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" | |
20 | #include "xfs_types.h" | |
21 | #include "xfs_bit.h" | |
22 | #include "xfs_log.h" | |
23 | #include "xfs_inum.h" | |
24 | #include "xfs_trans.h" | |
25 | #include "xfs_trans_priv.h" | |
26 | #include "xfs_log_priv.h" | |
27 | #include "xfs_sb.h" | |
28 | #include "xfs_ag.h" | |
71e330b5 DC |
29 | #include "xfs_mount.h" |
30 | #include "xfs_error.h" | |
31 | #include "xfs_alloc.h" | |
32 | ||
33 | /* | |
34 | * Perform initial CIL structure initialisation. If the CIL is not | |
35 | * enabled in this filesystem, ensure the log->l_cilp is null so | |
36 | * we can check this conditional to determine if we are doing delayed | |
37 | * logging or not. | |
38 | */ | |
39 | int | |
40 | xlog_cil_init( | |
41 | struct log *log) | |
42 | { | |
43 | struct xfs_cil *cil; | |
44 | struct xfs_cil_ctx *ctx; | |
45 | ||
46 | log->l_cilp = NULL; | |
47 | if (!(log->l_mp->m_flags & XFS_MOUNT_DELAYLOG)) | |
48 | return 0; | |
49 | ||
50 | cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL); | |
51 | if (!cil) | |
52 | return ENOMEM; | |
53 | ||
54 | ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL); | |
55 | if (!ctx) { | |
56 | kmem_free(cil); | |
57 | return ENOMEM; | |
58 | } | |
59 | ||
60 | INIT_LIST_HEAD(&cil->xc_cil); | |
61 | INIT_LIST_HEAD(&cil->xc_committing); | |
62 | spin_lock_init(&cil->xc_cil_lock); | |
63 | init_rwsem(&cil->xc_ctx_lock); | |
eb40a875 | 64 | init_waitqueue_head(&cil->xc_commit_wait); |
71e330b5 DC |
65 | |
66 | INIT_LIST_HEAD(&ctx->committing); | |
67 | INIT_LIST_HEAD(&ctx->busy_extents); | |
68 | ctx->sequence = 1; | |
69 | ctx->cil = cil; | |
70 | cil->xc_ctx = ctx; | |
a44f13ed | 71 | cil->xc_current_sequence = ctx->sequence; |
71e330b5 DC |
72 | |
73 | cil->xc_log = log; | |
74 | log->l_cilp = cil; | |
75 | return 0; | |
76 | } | |
77 | ||
78 | void | |
79 | xlog_cil_destroy( | |
80 | struct log *log) | |
81 | { | |
82 | if (!log->l_cilp) | |
83 | return; | |
84 | ||
85 | if (log->l_cilp->xc_ctx) { | |
86 | if (log->l_cilp->xc_ctx->ticket) | |
87 | xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket); | |
88 | kmem_free(log->l_cilp->xc_ctx); | |
89 | } | |
90 | ||
91 | ASSERT(list_empty(&log->l_cilp->xc_cil)); | |
92 | kmem_free(log->l_cilp); | |
93 | } | |
94 | ||
95 | /* | |
96 | * Allocate a new ticket. Failing to get a new ticket makes it really hard to | |
97 | * recover, so we don't allow failure here. Also, we allocate in a context that | |
98 | * we don't want to be issuing transactions from, so we need to tell the | |
99 | * allocation code this as well. | |
100 | * | |
101 | * We don't reserve any space for the ticket - we are going to steal whatever | |
102 | * space we require from transactions as they commit. To ensure we reserve all | |
103 | * the space required, we need to set the current reservation of the ticket to | |
104 | * zero so that we know to steal the initial transaction overhead from the | |
105 | * first transaction commit. | |
106 | */ | |
107 | static struct xlog_ticket * | |
108 | xlog_cil_ticket_alloc( | |
109 | struct log *log) | |
110 | { | |
111 | struct xlog_ticket *tic; | |
112 | ||
113 | tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0, | |
114 | KM_SLEEP|KM_NOFS); | |
115 | tic->t_trans_type = XFS_TRANS_CHECKPOINT; | |
116 | ||
117 | /* | |
118 | * set the current reservation to zero so we know to steal the basic | |
119 | * transaction overhead reservation from the first transaction commit. | |
120 | */ | |
121 | tic->t_curr_res = 0; | |
122 | return tic; | |
123 | } | |
124 | ||
125 | /* | |
126 | * After the first stage of log recovery is done, we know where the head and | |
127 | * tail of the log are. We need this log initialisation done before we can | |
128 | * initialise the first CIL checkpoint context. | |
129 | * | |
130 | * Here we allocate a log ticket to track space usage during a CIL push. This | |
131 | * ticket is passed to xlog_write() directly so that we don't slowly leak log | |
132 | * space by failing to account for space used by log headers and additional | |
133 | * region headers for split regions. | |
134 | */ | |
135 | void | |
136 | xlog_cil_init_post_recovery( | |
137 | struct log *log) | |
138 | { | |
139 | if (!log->l_cilp) | |
140 | return; | |
141 | ||
142 | log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log); | |
143 | log->l_cilp->xc_ctx->sequence = 1; | |
144 | log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle, | |
145 | log->l_curr_block); | |
146 | } | |
147 | ||
71e330b5 DC |
148 | /* |
149 | * Format log item into a flat buffers | |
150 | * | |
151 | * For delayed logging, we need to hold a formatted buffer containing all the | |
152 | * changes on the log item. This enables us to relog the item in memory and | |
153 | * write it out asynchronously without needing to relock the object that was | |
154 | * modified at the time it gets written into the iclog. | |
155 | * | |
156 | * This function builds a vector for the changes in each log item in the | |
157 | * transaction. It then works out the length of the buffer needed for each log | |
158 | * item, allocates them and formats the vector for the item into the buffer. | |
159 | * The buffer is then attached to the log item are then inserted into the | |
160 | * Committed Item List for tracking until the next checkpoint is written out. | |
161 | * | |
162 | * We don't set up region headers during this process; we simply copy the | |
163 | * regions into the flat buffer. We can do this because we still have to do a | |
164 | * formatting step to write the regions into the iclog buffer. Writing the | |
165 | * ophdrs during the iclog write means that we can support splitting large | |
166 | * regions across iclog boundares without needing a change in the format of the | |
167 | * item/region encapsulation. | |
168 | * | |
169 | * Hence what we need to do now is change the rewrite the vector array to point | |
170 | * to the copied region inside the buffer we just allocated. This allows us to | |
171 | * format the regions into the iclog as though they are being formatted | |
172 | * directly out of the objects themselves. | |
173 | */ | |
174 | static void | |
175 | xlog_cil_format_items( | |
176 | struct log *log, | |
3b93c7aa | 177 | struct xfs_log_vec *log_vector) |
71e330b5 DC |
178 | { |
179 | struct xfs_log_vec *lv; | |
180 | ||
71e330b5 DC |
181 | ASSERT(log_vector); |
182 | for (lv = log_vector; lv; lv = lv->lv_next) { | |
183 | void *ptr; | |
184 | int index; | |
185 | int len = 0; | |
186 | ||
187 | /* build the vector array and calculate it's length */ | |
188 | IOP_FORMAT(lv->lv_item, lv->lv_iovecp); | |
189 | for (index = 0; index < lv->lv_niovecs; index++) | |
190 | len += lv->lv_iovecp[index].i_len; | |
191 | ||
192 | lv->lv_buf_len = len; | |
d1583a38 | 193 | lv->lv_buf = kmem_alloc(lv->lv_buf_len, KM_SLEEP|KM_NOFS); |
71e330b5 DC |
194 | ptr = lv->lv_buf; |
195 | ||
196 | for (index = 0; index < lv->lv_niovecs; index++) { | |
197 | struct xfs_log_iovec *vec = &lv->lv_iovecp[index]; | |
198 | ||
199 | memcpy(ptr, vec->i_addr, vec->i_len); | |
200 | vec->i_addr = ptr; | |
201 | ptr += vec->i_len; | |
202 | } | |
203 | ASSERT(ptr == lv->lv_buf + lv->lv_buf_len); | |
3b93c7aa DC |
204 | } |
205 | } | |
71e330b5 | 206 | |
d1583a38 DC |
207 | /* |
208 | * Prepare the log item for insertion into the CIL. Calculate the difference in | |
209 | * log space and vectors it will consume, and if it is a new item pin it as | |
210 | * well. | |
211 | */ | |
212 | STATIC void | |
213 | xfs_cil_prepare_item( | |
214 | struct log *log, | |
215 | struct xfs_log_vec *lv, | |
216 | int *len, | |
217 | int *diff_iovecs) | |
218 | { | |
219 | struct xfs_log_vec *old = lv->lv_item->li_lv; | |
220 | ||
221 | if (old) { | |
222 | /* existing lv on log item, space used is a delta */ | |
223 | ASSERT(!list_empty(&lv->lv_item->li_cil)); | |
224 | ASSERT(old->lv_buf && old->lv_buf_len && old->lv_niovecs); | |
225 | ||
226 | *len += lv->lv_buf_len - old->lv_buf_len; | |
227 | *diff_iovecs += lv->lv_niovecs - old->lv_niovecs; | |
228 | kmem_free(old->lv_buf); | |
229 | kmem_free(old); | |
230 | } else { | |
231 | /* new lv, must pin the log item */ | |
232 | ASSERT(!lv->lv_item->li_lv); | |
233 | ASSERT(list_empty(&lv->lv_item->li_cil)); | |
234 | ||
235 | *len += lv->lv_buf_len; | |
236 | *diff_iovecs += lv->lv_niovecs; | |
237 | IOP_PIN(lv->lv_item); | |
238 | ||
239 | } | |
240 | ||
241 | /* attach new log vector to log item */ | |
242 | lv->lv_item->li_lv = lv; | |
243 | ||
244 | /* | |
245 | * If this is the first time the item is being committed to the | |
246 | * CIL, store the sequence number on the log item so we can | |
247 | * tell in future commits whether this is the first checkpoint | |
248 | * the item is being committed into. | |
249 | */ | |
250 | if (!lv->lv_item->li_seq) | |
251 | lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Insert the log items into the CIL and calculate the difference in space | |
256 | * consumed by the item. Add the space to the checkpoint ticket and calculate | |
257 | * if the change requires additional log metadata. If it does, take that space | |
258 | * as well. Remove the amount of space we addded to the checkpoint ticket from | |
259 | * the current transaction ticket so that the accounting works out correctly. | |
260 | */ | |
3b93c7aa DC |
261 | static void |
262 | xlog_cil_insert_items( | |
263 | struct log *log, | |
264 | struct xfs_log_vec *log_vector, | |
d1583a38 | 265 | struct xlog_ticket *ticket) |
3b93c7aa | 266 | { |
d1583a38 DC |
267 | struct xfs_cil *cil = log->l_cilp; |
268 | struct xfs_cil_ctx *ctx = cil->xc_ctx; | |
269 | struct xfs_log_vec *lv; | |
270 | int len = 0; | |
271 | int diff_iovecs = 0; | |
272 | int iclog_space; | |
3b93c7aa DC |
273 | |
274 | ASSERT(log_vector); | |
d1583a38 DC |
275 | |
276 | /* | |
277 | * Do all the accounting aggregation and switching of log vectors | |
278 | * around in a separate loop to the insertion of items into the CIL. | |
279 | * Then we can do a separate loop to update the CIL within a single | |
280 | * lock/unlock pair. This reduces the number of round trips on the CIL | |
281 | * lock from O(nr_logvectors) to O(1) and greatly reduces the overall | |
282 | * hold time for the transaction commit. | |
283 | * | |
284 | * If this is the first time the item is being placed into the CIL in | |
285 | * this context, pin it so it can't be written to disk until the CIL is | |
286 | * flushed to the iclog and the iclog written to disk. | |
287 | * | |
288 | * We can do this safely because the context can't checkpoint until we | |
289 | * are done so it doesn't matter exactly how we update the CIL. | |
290 | */ | |
291 | for (lv = log_vector; lv; lv = lv->lv_next) | |
292 | xfs_cil_prepare_item(log, lv, &len, &diff_iovecs); | |
293 | ||
294 | /* account for space used by new iovec headers */ | |
295 | len += diff_iovecs * sizeof(xlog_op_header_t); | |
296 | ||
297 | spin_lock(&cil->xc_cil_lock); | |
298 | ||
299 | /* move the items to the tail of the CIL */ | |
3b93c7aa | 300 | for (lv = log_vector; lv; lv = lv->lv_next) |
d1583a38 DC |
301 | list_move_tail(&lv->lv_item->li_cil, &cil->xc_cil); |
302 | ||
303 | ctx->nvecs += diff_iovecs; | |
304 | ||
305 | /* | |
306 | * Now transfer enough transaction reservation to the context ticket | |
307 | * for the checkpoint. The context ticket is special - the unit | |
308 | * reservation has to grow as well as the current reservation as we | |
309 | * steal from tickets so we can correctly determine the space used | |
310 | * during the transaction commit. | |
311 | */ | |
312 | if (ctx->ticket->t_curr_res == 0) { | |
313 | /* first commit in checkpoint, steal the header reservation */ | |
314 | ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len); | |
315 | ctx->ticket->t_curr_res = ctx->ticket->t_unit_res; | |
316 | ticket->t_curr_res -= ctx->ticket->t_unit_res; | |
317 | } | |
318 | ||
319 | /* do we need space for more log record headers? */ | |
320 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
321 | if (len > 0 && (ctx->space_used / iclog_space != | |
322 | (ctx->space_used + len) / iclog_space)) { | |
323 | int hdrs; | |
324 | ||
325 | hdrs = (len + iclog_space - 1) / iclog_space; | |
326 | /* need to take into account split region headers, too */ | |
327 | hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header); | |
328 | ctx->ticket->t_unit_res += hdrs; | |
329 | ctx->ticket->t_curr_res += hdrs; | |
330 | ticket->t_curr_res -= hdrs; | |
331 | ASSERT(ticket->t_curr_res >= len); | |
332 | } | |
333 | ticket->t_curr_res -= len; | |
334 | ctx->space_used += len; | |
335 | ||
336 | spin_unlock(&cil->xc_cil_lock); | |
71e330b5 DC |
337 | } |
338 | ||
339 | static void | |
340 | xlog_cil_free_logvec( | |
341 | struct xfs_log_vec *log_vector) | |
342 | { | |
343 | struct xfs_log_vec *lv; | |
344 | ||
345 | for (lv = log_vector; lv; ) { | |
346 | struct xfs_log_vec *next = lv->lv_next; | |
347 | kmem_free(lv->lv_buf); | |
348 | kmem_free(lv); | |
349 | lv = next; | |
350 | } | |
351 | } | |
352 | ||
71e330b5 DC |
353 | /* |
354 | * Mark all items committed and clear busy extents. We free the log vector | |
355 | * chains in a separate pass so that we unpin the log items as quickly as | |
356 | * possible. | |
357 | */ | |
358 | static void | |
359 | xlog_cil_committed( | |
360 | void *args, | |
361 | int abort) | |
362 | { | |
363 | struct xfs_cil_ctx *ctx = args; | |
71e330b5 DC |
364 | struct xfs_busy_extent *busyp, *n; |
365 | ||
0e57f6a3 DC |
366 | xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain, |
367 | ctx->start_lsn, abort); | |
71e330b5 DC |
368 | |
369 | list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list) | |
370 | xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp); | |
371 | ||
372 | spin_lock(&ctx->cil->xc_cil_lock); | |
373 | list_del(&ctx->committing); | |
374 | spin_unlock(&ctx->cil->xc_cil_lock); | |
375 | ||
376 | xlog_cil_free_logvec(ctx->lv_chain); | |
377 | kmem_free(ctx); | |
378 | } | |
379 | ||
380 | /* | |
a44f13ed DC |
381 | * Push the Committed Item List to the log. If @push_seq flag is zero, then it |
382 | * is a background flush and so we can chose to ignore it. Otherwise, if the | |
383 | * current sequence is the same as @push_seq we need to do a flush. If | |
384 | * @push_seq is less than the current sequence, then it has already been | |
385 | * flushed and we don't need to do anything - the caller will wait for it to | |
386 | * complete if necessary. | |
387 | * | |
388 | * @push_seq is a value rather than a flag because that allows us to do an | |
389 | * unlocked check of the sequence number for a match. Hence we can allows log | |
390 | * forces to run racily and not issue pushes for the same sequence twice. If we | |
391 | * get a race between multiple pushes for the same sequence they will block on | |
392 | * the first one and then abort, hence avoiding needless pushes. | |
71e330b5 | 393 | */ |
a44f13ed | 394 | STATIC int |
71e330b5 DC |
395 | xlog_cil_push( |
396 | struct log *log, | |
a44f13ed | 397 | xfs_lsn_t push_seq) |
71e330b5 DC |
398 | { |
399 | struct xfs_cil *cil = log->l_cilp; | |
400 | struct xfs_log_vec *lv; | |
401 | struct xfs_cil_ctx *ctx; | |
402 | struct xfs_cil_ctx *new_ctx; | |
403 | struct xlog_in_core *commit_iclog; | |
404 | struct xlog_ticket *tic; | |
405 | int num_lv; | |
406 | int num_iovecs; | |
407 | int len; | |
408 | int error = 0; | |
409 | struct xfs_trans_header thdr; | |
410 | struct xfs_log_iovec lhdr; | |
411 | struct xfs_log_vec lvhdr = { NULL }; | |
412 | xfs_lsn_t commit_lsn; | |
413 | ||
414 | if (!cil) | |
415 | return 0; | |
416 | ||
a44f13ed DC |
417 | ASSERT(!push_seq || push_seq <= cil->xc_ctx->sequence); |
418 | ||
71e330b5 DC |
419 | new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS); |
420 | new_ctx->ticket = xlog_cil_ticket_alloc(log); | |
421 | ||
80168676 DC |
422 | /* |
423 | * Lock out transaction commit, but don't block for background pushes | |
424 | * unless we are well over the CIL space limit. See the definition of | |
425 | * XLOG_CIL_HARD_SPACE_LIMIT() for the full explanation of the logic | |
426 | * used here. | |
427 | */ | |
df806158 | 428 | if (!down_write_trylock(&cil->xc_ctx_lock)) { |
80168676 DC |
429 | if (!push_seq && |
430 | cil->xc_ctx->space_used < XLOG_CIL_HARD_SPACE_LIMIT(log)) | |
df806158 DC |
431 | goto out_free_ticket; |
432 | down_write(&cil->xc_ctx_lock); | |
433 | } | |
71e330b5 DC |
434 | ctx = cil->xc_ctx; |
435 | ||
436 | /* check if we've anything to push */ | |
437 | if (list_empty(&cil->xc_cil)) | |
438 | goto out_skip; | |
439 | ||
df806158 | 440 | /* check for spurious background flush */ |
a44f13ed DC |
441 | if (!push_seq && cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) |
442 | goto out_skip; | |
443 | ||
444 | /* check for a previously pushed seqeunce */ | |
80168676 | 445 | if (push_seq && push_seq < cil->xc_ctx->sequence) |
df806158 DC |
446 | goto out_skip; |
447 | ||
71e330b5 DC |
448 | /* |
449 | * pull all the log vectors off the items in the CIL, and | |
450 | * remove the items from the CIL. We don't need the CIL lock | |
451 | * here because it's only needed on the transaction commit | |
452 | * side which is currently locked out by the flush lock. | |
453 | */ | |
454 | lv = NULL; | |
455 | num_lv = 0; | |
456 | num_iovecs = 0; | |
457 | len = 0; | |
458 | while (!list_empty(&cil->xc_cil)) { | |
459 | struct xfs_log_item *item; | |
460 | int i; | |
461 | ||
462 | item = list_first_entry(&cil->xc_cil, | |
463 | struct xfs_log_item, li_cil); | |
464 | list_del_init(&item->li_cil); | |
465 | if (!ctx->lv_chain) | |
466 | ctx->lv_chain = item->li_lv; | |
467 | else | |
468 | lv->lv_next = item->li_lv; | |
469 | lv = item->li_lv; | |
470 | item->li_lv = NULL; | |
471 | ||
472 | num_lv++; | |
473 | num_iovecs += lv->lv_niovecs; | |
474 | for (i = 0; i < lv->lv_niovecs; i++) | |
475 | len += lv->lv_iovecp[i].i_len; | |
476 | } | |
477 | ||
478 | /* | |
479 | * initialise the new context and attach it to the CIL. Then attach | |
480 | * the current context to the CIL committing lsit so it can be found | |
481 | * during log forces to extract the commit lsn of the sequence that | |
482 | * needs to be forced. | |
483 | */ | |
484 | INIT_LIST_HEAD(&new_ctx->committing); | |
485 | INIT_LIST_HEAD(&new_ctx->busy_extents); | |
486 | new_ctx->sequence = ctx->sequence + 1; | |
487 | new_ctx->cil = cil; | |
488 | cil->xc_ctx = new_ctx; | |
489 | ||
a44f13ed DC |
490 | /* |
491 | * mirror the new sequence into the cil structure so that we can do | |
492 | * unlocked checks against the current sequence in log forces without | |
493 | * risking deferencing a freed context pointer. | |
494 | */ | |
495 | cil->xc_current_sequence = new_ctx->sequence; | |
496 | ||
71e330b5 DC |
497 | /* |
498 | * The switch is now done, so we can drop the context lock and move out | |
499 | * of a shared context. We can't just go straight to the commit record, | |
500 | * though - we need to synchronise with previous and future commits so | |
501 | * that the commit records are correctly ordered in the log to ensure | |
502 | * that we process items during log IO completion in the correct order. | |
503 | * | |
504 | * For example, if we get an EFI in one checkpoint and the EFD in the | |
505 | * next (e.g. due to log forces), we do not want the checkpoint with | |
506 | * the EFD to be committed before the checkpoint with the EFI. Hence | |
507 | * we must strictly order the commit records of the checkpoints so | |
508 | * that: a) the checkpoint callbacks are attached to the iclogs in the | |
509 | * correct order; and b) the checkpoints are replayed in correct order | |
510 | * in log recovery. | |
511 | * | |
512 | * Hence we need to add this context to the committing context list so | |
513 | * that higher sequences will wait for us to write out a commit record | |
514 | * before they do. | |
515 | */ | |
516 | spin_lock(&cil->xc_cil_lock); | |
517 | list_add(&ctx->committing, &cil->xc_committing); | |
518 | spin_unlock(&cil->xc_cil_lock); | |
519 | up_write(&cil->xc_ctx_lock); | |
520 | ||
521 | /* | |
522 | * Build a checkpoint transaction header and write it to the log to | |
523 | * begin the transaction. We need to account for the space used by the | |
524 | * transaction header here as it is not accounted for in xlog_write(). | |
525 | * | |
526 | * The LSN we need to pass to the log items on transaction commit is | |
527 | * the LSN reported by the first log vector write. If we use the commit | |
528 | * record lsn then we can move the tail beyond the grant write head. | |
529 | */ | |
530 | tic = ctx->ticket; | |
531 | thdr.th_magic = XFS_TRANS_HEADER_MAGIC; | |
532 | thdr.th_type = XFS_TRANS_CHECKPOINT; | |
533 | thdr.th_tid = tic->t_tid; | |
534 | thdr.th_num_items = num_iovecs; | |
4e0d5f92 | 535 | lhdr.i_addr = &thdr; |
71e330b5 DC |
536 | lhdr.i_len = sizeof(xfs_trans_header_t); |
537 | lhdr.i_type = XLOG_REG_TYPE_TRANSHDR; | |
538 | tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t); | |
539 | ||
540 | lvhdr.lv_niovecs = 1; | |
541 | lvhdr.lv_iovecp = &lhdr; | |
542 | lvhdr.lv_next = ctx->lv_chain; | |
543 | ||
544 | error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); | |
545 | if (error) | |
546 | goto out_abort; | |
547 | ||
548 | /* | |
549 | * now that we've written the checkpoint into the log, strictly | |
550 | * order the commit records so replay will get them in the right order. | |
551 | */ | |
552 | restart: | |
553 | spin_lock(&cil->xc_cil_lock); | |
554 | list_for_each_entry(new_ctx, &cil->xc_committing, committing) { | |
555 | /* | |
556 | * Higher sequences will wait for this one so skip them. | |
557 | * Don't wait for own own sequence, either. | |
558 | */ | |
559 | if (new_ctx->sequence >= ctx->sequence) | |
560 | continue; | |
561 | if (!new_ctx->commit_lsn) { | |
562 | /* | |
563 | * It is still being pushed! Wait for the push to | |
564 | * complete, then start again from the beginning. | |
565 | */ | |
eb40a875 | 566 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
567 | goto restart; |
568 | } | |
569 | } | |
570 | spin_unlock(&cil->xc_cil_lock); | |
571 | ||
572 | commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); | |
573 | if (error || commit_lsn == -1) | |
574 | goto out_abort; | |
575 | ||
576 | /* attach all the transactions w/ busy extents to iclog */ | |
577 | ctx->log_cb.cb_func = xlog_cil_committed; | |
578 | ctx->log_cb.cb_arg = ctx; | |
579 | error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb); | |
580 | if (error) | |
581 | goto out_abort; | |
582 | ||
583 | /* | |
584 | * now the checkpoint commit is complete and we've attached the | |
585 | * callbacks to the iclog we can assign the commit LSN to the context | |
586 | * and wake up anyone who is waiting for the commit to complete. | |
587 | */ | |
588 | spin_lock(&cil->xc_cil_lock); | |
589 | ctx->commit_lsn = commit_lsn; | |
eb40a875 | 590 | wake_up_all(&cil->xc_commit_wait); |
71e330b5 DC |
591 | spin_unlock(&cil->xc_cil_lock); |
592 | ||
593 | /* release the hounds! */ | |
594 | return xfs_log_release_iclog(log->l_mp, commit_iclog); | |
595 | ||
596 | out_skip: | |
597 | up_write(&cil->xc_ctx_lock); | |
df806158 | 598 | out_free_ticket: |
71e330b5 DC |
599 | xfs_log_ticket_put(new_ctx->ticket); |
600 | kmem_free(new_ctx); | |
601 | return 0; | |
602 | ||
603 | out_abort: | |
604 | xlog_cil_committed(ctx, XFS_LI_ABORTED); | |
605 | return XFS_ERROR(EIO); | |
606 | } | |
607 | ||
a44f13ed DC |
608 | /* |
609 | * Commit a transaction with the given vector to the Committed Item List. | |
610 | * | |
611 | * To do this, we need to format the item, pin it in memory if required and | |
612 | * account for the space used by the transaction. Once we have done that we | |
613 | * need to release the unused reservation for the transaction, attach the | |
614 | * transaction to the checkpoint context so we carry the busy extents through | |
615 | * to checkpoint completion, and then unlock all the items in the transaction. | |
616 | * | |
617 | * For more specific information about the order of operations in | |
618 | * xfs_log_commit_cil() please refer to the comments in | |
619 | * xfs_trans_commit_iclog(). | |
620 | * | |
621 | * Called with the context lock already held in read mode to lock out | |
622 | * background commit, returns without it held once background commits are | |
623 | * allowed again. | |
624 | */ | |
625 | int | |
626 | xfs_log_commit_cil( | |
627 | struct xfs_mount *mp, | |
628 | struct xfs_trans *tp, | |
629 | struct xfs_log_vec *log_vector, | |
630 | xfs_lsn_t *commit_lsn, | |
631 | int flags) | |
632 | { | |
633 | struct log *log = mp->m_log; | |
634 | int log_flags = 0; | |
635 | int push = 0; | |
636 | ||
637 | if (flags & XFS_TRANS_RELEASE_LOG_RES) | |
638 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
639 | ||
640 | if (XLOG_FORCED_SHUTDOWN(log)) { | |
641 | xlog_cil_free_logvec(log_vector); | |
642 | return XFS_ERROR(EIO); | |
643 | } | |
644 | ||
3b93c7aa DC |
645 | /* |
646 | * do all the hard work of formatting items (including memory | |
647 | * allocation) outside the CIL context lock. This prevents stalling CIL | |
648 | * pushes when we are low on memory and a transaction commit spends a | |
649 | * lot of time in memory reclaim. | |
650 | */ | |
651 | xlog_cil_format_items(log, log_vector); | |
652 | ||
a44f13ed DC |
653 | /* lock out background commit */ |
654 | down_read(&log->l_cilp->xc_ctx_lock); | |
d1583a38 DC |
655 | if (commit_lsn) |
656 | *commit_lsn = log->l_cilp->xc_ctx->sequence; | |
657 | ||
658 | xlog_cil_insert_items(log, log_vector, tp->t_ticket); | |
a44f13ed DC |
659 | |
660 | /* check we didn't blow the reservation */ | |
661 | if (tp->t_ticket->t_curr_res < 0) | |
662 | xlog_print_tic_res(log->l_mp, tp->t_ticket); | |
663 | ||
664 | /* attach the transaction to the CIL if it has any busy extents */ | |
665 | if (!list_empty(&tp->t_busy)) { | |
666 | spin_lock(&log->l_cilp->xc_cil_lock); | |
667 | list_splice_init(&tp->t_busy, | |
668 | &log->l_cilp->xc_ctx->busy_extents); | |
669 | spin_unlock(&log->l_cilp->xc_cil_lock); | |
670 | } | |
671 | ||
672 | tp->t_commit_lsn = *commit_lsn; | |
673 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
674 | xfs_trans_unreserve_and_mod_sb(tp); | |
675 | ||
676 | /* | |
677 | * Once all the items of the transaction have been copied to the CIL, | |
678 | * the items can be unlocked and freed. | |
679 | * | |
680 | * This needs to be done before we drop the CIL context lock because we | |
681 | * have to update state in the log items and unlock them before they go | |
682 | * to disk. If we don't, then the CIL checkpoint can race with us and | |
683 | * we can run checkpoint completion before we've updated and unlocked | |
684 | * the log items. This affects (at least) processing of stale buffers, | |
685 | * inodes and EFIs. | |
686 | */ | |
687 | xfs_trans_free_items(tp, *commit_lsn, 0); | |
688 | ||
689 | /* check for background commit before unlock */ | |
690 | if (log->l_cilp->xc_ctx->space_used > XLOG_CIL_SPACE_LIMIT(log)) | |
691 | push = 1; | |
692 | ||
693 | up_read(&log->l_cilp->xc_ctx_lock); | |
694 | ||
695 | /* | |
696 | * We need to push CIL every so often so we don't cache more than we | |
697 | * can fit in the log. The limit really is that a checkpoint can't be | |
698 | * more than half the log (the current checkpoint is not allowed to | |
699 | * overwrite the previous checkpoint), but commit latency and memory | |
700 | * usage limit this to a smaller size in most cases. | |
701 | */ | |
702 | if (push) | |
703 | xlog_cil_push(log, 0); | |
704 | return 0; | |
705 | } | |
706 | ||
71e330b5 DC |
707 | /* |
708 | * Conditionally push the CIL based on the sequence passed in. | |
709 | * | |
710 | * We only need to push if we haven't already pushed the sequence | |
711 | * number given. Hence the only time we will trigger a push here is | |
712 | * if the push sequence is the same as the current context. | |
713 | * | |
714 | * We return the current commit lsn to allow the callers to determine if a | |
715 | * iclog flush is necessary following this call. | |
716 | * | |
717 | * XXX: Initially, just push the CIL unconditionally and return whatever | |
718 | * commit lsn is there. It'll be empty, so this is broken for now. | |
719 | */ | |
720 | xfs_lsn_t | |
a44f13ed | 721 | xlog_cil_force_lsn( |
71e330b5 | 722 | struct log *log, |
a44f13ed | 723 | xfs_lsn_t sequence) |
71e330b5 DC |
724 | { |
725 | struct xfs_cil *cil = log->l_cilp; | |
726 | struct xfs_cil_ctx *ctx; | |
727 | xfs_lsn_t commit_lsn = NULLCOMMITLSN; | |
728 | ||
a44f13ed DC |
729 | ASSERT(sequence <= cil->xc_current_sequence); |
730 | ||
731 | /* | |
732 | * check to see if we need to force out the current context. | |
733 | * xlog_cil_push() handles racing pushes for the same sequence, | |
734 | * so no need to deal with it here. | |
735 | */ | |
736 | if (sequence == cil->xc_current_sequence) | |
737 | xlog_cil_push(log, sequence); | |
71e330b5 DC |
738 | |
739 | /* | |
740 | * See if we can find a previous sequence still committing. | |
71e330b5 DC |
741 | * We need to wait for all previous sequence commits to complete |
742 | * before allowing the force of push_seq to go ahead. Hence block | |
743 | * on commits for those as well. | |
744 | */ | |
a44f13ed | 745 | restart: |
71e330b5 | 746 | spin_lock(&cil->xc_cil_lock); |
71e330b5 | 747 | list_for_each_entry(ctx, &cil->xc_committing, committing) { |
a44f13ed | 748 | if (ctx->sequence > sequence) |
71e330b5 DC |
749 | continue; |
750 | if (!ctx->commit_lsn) { | |
751 | /* | |
752 | * It is still being pushed! Wait for the push to | |
753 | * complete, then start again from the beginning. | |
754 | */ | |
eb40a875 | 755 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
756 | goto restart; |
757 | } | |
a44f13ed | 758 | if (ctx->sequence != sequence) |
71e330b5 DC |
759 | continue; |
760 | /* found it! */ | |
761 | commit_lsn = ctx->commit_lsn; | |
762 | } | |
763 | spin_unlock(&cil->xc_cil_lock); | |
764 | return commit_lsn; | |
765 | } | |
ccf7c23f DC |
766 | |
767 | /* | |
768 | * Check if the current log item was first committed in this sequence. | |
769 | * We can't rely on just the log item being in the CIL, we have to check | |
770 | * the recorded commit sequence number. | |
771 | * | |
772 | * Note: for this to be used in a non-racy manner, it has to be called with | |
773 | * CIL flushing locked out. As a result, it should only be used during the | |
774 | * transaction commit process when deciding what to format into the item. | |
775 | */ | |
776 | bool | |
777 | xfs_log_item_in_current_chkpt( | |
778 | struct xfs_log_item *lip) | |
779 | { | |
780 | struct xfs_cil_ctx *ctx; | |
781 | ||
782 | if (!(lip->li_mountp->m_flags & XFS_MOUNT_DELAYLOG)) | |
783 | return false; | |
784 | if (list_empty(&lip->li_cil)) | |
785 | return false; | |
786 | ||
787 | ctx = lip->li_mountp->m_log->l_cilp->xc_ctx; | |
788 | ||
789 | /* | |
790 | * li_seq is written on the first commit of a log item to record the | |
791 | * first checkpoint it is written to. Hence if it is different to the | |
792 | * current sequence, we're in a new checkpoint. | |
793 | */ | |
794 | if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0) | |
795 | return false; | |
796 | return true; | |
797 | } |