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