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" | |
71e330b5 | 21 | #include "xfs_log.h" |
71e330b5 DC |
22 | #include "xfs_trans.h" |
23 | #include "xfs_trans_priv.h" | |
24 | #include "xfs_log_priv.h" | |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
71e330b5 DC |
27 | #include "xfs_mount.h" |
28 | #include "xfs_error.h" | |
29 | #include "xfs_alloc.h" | |
efc27b52 | 30 | #include "xfs_extent_busy.h" |
e84661aa | 31 | #include "xfs_discard.h" |
71e330b5 | 32 | |
71e330b5 DC |
33 | /* |
34 | * Allocate a new ticket. Failing to get a new ticket makes it really hard to | |
35 | * recover, so we don't allow failure here. Also, we allocate in a context that | |
36 | * we don't want to be issuing transactions from, so we need to tell the | |
37 | * allocation code this as well. | |
38 | * | |
39 | * We don't reserve any space for the ticket - we are going to steal whatever | |
40 | * space we require from transactions as they commit. To ensure we reserve all | |
41 | * the space required, we need to set the current reservation of the ticket to | |
42 | * zero so that we know to steal the initial transaction overhead from the | |
43 | * first transaction commit. | |
44 | */ | |
45 | static struct xlog_ticket * | |
46 | xlog_cil_ticket_alloc( | |
f7bdf03a | 47 | struct xlog *log) |
71e330b5 DC |
48 | { |
49 | struct xlog_ticket *tic; | |
50 | ||
51 | tic = xlog_ticket_alloc(log, 0, 1, XFS_TRANSACTION, 0, | |
52 | KM_SLEEP|KM_NOFS); | |
53 | tic->t_trans_type = XFS_TRANS_CHECKPOINT; | |
54 | ||
55 | /* | |
56 | * set the current reservation to zero so we know to steal the basic | |
57 | * transaction overhead reservation from the first transaction commit. | |
58 | */ | |
59 | tic->t_curr_res = 0; | |
60 | return tic; | |
61 | } | |
62 | ||
63 | /* | |
64 | * After the first stage of log recovery is done, we know where the head and | |
65 | * tail of the log are. We need this log initialisation done before we can | |
66 | * initialise the first CIL checkpoint context. | |
67 | * | |
68 | * Here we allocate a log ticket to track space usage during a CIL push. This | |
69 | * ticket is passed to xlog_write() directly so that we don't slowly leak log | |
70 | * space by failing to account for space used by log headers and additional | |
71 | * region headers for split regions. | |
72 | */ | |
73 | void | |
74 | xlog_cil_init_post_recovery( | |
f7bdf03a | 75 | struct xlog *log) |
71e330b5 | 76 | { |
71e330b5 DC |
77 | log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log); |
78 | log->l_cilp->xc_ctx->sequence = 1; | |
79 | log->l_cilp->xc_ctx->commit_lsn = xlog_assign_lsn(log->l_curr_cycle, | |
80 | log->l_curr_block); | |
81 | } | |
82 | ||
71e330b5 DC |
83 | /* |
84 | * Format log item into a flat buffers | |
85 | * | |
86 | * For delayed logging, we need to hold a formatted buffer containing all the | |
87 | * changes on the log item. This enables us to relog the item in memory and | |
88 | * write it out asynchronously without needing to relock the object that was | |
89 | * modified at the time it gets written into the iclog. | |
90 | * | |
91 | * This function builds a vector for the changes in each log item in the | |
92 | * transaction. It then works out the length of the buffer needed for each log | |
93 | * item, allocates them and formats the vector for the item into the buffer. | |
94 | * The buffer is then attached to the log item are then inserted into the | |
95 | * Committed Item List for tracking until the next checkpoint is written out. | |
96 | * | |
97 | * We don't set up region headers during this process; we simply copy the | |
98 | * regions into the flat buffer. We can do this because we still have to do a | |
99 | * formatting step to write the regions into the iclog buffer. Writing the | |
100 | * ophdrs during the iclog write means that we can support splitting large | |
101 | * regions across iclog boundares without needing a change in the format of the | |
102 | * item/region encapsulation. | |
103 | * | |
104 | * Hence what we need to do now is change the rewrite the vector array to point | |
105 | * to the copied region inside the buffer we just allocated. This allows us to | |
106 | * format the regions into the iclog as though they are being formatted | |
107 | * directly out of the objects themselves. | |
108 | */ | |
0244b960 CH |
109 | static struct xfs_log_vec * |
110 | xlog_cil_prepare_log_vecs( | |
111 | struct xfs_trans *tp) | |
71e330b5 | 112 | { |
0244b960 CH |
113 | struct xfs_log_item_desc *lidp; |
114 | struct xfs_log_vec *lv = NULL; | |
115 | struct xfs_log_vec *ret_lv = NULL; | |
71e330b5 | 116 | |
0244b960 CH |
117 | |
118 | /* Bail out if we didn't find a log item. */ | |
119 | if (list_empty(&tp->t_items)) { | |
120 | ASSERT(0); | |
121 | return NULL; | |
122 | } | |
123 | ||
124 | list_for_each_entry(lidp, &tp->t_items, lid_trans) { | |
125 | struct xfs_log_vec *new_lv; | |
71e330b5 DC |
126 | void *ptr; |
127 | int index; | |
128 | int len = 0; | |
b3934213 | 129 | uint niovecs; |
fd63875c | 130 | bool ordered = false; |
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 | ||
fd63875c DC |
141 | /* |
142 | * Ordered items need to be tracked but we do not wish to write | |
143 | * them. We need a logvec to track the object, but we do not | |
144 | * need an iovec or buffer to be allocated for copying data. | |
145 | */ | |
146 | if (niovecs == XFS_LOG_VEC_ORDERED) { | |
147 | ordered = true; | |
148 | niovecs = 0; | |
149 | } | |
150 | ||
0244b960 | 151 | new_lv = kmem_zalloc(sizeof(*new_lv) + |
b3934213 | 152 | niovecs * sizeof(struct xfs_log_iovec), |
ac14876c | 153 | KM_SLEEP|KM_NOFS); |
0244b960 | 154 | |
fd63875c DC |
155 | new_lv->lv_item = lidp->lid_item; |
156 | new_lv->lv_niovecs = niovecs; | |
157 | if (ordered) { | |
158 | /* track as an ordered logvec */ | |
159 | new_lv->lv_buf_len = XFS_LOG_VEC_ORDERED; | |
160 | goto next; | |
161 | } | |
162 | ||
0244b960 CH |
163 | /* The allocated iovec region lies beyond the log vector. */ |
164 | new_lv->lv_iovecp = (struct xfs_log_iovec *)&new_lv[1]; | |
0244b960 | 165 | |
71e330b5 | 166 | /* build the vector array and calculate it's length */ |
0244b960 CH |
167 | IOP_FORMAT(new_lv->lv_item, new_lv->lv_iovecp); |
168 | for (index = 0; index < new_lv->lv_niovecs; index++) | |
169 | len += new_lv->lv_iovecp[index].i_len; | |
71e330b5 | 170 | |
0244b960 CH |
171 | new_lv->lv_buf_len = len; |
172 | new_lv->lv_buf = kmem_alloc(new_lv->lv_buf_len, | |
173 | KM_SLEEP|KM_NOFS); | |
174 | ptr = new_lv->lv_buf; | |
71e330b5 | 175 | |
0244b960 CH |
176 | for (index = 0; index < new_lv->lv_niovecs; index++) { |
177 | struct xfs_log_iovec *vec = &new_lv->lv_iovecp[index]; | |
71e330b5 DC |
178 | |
179 | memcpy(ptr, vec->i_addr, vec->i_len); | |
180 | vec->i_addr = ptr; | |
181 | ptr += vec->i_len; | |
182 | } | |
0244b960 CH |
183 | ASSERT(ptr == new_lv->lv_buf + new_lv->lv_buf_len); |
184 | ||
fd63875c | 185 | next: |
0244b960 CH |
186 | if (!ret_lv) |
187 | ret_lv = new_lv; | |
188 | else | |
189 | lv->lv_next = new_lv; | |
190 | lv = new_lv; | |
3b93c7aa | 191 | } |
0244b960 CH |
192 | |
193 | return ret_lv; | |
3b93c7aa | 194 | } |
71e330b5 | 195 | |
d1583a38 DC |
196 | /* |
197 | * Prepare the log item for insertion into the CIL. Calculate the difference in | |
198 | * log space and vectors it will consume, and if it is a new item pin it as | |
199 | * well. | |
200 | */ | |
201 | STATIC void | |
202 | xfs_cil_prepare_item( | |
f7bdf03a | 203 | struct xlog *log, |
d1583a38 DC |
204 | struct xfs_log_vec *lv, |
205 | int *len, | |
206 | int *diff_iovecs) | |
207 | { | |
208 | struct xfs_log_vec *old = lv->lv_item->li_lv; | |
209 | ||
210 | if (old) { | |
211 | /* existing lv on log item, space used is a delta */ | |
fd63875c DC |
212 | ASSERT((old->lv_buf && old->lv_buf_len && old->lv_niovecs) || |
213 | old->lv_buf_len == XFS_LOG_VEC_ORDERED); | |
214 | ||
215 | /* | |
216 | * If the new item is ordered, keep the old one that is already | |
217 | * tracking dirty or ordered regions | |
218 | */ | |
219 | if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { | |
220 | ASSERT(!lv->lv_buf); | |
221 | kmem_free(lv); | |
222 | return; | |
223 | } | |
d1583a38 DC |
224 | |
225 | *len += lv->lv_buf_len - old->lv_buf_len; | |
226 | *diff_iovecs += lv->lv_niovecs - old->lv_niovecs; | |
227 | kmem_free(old->lv_buf); | |
228 | kmem_free(old); | |
229 | } else { | |
230 | /* new lv, must pin the log item */ | |
231 | ASSERT(!lv->lv_item->li_lv); | |
d1583a38 | 232 | |
fd63875c DC |
233 | if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) { |
234 | *len += lv->lv_buf_len; | |
235 | *diff_iovecs += lv->lv_niovecs; | |
236 | } | |
d1583a38 DC |
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 | |
42b2aa86 | 258 | * as well. Remove the amount of space we added to the checkpoint ticket from |
d1583a38 DC |
259 | * the current transaction ticket so that the accounting works out correctly. |
260 | */ | |
3b93c7aa DC |
261 | static void |
262 | xlog_cil_insert_items( | |
f7bdf03a | 263 | struct xlog *log, |
3b93c7aa | 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 | */ | |
d1583a38 | 291 | spin_lock(&cil->xc_cil_lock); |
fd63875c DC |
292 | for (lv = log_vector; lv; ) { |
293 | struct xfs_log_vec *next = lv->lv_next; | |
d1583a38 | 294 | |
fd63875c DC |
295 | ASSERT(lv->lv_item->li_lv || list_empty(&lv->lv_item->li_cil)); |
296 | lv->lv_next = NULL; | |
297 | ||
298 | /* | |
299 | * xfs_cil_prepare_item() may free the lv, so move the item on | |
300 | * the CIL first. | |
301 | */ | |
d1583a38 | 302 | list_move_tail(&lv->lv_item->li_cil, &cil->xc_cil); |
fd63875c DC |
303 | xfs_cil_prepare_item(log, lv, &len, &diff_iovecs); |
304 | lv = next; | |
305 | } | |
d1583a38 | 306 | |
fd63875c DC |
307 | /* account for space used by new iovec headers */ |
308 | len += diff_iovecs * sizeof(xlog_op_header_t); | |
d1583a38 DC |
309 | ctx->nvecs += diff_iovecs; |
310 | ||
311 | /* | |
312 | * Now transfer enough transaction reservation to the context ticket | |
313 | * for the checkpoint. The context ticket is special - the unit | |
314 | * reservation has to grow as well as the current reservation as we | |
315 | * steal from tickets so we can correctly determine the space used | |
316 | * during the transaction commit. | |
317 | */ | |
318 | if (ctx->ticket->t_curr_res == 0) { | |
319 | /* first commit in checkpoint, steal the header reservation */ | |
320 | ASSERT(ticket->t_curr_res >= ctx->ticket->t_unit_res + len); | |
321 | ctx->ticket->t_curr_res = ctx->ticket->t_unit_res; | |
322 | ticket->t_curr_res -= ctx->ticket->t_unit_res; | |
323 | } | |
324 | ||
325 | /* do we need space for more log record headers? */ | |
326 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
327 | if (len > 0 && (ctx->space_used / iclog_space != | |
328 | (ctx->space_used + len) / iclog_space)) { | |
329 | int hdrs; | |
330 | ||
331 | hdrs = (len + iclog_space - 1) / iclog_space; | |
332 | /* need to take into account split region headers, too */ | |
333 | hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header); | |
334 | ctx->ticket->t_unit_res += hdrs; | |
335 | ctx->ticket->t_curr_res += hdrs; | |
336 | ticket->t_curr_res -= hdrs; | |
337 | ASSERT(ticket->t_curr_res >= len); | |
338 | } | |
339 | ticket->t_curr_res -= len; | |
340 | ctx->space_used += len; | |
341 | ||
342 | spin_unlock(&cil->xc_cil_lock); | |
71e330b5 DC |
343 | } |
344 | ||
345 | static void | |
346 | xlog_cil_free_logvec( | |
347 | struct xfs_log_vec *log_vector) | |
348 | { | |
349 | struct xfs_log_vec *lv; | |
350 | ||
351 | for (lv = log_vector; lv; ) { | |
352 | struct xfs_log_vec *next = lv->lv_next; | |
353 | kmem_free(lv->lv_buf); | |
354 | kmem_free(lv); | |
355 | lv = next; | |
356 | } | |
357 | } | |
358 | ||
71e330b5 DC |
359 | /* |
360 | * Mark all items committed and clear busy extents. We free the log vector | |
361 | * chains in a separate pass so that we unpin the log items as quickly as | |
362 | * possible. | |
363 | */ | |
364 | static void | |
365 | xlog_cil_committed( | |
366 | void *args, | |
367 | int abort) | |
368 | { | |
369 | struct xfs_cil_ctx *ctx = args; | |
e84661aa | 370 | struct xfs_mount *mp = ctx->cil->xc_log->l_mp; |
71e330b5 | 371 | |
0e57f6a3 DC |
372 | xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain, |
373 | ctx->start_lsn, abort); | |
71e330b5 | 374 | |
4ecbfe63 DC |
375 | xfs_extent_busy_sort(&ctx->busy_extents); |
376 | xfs_extent_busy_clear(mp, &ctx->busy_extents, | |
e84661aa | 377 | (mp->m_flags & XFS_MOUNT_DISCARD) && !abort); |
71e330b5 DC |
378 | |
379 | spin_lock(&ctx->cil->xc_cil_lock); | |
380 | list_del(&ctx->committing); | |
381 | spin_unlock(&ctx->cil->xc_cil_lock); | |
382 | ||
383 | xlog_cil_free_logvec(ctx->lv_chain); | |
e84661aa CH |
384 | |
385 | if (!list_empty(&ctx->busy_extents)) { | |
386 | ASSERT(mp->m_flags & XFS_MOUNT_DISCARD); | |
387 | ||
388 | xfs_discard_extents(mp, &ctx->busy_extents); | |
4ecbfe63 | 389 | xfs_extent_busy_clear(mp, &ctx->busy_extents, false); |
e84661aa CH |
390 | } |
391 | ||
71e330b5 DC |
392 | kmem_free(ctx); |
393 | } | |
394 | ||
395 | /* | |
a44f13ed DC |
396 | * Push the Committed Item List to the log. If @push_seq flag is zero, then it |
397 | * is a background flush and so we can chose to ignore it. Otherwise, if the | |
398 | * current sequence is the same as @push_seq we need to do a flush. If | |
399 | * @push_seq is less than the current sequence, then it has already been | |
400 | * flushed and we don't need to do anything - the caller will wait for it to | |
401 | * complete if necessary. | |
402 | * | |
403 | * @push_seq is a value rather than a flag because that allows us to do an | |
404 | * unlocked check of the sequence number for a match. Hence we can allows log | |
405 | * forces to run racily and not issue pushes for the same sequence twice. If we | |
406 | * get a race between multiple pushes for the same sequence they will block on | |
407 | * the first one and then abort, hence avoiding needless pushes. | |
71e330b5 | 408 | */ |
a44f13ed | 409 | STATIC int |
71e330b5 | 410 | xlog_cil_push( |
f7bdf03a | 411 | struct xlog *log) |
71e330b5 DC |
412 | { |
413 | struct xfs_cil *cil = log->l_cilp; | |
414 | struct xfs_log_vec *lv; | |
415 | struct xfs_cil_ctx *ctx; | |
416 | struct xfs_cil_ctx *new_ctx; | |
417 | struct xlog_in_core *commit_iclog; | |
418 | struct xlog_ticket *tic; | |
71e330b5 | 419 | int num_iovecs; |
71e330b5 DC |
420 | int error = 0; |
421 | struct xfs_trans_header thdr; | |
422 | struct xfs_log_iovec lhdr; | |
423 | struct xfs_log_vec lvhdr = { NULL }; | |
424 | xfs_lsn_t commit_lsn; | |
4c2d542f | 425 | xfs_lsn_t push_seq; |
71e330b5 DC |
426 | |
427 | if (!cil) | |
428 | return 0; | |
429 | ||
71e330b5 DC |
430 | new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS); |
431 | new_ctx->ticket = xlog_cil_ticket_alloc(log); | |
432 | ||
4c2d542f | 433 | down_write(&cil->xc_ctx_lock); |
71e330b5 DC |
434 | ctx = cil->xc_ctx; |
435 | ||
4c2d542f DC |
436 | spin_lock(&cil->xc_cil_lock); |
437 | push_seq = cil->xc_push_seq; | |
438 | ASSERT(push_seq <= ctx->sequence); | |
71e330b5 | 439 | |
4c2d542f DC |
440 | /* |
441 | * Check if we've anything to push. If there is nothing, then we don't | |
442 | * move on to a new sequence number and so we have to be able to push | |
443 | * this sequence again later. | |
444 | */ | |
445 | if (list_empty(&cil->xc_cil)) { | |
446 | cil->xc_push_seq = 0; | |
447 | spin_unlock(&cil->xc_cil_lock); | |
a44f13ed | 448 | goto out_skip; |
4c2d542f DC |
449 | } |
450 | spin_unlock(&cil->xc_cil_lock); | |
451 | ||
a44f13ed DC |
452 | |
453 | /* check for a previously pushed seqeunce */ | |
4c2d542f | 454 | if (push_seq < cil->xc_ctx->sequence) |
df806158 DC |
455 | goto out_skip; |
456 | ||
71e330b5 DC |
457 | /* |
458 | * pull all the log vectors off the items in the CIL, and | |
459 | * remove the items from the CIL. We don't need the CIL lock | |
460 | * here because it's only needed on the transaction commit | |
461 | * side which is currently locked out by the flush lock. | |
462 | */ | |
463 | lv = NULL; | |
71e330b5 | 464 | num_iovecs = 0; |
71e330b5 DC |
465 | while (!list_empty(&cil->xc_cil)) { |
466 | struct xfs_log_item *item; | |
71e330b5 DC |
467 | |
468 | item = list_first_entry(&cil->xc_cil, | |
469 | struct xfs_log_item, li_cil); | |
470 | list_del_init(&item->li_cil); | |
471 | if (!ctx->lv_chain) | |
472 | ctx->lv_chain = item->li_lv; | |
473 | else | |
474 | lv->lv_next = item->li_lv; | |
475 | lv = item->li_lv; | |
476 | item->li_lv = NULL; | |
71e330b5 | 477 | num_iovecs += lv->lv_niovecs; |
71e330b5 DC |
478 | } |
479 | ||
480 | /* | |
481 | * initialise the new context and attach it to the CIL. Then attach | |
482 | * the current context to the CIL committing lsit so it can be found | |
483 | * during log forces to extract the commit lsn of the sequence that | |
484 | * needs to be forced. | |
485 | */ | |
486 | INIT_LIST_HEAD(&new_ctx->committing); | |
487 | INIT_LIST_HEAD(&new_ctx->busy_extents); | |
488 | new_ctx->sequence = ctx->sequence + 1; | |
489 | new_ctx->cil = cil; | |
490 | cil->xc_ctx = new_ctx; | |
491 | ||
a44f13ed DC |
492 | /* |
493 | * mirror the new sequence into the cil structure so that we can do | |
494 | * unlocked checks against the current sequence in log forces without | |
495 | * risking deferencing a freed context pointer. | |
496 | */ | |
497 | cil->xc_current_sequence = new_ctx->sequence; | |
498 | ||
71e330b5 DC |
499 | /* |
500 | * The switch is now done, so we can drop the context lock and move out | |
501 | * of a shared context. We can't just go straight to the commit record, | |
502 | * though - we need to synchronise with previous and future commits so | |
503 | * that the commit records are correctly ordered in the log to ensure | |
504 | * that we process items during log IO completion in the correct order. | |
505 | * | |
506 | * For example, if we get an EFI in one checkpoint and the EFD in the | |
507 | * next (e.g. due to log forces), we do not want the checkpoint with | |
508 | * the EFD to be committed before the checkpoint with the EFI. Hence | |
509 | * we must strictly order the commit records of the checkpoints so | |
510 | * that: a) the checkpoint callbacks are attached to the iclogs in the | |
511 | * correct order; and b) the checkpoints are replayed in correct order | |
512 | * in log recovery. | |
513 | * | |
514 | * Hence we need to add this context to the committing context list so | |
515 | * that higher sequences will wait for us to write out a commit record | |
516 | * before they do. | |
517 | */ | |
518 | spin_lock(&cil->xc_cil_lock); | |
519 | list_add(&ctx->committing, &cil->xc_committing); | |
520 | spin_unlock(&cil->xc_cil_lock); | |
521 | up_write(&cil->xc_ctx_lock); | |
522 | ||
523 | /* | |
524 | * Build a checkpoint transaction header and write it to the log to | |
525 | * begin the transaction. We need to account for the space used by the | |
526 | * transaction header here as it is not accounted for in xlog_write(). | |
527 | * | |
528 | * The LSN we need to pass to the log items on transaction commit is | |
529 | * the LSN reported by the first log vector write. If we use the commit | |
530 | * record lsn then we can move the tail beyond the grant write head. | |
531 | */ | |
532 | tic = ctx->ticket; | |
533 | thdr.th_magic = XFS_TRANS_HEADER_MAGIC; | |
534 | thdr.th_type = XFS_TRANS_CHECKPOINT; | |
535 | thdr.th_tid = tic->t_tid; | |
536 | thdr.th_num_items = num_iovecs; | |
4e0d5f92 | 537 | lhdr.i_addr = &thdr; |
71e330b5 DC |
538 | lhdr.i_len = sizeof(xfs_trans_header_t); |
539 | lhdr.i_type = XLOG_REG_TYPE_TRANSHDR; | |
540 | tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t); | |
541 | ||
542 | lvhdr.lv_niovecs = 1; | |
543 | lvhdr.lv_iovecp = &lhdr; | |
544 | lvhdr.lv_next = ctx->lv_chain; | |
545 | ||
546 | error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); | |
547 | if (error) | |
7db37c5e | 548 | goto out_abort_free_ticket; |
71e330b5 DC |
549 | |
550 | /* | |
551 | * now that we've written the checkpoint into the log, strictly | |
552 | * order the commit records so replay will get them in the right order. | |
553 | */ | |
554 | restart: | |
555 | spin_lock(&cil->xc_cil_lock); | |
556 | list_for_each_entry(new_ctx, &cil->xc_committing, committing) { | |
557 | /* | |
558 | * Higher sequences will wait for this one so skip them. | |
559 | * Don't wait for own own sequence, either. | |
560 | */ | |
561 | if (new_ctx->sequence >= ctx->sequence) | |
562 | continue; | |
563 | if (!new_ctx->commit_lsn) { | |
564 | /* | |
565 | * It is still being pushed! Wait for the push to | |
566 | * complete, then start again from the beginning. | |
567 | */ | |
eb40a875 | 568 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
569 | goto restart; |
570 | } | |
571 | } | |
572 | spin_unlock(&cil->xc_cil_lock); | |
573 | ||
7db37c5e | 574 | /* xfs_log_done always frees the ticket on error. */ |
71e330b5 | 575 | commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); |
7db37c5e | 576 | if (commit_lsn == -1) |
71e330b5 DC |
577 | goto out_abort; |
578 | ||
579 | /* attach all the transactions w/ busy extents to iclog */ | |
580 | ctx->log_cb.cb_func = xlog_cil_committed; | |
581 | ctx->log_cb.cb_arg = ctx; | |
582 | error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb); | |
583 | if (error) | |
584 | goto out_abort; | |
585 | ||
586 | /* | |
587 | * now the checkpoint commit is complete and we've attached the | |
588 | * callbacks to the iclog we can assign the commit LSN to the context | |
589 | * and wake up anyone who is waiting for the commit to complete. | |
590 | */ | |
591 | spin_lock(&cil->xc_cil_lock); | |
592 | ctx->commit_lsn = commit_lsn; | |
eb40a875 | 593 | wake_up_all(&cil->xc_commit_wait); |
71e330b5 DC |
594 | spin_unlock(&cil->xc_cil_lock); |
595 | ||
596 | /* release the hounds! */ | |
597 | return xfs_log_release_iclog(log->l_mp, commit_iclog); | |
598 | ||
599 | out_skip: | |
600 | up_write(&cil->xc_ctx_lock); | |
601 | xfs_log_ticket_put(new_ctx->ticket); | |
602 | kmem_free(new_ctx); | |
603 | return 0; | |
604 | ||
7db37c5e DC |
605 | out_abort_free_ticket: |
606 | xfs_log_ticket_put(tic); | |
71e330b5 DC |
607 | out_abort: |
608 | xlog_cil_committed(ctx, XFS_LI_ABORTED); | |
609 | return XFS_ERROR(EIO); | |
610 | } | |
611 | ||
4c2d542f DC |
612 | static void |
613 | xlog_cil_push_work( | |
614 | struct work_struct *work) | |
615 | { | |
616 | struct xfs_cil *cil = container_of(work, struct xfs_cil, | |
617 | xc_push_work); | |
618 | xlog_cil_push(cil->xc_log); | |
619 | } | |
620 | ||
621 | /* | |
622 | * We need to push CIL every so often so we don't cache more than we can fit in | |
623 | * the log. The limit really is that a checkpoint can't be more than half the | |
624 | * log (the current checkpoint is not allowed to overwrite the previous | |
625 | * checkpoint), but commit latency and memory usage limit this to a smaller | |
626 | * size. | |
627 | */ | |
628 | static void | |
629 | xlog_cil_push_background( | |
f7bdf03a | 630 | struct xlog *log) |
4c2d542f DC |
631 | { |
632 | struct xfs_cil *cil = log->l_cilp; | |
633 | ||
634 | /* | |
635 | * The cil won't be empty because we are called while holding the | |
636 | * context lock so whatever we added to the CIL will still be there | |
637 | */ | |
638 | ASSERT(!list_empty(&cil->xc_cil)); | |
639 | ||
640 | /* | |
641 | * don't do a background push if we haven't used up all the | |
642 | * space available yet. | |
643 | */ | |
644 | if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log)) | |
645 | return; | |
646 | ||
647 | spin_lock(&cil->xc_cil_lock); | |
648 | if (cil->xc_push_seq < cil->xc_current_sequence) { | |
649 | cil->xc_push_seq = cil->xc_current_sequence; | |
650 | queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work); | |
651 | } | |
652 | spin_unlock(&cil->xc_cil_lock); | |
653 | ||
654 | } | |
655 | ||
656 | static void | |
657 | xlog_cil_push_foreground( | |
f7bdf03a | 658 | struct xlog *log, |
4c2d542f DC |
659 | xfs_lsn_t push_seq) |
660 | { | |
661 | struct xfs_cil *cil = log->l_cilp; | |
662 | ||
663 | if (!cil) | |
664 | return; | |
665 | ||
666 | ASSERT(push_seq && push_seq <= cil->xc_current_sequence); | |
667 | ||
668 | /* start on any pending background push to minimise wait time on it */ | |
669 | flush_work(&cil->xc_push_work); | |
670 | ||
671 | /* | |
672 | * If the CIL is empty or we've already pushed the sequence then | |
673 | * there's no work we need to do. | |
674 | */ | |
675 | spin_lock(&cil->xc_cil_lock); | |
676 | if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) { | |
677 | spin_unlock(&cil->xc_cil_lock); | |
678 | return; | |
679 | } | |
680 | ||
681 | cil->xc_push_seq = push_seq; | |
682 | spin_unlock(&cil->xc_cil_lock); | |
683 | ||
684 | /* do the push now */ | |
685 | xlog_cil_push(log); | |
686 | } | |
687 | ||
a44f13ed DC |
688 | /* |
689 | * Commit a transaction with the given vector to the Committed Item List. | |
690 | * | |
691 | * To do this, we need to format the item, pin it in memory if required and | |
692 | * account for the space used by the transaction. Once we have done that we | |
693 | * need to release the unused reservation for the transaction, attach the | |
694 | * transaction to the checkpoint context so we carry the busy extents through | |
695 | * to checkpoint completion, and then unlock all the items in the transaction. | |
696 | * | |
a44f13ed DC |
697 | * Called with the context lock already held in read mode to lock out |
698 | * background commit, returns without it held once background commits are | |
699 | * allowed again. | |
700 | */ | |
0244b960 | 701 | int |
a44f13ed DC |
702 | xfs_log_commit_cil( |
703 | struct xfs_mount *mp, | |
704 | struct xfs_trans *tp, | |
a44f13ed DC |
705 | xfs_lsn_t *commit_lsn, |
706 | int flags) | |
707 | { | |
f7bdf03a | 708 | struct xlog *log = mp->m_log; |
a44f13ed | 709 | int log_flags = 0; |
0244b960 | 710 | struct xfs_log_vec *log_vector; |
a44f13ed DC |
711 | |
712 | if (flags & XFS_TRANS_RELEASE_LOG_RES) | |
713 | log_flags = XFS_LOG_REL_PERM_RESERV; | |
714 | ||
3b93c7aa | 715 | /* |
0244b960 | 716 | * Do all the hard work of formatting items (including memory |
3b93c7aa DC |
717 | * allocation) outside the CIL context lock. This prevents stalling CIL |
718 | * pushes when we are low on memory and a transaction commit spends a | |
719 | * lot of time in memory reclaim. | |
720 | */ | |
0244b960 CH |
721 | log_vector = xlog_cil_prepare_log_vecs(tp); |
722 | if (!log_vector) | |
723 | return ENOMEM; | |
3b93c7aa | 724 | |
a44f13ed DC |
725 | /* lock out background commit */ |
726 | down_read(&log->l_cilp->xc_ctx_lock); | |
d1583a38 DC |
727 | if (commit_lsn) |
728 | *commit_lsn = log->l_cilp->xc_ctx->sequence; | |
729 | ||
fd63875c | 730 | /* xlog_cil_insert_items() destroys log_vector list */ |
d1583a38 | 731 | xlog_cil_insert_items(log, log_vector, tp->t_ticket); |
a44f13ed DC |
732 | |
733 | /* check we didn't blow the reservation */ | |
734 | if (tp->t_ticket->t_curr_res < 0) | |
735 | xlog_print_tic_res(log->l_mp, tp->t_ticket); | |
736 | ||
737 | /* attach the transaction to the CIL if it has any busy extents */ | |
738 | if (!list_empty(&tp->t_busy)) { | |
739 | spin_lock(&log->l_cilp->xc_cil_lock); | |
740 | list_splice_init(&tp->t_busy, | |
741 | &log->l_cilp->xc_ctx->busy_extents); | |
742 | spin_unlock(&log->l_cilp->xc_cil_lock); | |
743 | } | |
744 | ||
745 | tp->t_commit_lsn = *commit_lsn; | |
746 | xfs_log_done(mp, tp->t_ticket, NULL, log_flags); | |
747 | xfs_trans_unreserve_and_mod_sb(tp); | |
748 | ||
749 | /* | |
750 | * Once all the items of the transaction have been copied to the CIL, | |
751 | * the items can be unlocked and freed. | |
752 | * | |
753 | * This needs to be done before we drop the CIL context lock because we | |
754 | * have to update state in the log items and unlock them before they go | |
755 | * to disk. If we don't, then the CIL checkpoint can race with us and | |
756 | * we can run checkpoint completion before we've updated and unlocked | |
757 | * the log items. This affects (at least) processing of stale buffers, | |
758 | * inodes and EFIs. | |
759 | */ | |
760 | xfs_trans_free_items(tp, *commit_lsn, 0); | |
761 | ||
4c2d542f | 762 | xlog_cil_push_background(log); |
a44f13ed DC |
763 | |
764 | up_read(&log->l_cilp->xc_ctx_lock); | |
0244b960 | 765 | return 0; |
a44f13ed DC |
766 | } |
767 | ||
71e330b5 DC |
768 | /* |
769 | * Conditionally push the CIL based on the sequence passed in. | |
770 | * | |
771 | * We only need to push if we haven't already pushed the sequence | |
772 | * number given. Hence the only time we will trigger a push here is | |
773 | * if the push sequence is the same as the current context. | |
774 | * | |
775 | * We return the current commit lsn to allow the callers to determine if a | |
776 | * iclog flush is necessary following this call. | |
71e330b5 DC |
777 | */ |
778 | xfs_lsn_t | |
a44f13ed | 779 | xlog_cil_force_lsn( |
f7bdf03a | 780 | struct xlog *log, |
a44f13ed | 781 | xfs_lsn_t sequence) |
71e330b5 DC |
782 | { |
783 | struct xfs_cil *cil = log->l_cilp; | |
784 | struct xfs_cil_ctx *ctx; | |
785 | xfs_lsn_t commit_lsn = NULLCOMMITLSN; | |
786 | ||
a44f13ed DC |
787 | ASSERT(sequence <= cil->xc_current_sequence); |
788 | ||
789 | /* | |
790 | * check to see if we need to force out the current context. | |
791 | * xlog_cil_push() handles racing pushes for the same sequence, | |
792 | * so no need to deal with it here. | |
793 | */ | |
4c2d542f | 794 | xlog_cil_push_foreground(log, sequence); |
71e330b5 DC |
795 | |
796 | /* | |
797 | * See if we can find a previous sequence still committing. | |
71e330b5 DC |
798 | * We need to wait for all previous sequence commits to complete |
799 | * before allowing the force of push_seq to go ahead. Hence block | |
800 | * on commits for those as well. | |
801 | */ | |
a44f13ed | 802 | restart: |
71e330b5 | 803 | spin_lock(&cil->xc_cil_lock); |
71e330b5 | 804 | list_for_each_entry(ctx, &cil->xc_committing, committing) { |
a44f13ed | 805 | if (ctx->sequence > sequence) |
71e330b5 DC |
806 | continue; |
807 | if (!ctx->commit_lsn) { | |
808 | /* | |
809 | * It is still being pushed! Wait for the push to | |
810 | * complete, then start again from the beginning. | |
811 | */ | |
eb40a875 | 812 | xlog_wait(&cil->xc_commit_wait, &cil->xc_cil_lock); |
71e330b5 DC |
813 | goto restart; |
814 | } | |
a44f13ed | 815 | if (ctx->sequence != sequence) |
71e330b5 DC |
816 | continue; |
817 | /* found it! */ | |
818 | commit_lsn = ctx->commit_lsn; | |
819 | } | |
820 | spin_unlock(&cil->xc_cil_lock); | |
821 | return commit_lsn; | |
822 | } | |
ccf7c23f DC |
823 | |
824 | /* | |
825 | * Check if the current log item was first committed in this sequence. | |
826 | * We can't rely on just the log item being in the CIL, we have to check | |
827 | * the recorded commit sequence number. | |
828 | * | |
829 | * Note: for this to be used in a non-racy manner, it has to be called with | |
830 | * CIL flushing locked out. As a result, it should only be used during the | |
831 | * transaction commit process when deciding what to format into the item. | |
832 | */ | |
833 | bool | |
834 | xfs_log_item_in_current_chkpt( | |
835 | struct xfs_log_item *lip) | |
836 | { | |
837 | struct xfs_cil_ctx *ctx; | |
838 | ||
ccf7c23f DC |
839 | if (list_empty(&lip->li_cil)) |
840 | return false; | |
841 | ||
842 | ctx = lip->li_mountp->m_log->l_cilp->xc_ctx; | |
843 | ||
844 | /* | |
845 | * li_seq is written on the first commit of a log item to record the | |
846 | * first checkpoint it is written to. Hence if it is different to the | |
847 | * current sequence, we're in a new checkpoint. | |
848 | */ | |
849 | if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0) | |
850 | return false; | |
851 | return true; | |
852 | } | |
4c2d542f DC |
853 | |
854 | /* | |
855 | * Perform initial CIL structure initialisation. | |
856 | */ | |
857 | int | |
858 | xlog_cil_init( | |
f7bdf03a | 859 | struct xlog *log) |
4c2d542f DC |
860 | { |
861 | struct xfs_cil *cil; | |
862 | struct xfs_cil_ctx *ctx; | |
863 | ||
864 | cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL); | |
865 | if (!cil) | |
866 | return ENOMEM; | |
867 | ||
868 | ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL); | |
869 | if (!ctx) { | |
870 | kmem_free(cil); | |
871 | return ENOMEM; | |
872 | } | |
873 | ||
874 | INIT_WORK(&cil->xc_push_work, xlog_cil_push_work); | |
875 | INIT_LIST_HEAD(&cil->xc_cil); | |
876 | INIT_LIST_HEAD(&cil->xc_committing); | |
877 | spin_lock_init(&cil->xc_cil_lock); | |
878 | init_rwsem(&cil->xc_ctx_lock); | |
879 | init_waitqueue_head(&cil->xc_commit_wait); | |
880 | ||
881 | INIT_LIST_HEAD(&ctx->committing); | |
882 | INIT_LIST_HEAD(&ctx->busy_extents); | |
883 | ctx->sequence = 1; | |
884 | ctx->cil = cil; | |
885 | cil->xc_ctx = ctx; | |
886 | cil->xc_current_sequence = ctx->sequence; | |
887 | ||
888 | cil->xc_log = log; | |
889 | log->l_cilp = cil; | |
890 | return 0; | |
891 | } | |
892 | ||
893 | void | |
894 | xlog_cil_destroy( | |
f7bdf03a | 895 | struct xlog *log) |
4c2d542f DC |
896 | { |
897 | if (log->l_cilp->xc_ctx) { | |
898 | if (log->l_cilp->xc_ctx->ticket) | |
899 | xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket); | |
900 | kmem_free(log->l_cilp->xc_ctx); | |
901 | } | |
902 | ||
903 | ASSERT(list_empty(&log->l_cilp->xc_cil)); | |
904 | kmem_free(log->l_cilp); | |
905 | } | |
906 |