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" | |
71e330b5 DC |
23 | #include "xfs_trans.h" |
24 | #include "xfs_trans_priv.h" | |
25 | #include "xfs_log_priv.h" | |
26 | #include "xfs_sb.h" | |
27 | #include "xfs_ag.h" | |
71e330b5 DC |
28 | #include "xfs_mount.h" |
29 | #include "xfs_error.h" | |
30 | #include "xfs_alloc.h" | |
efc27b52 | 31 | #include "xfs_extent_busy.h" |
e84661aa | 32 | #include "xfs_discard.h" |
71e330b5 | 33 | |
71e330b5 DC |
34 | /* |
35 | * Allocate a new ticket. Failing to get a new ticket makes it really hard to | |
36 | * recover, so we don't allow failure here. Also, we allocate in a context that | |
37 | * we don't want to be issuing transactions from, so we need to tell the | |
38 | * allocation code this as well. | |
39 | * | |
40 | * We don't reserve any space for the ticket - we are going to steal whatever | |
41 | * space we require from transactions as they commit. To ensure we reserve all | |
42 | * the space required, we need to set the current reservation of the ticket to | |
43 | * zero so that we know to steal the initial transaction overhead from the | |
44 | * first transaction commit. | |
45 | */ | |
46 | static struct xlog_ticket * | |
47 | xlog_cil_ticket_alloc( | |
48 | struct log *log) | |
49 | { | |
50 | struct xlog_ticket *tic; | |
51 | ||
52 | tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0, | |
53 | KM_SLEEP|KM_NOFS); | |
54 | tic->t_trans_type = XFS_TRANS_CHECKPOINT; | |
55 | ||
56 | /* | |
57 | * set the current reservation to zero so we know to steal the basic | |
58 | * transaction overhead reservation from the first transaction commit. | |
59 | */ | |
60 | tic->t_curr_res = 0; | |
61 | return tic; | |
62 | } | |
63 | ||
64 | /* | |
65 | * After the first stage of log recovery is done, we know where the head and | |
66 | * tail of the log are. We need this log initialisation done before we can | |
67 | * initialise the first CIL checkpoint context. | |
68 | * | |
69 | * Here we allocate a log ticket to track space usage during a CIL push. This | |
70 | * ticket is passed to xlog_write() directly so that we don't slowly leak log | |
71 | * space by failing to account for space used by log headers and additional | |
72 | * region headers for split regions. | |
73 | */ | |
74 | void | |
75 | xlog_cil_init_post_recovery( | |
76 | struct log *log) | |
77 | { | |
71e330b5 DC |
78 | log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log); |
79 | log->l_cilp->xc_ctx->sequence = 1; | |
80 | log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle, | |
81 | log->l_curr_block); | |
82 | } | |
83 | ||
71e330b5 DC |
84 | /* |
85 | * Format log item into a flat buffers | |
86 | * | |
87 | * For delayed logging, we need to hold a formatted buffer containing all the | |
88 | * changes on the log item. This enables us to relog the item in memory and | |
89 | * write it out asynchronously without needing to relock the object that was | |
90 | * modified at the time it gets written into the iclog. | |
91 | * | |
92 | * This function builds a vector for the changes in each log item in the | |
93 | * transaction. It then works out the length of the buffer needed for each log | |
94 | * item, allocates them and formats the vector for the item into the buffer. | |
95 | * The buffer is then attached to the log item are then inserted into the | |
96 | * Committed Item List for tracking until the next checkpoint is written out. | |
97 | * | |
98 | * We don't set up region headers during this process; we simply copy the | |
99 | * regions into the flat buffer. We can do this because we still have to do a | |
100 | * formatting step to write the regions into the iclog buffer. Writing the | |
101 | * ophdrs during the iclog write means that we can support splitting large | |
102 | * regions across iclog boundares without needing a change in the format of the | |
103 | * item/region encapsulation. | |
104 | * | |
105 | * Hence what we need to do now is change the rewrite the vector array to point | |
106 | * to the copied region inside the buffer we just allocated. This allows us to | |
107 | * format the regions into the iclog as though they are being formatted | |
108 | * directly out of the objects themselves. | |
109 | */ | |
0244b960 CH |
110 | static struct xfs_log_vec * |
111 | xlog_cil_prepare_log_vecs( | |
112 | struct xfs_trans *tp) | |
71e330b5 | 113 | { |
0244b960 CH |
114 | struct xfs_log_item_desc *lidp; |
115 | struct xfs_log_vec *lv = NULL; | |
116 | struct xfs_log_vec *ret_lv = NULL; | |
71e330b5 | 117 | |
0244b960 CH |
118 | |
119 | /* Bail out if we didn't find a log item. */ | |
120 | if (list_empty(&tp->t_items)) { | |
121 | ASSERT(0); | |
122 | return NULL; | |
123 | } | |
124 | ||
125 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { | |
126 | struct xfs_log_vec *new_lv; | |
71e330b5 DC |
127 | void *ptr; |
128 | int index; | |
129 | int len = 0; | |
b3934213 | 130 | uint niovecs; |
71e330b5 | 131 | |
0244b960 CH |
132 | /* Skip items which aren't dirty in this transaction. */ |
133 | if (!(lidp->lid_flags & XFS_LID_DIRTY)) | |
134 | continue; | |
135 | ||
136 | /* Skip items that do not have any vectors for writing */ | |
b3934213 CH |
137 | niovecs = IOP_SIZE(lidp->lid_item); |
138 | if (!niovecs) | |
0244b960 CH |
139 | continue; |
140 | ||
141 | new_lv = kmem_zalloc(sizeof(*new_lv) + | |
b3934213 | 142 | niovecs * sizeof(struct xfs_log_iovec), |
0244b960 CH |
143 | KM_SLEEP); |
144 | ||
145 | /* The allocated iovec region lies beyond the log vector. */ | |
146 | new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1]; | |
b3934213 | 147 | new_lv->lv_niovecs = niovecs; |
0244b960 CH |
148 | new_lv->lv_item = lidp->lid_item; |
149 | ||
71e330b5 | 150 | /* build the vector array and calculate it's length */ |
0244b960 CH |
151 | IOP_FORMAT(new_lv->lv_item, new_lv->lv_iovecp); |
152 | for (index = 0; index < new_lv->lv_niovecs; index++) | |
153 | len += new_lv->lv_iovecp[index].i_len; | |
71e330b5 | 154 | |
0244b960 CH |
155 | new_lv->lv_buf_len = len; |
156 | new_lv->lv_buf = kmem_alloc(new_lv->lv_buf_len, | |
157 | KM_SLEEP|KM_NOFS); | |
158 | ptr = new_lv->lv_buf; | |
71e330b5 | 159 | |
0244b960 CH |
160 | for (index = 0; index < new_lv->lv_niovecs; index++) { |
161 | struct xfs_log_iovec *vec = &new_lv->lv_iovecp[index]; | |
71e330b5 DC |
162 | |
163 | memcpy(ptr, vec->i_addr, vec->i_len); | |
164 | vec->i_addr = ptr; | |
165 | ptr += vec->i_len; | |
166 | } | |
0244b960 CH |
167 | ASSERT(ptr == new_lv->lv_buf + new_lv->lv_buf_len); |
168 | ||
169 | if (!ret_lv) | |
170 | ret_lv = new_lv; | |
171 | else | |
172 | lv->lv_next = new_lv; | |
173 | lv = new_lv; | |
3b93c7aa | 174 | } |
0244b960 CH |
175 | |
176 | return ret_lv; | |
3b93c7aa | 177 | } |
71e330b5 | 178 | |
d1583a38 DC |
179 | /* |
180 | * Prepare the log item for insertion into the CIL. Calculate the difference in | |
181 | * log space and vectors it will consume, and if it is a new item pin it as | |
182 | * well. | |
183 | */ | |
184 | STATIC void | |
185 | xfs_cil_prepare_item( | |
186 | struct log *log, | |
187 | struct xfs_log_vec *lv, | |
188 | int *len, | |
189 | int *diff_iovecs) | |
190 | { | |
191 | struct xfs_log_vec *old = lv->lv_item->li_lv; | |
192 | ||
193 | if (old) { | |
194 | /* existing lv on log item, space used is a delta */ | |
195 | ASSERT(!list_empty(&lv->lv_item->li_cil)); | |
196 | ASSERT(old->lv_buf && old->lv_buf_len && old->lv_niovecs); | |
197 | ||
198 | *len += lv->lv_buf_len - old->lv_buf_len; | |
199 | *diff_iovecs += lv->lv_niovecs - old->lv_niovecs; | |
200 | kmem_free(old->lv_buf); | |
201 | kmem_free(old); | |
202 | } else { | |
203 | /* new lv, must pin the log item */ | |
204 | ASSERT(!lv->lv_item->li_lv); | |
205 | ASSERT(list_empty(&lv->lv_item->li_cil)); | |
206 | ||
207 | *len += lv->lv_buf_len; | |
208 | *diff_iovecs += lv->lv_niovecs; | |
209 | IOP_PIN(lv->lv_item); | |
210 | ||
211 | } | |
212 | ||
213 | /* attach new log vector to log item */ | |
214 | lv->lv_item->li_lv = lv; | |
215 | ||
216 | /* | |
217 | * If this is the first time the item is being committed to the | |
218 | * CIL, store the sequence number on the log item so we can | |
219 | * tell in future commits whether this is the first checkpoint | |
220 | * the item is being committed into. | |
221 | */ | |
222 | if (!lv->lv_item->li_seq) | |
223 | lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Insert the log items into the CIL and calculate the difference in space | |
228 | * consumed by the item. Add the space to the checkpoint ticket and calculate | |
229 | * if the change requires additional log metadata. If it does, take that space | |
42b2aa86 | 230 | * as well. Remove the amount of space we added to the checkpoint ticket from |
d1583a38 DC |
231 | * the current transaction ticket so that the accounting works out correctly. |
232 | */ | |
3b93c7aa DC |
233 | static void |
234 | xlog_cil_insert_items( | |
235 | struct log *log, | |
236 | struct xfs_log_vec *log_vector, | |
d1583a38 | 237 | struct xlog_ticket *ticket) |
3b93c7aa | 238 | { |
d1583a38 DC |
239 | struct xfs_cil *cil = log->l_cilp; |
240 | struct xfs_cil_ctx *ctx = cil->xc_ctx; | |
241 | struct xfs_log_vec *lv; | |
242 | int len = 0; | |
243 | int diff_iovecs = 0; | |
244 | int iclog_space; | |
3b93c7aa DC |
245 | |
246 | ASSERT(log_vector); | |
d1583a38 DC |
247 | |
248 | /* | |
249 | * Do all the accounting aggregation and switching of log vectors | |
250 | * around in a separate loop to the insertion of items into the CIL. | |
251 | * Then we can do a separate loop to update the CIL within a single | |
252 | * lock/unlock pair. This reduces the number of round trips on the CIL | |
253 | * lock from O(nr_logvectors) to O(1) and greatly reduces the overall | |
254 | * hold time for the transaction commit. | |
255 | * | |
256 | * If this is the first time the item is being placed into the CIL in | |
257 | * this context, pin it so it can't be written to disk until the CIL is | |
258 | * flushed to the iclog and the iclog written to disk. | |
259 | * | |
260 | * We can do this safely because the context can't checkpoint until we | |
261 | * are done so it doesn't matter exactly how we update the CIL. | |
262 | */ | |
263 | for (lv = log_vector; lv; lv = lv->lv_next) | |
264 | xfs_cil_prepare_item(log, lv, &len, &diff_iovecs); | |
265 | ||
266 | /* account for space used by new iovec headers */ | |
267 | len += diff_iovecs * sizeof(xlog_op_header_t); | |
268 | ||
269 | spin_lock(&cil->xc_cil_lock); | |
270 | ||
271 | /* move the items to the tail of the CIL */ | |
3b93c7aa | 272 | for (lv = log_vector; lv; lv = lv->lv_next) |
d1583a38 DC |
273 | list_move_tail(&lv->lv_item->li_cil, &cil->xc_cil); |
274 | ||
275 | ctx->nvecs += diff_iovecs; | |
276 | ||
277 | /* | |
278 | * Now transfer enough transaction reservation to the context ticket | |
279 | * for the checkpoint. The context ticket is special - the unit | |
280 | * reservation has to grow as well as the current reservation as we | |
281 | * steal from tickets so we can correctly determine the space used | |
282 | * during the transaction commit. | |
283 | */ | |
284 | if (ctx->ticket->t_curr_res == 0) { | |
285 | /* first commit in checkpoint, steal the header reservation */ | |
286 | ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len); | |
287 | ctx->ticket->t_curr_res = ctx->ticket->t_unit_res; | |
288 | ticket->t_curr_res -= ctx->ticket->t_unit_res; | |
289 | } | |
290 | ||
291 | /* do we need space for more log record headers? */ | |
292 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
293 | if (len > 0 && (ctx->space_used / iclog_space != | |
294 | (ctx->space_used + len) / iclog_space)) { | |
295 | int hdrs; | |
296 | ||
297 | hdrs = (len + iclog_space - 1) / iclog_space; | |
298 | /* need to take into account split region headers, too */ | |
299 | hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header); | |
300 | ctx->ticket->t_unit_res += hdrs; | |
301 | ctx->ticket->t_curr_res += hdrs; | |
302 | ticket->t_curr_res -= hdrs; | |
303 | ASSERT(ticket->t_curr_res >= len); | |
304 | } | |
305 | ticket->t_curr_res -= len; | |
306 | ctx->space_used += len; | |
307 | ||
308 | spin_unlock(&cil->xc_cil_lock); | |
71e330b5 DC |
309 | } |
310 | ||
311 | static void | |
312 | xlog_cil_free_logvec( | |
313 | struct xfs_log_vec *log_vector) | |
314 | { | |
315 | struct xfs_log_vec *lv; | |
316 | ||
317 | for (lv = log_vector; lv; ) { | |
318 | struct xfs_log_vec *next = lv->lv_next; | |
319 | kmem_free(lv->lv_buf); | |
320 | kmem_free(lv); | |
321 | lv = next; | |
322 | } | |
323 | } | |
324 | ||
71e330b5 DC |
325 | /* |
326 | * Mark all items committed and clear busy extents. We free the log vector | |
327 | * chains in a separate pass so that we unpin the log items as quickly as | |
328 | * possible. | |
329 | */ | |
330 | static void | |
331 | xlog_cil_committed( | |
332 | void *args, | |
333 | int abort) | |
334 | { | |
335 | struct xfs_cil_ctx *ctx = args; | |
e84661aa | 336 | struct xfs_mount *mp = ctx->cil->xc_log->l_mp; |
71e330b5 | 337 | |
0e57f6a3 DC |
338 | xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain, |
339 | ctx->start_lsn, abort); | |
71e330b5 | 340 | |
8a072a4d | 341 | xfs_alloc_busy_sort(&ctx->busy_extents); |
e84661aa CH |
342 | xfs_alloc_busy_clear(mp, &ctx->busy_extents, |
343 | (mp->m_flags & XFS_MOUNT_DISCARD) && !abort); | |
71e330b5 DC |
344 | |
345 | spin_lock(&ctx->cil->xc_cil_lock); | |
346 | list_del(&ctx->committing); | |
347 | spin_unlock(&ctx->cil->xc_cil_lock); | |
348 | ||
349 | xlog_cil_free_logvec(ctx->lv_chain); | |
e84661aa CH |
350 | |
351 | if (!list_empty(&ctx->busy_extents)) { | |
352 | ASSERT(mp->m_flags & XFS_MOUNT_DISCARD); | |
353 | ||
354 | xfs_discard_extents(mp, &ctx->busy_extents); | |
355 | xfs_alloc_busy_clear(mp, &ctx->busy_extents, false); | |
356 | } | |
357 | ||
71e330b5 DC |
358 | kmem_free(ctx); |
359 | } | |
360 | ||
361 | /* | |
a44f13ed DC |
362 | * Push the Committed Item List to the log. If @push_seq flag is zero, then it |
363 | * is a background flush and so we can chose to ignore it. Otherwise, if the | |
364 | * current sequence is the same as @push_seq we need to do a flush. If | |
365 | * @push_seq is less than the current sequence, then it has already been | |
366 | * flushed and we don't need to do anything - the caller will wait for it to | |
367 | * complete if necessary. | |
368 | * | |
369 | * @push_seq is a value rather than a flag because that allows us to do an | |
370 | * unlocked check of the sequence number for a match. Hence we can allows log | |
371 | * forces to run racily and not issue pushes for the same sequence twice. If we | |
372 | * get a race between multiple pushes for the same sequence they will block on | |
373 | * the first one and then abort, hence avoiding needless pushes. | |
71e330b5 | 374 | */ |
a44f13ed | 375 | STATIC int |
71e330b5 | 376 | xlog_cil_push( |
4c2d542f | 377 | struct log *log) |
71e330b5 DC |
378 | { |
379 | struct xfs_cil *cil = log->l_cilp; | |
380 | struct xfs_log_vec *lv; | |
381 | struct xfs_cil_ctx *ctx; | |
382 | struct xfs_cil_ctx *new_ctx; | |
383 | struct xlog_in_core *commit_iclog; | |
384 | struct xlog_ticket *tic; | |
385 | int num_lv; | |
386 | int num_iovecs; | |
387 | int len; | |
388 | int error = 0; | |
389 | struct xfs_trans_header thdr; | |
390 | struct xfs_log_iovec lhdr; | |
391 | struct xfs_log_vec lvhdr = { NULL }; | |
392 | xfs_lsn_t commit_lsn; | |
4c2d542f | 393 | xfs_lsn_t push_seq; |
71e330b5 DC |
394 | |
395 | if (!cil) | |
396 | return 0; | |
397 | ||
71e330b5 DC |
398 | new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS); |
399 | new_ctx->ticket = xlog_cil_ticket_alloc(log); | |
400 | ||
4c2d542f | 401 | down_write(&cil->xc_ctx_lock); |
71e330b5 DC |
402 | ctx = cil->xc_ctx; |
403 | ||
4c2d542f DC |
404 | spin_lock(&cil->xc_cil_lock); |
405 | push_seq = cil->xc_push_seq; | |
406 | ASSERT(push_seq <= ctx->sequence); | |
71e330b5 | 407 | |
4c2d542f DC |
408 | /* |
409 | * Check if we've anything to push. If there is nothing, then we don't | |
410 | * move on to a new sequence number and so we have to be able to push | |
411 | * this sequence again later. | |
412 | */ | |
413 | if (list_empty(&cil->xc_cil)) { | |
414 | cil->xc_push_seq = 0; | |
415 | spin_unlock(&cil->xc_cil_lock); | |
a44f13ed | 416 | goto out_skip; |
4c2d542f DC |
417 | } |
418 | spin_unlock(&cil->xc_cil_lock); | |
419 | ||
a44f13ed DC |
420 | |
421 | /* check for a previously pushed seqeunce */ | |
4c2d542f | 422 | if (push_seq < cil->xc_ctx->sequence) |
df806158 DC |
423 | goto out_skip; |
424 | ||
71e330b5 DC |
425 | /* |
426 | * pull all the log vectors off the items in the CIL, and | |
427 | * remove the items from the CIL. We don't need the CIL lock | |
428 | * here because it's only needed on the transaction commit | |
429 | * side which is currently locked out by the flush lock. | |
430 | */ | |
431 | lv = NULL; | |
432 | num_lv = 0; | |
433 | num_iovecs = 0; | |
434 | len = 0; | |
435 | while (!list_empty(&cil->xc_cil)) { | |
436 | struct xfs_log_item *item; | |
437 | int i; | |
438 | ||
439 | item = list_first_entry(&cil->xc_cil, | |
440 | struct xfs_log_item, li_cil); | |
441 | list_del_init(&item->li_cil); | |
442 | if (!ctx->lv_chain) | |
443 | ctx->lv_chain = item->li_lv; | |
444 | else | |
445 | lv->lv_next = item->li_lv; | |
446 | lv = item->li_lv; | |
447 | item->li_lv = NULL; | |
448 | ||
449 | num_lv++; | |
450 | num_iovecs += lv->lv_niovecs; | |
451 | for (i = 0; i < lv->lv_niovecs; i++) | |
452 | len += lv->lv_iovecp[i].i_len; | |
453 | } | |
454 | ||
455 | /* | |
456 | * initialise the new context and attach it to the CIL. Then attach | |
457 | * the current context to the CIL committing lsit so it can be found | |
458 | * during log forces to extract the commit lsn of the sequence that | |
459 | * needs to be forced. | |
460 | */ | |
461 | INIT_LIST_HEAD(&new_ctx->committing); | |
462 | INIT_LIST_HEAD(&new_ctx->busy_extents); | |
463 | new_ctx->sequence = ctx->sequence + 1; | |
464 | new_ctx->cil = cil; | |
465 | cil->xc_ctx = new_ctx; | |
466 | ||
a44f13ed DC |
467 | /* |
468 | * mirror the new sequence into the cil structure so that we can do | |
469 | * unlocked checks against the current sequence in log forces without | |
470 | * risking deferencing a freed context pointer. | |
471 | */ | |
472 | cil->xc_current_sequence = new_ctx->sequence; | |
473 | ||
71e330b5 DC |
474 | /* |
475 | * The switch is now done, so we can drop the context lock and move out | |
476 | * of a shared context. We can't just go straight to the commit record, | |
477 | * though - we need to synchronise with previous and future commits so | |
478 | * that the commit records are correctly ordered in the log to ensure | |
479 | * that we process items during log IO completion in the correct order. | |
480 | * | |
481 | * For example, if we get an EFI in one checkpoint and the EFD in the | |
482 | * next (e.g. due to log forces), we do not want the checkpoint with | |
483 | * the EFD to be committed before the checkpoint with the EFI. Hence | |
484 | * we must strictly order the commit records of the checkpoints so | |
485 | * that: a) the checkpoint callbacks are attached to the iclogs in the | |
486 | * correct order; and b) the checkpoints are replayed in correct order | |
487 | * in log recovery. | |
488 | * | |
489 | * Hence we need to add this context to the committing context list so | |
490 | * that higher sequences will wait for us to write out a commit record | |
491 | * before they do. | |
492 | */ | |
493 | spin_lock(&cil->xc_cil_lock); | |
494 | list_add(&ctx->committing, &cil->xc_committing); | |
495 | spin_unlock(&cil->xc_cil_lock); | |
496 | up_write(&cil->xc_ctx_lock); | |
497 | ||
498 | /* | |
499 | * Build a checkpoint transaction header and write it to the log to | |
500 | * begin the transaction. We need to account for the space used by the | |
501 | * transaction header here as it is not accounted for in xlog_write(). | |
502 | * | |
503 | * The LSN we need to pass to the log items on transaction commit is | |
504 | * the LSN reported by the first log vector write. If we use the commit | |
505 | * record lsn then we can move the tail beyond the grant write head. | |
506 | */ | |
507 | tic = ctx->ticket; | |
508 | thdr.th_magic = XFS_TRANS_HEADER_MAGIC; | |
509 | thdr.th_type = XFS_TRANS_CHECKPOINT; | |
510 | thdr.th_tid = tic->t_tid; | |
511 | thdr.th_num_items = num_iovecs; | |
4e0d5f92 | 512 | lhdr.i_addr = &thdr; |
71e330b5 DC |
513 | lhdr.i_len = sizeof(xfs_trans_header_t); |
514 | lhdr.i_type = XLOG_REG_TYPE_TRANSHDR; | |
515 | tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t); | |
516 | ||
517 | lvhdr.lv_niovecs = 1; | |
518 | lvhdr.lv_iovecp = &lhdr; | |
519 | lvhdr.lv_next = ctx->lv_chain; | |
520 | ||
521 | error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); | |
522 | if (error) | |
7db37c5e | 523 | goto out_abort_free_ticket; |
71e330b5 DC |
524 | |
525 | /* | |
526 | * now that we've written the checkpoint into the log, strictly | |
527 | * order the commit records so replay will get them in the right order. | |
528 | */ | |
529 | restart: | |
530 | spin_lock(&cil->xc_cil_lock); | |
531 | list_for_each_entry(new_ctx, &cil->xc_committing, committing) { | |
532 | /* | |
533 | * Higher sequences will wait for this one so skip them. | |
534 | * Don't wait for own own sequence, either. | |
535 | */ | |
536 | if (new_ctx->sequence >= ctx->sequence) | |
537 | continue; | |
538 | if (!new_ctx->commit_lsn) { | |
539 | /* | |
540 | * It is still being pushed! Wait for the push to | |
541 | * complete, then start again from the beginning. | |
542 | */ | |
eb40a875 | 543 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
544 | goto restart; |
545 | } | |
546 | } | |
547 | spin_unlock(&cil->xc_cil_lock); | |
548 | ||
7db37c5e | 549 | /* xfs_log_done always frees the ticket on error. */ |
71e330b5 | 550 | commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); |
7db37c5e | 551 | if (commit_lsn == -1) |
71e330b5 DC |
552 | goto out_abort; |
553 | ||
554 | /* attach all the transactions w/ busy extents to iclog */ | |
555 | ctx->log_cb.cb_func = xlog_cil_committed; | |
556 | ctx->log_cb.cb_arg = ctx; | |
557 | error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb); | |
558 | if (error) | |
559 | goto out_abort; | |
560 | ||
561 | /* | |
562 | * now the checkpoint commit is complete and we've attached the | |
563 | * callbacks to the iclog we can assign the commit LSN to the context | |
564 | * and wake up anyone who is waiting for the commit to complete. | |
565 | */ | |
566 | spin_lock(&cil->xc_cil_lock); | |
567 | ctx->commit_lsn = commit_lsn; | |
eb40a875 | 568 | wake_up_all(&cil->xc_commit_wait); |
71e330b5 DC |
569 | spin_unlock(&cil->xc_cil_lock); |
570 | ||
571 | /* release the hounds! */ | |
572 | return xfs_log_release_iclog(log->l_mp, commit_iclog); | |
573 | ||
574 | out_skip: | |
575 | up_write(&cil->xc_ctx_lock); | |
576 | xfs_log_ticket_put(new_ctx->ticket); | |
577 | kmem_free(new_ctx); | |
578 | return 0; | |
579 | ||
7db37c5e DC |
580 | out_abort_free_ticket: |
581 | xfs_log_ticket_put(tic); | |
71e330b5 DC |
582 | out_abort: |
583 | xlog_cil_committed(ctx, XFS_LI_ABORTED); | |
584 | return XFS_ERROR(EIO); | |
585 | } | |
586 | ||
4c2d542f DC |
587 | static void |
588 | xlog_cil_push_work( | |
589 | struct work_struct *work) | |
590 | { | |
591 | struct xfs_cil *cil = container_of(work, struct xfs_cil, | |
592 | xc_push_work); | |
593 | xlog_cil_push(cil->xc_log); | |
594 | } | |
595 | ||
596 | /* | |
597 | * We need to push CIL every so often so we don't cache more than we can fit in | |
598 | * the log. The limit really is that a checkpoint can't be more than half the | |
599 | * log (the current checkpoint is not allowed to overwrite the previous | |
600 | * checkpoint), but commit latency and memory usage limit this to a smaller | |
601 | * size. | |
602 | */ | |
603 | static void | |
604 | xlog_cil_push_background( | |
605 | struct log *log) | |
606 | { | |
607 | struct xfs_cil *cil = log->l_cilp; | |
608 | ||
609 | /* | |
610 | * The cil won't be empty because we are called while holding the | |
611 | * context lock so whatever we added to the CIL will still be there | |
612 | */ | |
613 | ASSERT(!list_empty(&cil->xc_cil)); | |
614 | ||
615 | /* | |
616 | * don't do a background push if we haven't used up all the | |
617 | * space available yet. | |
618 | */ | |
619 | if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) | |
620 | return; | |
621 | ||
622 | spin_lock(&cil->xc_cil_lock); | |
623 | if (cil->xc_push_seq < cil->xc_current_sequence) { | |
624 | cil->xc_push_seq = cil->xc_current_sequence; | |
625 | queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work); | |
626 | } | |
627 | spin_unlock(&cil->xc_cil_lock); | |
628 | ||
629 | } | |
630 | ||
631 | static void | |
632 | xlog_cil_push_foreground( | |
633 | struct log *log, | |
634 | xfs_lsn_t push_seq) | |
635 | { | |
636 | struct xfs_cil *cil = log->l_cilp; | |
637 | ||
638 | if (!cil) | |
639 | return; | |
640 | ||
641 | ASSERT(push_seq && push_seq <= cil->xc_current_sequence); | |
642 | ||
643 | /* start on any pending background push to minimise wait time on it */ | |
644 | flush_work(&cil->xc_push_work); | |
645 | ||
646 | /* | |
647 | * If the CIL is empty or we've already pushed the sequence then | |
648 | * there's no work we need to do. | |
649 | */ | |
650 | spin_lock(&cil->xc_cil_lock); | |
651 | if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) { | |
652 | spin_unlock(&cil->xc_cil_lock); | |
653 | return; | |
654 | } | |
655 | ||
656 | cil->xc_push_seq = push_seq; | |
657 | spin_unlock(&cil->xc_cil_lock); | |
658 | ||
659 | /* do the push now */ | |
660 | xlog_cil_push(log); | |
661 | } | |
662 | ||
a44f13ed DC |
663 | /* |
664 | * Commit a transaction with the given vector to the Committed Item List. | |
665 | * | |
666 | * To do this, we need to format the item, pin it in memory if required and | |
667 | * account for the space used by the transaction. Once we have done that we | |
668 | * need to release the unused reservation for the transaction, attach the | |
669 | * transaction to the checkpoint context so we carry the busy extents through | |
670 | * to checkpoint completion, and then unlock all the items in the transaction. | |
671 | * | |
672 | * For more specific information about the order of operations in | |
673 | * xfs_log_commit_cil() please refer to the comments in | |
674 | * xfs_trans_commit_iclog(). | |
675 | * | |
676 | * Called with the context lock already held in read mode to lock out | |
677 | * background commit, returns without it held once background commits are | |
678 | * allowed again. | |
679 | */ | |
0244b960 | 680 | int |
a44f13ed DC |
681 | xfs_log_commit_cil( |
682 | struct xfs_mount *mp, | |
683 | struct xfs_trans *tp, | |
a44f13ed DC |
684 | xfs_lsn_t *commit_lsn, |
685 | int flags) | |
686 | { | |
687 | struct log *log = mp->m_log; | |
688 | int log_flags = 0; | |
0244b960 | 689 | struct xfs_log_vec *log_vector; |
a44f13ed DC |
690 | |
691 | if (flags & XFS_TRANS_RELEASE_LOG_RES) | |
692 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
693 | ||
3b93c7aa | 694 | /* |
0244b960 | 695 | * Do all the hard work of formatting items (including memory |
3b93c7aa DC |
696 | * allocation) outside the CIL context lock. This prevents stalling CIL |
697 | * pushes when we are low on memory and a transaction commit spends a | |
698 | * lot of time in memory reclaim. | |
699 | */ | |
0244b960 CH |
700 | log_vector = xlog_cil_prepare_log_vecs(tp); |
701 | if (!log_vector) | |
702 | return ENOMEM; | |
3b93c7aa | 703 | |
a44f13ed DC |
704 | /* lock out background commit */ |
705 | down_read(&log->l_cilp->xc_ctx_lock); | |
d1583a38 DC |
706 | if (commit_lsn) |
707 | *commit_lsn = log->l_cilp->xc_ctx->sequence; | |
708 | ||
709 | xlog_cil_insert_items(log, log_vector, tp->t_ticket); | |
a44f13ed DC |
710 | |
711 | /* check we didn't blow the reservation */ | |
712 | if (tp->t_ticket->t_curr_res < 0) | |
713 | xlog_print_tic_res(log->l_mp, tp->t_ticket); | |
714 | ||
715 | /* attach the transaction to the CIL if it has any busy extents */ | |
716 | if (!list_empty(&tp->t_busy)) { | |
717 | spin_lock(&log->l_cilp->xc_cil_lock); | |
718 | list_splice_init(&tp->t_busy, | |
719 | &log->l_cilp->xc_ctx->busy_extents); | |
720 | spin_unlock(&log->l_cilp->xc_cil_lock); | |
721 | } | |
722 | ||
723 | tp->t_commit_lsn = *commit_lsn; | |
724 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
725 | xfs_trans_unreserve_and_mod_sb(tp); | |
726 | ||
727 | /* | |
728 | * Once all the items of the transaction have been copied to the CIL, | |
729 | * the items can be unlocked and freed. | |
730 | * | |
731 | * This needs to be done before we drop the CIL context lock because we | |
732 | * have to update state in the log items and unlock them before they go | |
733 | * to disk. If we don't, then the CIL checkpoint can race with us and | |
734 | * we can run checkpoint completion before we've updated and unlocked | |
735 | * the log items. This affects (at least) processing of stale buffers, | |
736 | * inodes and EFIs. | |
737 | */ | |
738 | xfs_trans_free_items(tp, *commit_lsn, 0); | |
739 | ||
4c2d542f | 740 | xlog_cil_push_background(log); |
a44f13ed DC |
741 | |
742 | up_read(&log->l_cilp->xc_ctx_lock); | |
0244b960 | 743 | return 0; |
a44f13ed DC |
744 | } |
745 | ||
71e330b5 DC |
746 | /* |
747 | * Conditionally push the CIL based on the sequence passed in. | |
748 | * | |
749 | * We only need to push if we haven't already pushed the sequence | |
750 | * number given. Hence the only time we will trigger a push here is | |
751 | * if the push sequence is the same as the current context. | |
752 | * | |
753 | * We return the current commit lsn to allow the callers to determine if a | |
754 | * iclog flush is necessary following this call. | |
71e330b5 DC |
755 | */ |
756 | xfs_lsn_t | |
a44f13ed | 757 | xlog_cil_force_lsn( |
71e330b5 | 758 | struct log *log, |
a44f13ed | 759 | xfs_lsn_t sequence) |
71e330b5 DC |
760 | { |
761 | struct xfs_cil *cil = log->l_cilp; | |
762 | struct xfs_cil_ctx *ctx; | |
763 | xfs_lsn_t commit_lsn = NULLCOMMITLSN; | |
764 | ||
a44f13ed DC |
765 | ASSERT(sequence <= cil->xc_current_sequence); |
766 | ||
767 | /* | |
768 | * check to see if we need to force out the current context. | |
769 | * xlog_cil_push() handles racing pushes for the same sequence, | |
770 | * so no need to deal with it here. | |
771 | */ | |
4c2d542f | 772 | xlog_cil_push_foreground(log, sequence); |
71e330b5 DC |
773 | |
774 | /* | |
775 | * See if we can find a previous sequence still committing. | |
71e330b5 DC |
776 | * We need to wait for all previous sequence commits to complete |
777 | * before allowing the force of push_seq to go ahead. Hence block | |
778 | * on commits for those as well. | |
779 | */ | |
a44f13ed | 780 | restart: |
71e330b5 | 781 | spin_lock(&cil->xc_cil_lock); |
71e330b5 | 782 | list_for_each_entry(ctx, &cil->xc_committing, committing) { |
a44f13ed | 783 | if (ctx->sequence > sequence) |
71e330b5 DC |
784 | continue; |
785 | if (!ctx->commit_lsn) { | |
786 | /* | |
787 | * It is still being pushed! Wait for the push to | |
788 | * complete, then start again from the beginning. | |
789 | */ | |
eb40a875 | 790 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
791 | goto restart; |
792 | } | |
a44f13ed | 793 | if (ctx->sequence != sequence) |
71e330b5 DC |
794 | continue; |
795 | /* found it! */ | |
796 | commit_lsn = ctx->commit_lsn; | |
797 | } | |
798 | spin_unlock(&cil->xc_cil_lock); | |
799 | return commit_lsn; | |
800 | } | |
ccf7c23f DC |
801 | |
802 | /* | |
803 | * Check if the current log item was first committed in this sequence. | |
804 | * We can't rely on just the log item being in the CIL, we have to check | |
805 | * the recorded commit sequence number. | |
806 | * | |
807 | * Note: for this to be used in a non-racy manner, it has to be called with | |
808 | * CIL flushing locked out. As a result, it should only be used during the | |
809 | * transaction commit process when deciding what to format into the item. | |
810 | */ | |
811 | bool | |
812 | xfs_log_item_in_current_chkpt( | |
813 | struct xfs_log_item *lip) | |
814 | { | |
815 | struct xfs_cil_ctx *ctx; | |
816 | ||
ccf7c23f DC |
817 | if (list_empty(&lip->li_cil)) |
818 | return false; | |
819 | ||
820 | ctx = lip->li_mountp->m_log->l_cilp->xc_ctx; | |
821 | ||
822 | /* | |
823 | * li_seq is written on the first commit of a log item to record the | |
824 | * first checkpoint it is written to. Hence if it is different to the | |
825 | * current sequence, we're in a new checkpoint. | |
826 | */ | |
827 | if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0) | |
828 | return false; | |
829 | return true; | |
830 | } | |
4c2d542f DC |
831 | |
832 | /* | |
833 | * Perform initial CIL structure initialisation. | |
834 | */ | |
835 | int | |
836 | xlog_cil_init( | |
837 | struct log *log) | |
838 | { | |
839 | struct xfs_cil *cil; | |
840 | struct xfs_cil_ctx *ctx; | |
841 | ||
842 | cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL); | |
843 | if (!cil) | |
844 | return ENOMEM; | |
845 | ||
846 | ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL); | |
847 | if (!ctx) { | |
848 | kmem_free(cil); | |
849 | return ENOMEM; | |
850 | } | |
851 | ||
852 | INIT_WORK(&cil->xc_push_work, xlog_cil_push_work); | |
853 | INIT_LIST_HEAD(&cil->xc_cil); | |
854 | INIT_LIST_HEAD(&cil->xc_committing); | |
855 | spin_lock_init(&cil->xc_cil_lock); | |
856 | init_rwsem(&cil->xc_ctx_lock); | |
857 | init_waitqueue_head(&cil->xc_commit_wait); | |
858 | ||
859 | INIT_LIST_HEAD(&ctx->committing); | |
860 | INIT_LIST_HEAD(&ctx->busy_extents); | |
861 | ctx->sequence = 1; | |
862 | ctx->cil = cil; | |
863 | cil->xc_ctx = ctx; | |
864 | cil->xc_current_sequence = ctx->sequence; | |
865 | ||
866 | cil->xc_log = log; | |
867 | log->l_cilp = cil; | |
868 | return 0; | |
869 | } | |
870 | ||
871 | void | |
872 | xlog_cil_destroy( | |
873 | struct log *log) | |
874 | { | |
875 | if (log->l_cilp->xc_ctx) { | |
876 | if (log->l_cilp->xc_ctx->ticket) | |
877 | xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket); | |
878 | kmem_free(log->l_cilp->xc_ctx); | |
879 | } | |
880 | ||
881 | ASSERT(list_empty(&log->l_cilp->xc_cil)); | |
882 | kmem_free(log->l_cilp); | |
883 | } | |
884 |