blob: b3425b34e3d5782d95252c71acadfffdbd24ad4c [file] [log] [blame]
Dave Chinner71e330b2010-05-21 14:37:18 +10001/*
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"
Dave Chinner239880e2013-10-23 10:50:10 +110020#include "xfs_log_format.h"
Dave Chinner70a98832013-10-23 10:36:05 +110021#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110022#include "xfs_trans_resv.h"
Dave Chinner71e330b2010-05-21 14:37:18 +100023#include "xfs_sb.h"
24#include "xfs_ag.h"
Dave Chinner71e330b2010-05-21 14:37:18 +100025#include "xfs_mount.h"
26#include "xfs_error.h"
27#include "xfs_alloc.h"
Dave Chinnerefc27b52012-04-29 10:39:43 +000028#include "xfs_extent_busy.h"
Christoph Hellwige84661a2011-05-20 13:45:32 +000029#include "xfs_discard.h"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_trans.h"
31#include "xfs_trans_priv.h"
32#include "xfs_log.h"
33#include "xfs_log_priv.h"
Dave Chinner71e330b2010-05-21 14:37:18 +100034
35/*
Dave Chinner71e330b2010-05-21 14:37:18 +100036 * 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 */
47static struct xlog_ticket *
48xlog_cil_ticket_alloc(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -050049 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +100050{
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 */
75void
76xlog_cil_init_post_recovery(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -050077 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +100078{
Dave Chinner71e330b2010-05-21 14:37:18 +100079 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
85/*
Dave Chinner991aaf62013-08-12 20:50:07 +100086 * Prepare the log item for insertion into the CIL. Calculate the difference in
87 * log space and vectors it will consume, and if it is a new item pin it as
88 * well.
89 */
90STATIC void
91xfs_cil_prepare_item(
92 struct xlog *log,
93 struct xfs_log_vec *lv,
94 struct xfs_log_vec *old_lv,
95 int *diff_len,
96 int *diff_iovecs)
97{
98 /* Account for the new LV being passed in */
99 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
Dave Chinner110dc242014-05-20 08:18:09 +1000100 *diff_len += lv->lv_bytes;
Dave Chinner991aaf62013-08-12 20:50:07 +1000101 *diff_iovecs += lv->lv_niovecs;
102 }
103
104 /*
105 * If there is no old LV, this is the first time we've seen the item in
106 * this CIL context and so we need to pin it. If we are replacing the
107 * old_lv, then remove the space it accounts for and free it.
108 */
109 if (!old_lv)
110 lv->lv_item->li_ops->iop_pin(lv->lv_item);
111 else if (old_lv != lv) {
112 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
113
Dave Chinner110dc242014-05-20 08:18:09 +1000114 *diff_len -= old_lv->lv_bytes;
Dave Chinner991aaf62013-08-12 20:50:07 +1000115 *diff_iovecs -= old_lv->lv_niovecs;
116 kmem_free(old_lv);
117 }
118
119 /* attach new log vector to log item */
120 lv->lv_item->li_lv = lv;
121
122 /*
123 * If this is the first time the item is being committed to the
124 * CIL, store the sequence number on the log item so we can
125 * tell in future commits whether this is the first checkpoint
126 * the item is being committed into.
127 */
128 if (!lv->lv_item->li_seq)
129 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
130}
131
132/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000133 * Format log item into a flat buffers
134 *
135 * For delayed logging, we need to hold a formatted buffer containing all the
136 * changes on the log item. This enables us to relog the item in memory and
137 * write it out asynchronously without needing to relock the object that was
138 * modified at the time it gets written into the iclog.
139 *
140 * This function builds a vector for the changes in each log item in the
141 * transaction. It then works out the length of the buffer needed for each log
142 * item, allocates them and formats the vector for the item into the buffer.
143 * The buffer is then attached to the log item are then inserted into the
144 * Committed Item List for tracking until the next checkpoint is written out.
145 *
146 * We don't set up region headers during this process; we simply copy the
147 * regions into the flat buffer. We can do this because we still have to do a
148 * formatting step to write the regions into the iclog buffer. Writing the
149 * ophdrs during the iclog write means that we can support splitting large
150 * regions across iclog boundares without needing a change in the format of the
151 * item/region encapsulation.
152 *
153 * Hence what we need to do now is change the rewrite the vector array to point
154 * to the copied region inside the buffer we just allocated. This allows us to
155 * format the regions into the iclog as though they are being formatted
156 * directly out of the objects themselves.
157 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000158static void
159xlog_cil_insert_format_items(
160 struct xlog *log,
161 struct xfs_trans *tp,
162 int *diff_len,
163 int *diff_iovecs)
Dave Chinner71e330b2010-05-21 14:37:18 +1000164{
Christoph Hellwig0244b962011-12-06 21:58:08 +0000165 struct xfs_log_item_desc *lidp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000166
Christoph Hellwig0244b962011-12-06 21:58:08 +0000167
168 /* Bail out if we didn't find a log item. */
169 if (list_empty(&tp->t_items)) {
170 ASSERT(0);
Dave Chinner991aaf62013-08-12 20:50:07 +1000171 return;
Christoph Hellwig0244b962011-12-06 21:58:08 +0000172 }
173
174 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner166d1362013-08-12 20:50:04 +1000175 struct xfs_log_item *lip = lidp->lid_item;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000176 struct xfs_log_vec *lv;
Dave Chinner991aaf62013-08-12 20:50:07 +1000177 struct xfs_log_vec *old_lv;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000178 int niovecs = 0;
179 int nbytes = 0;
180 int buf_size;
Dave Chinnerfd638752013-06-27 16:04:51 +1000181 bool ordered = false;
Dave Chinner71e330b2010-05-21 14:37:18 +1000182
Christoph Hellwig0244b962011-12-06 21:58:08 +0000183 /* Skip items which aren't dirty in this transaction. */
184 if (!(lidp->lid_flags & XFS_LID_DIRTY))
185 continue;
186
Dave Chinner166d1362013-08-12 20:50:04 +1000187 /* get number of vecs and size of data to be stored */
188 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
189
Christoph Hellwig0244b962011-12-06 21:58:08 +0000190 /* Skip items that do not have any vectors for writing */
Christoph Hellwigb3934212011-12-06 21:58:09 +0000191 if (!niovecs)
Christoph Hellwig0244b962011-12-06 21:58:08 +0000192 continue;
193
Dave Chinnerfd638752013-06-27 16:04:51 +1000194 /*
195 * Ordered items need to be tracked but we do not wish to write
196 * them. We need a logvec to track the object, but we do not
197 * need an iovec or buffer to be allocated for copying data.
198 */
199 if (niovecs == XFS_LOG_VEC_ORDERED) {
200 ordered = true;
201 niovecs = 0;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000202 nbytes = 0;
Dave Chinnerfd638752013-06-27 16:04:51 +1000203 }
204
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100205 /*
206 * We 64-bit align the length of each iovec so that the start
207 * of the next one is naturally aligned. We'll need to
Dave Chinner3895e512014-02-10 10:37:18 +1100208 * account for that slack space here. Then round nbytes up
209 * to 64-bit alignment so that the initial buffer alignment is
210 * easy to calculate and verify.
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100211 */
212 nbytes += niovecs * sizeof(uint64_t);
Dave Chinner3895e512014-02-10 10:37:18 +1100213 nbytes = round_up(nbytes, sizeof(uint64_t));
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100214
Dave Chinner991aaf62013-08-12 20:50:07 +1000215 /* grab the old item if it exists for reservation accounting */
216 old_lv = lip->li_lv;
217
Dave Chinner3895e512014-02-10 10:37:18 +1100218 /*
219 * The data buffer needs to start 64-bit aligned, so round up
220 * that space to ensure we can align it appropriately and not
221 * overrun the buffer.
222 */
223 buf_size = nbytes +
224 round_up((sizeof(struct xfs_log_vec) +
225 niovecs * sizeof(struct xfs_log_iovec)),
226 sizeof(uint64_t));
Christoph Hellwig0244b962011-12-06 21:58:08 +0000227
Dave Chinnerf5baac32013-08-12 20:50:06 +1000228 /* compare to existing item size */
229 if (lip->li_lv && buf_size <= lip->li_lv->lv_size) {
230 /* same or smaller, optimise common overwrite case */
231 lv = lip->li_lv;
232 lv->lv_next = NULL;
233
234 if (ordered)
235 goto insert;
236
Dave Chinner991aaf62013-08-12 20:50:07 +1000237 /*
238 * set the item up as though it is a new insertion so
239 * that the space reservation accounting is correct.
240 */
241 *diff_iovecs -= lv->lv_niovecs;
Dave Chinner110dc242014-05-20 08:18:09 +1000242 *diff_len -= lv->lv_bytes;
Christoph Hellwig9597df62013-12-13 11:00:42 +1100243 } else {
244 /* allocate new data chunk */
245 lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS);
246 lv->lv_item = lip;
247 lv->lv_size = buf_size;
248 if (ordered) {
249 /* track as an ordered logvec */
250 ASSERT(lip->li_lv == NULL);
251 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
252 goto insert;
253 }
254 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
Dave Chinnerf5baac32013-08-12 20:50:06 +1000255 }
256
Christoph Hellwig9597df62013-12-13 11:00:42 +1100257 /* Ensure the lv is set up according to ->iop_size */
Dave Chinner7492c5b2013-08-12 20:50:05 +1000258 lv->lv_niovecs = niovecs;
Christoph Hellwig0244b962011-12-06 21:58:08 +0000259
Dave Chinner7492c5b2013-08-12 20:50:05 +1000260 /* The allocated data region lies beyond the iovec region */
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100261 lv->lv_buf_len = 0;
Dave Chinner110dc242014-05-20 08:18:09 +1000262 lv->lv_bytes = 0;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000263 lv->lv_buf = (char *)lv + buf_size - nbytes;
Dave Chinner3895e512014-02-10 10:37:18 +1100264 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
265
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100266 lip->li_ops->iop_format(lip, lv);
Dave Chinner7492c5b2013-08-12 20:50:05 +1000267insert:
Dave Chinnerf5baac32013-08-12 20:50:06 +1000268 ASSERT(lv->lv_buf_len <= nbytes);
Dave Chinner991aaf62013-08-12 20:50:07 +1000269 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
Dave Chinner71e330b2010-05-21 14:37:18 +1000270 }
Dave Chinnerd1583a32010-09-24 18:14:13 +1000271}
272
273/*
274 * Insert the log items into the CIL and calculate the difference in space
275 * consumed by the item. Add the space to the checkpoint ticket and calculate
276 * if the change requires additional log metadata. If it does, take that space
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800277 * as well. Remove the amount of space we added to the checkpoint ticket from
Dave Chinnerd1583a32010-09-24 18:14:13 +1000278 * the current transaction ticket so that the accounting works out correctly.
279 */
Dave Chinner71e330b2010-05-21 14:37:18 +1000280static void
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000281xlog_cil_insert_items(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500282 struct xlog *log,
Dave Chinner991aaf62013-08-12 20:50:07 +1000283 struct xfs_trans *tp)
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000284{
Dave Chinnerd1583a32010-09-24 18:14:13 +1000285 struct xfs_cil *cil = log->l_cilp;
286 struct xfs_cil_ctx *ctx = cil->xc_ctx;
Dave Chinner991aaf62013-08-12 20:50:07 +1000287 struct xfs_log_item_desc *lidp;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000288 int len = 0;
289 int diff_iovecs = 0;
290 int iclog_space;
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000291
Dave Chinner991aaf62013-08-12 20:50:07 +1000292 ASSERT(tp);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000293
294 /*
Dave Chinnerd1583a32010-09-24 18:14:13 +1000295 * We can do this safely because the context can't checkpoint until we
296 * are done so it doesn't matter exactly how we update the CIL.
297 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000298 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
299
300 /*
301 * Now (re-)position everything modified at the tail of the CIL.
302 * We do this here so we only need to take the CIL lock once during
303 * the transaction commit.
304 */
Dave Chinnerfd638752013-06-27 16:04:51 +1000305 spin_lock(&cil->xc_cil_lock);
Dave Chinner991aaf62013-08-12 20:50:07 +1000306 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
307 struct xfs_log_item *lip = lidp->lid_item;
Dave Chinnerfd638752013-06-27 16:04:51 +1000308
Dave Chinner991aaf62013-08-12 20:50:07 +1000309 /* Skip items which aren't dirty in this transaction. */
310 if (!(lidp->lid_flags & XFS_LID_DIRTY))
311 continue;
Dave Chinnerfd638752013-06-27 16:04:51 +1000312
Dave Chinner991aaf62013-08-12 20:50:07 +1000313 list_move_tail(&lip->li_cil, &cil->xc_cil);
Dave Chinnerfd638752013-06-27 16:04:51 +1000314 }
Dave Chinnerd1583a32010-09-24 18:14:13 +1000315
316 /* account for space used by new iovec headers */
317 len += diff_iovecs * sizeof(xlog_op_header_t);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000318 ctx->nvecs += diff_iovecs;
319
Dave Chinner991aaf62013-08-12 20:50:07 +1000320 /* attach the transaction to the CIL if it has any busy extents */
321 if (!list_empty(&tp->t_busy))
322 list_splice_init(&tp->t_busy, &ctx->busy_extents);
323
Dave Chinnerd1583a32010-09-24 18:14:13 +1000324 /*
325 * Now transfer enough transaction reservation to the context ticket
326 * for the checkpoint. The context ticket is special - the unit
327 * reservation has to grow as well as the current reservation as we
328 * steal from tickets so we can correctly determine the space used
329 * during the transaction commit.
330 */
331 if (ctx->ticket->t_curr_res == 0) {
Dave Chinnerd1583a32010-09-24 18:14:13 +1000332 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
Dave Chinner991aaf62013-08-12 20:50:07 +1000333 tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000334 }
335
336 /* do we need space for more log record headers? */
337 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
338 if (len > 0 && (ctx->space_used / iclog_space !=
339 (ctx->space_used + len) / iclog_space)) {
340 int hdrs;
341
342 hdrs = (len + iclog_space - 1) / iclog_space;
343 /* need to take into account split region headers, too */
344 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
345 ctx->ticket->t_unit_res += hdrs;
346 ctx->ticket->t_curr_res += hdrs;
Dave Chinner991aaf62013-08-12 20:50:07 +1000347 tp->t_ticket->t_curr_res -= hdrs;
348 ASSERT(tp->t_ticket->t_curr_res >= len);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000349 }
Dave Chinner991aaf62013-08-12 20:50:07 +1000350 tp->t_ticket->t_curr_res -= len;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000351 ctx->space_used += len;
352
353 spin_unlock(&cil->xc_cil_lock);
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000354}
355
356static void
Dave Chinner71e330b2010-05-21 14:37:18 +1000357xlog_cil_free_logvec(
358 struct xfs_log_vec *log_vector)
359{
360 struct xfs_log_vec *lv;
361
362 for (lv = log_vector; lv; ) {
363 struct xfs_log_vec *next = lv->lv_next;
Dave Chinner71e330b2010-05-21 14:37:18 +1000364 kmem_free(lv);
365 lv = next;
366 }
367}
368
369/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000370 * Mark all items committed and clear busy extents. We free the log vector
371 * chains in a separate pass so that we unpin the log items as quickly as
372 * possible.
373 */
374static void
375xlog_cil_committed(
376 void *args,
377 int abort)
378{
379 struct xfs_cil_ctx *ctx = args;
Christoph Hellwige84661a2011-05-20 13:45:32 +0000380 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000381
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100382 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
383 ctx->start_lsn, abort);
Dave Chinner71e330b2010-05-21 14:37:18 +1000384
Dave Chinner4ecbfe62012-04-29 10:41:10 +0000385 xfs_extent_busy_sort(&ctx->busy_extents);
386 xfs_extent_busy_clear(mp, &ctx->busy_extents,
Christoph Hellwige84661a2011-05-20 13:45:32 +0000387 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
Dave Chinner71e330b2010-05-21 14:37:18 +1000388
Dave Chinnerac983512014-05-07 08:05:50 +1000389 /*
390 * If we are aborting the commit, wake up anyone waiting on the
391 * committing list. If we don't, then a shutdown we can leave processes
392 * waiting in xlog_cil_force_lsn() waiting on a sequence commit that
393 * will never happen because we aborted it.
394 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000395 spin_lock(&ctx->cil->xc_push_lock);
Dave Chinnerac983512014-05-07 08:05:50 +1000396 if (abort)
397 wake_up_all(&ctx->cil->xc_commit_wait);
Dave Chinner71e330b2010-05-21 14:37:18 +1000398 list_del(&ctx->committing);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000399 spin_unlock(&ctx->cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000400
401 xlog_cil_free_logvec(ctx->lv_chain);
Christoph Hellwige84661a2011-05-20 13:45:32 +0000402
403 if (!list_empty(&ctx->busy_extents)) {
404 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
405
406 xfs_discard_extents(mp, &ctx->busy_extents);
Dave Chinner4ecbfe62012-04-29 10:41:10 +0000407 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
Christoph Hellwige84661a2011-05-20 13:45:32 +0000408 }
409
Dave Chinner71e330b2010-05-21 14:37:18 +1000410 kmem_free(ctx);
411}
412
413/*
Dave Chinnera44f13e2010-08-24 11:40:03 +1000414 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
415 * is a background flush and so we can chose to ignore it. Otherwise, if the
416 * current sequence is the same as @push_seq we need to do a flush. If
417 * @push_seq is less than the current sequence, then it has already been
418 * flushed and we don't need to do anything - the caller will wait for it to
419 * complete if necessary.
420 *
421 * @push_seq is a value rather than a flag because that allows us to do an
422 * unlocked check of the sequence number for a match. Hence we can allows log
423 * forces to run racily and not issue pushes for the same sequence twice. If we
424 * get a race between multiple pushes for the same sequence they will block on
425 * the first one and then abort, hence avoiding needless pushes.
Dave Chinner71e330b2010-05-21 14:37:18 +1000426 */
Dave Chinnera44f13e2010-08-24 11:40:03 +1000427STATIC int
Dave Chinner71e330b2010-05-21 14:37:18 +1000428xlog_cil_push(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500429 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +1000430{
431 struct xfs_cil *cil = log->l_cilp;
432 struct xfs_log_vec *lv;
433 struct xfs_cil_ctx *ctx;
434 struct xfs_cil_ctx *new_ctx;
435 struct xlog_in_core *commit_iclog;
436 struct xlog_ticket *tic;
Dave Chinner71e330b2010-05-21 14:37:18 +1000437 int num_iovecs;
Dave Chinner71e330b2010-05-21 14:37:18 +1000438 int error = 0;
439 struct xfs_trans_header thdr;
440 struct xfs_log_iovec lhdr;
441 struct xfs_log_vec lvhdr = { NULL };
442 xfs_lsn_t commit_lsn;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000443 xfs_lsn_t push_seq;
Dave Chinner71e330b2010-05-21 14:37:18 +1000444
445 if (!cil)
446 return 0;
447
Dave Chinner71e330b2010-05-21 14:37:18 +1000448 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
449 new_ctx->ticket = xlog_cil_ticket_alloc(log);
450
Dave Chinner4c2d5422012-04-23 17:54:32 +1000451 down_write(&cil->xc_ctx_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000452 ctx = cil->xc_ctx;
453
Dave Chinner4bb928c2013-08-12 20:50:08 +1000454 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000455 push_seq = cil->xc_push_seq;
456 ASSERT(push_seq <= ctx->sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000457
Dave Chinner4c2d5422012-04-23 17:54:32 +1000458 /*
459 * Check if we've anything to push. If there is nothing, then we don't
460 * move on to a new sequence number and so we have to be able to push
461 * this sequence again later.
462 */
463 if (list_empty(&cil->xc_cil)) {
464 cil->xc_push_seq = 0;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000465 spin_unlock(&cil->xc_push_lock);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000466 goto out_skip;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000467 }
Dave Chinner4bb928c2013-08-12 20:50:08 +1000468 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000469
Dave Chinnera44f13e2010-08-24 11:40:03 +1000470
471 /* check for a previously pushed seqeunce */
Dave Chinner4c2d5422012-04-23 17:54:32 +1000472 if (push_seq < cil->xc_ctx->sequence)
Dave Chinnerdf806152010-05-17 15:52:13 +1000473 goto out_skip;
474
Dave Chinner71e330b2010-05-21 14:37:18 +1000475 /*
476 * pull all the log vectors off the items in the CIL, and
477 * remove the items from the CIL. We don't need the CIL lock
478 * here because it's only needed on the transaction commit
479 * side which is currently locked out by the flush lock.
480 */
481 lv = NULL;
Dave Chinner71e330b2010-05-21 14:37:18 +1000482 num_iovecs = 0;
Dave Chinner71e330b2010-05-21 14:37:18 +1000483 while (!list_empty(&cil->xc_cil)) {
484 struct xfs_log_item *item;
Dave Chinner71e330b2010-05-21 14:37:18 +1000485
486 item = list_first_entry(&cil->xc_cil,
487 struct xfs_log_item, li_cil);
488 list_del_init(&item->li_cil);
489 if (!ctx->lv_chain)
490 ctx->lv_chain = item->li_lv;
491 else
492 lv->lv_next = item->li_lv;
493 lv = item->li_lv;
494 item->li_lv = NULL;
Dave Chinner71e330b2010-05-21 14:37:18 +1000495 num_iovecs += lv->lv_niovecs;
Dave Chinner71e330b2010-05-21 14:37:18 +1000496 }
497
498 /*
499 * initialise the new context and attach it to the CIL. Then attach
500 * the current context to the CIL committing lsit so it can be found
501 * during log forces to extract the commit lsn of the sequence that
502 * needs to be forced.
503 */
504 INIT_LIST_HEAD(&new_ctx->committing);
505 INIT_LIST_HEAD(&new_ctx->busy_extents);
506 new_ctx->sequence = ctx->sequence + 1;
507 new_ctx->cil = cil;
508 cil->xc_ctx = new_ctx;
509
510 /*
511 * The switch is now done, so we can drop the context lock and move out
512 * of a shared context. We can't just go straight to the commit record,
513 * though - we need to synchronise with previous and future commits so
514 * that the commit records are correctly ordered in the log to ensure
515 * that we process items during log IO completion in the correct order.
516 *
517 * For example, if we get an EFI in one checkpoint and the EFD in the
518 * next (e.g. due to log forces), we do not want the checkpoint with
519 * the EFD to be committed before the checkpoint with the EFI. Hence
520 * we must strictly order the commit records of the checkpoints so
521 * that: a) the checkpoint callbacks are attached to the iclogs in the
522 * correct order; and b) the checkpoints are replayed in correct order
523 * in log recovery.
524 *
525 * Hence we need to add this context to the committing context list so
526 * that higher sequences will wait for us to write out a commit record
527 * before they do.
Dave Chinnerf876e442014-02-27 16:40:42 +1100528 *
529 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
530 * structure atomically with the addition of this sequence to the
531 * committing list. This also ensures that we can do unlocked checks
532 * against the current sequence in log forces without risking
533 * deferencing a freed context pointer.
Dave Chinner71e330b2010-05-21 14:37:18 +1000534 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000535 spin_lock(&cil->xc_push_lock);
Dave Chinnerf876e442014-02-27 16:40:42 +1100536 cil->xc_current_sequence = new_ctx->sequence;
Dave Chinner71e330b2010-05-21 14:37:18 +1000537 list_add(&ctx->committing, &cil->xc_committing);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000538 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000539 up_write(&cil->xc_ctx_lock);
540
541 /*
542 * Build a checkpoint transaction header and write it to the log to
543 * begin the transaction. We need to account for the space used by the
544 * transaction header here as it is not accounted for in xlog_write().
545 *
546 * The LSN we need to pass to the log items on transaction commit is
547 * the LSN reported by the first log vector write. If we use the commit
548 * record lsn then we can move the tail beyond the grant write head.
549 */
550 tic = ctx->ticket;
551 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
552 thdr.th_type = XFS_TRANS_CHECKPOINT;
553 thdr.th_tid = tic->t_tid;
554 thdr.th_num_items = num_iovecs;
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000555 lhdr.i_addr = &thdr;
Dave Chinner71e330b2010-05-21 14:37:18 +1000556 lhdr.i_len = sizeof(xfs_trans_header_t);
557 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
558 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
559
560 lvhdr.lv_niovecs = 1;
561 lvhdr.lv_iovecp = &lhdr;
562 lvhdr.lv_next = ctx->lv_chain;
563
564 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
565 if (error)
Dave Chinner7db37c52011-01-27 12:02:00 +1100566 goto out_abort_free_ticket;
Dave Chinner71e330b2010-05-21 14:37:18 +1000567
568 /*
569 * now that we've written the checkpoint into the log, strictly
570 * order the commit records so replay will get them in the right order.
571 */
572restart:
Dave Chinner4bb928c2013-08-12 20:50:08 +1000573 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000574 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
575 /*
Dave Chinnerac983512014-05-07 08:05:50 +1000576 * Avoid getting stuck in this loop because we were woken by the
577 * shutdown, but then went back to sleep once already in the
578 * shutdown state.
579 */
580 if (XLOG_FORCED_SHUTDOWN(log)) {
581 spin_unlock(&cil->xc_push_lock);
582 goto out_abort_free_ticket;
583 }
584
585 /*
Dave Chinner71e330b2010-05-21 14:37:18 +1000586 * Higher sequences will wait for this one so skip them.
Dave Chinnerac983512014-05-07 08:05:50 +1000587 * Don't wait for our own sequence, either.
Dave Chinner71e330b2010-05-21 14:37:18 +1000588 */
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 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000596 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000597 goto restart;
598 }
599 }
Dave Chinner4bb928c2013-08-12 20:50:08 +1000600 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000601
Dave Chinner7db37c52011-01-27 12:02:00 +1100602 /* xfs_log_done always frees the ticket on error. */
Dave Chinner71e330b2010-05-21 14:37:18 +1000603 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
Dave Chinner7db37c52011-01-27 12:02:00 +1100604 if (commit_lsn == -1)
Dave Chinner71e330b2010-05-21 14:37:18 +1000605 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 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000619 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000620 ctx->commit_lsn = commit_lsn;
Dave Chinnereb40a872010-12-21 12:09:01 +1100621 wake_up_all(&cil->xc_commit_wait);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000622 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000623
624 /* release the hounds! */
625 return xfs_log_release_iclog(log->l_mp, commit_iclog);
626
627out_skip:
628 up_write(&cil->xc_ctx_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000629 xfs_log_ticket_put(new_ctx->ticket);
630 kmem_free(new_ctx);
631 return 0;
632
Dave Chinner7db37c52011-01-27 12:02:00 +1100633out_abort_free_ticket:
634 xfs_log_ticket_put(tic);
Dave Chinner71e330b2010-05-21 14:37:18 +1000635out_abort:
636 xlog_cil_committed(ctx, XFS_LI_ABORTED);
637 return XFS_ERROR(EIO);
638}
639
Dave Chinner4c2d5422012-04-23 17:54:32 +1000640static void
641xlog_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 */
656static void
657xlog_cil_push_background(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500658 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000659{
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
Dave Chinner4bb928c2013-08-12 20:50:08 +1000675 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000676 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 }
Dave Chinner4bb928c2013-08-12 20:50:08 +1000680 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000681
682}
683
Dave Chinnerf876e442014-02-27 16:40:42 +1100684/*
685 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
686 * number that is passed. When it returns, the work will be queued for
687 * @push_seq, but it won't be completed. The caller is expected to do any
688 * waiting for push_seq to complete if it is required.
689 */
Dave Chinner4c2d5422012-04-23 17:54:32 +1000690static void
Dave Chinnerf876e442014-02-27 16:40:42 +1100691xlog_cil_push_now(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500692 struct xlog *log,
Dave Chinner4c2d5422012-04-23 17:54:32 +1000693 xfs_lsn_t push_seq)
694{
695 struct xfs_cil *cil = log->l_cilp;
696
697 if (!cil)
698 return;
699
700 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
701
702 /* start on any pending background push to minimise wait time on it */
703 flush_work(&cil->xc_push_work);
704
705 /*
706 * If the CIL is empty or we've already pushed the sequence then
707 * there's no work we need to do.
708 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000709 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000710 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
Dave Chinner4bb928c2013-08-12 20:50:08 +1000711 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000712 return;
713 }
714
715 cil->xc_push_seq = push_seq;
Dave Chinnerf876e442014-02-27 16:40:42 +1100716 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000717 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000718}
719
Dave Chinner2c6e24c2013-10-15 09:17:49 +1100720bool
721xlog_cil_empty(
722 struct xlog *log)
723{
724 struct xfs_cil *cil = log->l_cilp;
725 bool empty = false;
726
727 spin_lock(&cil->xc_push_lock);
728 if (list_empty(&cil->xc_cil))
729 empty = true;
730 spin_unlock(&cil->xc_push_lock);
731 return empty;
732}
733
Dave Chinner71e330b2010-05-21 14:37:18 +1000734/*
Dave Chinnera44f13e2010-08-24 11:40:03 +1000735 * Commit a transaction with the given vector to the Committed Item List.
736 *
737 * To do this, we need to format the item, pin it in memory if required and
738 * account for the space used by the transaction. Once we have done that we
739 * need to release the unused reservation for the transaction, attach the
740 * transaction to the checkpoint context so we carry the busy extents through
741 * to checkpoint completion, and then unlock all the items in the transaction.
742 *
Dave Chinnera44f13e2010-08-24 11:40:03 +1000743 * Called with the context lock already held in read mode to lock out
744 * background commit, returns without it held once background commits are
745 * allowed again.
746 */
Jie Liuc6f97262014-02-07 15:26:07 +1100747void
Dave Chinnera44f13e2010-08-24 11:40:03 +1000748xfs_log_commit_cil(
749 struct xfs_mount *mp,
750 struct xfs_trans *tp,
Dave Chinnera44f13e2010-08-24 11:40:03 +1000751 xfs_lsn_t *commit_lsn,
752 int flags)
753{
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500754 struct xlog *log = mp->m_log;
Dave Chinner991aaf62013-08-12 20:50:07 +1000755 struct xfs_cil *cil = log->l_cilp;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000756 int log_flags = 0;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000757
758 if (flags & XFS_TRANS_RELEASE_LOG_RES)
759 log_flags = XFS_LOG_REL_PERM_RESERV;
760
Dave Chinnerf5baac32013-08-12 20:50:06 +1000761 /* lock out background commit */
Dave Chinner991aaf62013-08-12 20:50:07 +1000762 down_read(&cil->xc_ctx_lock);
Dave Chinnerf5baac32013-08-12 20:50:06 +1000763
Dave Chinner991aaf62013-08-12 20:50:07 +1000764 xlog_cil_insert_items(log, tp);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000765
766 /* check we didn't blow the reservation */
767 if (tp->t_ticket->t_curr_res < 0)
Dave Chinner991aaf62013-08-12 20:50:07 +1000768 xlog_print_tic_res(mp, tp->t_ticket);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000769
Dave Chinner991aaf62013-08-12 20:50:07 +1000770 tp->t_commit_lsn = cil->xc_ctx->sequence;
771 if (commit_lsn)
772 *commit_lsn = tp->t_commit_lsn;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000773
Dave Chinnera44f13e2010-08-24 11:40:03 +1000774 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
775 xfs_trans_unreserve_and_mod_sb(tp);
776
777 /*
778 * Once all the items of the transaction have been copied to the CIL,
779 * the items can be unlocked and freed.
780 *
781 * This needs to be done before we drop the CIL context lock because we
782 * have to update state in the log items and unlock them before they go
783 * to disk. If we don't, then the CIL checkpoint can race with us and
784 * we can run checkpoint completion before we've updated and unlocked
785 * the log items. This affects (at least) processing of stale buffers,
786 * inodes and EFIs.
787 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000788 xfs_trans_free_items(tp, tp->t_commit_lsn, 0);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000789
Dave Chinner4c2d5422012-04-23 17:54:32 +1000790 xlog_cil_push_background(log);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000791
Dave Chinner991aaf62013-08-12 20:50:07 +1000792 up_read(&cil->xc_ctx_lock);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000793}
794
795/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000796 * Conditionally push the CIL based on the sequence passed in.
797 *
798 * We only need to push if we haven't already pushed the sequence
799 * number given. Hence the only time we will trigger a push here is
800 * if the push sequence is the same as the current context.
801 *
802 * We return the current commit lsn to allow the callers to determine if a
803 * iclog flush is necessary following this call.
Dave Chinner71e330b2010-05-21 14:37:18 +1000804 */
805xfs_lsn_t
Dave Chinnera44f13e2010-08-24 11:40:03 +1000806xlog_cil_force_lsn(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500807 struct xlog *log,
Dave Chinnera44f13e2010-08-24 11:40:03 +1000808 xfs_lsn_t sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000809{
810 struct xfs_cil *cil = log->l_cilp;
811 struct xfs_cil_ctx *ctx;
812 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
813
Dave Chinnera44f13e2010-08-24 11:40:03 +1000814 ASSERT(sequence <= cil->xc_current_sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000815
Dave Chinnera44f13e2010-08-24 11:40:03 +1000816 /*
817 * check to see if we need to force out the current context.
818 * xlog_cil_push() handles racing pushes for the same sequence,
819 * so no need to deal with it here.
820 */
Dave Chinnerf876e442014-02-27 16:40:42 +1100821restart:
822 xlog_cil_push_now(log, sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000823
824 /*
825 * See if we can find a previous sequence still committing.
Dave Chinner71e330b2010-05-21 14:37:18 +1000826 * We need to wait for all previous sequence commits to complete
827 * before allowing the force of push_seq to go ahead. Hence block
828 * on commits for those as well.
829 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000830 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000831 list_for_each_entry(ctx, &cil->xc_committing, committing) {
Dave Chinnerac983512014-05-07 08:05:50 +1000832 /*
833 * Avoid getting stuck in this loop because we were woken by the
834 * shutdown, but then went back to sleep once already in the
835 * shutdown state.
836 */
837 if (XLOG_FORCED_SHUTDOWN(log))
838 goto out_shutdown;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000839 if (ctx->sequence > sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000840 continue;
841 if (!ctx->commit_lsn) {
842 /*
843 * It is still being pushed! Wait for the push to
844 * complete, then start again from the beginning.
845 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000846 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000847 goto restart;
848 }
Dave Chinnera44f13e2010-08-24 11:40:03 +1000849 if (ctx->sequence != sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000850 continue;
851 /* found it! */
852 commit_lsn = ctx->commit_lsn;
853 }
Dave Chinnerf876e442014-02-27 16:40:42 +1100854
855 /*
856 * The call to xlog_cil_push_now() executes the push in the background.
857 * Hence by the time we have got here it our sequence may not have been
858 * pushed yet. This is true if the current sequence still matches the
859 * push sequence after the above wait loop and the CIL still contains
860 * dirty objects.
861 *
Dave Chinnerac983512014-05-07 08:05:50 +1000862 * When the push occurs, it will empty the CIL and atomically increment
863 * the currect sequence past the push sequence and move it into the
864 * committing list. Of course, if the CIL is clean at the time of the
865 * push, it won't have pushed the CIL at all, so in that case we should
866 * try the push for this sequence again from the start just in case.
Dave Chinnerf876e442014-02-27 16:40:42 +1100867 */
Dave Chinnerf876e442014-02-27 16:40:42 +1100868 if (sequence == cil->xc_current_sequence &&
869 !list_empty(&cil->xc_cil)) {
870 spin_unlock(&cil->xc_push_lock);
871 goto restart;
872 }
873
Dave Chinner4bb928c2013-08-12 20:50:08 +1000874 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000875 return commit_lsn;
Dave Chinnerac983512014-05-07 08:05:50 +1000876
877 /*
878 * We detected a shutdown in progress. We need to trigger the log force
879 * to pass through it's iclog state machine error handling, even though
880 * we are already in a shutdown state. Hence we can't return
881 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
882 * LSN is already stable), so we return a zero LSN instead.
883 */
884out_shutdown:
885 spin_unlock(&cil->xc_push_lock);
886 return 0;
Dave Chinner71e330b2010-05-21 14:37:18 +1000887}
Dave Chinnerccf7c232010-05-20 23:19:42 +1000888
889/*
890 * Check if the current log item was first committed in this sequence.
891 * We can't rely on just the log item being in the CIL, we have to check
892 * the recorded commit sequence number.
893 *
894 * Note: for this to be used in a non-racy manner, it has to be called with
895 * CIL flushing locked out. As a result, it should only be used during the
896 * transaction commit process when deciding what to format into the item.
897 */
898bool
899xfs_log_item_in_current_chkpt(
900 struct xfs_log_item *lip)
901{
902 struct xfs_cil_ctx *ctx;
903
Dave Chinnerccf7c232010-05-20 23:19:42 +1000904 if (list_empty(&lip->li_cil))
905 return false;
906
907 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
908
909 /*
910 * li_seq is written on the first commit of a log item to record the
911 * first checkpoint it is written to. Hence if it is different to the
912 * current sequence, we're in a new checkpoint.
913 */
914 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
915 return false;
916 return true;
917}
Dave Chinner4c2d5422012-04-23 17:54:32 +1000918
919/*
920 * Perform initial CIL structure initialisation.
921 */
922int
923xlog_cil_init(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500924 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000925{
926 struct xfs_cil *cil;
927 struct xfs_cil_ctx *ctx;
928
929 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
930 if (!cil)
931 return ENOMEM;
932
933 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
934 if (!ctx) {
935 kmem_free(cil);
936 return ENOMEM;
937 }
938
939 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
940 INIT_LIST_HEAD(&cil->xc_cil);
941 INIT_LIST_HEAD(&cil->xc_committing);
942 spin_lock_init(&cil->xc_cil_lock);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000943 spin_lock_init(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000944 init_rwsem(&cil->xc_ctx_lock);
945 init_waitqueue_head(&cil->xc_commit_wait);
946
947 INIT_LIST_HEAD(&ctx->committing);
948 INIT_LIST_HEAD(&ctx->busy_extents);
949 ctx->sequence = 1;
950 ctx->cil = cil;
951 cil->xc_ctx = ctx;
952 cil->xc_current_sequence = ctx->sequence;
953
954 cil->xc_log = log;
955 log->l_cilp = cil;
956 return 0;
957}
958
959void
960xlog_cil_destroy(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500961 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000962{
963 if (log->l_cilp->xc_ctx) {
964 if (log->l_cilp->xc_ctx->ticket)
965 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
966 kmem_free(log->l_cilp->xc_ctx);
967 }
968
969 ASSERT(list_empty(&log->l_cilp->xc_cil));
970 kmem_free(log->l_cilp);
971}
972