv); |
363 | lv = next; | |
364 | } | |
365 | } | |
366 | ||
71e330b5 DC |
367 | /* |
368 | * Mark all items committed and clear busy extents. We free the log vector | |
369 | * chains in a separate pass so that we unpin the log items as quickly as | |
370 | * possible. | |
371 | */ | |
372 | static void | |
373 | xlog_cil_committed( | |
374 | void *args, | |
375 | int abort) | |
376 | { | |
377 | struct xfs_cil_ctx *ctx = args; | |
e84661aa | 378 | struct xfs_mount *mp = ctx->cil->xc_log->l_mp; |
71e330b5 | 379 | |
0e57f6a3 DC |
380 | xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain, |
381 | ctx->start_lsn, abort); | |
71e330b5 | 382 | |
4ecbfe63 DC |
383 | xfs_extent_busy_sort(&ctx->busy_extents); |
384 | xfs_extent_busy_clear(mp, &ctx->busy_extents, | |
e84661aa | 385 | (mp->m_flags & XFS_MOUNT_DISCARD) && !abort); |
71e330b5 | 386 | |
ac983517 DC |
387 | /* |
388 | * If we are aborting the commit, wake up anyone waiting on the | |
389 | * committing list. If we don't, then a shutdown we can leave processes | |
390 | * waiting in xlog_cil_force_lsn() waiting on a sequence commit that | |
391 | * will never happen because we aborted it. | |
392 | */ | |
4bb928cd | 393 | spin_lock(&ctx->cil->xc_push_lock); |
ac983517 DC |
394 | if (abort) |
395 | wake_up_all(&ctx->cil->xc_commit_wait); | |
71e330b5 | 396 | list_del(&ctx->committing); |
4bb928cd | 397 | spin_unlock(&ctx->cil->xc_push_lock); |
71e330b5 DC |
398 | |
399 | xlog_cil_free_logvec(ctx->lv_chain); | |
e84661aa CH |
400 | |
401 | if (!list_empty(&ctx->busy_extents)) { | |
402 | ASSERT(mp->m_flags & XFS_MOUNT_DISCARD); | |
403 | ||
404 | xfs_discard_extents(mp, &ctx->busy_extents); | |
4ecbfe63 | 405 | xfs_extent_busy_clear(mp, &ctx->busy_extents, false); |
e84661aa CH |
406 | } |
407 | ||
71e330b5 DC |
408 | kmem_free(ctx); |
409 | } | |
410 | ||
411 | /* | |
a44f13ed DC |
412 | * Push the Committed Item List to the log. If @push_seq flag is zero, then it |
413 | * is a background flush and so we can chose to ignore it. Otherwise, if the | |
414 | * current sequence is the same as @push_seq we need to do a flush. If | |
415 | * @push_seq is less than the current sequence, then it has already been | |
416 | * flushed and we don't need to do anything - the caller will wait for it to | |
417 | * complete if necessary. | |
418 | * | |
419 | * @push_seq is a value rather than a flag because that allows us to do an | |
420 | * unlocked check of the sequence number for a match. Hence we can allows log | |
421 | * forces to run racily and not issue pushes for the same sequence twice. If we | |
422 | * get a race between multiple pushes for the same sequence they will block on | |
423 | * the first one and then abort, hence avoiding needless pushes. | |
71e330b5 | 424 | */ |
a44f13ed | 425 | STATIC int |
71e330b5 | 426 | xlog_cil_push( |
f7bdf03a | 427 | struct xlog *log) |
71e330b5 DC |
428 | { |
429 | struct xfs_cil *cil = log->l_cilp; | |
430 | struct xfs_log_vec *lv; | |
431 | struct xfs_cil_ctx *ctx; | |
432 | struct xfs_cil_ctx *new_ctx; | |
433 | struct xlog_in_core *commit_iclog; | |
434 | struct xlog_ticket *tic; | |
71e330b5 | 435 | int num_iovecs; |
71e330b5 DC |
436 | int error = 0; |
437 | struct xfs_trans_header thdr; | |
438 | struct xfs_log_iovec lhdr; | |
439 | struct xfs_log_vec lvhdr = { NULL }; | |
440 | xfs_lsn_t commit_lsn; | |
4c2d542f | 441 | xfs_lsn_t push_seq; |
71e330b5 DC |
442 | |
443 | if (!cil) | |
444 | return 0; | |
445 | ||
71e330b5 DC |
446 | new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS); |
447 | new_ctx->ticket = xlog_cil_ticket_alloc(log); | |
448 | ||
4c2d542f | 449 | down_write(&cil->xc_ctx_lock); |
71e330b5 DC |
450 | ctx = cil->xc_ctx; |
451 | ||
4bb928cd | 452 | spin_lock(&cil->xc_push_lock); |
4c2d542f DC |
453 | push_seq = cil->xc_push_seq; |
454 | ASSERT(push_seq <= ctx->sequence); | |
71e330b5 | 455 | |
4c2d542f DC |
456 | /* |
457 | * Check if we've anything to push. If there is nothing, then we don't | |
458 | * move on to a new sequence number and so we have to be able to push | |
459 | * this sequence again later. | |
460 | */ | |
461 | if (list_empty(&cil->xc_cil)) { | |
462 | cil->xc_push_seq = 0; | |
4bb928cd | 463 | spin_unlock(&cil->xc_push_lock); |
a44f13ed | 464 | goto out_skip; |
4c2d542f | 465 | } |
4bb928cd | 466 | spin_unlock(&cil->xc_push_lock); |
4c2d542f | 467 | |
a44f13ed DC |
468 | |
469 | /* check for a previously pushed seqeunce */ | |
4c2d542f | 470 | if (push_seq < cil->xc_ctx->sequence) |
df806158 DC |
471 | goto out_skip; |
472 | ||
71e330b5 DC |
473 | /* |
474 | * pull all the log vectors off the items in the CIL, and | |
475 | * remove the items from the CIL. We don't need the CIL lock | |
476 | * here because it's only needed on the transaction commit | |
477 | * side which is currently locked out by the flush lock. | |
478 | */ | |
479 | lv = NULL; | |
71e330b5 | 480 | num_iovecs = 0; |
71e330b5 DC |
481 | while (!list_empty(&cil->xc_cil)) { |
482 | struct xfs_log_item *item; | |
71e330b5 DC |
483 | |
484 | item = list_first_entry(&cil->xc_cil, | |
485 | struct xfs_log_item, li_cil); | |
486 | list_del_init(&item->li_cil); | |
487 | if (!ctx->lv_chain) | |
488 | ctx->lv_chain = item->li_lv; | |
489 | else | |
490 | lv->lv_next = item->li_lv; | |
491 | lv = item->li_lv; | |
492 | item->li_lv = NULL; | |
71e330b5 | 493 | num_iovecs += lv->lv_niovecs; |
71e330b5 DC |
494 | } |
495 | ||
496 | /* | |
497 | * initialise the new context and attach it to the CIL. Then attach | |
498 | * the current context to the CIL committing lsit so it can be found | |
499 | * during log forces to extract the commit lsn of the sequence that | |
500 | * needs to be forced. | |
501 | */ | |
502 | INIT_LIST_HEAD(&new_ctx->committing); | |
503 | INIT_LIST_HEAD(&new_ctx->busy_extents); | |
504 | new_ctx->sequence = ctx->sequence + 1; | |
505 | new_ctx->cil = cil; | |
506 | cil->xc_ctx = new_ctx; | |
507 | ||
508 | /* | |
509 | * The switch is now done, so we can drop the context lock and move out | |
510 | * of a shared context. We can't just go straight to the commit record, | |
511 | * though - we need to synchronise with previous and future commits so | |
512 | * that the commit records are correctly ordered in the log to ensure | |
513 | * that we process items during log IO completion in the correct order. | |
514 | * | |
515 | * For example, if we get an EFI in one checkpoint and the EFD in the | |
516 | * next (e.g. due to log forces), we do not want the checkpoint with | |
517 | * the EFD to be committed before the checkpoint with the EFI. Hence | |
518 | * we must strictly order the commit records of the checkpoints so | |
519 | * that: a) the checkpoint callbacks are attached to the iclogs in the | |
520 | * correct order; and b) the checkpoints are replayed in correct order | |
521 | * in log recovery. | |
522 | * | |
523 | * Hence we need to add this context to the committing context list so | |
524 | * that higher sequences will wait for us to write out a commit record | |
525 | * before they do. | |
f876e446 DC |
526 | * |
527 | * xfs_log_force_lsn requires us to mirror the new sequence into the cil | |
528 | * structure atomically with the addition of this sequence to the | |
529 | * committing list. This also ensures that we can do unlocked checks | |
530 | * against the current sequence in log forces without risking | |
531 | * deferencing a freed context pointer. | |
71e330b5 | 532 | */ |
4bb928cd | 533 | spin_lock(&cil->xc_push_lock); |
f876e446 | 534 | cil->xc_current_sequence = new_ctx->sequence; |
71e330b5 | 535 | list_add(&ctx->committing, &cil->xc_committing); |
4bb928cd | 536 | spin_unlock(&cil->xc_push_lock); |
71e330b5 DC |
537 | up_write(&cil->xc_ctx_lock); |
538 | ||
539 | /* | |
540 | * Build a checkpoint transaction header and write it to the log to | |
541 | * begin the transaction. We need to account for the space used by the | |
542 | * transaction header here as it is not accounted for in xlog_write(). | |
543 | * | |
544 | * The LSN we need to pass to the log items on transaction commit is | |
545 | * the LSN reported by the first log vector write. If we use the commit | |
546 | * record lsn then we can move the tail beyond the grant write head. | |
547 | */ | |
548 | tic = ctx->ticket; | |
549 | thdr.th_magic = XFS_TRANS_HEADER_MAGIC; | |
550 | thdr.th_type = XFS_TRANS_CHECKPOINT; | |
551 | thdr.th_tid = tic->t_tid; | |
552 | thdr.th_num_items = num_iovecs; | |
4e0d5f92 | 553 | lhdr.i_addr = &thdr; |
71e330b5 DC |
554 | lhdr.i_len = sizeof(xfs_trans_header_t); |
555 | lhdr.i_type = XLOG_REG_TYPE_TRANSHDR; | |
556 | tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t); | |
557 | ||
558 | lvhdr.lv_niovecs = 1; | |
559 | lvhdr.lv_iovecp = &lhdr; | |
560 | lvhdr.lv_next = ctx->lv_chain; | |
561 | ||
562 | error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); | |
563 | if (error) | |
7db37c5e | 564 | goto out_abort_free_ticket; |
71e330b5 DC |
565 | |
566 | /* | |
567 | * now that we've written the checkpoint into the log, strictly | |
568 | * order the commit records so replay will get them in the right order. | |
569 | */ | |
570 | restart: | |
4bb928cd | 571 | spin_lock(&cil->xc_push_lock); |
71e330b5 | 572 | list_for_each_entry(new_ctx, &cil->xc_committing, committing) { |
ac983517 DC |
573 | /* |
574 | * Avoid getting stuck in this loop because we were woken by the | |
575 | * shutdown, but then went back to sleep once already in the | |
576 | * shutdown state. | |
577 | */ | |
578 | if (XLOG_FORCED_SHUTDOWN(log)) { | |
579 | spin_unlock(&cil->xc_push_lock); | |
580 | goto out_abort_free_ticket; | |
581 | } | |
582 | ||
71e330b5 DC |
583 | /* |
584 | * Higher sequences will wait for this one so skip them. | |
ac983517 | 585 | * Don't wait for our own sequence, either. |
71e330b5 DC |
586 | */ |
587 | if (new_ctx->sequence >= ctx->sequence) | |
588 | continue; | |
589 | if (!new_ctx->commit_lsn) { | |
590 | /* | |
591 | * It is still being pushed! Wait for the push to | |
592 | * complete, then start again from the beginning. | |
593 | */ | |
4bb928cd | 594 | xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock); |
71e330b5 DC |
595 | goto restart; |
596 | } | |
597 | } | |
4bb928cd | 598 | spin_unlock(&cil->xc_push_lock); |
71e330b5 | 599 | |
7db37c5e | 600 | /* xfs_log_done always frees the ticket on error. */ |
71e330b5 | 601 | commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); |
7db37c5e | 602 | if (commit_lsn == -1) |
71e330b5 DC |
603 | goto out_abort; |
604 | ||
605 | /* attach all the transactions w/ busy extents to iclog */ | |
606 | ctx->log_cb.cb_func = xlog_cil_committed; | |
607 | ctx->log_cb.cb_arg = ctx; | |
608 | error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb); | |
609 | if (error) | |
610 | goto out_abort; | |
611 | ||
612 | /* | |
613 | * now the checkpoint commit is complete and we've attached the | |
614 | * callbacks to the iclog we can assign the commit LSN to the context | |
615 | * and wake up anyone who is waiting for the commit to complete. | |
616 | */ | |
4bb928cd | 617 | spin_lock(&cil->xc_push_lock); |
71e330b5 | 618 | ctx->commit_lsn = commit_lsn; |
eb40a875 | 619 | wake_up_all(&cil->xc_commit_wait); |
4bb928cd | 620 | spin_unlock(&cil->xc_push_lock); |
71e330b5 DC |
621 | |
622 | /* release the hounds! */ | |
623 | return xfs_log_release_iclog(log->l_mp, commit_iclog); | |
624 | ||
625 | out_skip: | |
626 | up_write(&cil->xc_ctx_lock); | |
627 | xfs_log_ticket_put(new_ctx->ticket); | |
628 | kmem_free(new_ctx); | |
629 | return 0; | |
630 | ||
7db37c5e DC |
631 | out_abort_free_ticket: |
632 | xfs_log_ticket_put(tic); | |
71e330b5 DC |
633 | out_abort: |
634 | xlog_cil_committed(ctx, XFS_LI_ABORTED); | |
2451337d | 635 | return -EIO; |
71e330b5 DC |
636 | } |
637 | ||
4c2d542f DC |
638 | static void |
639 | xlog_cil_push_work( | |
640 | struct work_struct *work) | |
641 | { | |
642 | struct xfs_cil *cil = container_of(work, struct xfs_cil, | |
643 | xc_push_work); | |
644 | xlog_cil_push(cil->xc_log); | |
645 | } | |
646 | ||
647 | /* | |
648 | * We need to push CIL every so often so we don't cache more than we can fit in | |
649 | * the log. The limit really is that a checkpoint can't be more than half the | |
650 | * log (the current checkpoint is not allowed to overwrite the previous | |
651 | * checkpoint), but commit latency and memory usage limit this to a smaller | |
652 | * size. | |
653 | */ | |
654 | static void | |
655 | xlog_cil_push_background( | |
f7bdf03a | 656 | struct xlog *log) |
4c2d542f DC |
657 | { |
658 | struct xfs_cil *cil = log->l_cilp; | |
659 | ||
660 | /* | |
661 | * The cil won't be empty because we are called while holding the | |
662 | * context lock so whatever we added to the CIL will still be there | |
663 | */ | |
664 | ASSERT(!list_empty(&cil->xc_cil)); | |
665 | ||
666 | /* | |
667 | * don't do a background push if we haven't used up all the | |
668 | * space available yet. | |
669 | */ | |
670 | if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) | |
671 | return; | |
672 | ||
4bb928cd | 673 | spin_lock(&cil->xc_push_lock); |
4c2d542f DC |
674 | if (cil->xc_push_seq < cil->xc_current_sequence) { |
675 | cil->xc_push_seq = cil->xc_current_sequence; | |
676 | queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work); | |
677 | } | |
4bb928cd | 678 | spin_unlock(&cil->xc_push_lock); |
4c2d542f DC |
679 | |
680 | } | |
681 | ||
f876e446 DC |
682 | /* |
683 | * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence | |
684 | * number that is passed. When it returns, the work will be queued for | |
685 | * @push_seq, but it won't be completed. The caller is expected to do any | |
686 | * waiting for push_seq to complete if it is required. | |
687 | */ | |
4c2d542f | 688 | static void |
f876e446 | 689 | xlog_cil_push_now( |
f7bdf03a | 690 | struct xlog *log, |
4c2d542f DC |
691 | xfs_lsn_t push_seq) |
692 | { | |
693 | struct xfs_cil *cil = log->l_cilp; | |
694 | ||
695 | if (!cil) | |
696 | return; | |
697 | ||
698 | ASSERT(push_seq && push_seq <= cil->xc_current_sequence); | |
699 | ||
700 | /* start on any pending background push to minimise wait time on it */ | |
701 | flush_work(&cil->xc_push_work); | |
702 | ||
703 | /* | |
704 | * If the CIL is empty or we've already pushed the sequence then | |
705 | * there's no work we need to do. | |
706 | */ | |
4bb928cd | 707 | spin_lock(&cil->xc_push_lock); |
4c2d542f | 708 | if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) { |
4bb928cd | 709 | spin_unlock(&cil->xc_push_lock); |
4c2d542f DC |
710 | return; |
711 | } | |
712 | ||
713 | cil->xc_push_seq = push_seq; | |
f876e446 | 714 | queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work); |
4bb928cd | 715 | spin_unlock(&cil->xc_push_lock); |
4c2d542f DC |
716 | } |
717 | ||
2c6e24ce DC |
718 | bool |
719 | xlog_cil_empty( | |
720 | struct xlog *log) | |
721 | { | |
722 | struct xfs_cil *cil = log->l_cilp; | |
723 | bool empty = false; | |
724 | ||
725 | spin_lock(&cil->xc_push_lock); | |
726 | if (list_empty(&cil->xc_cil)) | |
727 | empty = true; | |
728 | spin_unlock(&cil->xc_push_lock); | |
729 | return empty; | |
730 | } | |
731 | ||
a44f13ed DC |
732 | /* |
733 | * Commit a transaction with the given vector to the Committed Item List. | |
734 | * | |
735 | * To do this, we need to format the item, pin it in memory if required and | |
736 | * account for the space used by the transaction. Once we have done that we | |
737 | * need to release the unused reservation for the transaction, attach the | |
738 | * transaction to the checkpoint context so we carry the busy extents through | |
739 | * to checkpoint completion, and then unlock all the items in the transaction. | |
740 | * | |
a44f13ed DC |
741 | * Called with the context lock already held in read mode to lock out |
742 | * background commit, returns without it held once background commits are | |
743 | * allowed again. | |
744 | */ | |
c6f97264 | 745 | void |
a44f13ed DC |
746 | xfs_log_commit_cil( |
747 | struct xfs_mount *mp, | |
748 | struct xfs_trans *tp, | |
a44f13ed DC |
749 | xfs_lsn_t *commit_lsn, |
750 | int flags) | |
751 | { | |
f7bdf03a | 752 | struct xlog *log = mp->m_log; |
991aaf65 | 753 | struct xfs_cil *cil = log->l_cilp; |
a44f13ed | 754 | int log_flags = 0; |
a44f13ed DC |
755 | |
756 | if (flags & XFS_TRANS_RELEASE_LOG_RES) | |
757 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
758 | ||
f5baac35 | 759 | /* lock out background commit */ |
991aaf65 | 760 | down_read(&cil->xc_ctx_lock); |
f5baac35 | 761 | |
991aaf65 | 762 | xlog_cil_insert_items(log, tp); |
a44f13ed DC |
763 | |
764 | /* check we didn't blow the reservation */ | |
765 | if (tp->t_ticket->t_curr_res < 0) | |
991aaf65 | 766 | xlog_print_tic_res(mp, tp->t_ticket); |
a44f13ed | 767 | |
991aaf65 DC |
768 | tp->t_commit_lsn = cil->xc_ctx->sequence; |
769 | if (commit_lsn) | |
770 | *commit_lsn = tp->t_commit_lsn; | |
a44f13ed | 771 | |
a44f13ed DC |
772 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); |
773 | xfs_trans_unreserve_and_mod_sb(tp); | |
774 | ||
775 | /* | |
776 | * Once all the items of the transaction have been copied to the CIL, | |
777 | * the items can be unlocked and freed. | |
778 | * | |
779 | * This needs to be done before we drop the CIL context lock because we | |
780 | * have to update state in the log items and unlock them before they go | |
781 | * to disk. If we don't, then the CIL checkpoint can race with us and | |
782 | * we can run checkpoint completion before we've updated and unlocked | |
783 | * the log items. This affects (at least) processing of stale buffers, | |
784 | * inodes and EFIs. | |
785 | */ | |
991aaf65 | 786 | xfs_trans_free_items(tp, tp->t_commit_lsn, 0); |
a44f13ed | 787 | |
4c2d542f | 788 | xlog_cil_push_background(log); |
a44f13ed | 789 | |
991aaf65 | 790 | up_read(&cil->xc_ctx_lock); |
a44f13ed DC |
791 | } |
792 | ||
71e330b5 DC |
793 | /* |
794 | * Conditionally push the CIL based on the sequence passed in. | |
795 | * | |
796 | * We only need to push if we haven't already pushed the sequence | |
797 | * number given. Hence the only time we will trigger a push here is | |
798 | * if the push sequence is the same as the current context. | |
799 | * | |
800 | * We return the current commit lsn to allow the callers to determine if a | |
801 | * iclog flush is necessary following this call. | |
71e330b5 DC |
802 | */ |
803 | xfs_lsn_t | |
a44f13ed | 804 | xlog_cil_force_lsn( |
f7bdf03a | 805 | struct xlog *log, |
a44f13ed | 806 | xfs_lsn_t sequence) |
71e330b5 DC |
807 | { |
808 | struct xfs_cil *cil = log->l_cilp; | |
809 | struct xfs_cil_ctx *ctx; | |
810 | xfs_lsn_t commit_lsn = NULLCOMMITLSN; | |
811 | ||
a44f13ed DC |
812 | ASSERT(sequence <= cil->xc_current_sequence); |
813 | ||
814 | /* | |
815 | * check to see if we need to force out the current context. | |
816 | * xlog_cil_push() handles racing pushes for the same sequence, | |
817 | * so no need to deal with it here. | |
818 | */ | |
f876e446 DC |
819 | restart: |
820 | xlog_cil_push_now(log, sequence); | |
71e330b5 DC |
821 | |
822 | /* | |
823 | * See if we can find a previous sequence still committing. | |
71e330b5 DC |
824 | * We need to wait for all previous sequence commits to complete |
825 | * before allowing the force of push_seq to go ahead. Hence block | |
826 | * on commits for those as well. | |
827 | */ | |
4bb928cd | 828 | spin_lock(&cil->xc_push_lock); |
71e330b5 | 829 | list_for_each_entry(ctx, &cil->xc_committing, committing) { |
ac983517 DC |
830 | /* |
831 | * Avoid getting stuck in this loop because we were woken by the | |
832 | * shutdown, but then went back to sleep once already in the | |
833 | * shutdown state. | |
834 | */ | |
835 | if (XLOG_FORCED_SHUTDOWN(log)) | |
836 | goto out_shutdown; | |
a44f13ed | 837 | if (ctx->sequence > sequence) |
71e330b5 DC |
838 | continue; |
839 | if (!ctx->commit_lsn) { | |
840 | /* | |
841 | * It is still being pushed! Wait for the push to | |
842 | * complete, then start again from the beginning. | |
843 | */ | |
4bb928cd | 844 | xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock); |
71e330b5 DC |
845 | goto restart; |
846 | } | |
a44f13ed | 847 | if (ctx->sequence != sequence) |
71e330b5 DC |
848 | continue; |
849 | /* found it! */ | |
850 | commit_lsn = ctx->commit_lsn; | |
851 | } | |
f876e446 DC |
852 | |
853 | /* | |
854 | * The call to xlog_cil_push_now() executes the push in the background. | |
855 | * Hence by the time we have got here it our sequence may not have been | |
856 | * pushed yet. This is true if the current sequence still matches the | |
857 | * push sequence after the above wait loop and the CIL still contains | |
858 | * dirty objects. | |
859 | * | |
ac983517 DC |
860 | * When the push occurs, it will empty the CIL and atomically increment |
861 | * the currect sequence past the push sequence and move it into the | |
862 | * committing list. Of course, if the CIL is clean at the time of the | |
863 | * push, it won't have pushed the CIL at all, so in that case we should | |
864 | * try the push for this sequence again from the start just in case. | |
f876e446 | 865 | */ |
f876e446 DC |
866 | if (sequence == cil->xc_current_sequence && |
867 | !list_empty(&cil->xc_cil)) { | |
868 | spin_unlock(&cil->xc_push_lock); | |
869 | goto restart; | |
870 | } | |
871 | ||
4bb928cd | 872 | spin_unlock(&cil->xc_push_lock); |
71e330b5 | 873 | return commit_lsn; |
ac983517 DC |
874 | |
875 | /* | |
876 | * We detected a shutdown in progress. We need to trigger the log force | |
877 | * to pass through it's iclog state machine error handling, even though | |
878 | * we are already in a shutdown state. Hence we can't return | |
879 | * NULLCOMMITLSN here as that has special meaning to log forces (i.e. | |
880 | * LSN is already stable), so we return a zero LSN instead. | |
881 | */ | |
882 | out_shutdown: | |
883 | spin_unlock(&cil->xc_push_lock); | |
884 | return 0; | |
71e330b5 | 885 | } |
ccf7c23f DC |
886 | |
887 | /* | |
888 | * Check if the current log item was first committed in this sequence. | |
889 | * We can't rely on just the log item being in the CIL, we have to check | |
890 | * the recorded commit sequence number. | |
891 | * | |
892 | * Note: for this to be used in a non-racy manner, it has to be called with | |
893 | * CIL flushing locked out. As a result, it should only be used during the | |
894 | * transaction commit process when deciding what to format into the item. | |
895 | */ | |
896 | bool | |
897 | xfs_log_item_in_current_chkpt( | |
898 | struct xfs_log_item *lip) | |
899 | { | |
900 | struct xfs_cil_ctx *ctx; | |
901 | ||
ccf7c23f DC |
902 | if (list_empty(&lip->li_cil)) |
903 | return false; | |
904 | ||
905 | ctx = lip->li_mountp->m_log->l_cilp->xc_ctx; | |
906 | ||
907 | /* | |
908 | * li_seq is written on the first commit of a log item to record the | |
909 | * first checkpoint it is written to. Hence if it is different to the | |
910 | * current sequence, we're in a new checkpoint. | |
911 | */ | |
912 | if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0) | |
913 | return false; | |
914 | return true; | |
915 | } | |
4c2d542f DC |
916 | |
917 | /* | |
918 | * Perform initial CIL structure initialisation. | |
919 | */ | |
920 | int | |
921 | xlog_cil_init( | |
f7bdf03a | 922 | struct xlog *log) |
4c2d542f DC |
923 | { |
924 | struct xfs_cil *cil; | |
925 | struct xfs_cil_ctx *ctx; | |
926 | ||
927 | cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL); | |
928 | if (!cil) | |
2451337d | 929 | return -ENOMEM; |
4c2d542f DC |
930 | |
931 | ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL); | |
932 | if (!ctx) { | |
933 | kmem_free(cil); | |
2451337d | 934 | return -ENOMEM; |
4c2d542f DC |
935 | } |
936 | ||
937 | INIT_WORK(&cil->xc_push_work, xlog_cil_push_work); | |
938 | INIT_LIST_HEAD(&cil->xc_cil); | |
939 | INIT_LIST_HEAD(&cil->xc_committing); | |
940 | spin_lock_init(&cil->xc_cil_lock); | |
4bb928cd | 941 | spin_lock_init(&cil->xc_push_lock); |
4c2d542f DC |
942 | init_rwsem(&cil->xc_ctx_lock); |
943 | init_waitqueue_head(&cil->xc_commit_wait); | |
944 | ||
945 | INIT_LIST_HEAD(&ctx->committing); | |
946 | INIT_LIST_HEAD(&ctx->busy_extents); | |
947 | ctx->sequence = 1; | |
948 | ctx->cil = cil; | |
949 | cil->xc_ctx = ctx; | |
950 | cil->xc_current_sequence = ctx->sequence; | |
951 | ||
952 | cil->xc_log = log; | |
953 | log->l_cilp = cil; | |
954 | return 0; | |
955 | } | |
956 | ||
957 | void | |
958 | xlog_cil_destroy( | |
f7bdf03a | 959 | struct xlog *log) |
4c2d542f DC |
960 | { |
961 | if (log->l_cilp->xc_ctx) { | |
962 | if (log->l_cilp->xc_ctx->ticket) | |
963 | xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket); | |
964 | kmem_free(log->l_cilp->xc_ctx); | |
965 | } | |
966 | ||
967 | ASSERT(list_empty(&log->l_cilp->xc_cil)); | |
968 | kmem_free(log->l_cilp); | |
969 | } | |
970 |