blob: 45cc0ce18adf04cac50fe801884439a246285443 [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"
Christoph Hellwig4fb6e8a2014-11-28 14:25:04 +110020#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110021#include "xfs_log_format.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110022#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110023#include "xfs_trans_resv.h"
Dave Chinner71e330b2010-05-21 14:37:18 +100024#include "xfs_mount.h"
25#include "xfs_error.h"
26#include "xfs_alloc.h"
Dave Chinnerefc27b52012-04-29 10:39:43 +000027#include "xfs_extent_busy.h"
Christoph Hellwige84661a2011-05-20 13:45:32 +000028#include "xfs_discard.h"
Dave Chinner239880e2013-10-23 10:50:10 +110029#include "xfs_trans.h"
30#include "xfs_trans_priv.h"
31#include "xfs_log.h"
32#include "xfs_log_priv.h"
Dave Chinner71e330b2010-05-21 14:37:18 +100033
34/*
Dave Chinner71e330b2010-05-21 14:37:18 +100035 * 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 */
46static struct xlog_ticket *
47xlog_cil_ticket_alloc(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -050048 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +100049{
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 */
74void
75xlog_cil_init_post_recovery(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -050076 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +100077{
Dave Chinner71e330b2010-05-21 14:37:18 +100078 log->l_cilp->xc_ctx->ticket = xlog_cil_ticket_alloc(log);
79 log->l_cilp->xc_ctx->sequence = 1;
Dave Chinner71e330b2010-05-21 14:37:18 +100080}
81
82/*
Dave Chinner991aaf62013-08-12 20:50:07 +100083 * Prepare the log item for insertion into the CIL. Calculate the difference in
84 * log space and vectors it will consume, and if it is a new item pin it as
85 * well.
86 */
87STATIC void
88xfs_cil_prepare_item(
89 struct xlog *log,
90 struct xfs_log_vec *lv,
91 struct xfs_log_vec *old_lv,
92 int *diff_len,
93 int *diff_iovecs)
94{
95 /* Account for the new LV being passed in */
96 if (lv->lv_buf_len != XFS_LOG_VEC_ORDERED) {
Dave Chinner110dc242014-05-20 08:18:09 +100097 *diff_len += lv->lv_bytes;
Dave Chinner991aaf62013-08-12 20:50:07 +100098 *diff_iovecs += lv->lv_niovecs;
99 }
100
101 /*
102 * If there is no old LV, this is the first time we've seen the item in
103 * this CIL context and so we need to pin it. If we are replacing the
104 * old_lv, then remove the space it accounts for and free it.
105 */
106 if (!old_lv)
107 lv->lv_item->li_ops->iop_pin(lv->lv_item);
108 else if (old_lv != lv) {
109 ASSERT(lv->lv_buf_len != XFS_LOG_VEC_ORDERED);
110
Dave Chinner110dc242014-05-20 08:18:09 +1000111 *diff_len -= old_lv->lv_bytes;
Dave Chinner991aaf62013-08-12 20:50:07 +1000112 *diff_iovecs -= old_lv->lv_niovecs;
113 kmem_free(old_lv);
114 }
115
116 /* attach new log vector to log item */
117 lv->lv_item->li_lv = lv;
118
119 /*
120 * If this is the first time the item is being committed to the
121 * CIL, store the sequence number on the log item so we can
122 * tell in future commits whether this is the first checkpoint
123 * the item is being committed into.
124 */
125 if (!lv->lv_item->li_seq)
126 lv->lv_item->li_seq = log->l_cilp->xc_ctx->sequence;
127}
128
129/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000130 * Format log item into a flat buffers
131 *
132 * For delayed logging, we need to hold a formatted buffer containing all the
133 * changes on the log item. This enables us to relog the item in memory and
134 * write it out asynchronously without needing to relock the object that was
135 * modified at the time it gets written into the iclog.
136 *
137 * This function builds a vector for the changes in each log item in the
138 * transaction. It then works out the length of the buffer needed for each log
139 * item, allocates them and formats the vector for the item into the buffer.
140 * The buffer is then attached to the log item are then inserted into the
141 * Committed Item List for tracking until the next checkpoint is written out.
142 *
143 * We don't set up region headers during this process; we simply copy the
144 * regions into the flat buffer. We can do this because we still have to do a
145 * formatting step to write the regions into the iclog buffer. Writing the
146 * ophdrs during the iclog write means that we can support splitting large
147 * regions across iclog boundares without needing a change in the format of the
148 * item/region encapsulation.
149 *
150 * Hence what we need to do now is change the rewrite the vector array to point
151 * to the copied region inside the buffer we just allocated. This allows us to
152 * format the regions into the iclog as though they are being formatted
153 * directly out of the objects themselves.
154 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000155static void
156xlog_cil_insert_format_items(
157 struct xlog *log,
158 struct xfs_trans *tp,
159 int *diff_len,
160 int *diff_iovecs)
Dave Chinner71e330b2010-05-21 14:37:18 +1000161{
Christoph Hellwig0244b962011-12-06 21:58:08 +0000162 struct xfs_log_item_desc *lidp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000163
Christoph Hellwig0244b962011-12-06 21:58:08 +0000164
165 /* Bail out if we didn't find a log item. */
166 if (list_empty(&tp->t_items)) {
167 ASSERT(0);
Dave Chinner991aaf62013-08-12 20:50:07 +1000168 return;
Christoph Hellwig0244b962011-12-06 21:58:08 +0000169 }
170
171 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
Dave Chinner166d1362013-08-12 20:50:04 +1000172 struct xfs_log_item *lip = lidp->lid_item;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000173 struct xfs_log_vec *lv;
Dave Chinner991aaf62013-08-12 20:50:07 +1000174 struct xfs_log_vec *old_lv;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000175 int niovecs = 0;
176 int nbytes = 0;
177 int buf_size;
Dave Chinnerfd638752013-06-27 16:04:51 +1000178 bool ordered = false;
Dave Chinner71e330b2010-05-21 14:37:18 +1000179
Christoph Hellwig0244b962011-12-06 21:58:08 +0000180 /* Skip items which aren't dirty in this transaction. */
181 if (!(lidp->lid_flags & XFS_LID_DIRTY))
182 continue;
183
Dave Chinner166d1362013-08-12 20:50:04 +1000184 /* get number of vecs and size of data to be stored */
185 lip->li_ops->iop_size(lip, &niovecs, &nbytes);
186
Christoph Hellwig0244b962011-12-06 21:58:08 +0000187 /* Skip items that do not have any vectors for writing */
Christoph Hellwigb3934212011-12-06 21:58:09 +0000188 if (!niovecs)
Christoph Hellwig0244b962011-12-06 21:58:08 +0000189 continue;
190
Dave Chinnerfd638752013-06-27 16:04:51 +1000191 /*
192 * Ordered items need to be tracked but we do not wish to write
193 * them. We need a logvec to track the object, but we do not
194 * need an iovec or buffer to be allocated for copying data.
195 */
196 if (niovecs == XFS_LOG_VEC_ORDERED) {
197 ordered = true;
198 niovecs = 0;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000199 nbytes = 0;
Dave Chinnerfd638752013-06-27 16:04:51 +1000200 }
201
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100202 /*
203 * We 64-bit align the length of each iovec so that the start
204 * of the next one is naturally aligned. We'll need to
Dave Chinner3895e512014-02-10 10:37:18 +1100205 * account for that slack space here. Then round nbytes up
206 * to 64-bit alignment so that the initial buffer alignment is
207 * easy to calculate and verify.
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100208 */
209 nbytes += niovecs * sizeof(uint64_t);
Dave Chinner3895e512014-02-10 10:37:18 +1100210 nbytes = round_up(nbytes, sizeof(uint64_t));
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100211
Dave Chinner991aaf62013-08-12 20:50:07 +1000212 /* grab the old item if it exists for reservation accounting */
213 old_lv = lip->li_lv;
214
Dave Chinner3895e512014-02-10 10:37:18 +1100215 /*
216 * The data buffer needs to start 64-bit aligned, so round up
217 * that space to ensure we can align it appropriately and not
218 * overrun the buffer.
219 */
220 buf_size = nbytes +
221 round_up((sizeof(struct xfs_log_vec) +
222 niovecs * sizeof(struct xfs_log_iovec)),
223 sizeof(uint64_t));
Christoph Hellwig0244b962011-12-06 21:58:08 +0000224
Dave Chinnerf5baac32013-08-12 20:50:06 +1000225 /* compare to existing item size */
226 if (lip->li_lv && buf_size <= lip->li_lv->lv_size) {
227 /* same or smaller, optimise common overwrite case */
228 lv = lip->li_lv;
229 lv->lv_next = NULL;
230
231 if (ordered)
232 goto insert;
233
Dave Chinner991aaf62013-08-12 20:50:07 +1000234 /*
235 * set the item up as though it is a new insertion so
236 * that the space reservation accounting is correct.
237 */
238 *diff_iovecs -= lv->lv_niovecs;
Dave Chinner110dc242014-05-20 08:18:09 +1000239 *diff_len -= lv->lv_bytes;
Christoph Hellwig9597df62013-12-13 11:00:42 +1100240 } else {
241 /* allocate new data chunk */
242 lv = kmem_zalloc(buf_size, KM_SLEEP|KM_NOFS);
243 lv->lv_item = lip;
244 lv->lv_size = buf_size;
245 if (ordered) {
246 /* track as an ordered logvec */
247 ASSERT(lip->li_lv == NULL);
248 lv->lv_buf_len = XFS_LOG_VEC_ORDERED;
249 goto insert;
250 }
251 lv->lv_iovecp = (struct xfs_log_iovec *)&lv[1];
Dave Chinnerf5baac32013-08-12 20:50:06 +1000252 }
253
Christoph Hellwig9597df62013-12-13 11:00:42 +1100254 /* Ensure the lv is set up according to ->iop_size */
Dave Chinner7492c5b2013-08-12 20:50:05 +1000255 lv->lv_niovecs = niovecs;
Christoph Hellwig0244b962011-12-06 21:58:08 +0000256
Dave Chinner7492c5b2013-08-12 20:50:05 +1000257 /* The allocated data region lies beyond the iovec region */
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100258 lv->lv_buf_len = 0;
Dave Chinner110dc242014-05-20 08:18:09 +1000259 lv->lv_bytes = 0;
Dave Chinner7492c5b2013-08-12 20:50:05 +1000260 lv->lv_buf = (char *)lv + buf_size - nbytes;
Dave Chinner3895e512014-02-10 10:37:18 +1100261 ASSERT(IS_ALIGNED((unsigned long)lv->lv_buf, sizeof(uint64_t)));
262
Christoph Hellwigbde7cff2013-12-13 11:34:02 +1100263 lip->li_ops->iop_format(lip, lv);
Dave Chinner7492c5b2013-08-12 20:50:05 +1000264insert:
Dave Chinnerf5baac32013-08-12 20:50:06 +1000265 ASSERT(lv->lv_buf_len <= nbytes);
Dave Chinner991aaf62013-08-12 20:50:07 +1000266 xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);
Dave Chinner71e330b2010-05-21 14:37:18 +1000267 }
Dave Chinnerd1583a32010-09-24 18:14:13 +1000268}
269
270/*
271 * Insert the log items into the CIL and calculate the difference in space
272 * consumed by the item. Add the space to the checkpoint ticket and calculate
273 * if the change requires additional log metadata. If it does, take that space
Justin P. Mattock42b2aa82011-11-28 20:31:00 -0800274 * as well. Remove the amount of space we added to the checkpoint ticket from
Dave Chinnerd1583a32010-09-24 18:14:13 +1000275 * the current transaction ticket so that the accounting works out correctly.
276 */
Dave Chinner71e330b2010-05-21 14:37:18 +1000277static void
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000278xlog_cil_insert_items(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500279 struct xlog *log,
Dave Chinner991aaf62013-08-12 20:50:07 +1000280 struct xfs_trans *tp)
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000281{
Dave Chinnerd1583a32010-09-24 18:14:13 +1000282 struct xfs_cil *cil = log->l_cilp;
283 struct xfs_cil_ctx *ctx = cil->xc_ctx;
Dave Chinner991aaf62013-08-12 20:50:07 +1000284 struct xfs_log_item_desc *lidp;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000285 int len = 0;
286 int diff_iovecs = 0;
287 int iclog_space;
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000288
Dave Chinner991aaf62013-08-12 20:50:07 +1000289 ASSERT(tp);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000290
291 /*
Dave Chinnerd1583a32010-09-24 18:14:13 +1000292 * We can do this safely because the context can't checkpoint until we
293 * are done so it doesn't matter exactly how we update the CIL.
294 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000295 xlog_cil_insert_format_items(log, tp, &len, &diff_iovecs);
296
297 /*
298 * Now (re-)position everything modified at the tail of the CIL.
299 * We do this here so we only need to take the CIL lock once during
300 * the transaction commit.
301 */
Dave Chinnerfd638752013-06-27 16:04:51 +1000302 spin_lock(&cil->xc_cil_lock);
Dave Chinner991aaf62013-08-12 20:50:07 +1000303 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
304 struct xfs_log_item *lip = lidp->lid_item;
Dave Chinnerfd638752013-06-27 16:04:51 +1000305
Dave Chinner991aaf62013-08-12 20:50:07 +1000306 /* Skip items which aren't dirty in this transaction. */
307 if (!(lidp->lid_flags & XFS_LID_DIRTY))
308 continue;
Dave Chinnerfd638752013-06-27 16:04:51 +1000309
Dave Chinner991aaf62013-08-12 20:50:07 +1000310 list_move_tail(&lip->li_cil, &cil->xc_cil);
Dave Chinnerfd638752013-06-27 16:04:51 +1000311 }
Dave Chinnerd1583a32010-09-24 18:14:13 +1000312
313 /* account for space used by new iovec headers */
314 len += diff_iovecs * sizeof(xlog_op_header_t);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000315 ctx->nvecs += diff_iovecs;
316
Dave Chinner991aaf62013-08-12 20:50:07 +1000317 /* attach the transaction to the CIL if it has any busy extents */
318 if (!list_empty(&tp->t_busy))
319 list_splice_init(&tp->t_busy, &ctx->busy_extents);
320
Dave Chinnerd1583a32010-09-24 18:14:13 +1000321 /*
322 * Now transfer enough transaction reservation to the context ticket
323 * for the checkpoint. The context ticket is special - the unit
324 * reservation has to grow as well as the current reservation as we
325 * steal from tickets so we can correctly determine the space used
326 * during the transaction commit.
327 */
328 if (ctx->ticket->t_curr_res == 0) {
Dave Chinnerd1583a32010-09-24 18:14:13 +1000329 ctx->ticket->t_curr_res = ctx->ticket->t_unit_res;
Dave Chinner991aaf62013-08-12 20:50:07 +1000330 tp->t_ticket->t_curr_res -= ctx->ticket->t_unit_res;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000331 }
332
333 /* do we need space for more log record headers? */
334 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
335 if (len > 0 && (ctx->space_used / iclog_space !=
336 (ctx->space_used + len) / iclog_space)) {
337 int hdrs;
338
339 hdrs = (len + iclog_space - 1) / iclog_space;
340 /* need to take into account split region headers, too */
341 hdrs *= log->l_iclog_hsize + sizeof(struct xlog_op_header);
342 ctx->ticket->t_unit_res += hdrs;
343 ctx->ticket->t_curr_res += hdrs;
Dave Chinner991aaf62013-08-12 20:50:07 +1000344 tp->t_ticket->t_curr_res -= hdrs;
345 ASSERT(tp->t_ticket->t_curr_res >= len);
Dave Chinnerd1583a32010-09-24 18:14:13 +1000346 }
Dave Chinner991aaf62013-08-12 20:50:07 +1000347 tp->t_ticket->t_curr_res -= len;
Dave Chinnerd1583a32010-09-24 18:14:13 +1000348 ctx->space_used += len;
349
350 spin_unlock(&cil->xc_cil_lock);
Dave Chinner3b93c7a2010-08-24 11:45:53 +1000351}
352
353static void
Dave Chinner71e330b2010-05-21 14:37:18 +1000354xlog_cil_free_logvec(
355 struct xfs_log_vec *log_vector)
356{
357 struct xfs_log_vec *lv;
358
359 for (lv = log_vector; lv; ) {
360 struct xfs_log_vec *next = lv->lv_next;
Dave Chinner71e330b2010-05-21 14:37:18 +1000361 kmem_free(lv);
362 lv = next;
363 }
364}
365
366/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000367 * Mark all items committed and clear busy extents. We free the log vector
368 * chains in a separate pass so that we unpin the log items as quickly as
369 * possible.
370 */
371static void
372xlog_cil_committed(
373 void *args,
374 int abort)
375{
376 struct xfs_cil_ctx *ctx = args;
Christoph Hellwige84661a2011-05-20 13:45:32 +0000377 struct xfs_mount *mp = ctx->cil->xc_log->l_mp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000378
Dave Chinner0e57f6a2010-12-20 12:02:19 +1100379 xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
380 ctx->start_lsn, abort);
Dave Chinner71e330b2010-05-21 14:37:18 +1000381
Dave Chinner4ecbfe62012-04-29 10:41:10 +0000382 xfs_extent_busy_sort(&ctx->busy_extents);
383 xfs_extent_busy_clear(mp, &ctx->busy_extents,
Christoph Hellwige84661a2011-05-20 13:45:32 +0000384 (mp->m_flags & XFS_MOUNT_DISCARD) && !abort);
Dave Chinner71e330b2010-05-21 14:37:18 +1000385
Dave Chinnerac983512014-05-07 08:05:50 +1000386 /*
387 * If we are aborting the commit, wake up anyone waiting on the
388 * committing list. If we don't, then a shutdown we can leave processes
389 * waiting in xlog_cil_force_lsn() waiting on a sequence commit that
390 * will never happen because we aborted it.
391 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000392 spin_lock(&ctx->cil->xc_push_lock);
Dave Chinnerac983512014-05-07 08:05:50 +1000393 if (abort)
394 wake_up_all(&ctx->cil->xc_commit_wait);
Dave Chinner71e330b2010-05-21 14:37:18 +1000395 list_del(&ctx->committing);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000396 spin_unlock(&ctx->cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000397
398 xlog_cil_free_logvec(ctx->lv_chain);
Christoph Hellwige84661a2011-05-20 13:45:32 +0000399
400 if (!list_empty(&ctx->busy_extents)) {
401 ASSERT(mp->m_flags & XFS_MOUNT_DISCARD);
402
403 xfs_discard_extents(mp, &ctx->busy_extents);
Dave Chinner4ecbfe62012-04-29 10:41:10 +0000404 xfs_extent_busy_clear(mp, &ctx->busy_extents, false);
Christoph Hellwige84661a2011-05-20 13:45:32 +0000405 }
406
Dave Chinner71e330b2010-05-21 14:37:18 +1000407 kmem_free(ctx);
408}
409
410/*
Dave Chinnera44f13e2010-08-24 11:40:03 +1000411 * Push the Committed Item List to the log. If @push_seq flag is zero, then it
412 * is a background flush and so we can chose to ignore it. Otherwise, if the
413 * current sequence is the same as @push_seq we need to do a flush. If
414 * @push_seq is less than the current sequence, then it has already been
415 * flushed and we don't need to do anything - the caller will wait for it to
416 * complete if necessary.
417 *
418 * @push_seq is a value rather than a flag because that allows us to do an
419 * unlocked check of the sequence number for a match. Hence we can allows log
420 * forces to run racily and not issue pushes for the same sequence twice. If we
421 * get a race between multiple pushes for the same sequence they will block on
422 * the first one and then abort, hence avoiding needless pushes.
Dave Chinner71e330b2010-05-21 14:37:18 +1000423 */
Dave Chinnera44f13e2010-08-24 11:40:03 +1000424STATIC int
Dave Chinner71e330b2010-05-21 14:37:18 +1000425xlog_cil_push(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500426 struct xlog *log)
Dave Chinner71e330b2010-05-21 14:37:18 +1000427{
428 struct xfs_cil *cil = log->l_cilp;
429 struct xfs_log_vec *lv;
430 struct xfs_cil_ctx *ctx;
431 struct xfs_cil_ctx *new_ctx;
432 struct xlog_in_core *commit_iclog;
433 struct xlog_ticket *tic;
Dave Chinner71e330b2010-05-21 14:37:18 +1000434 int num_iovecs;
Dave Chinner71e330b2010-05-21 14:37:18 +1000435 int error = 0;
436 struct xfs_trans_header thdr;
437 struct xfs_log_iovec lhdr;
438 struct xfs_log_vec lvhdr = { NULL };
439 xfs_lsn_t commit_lsn;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000440 xfs_lsn_t push_seq;
Dave Chinner71e330b2010-05-21 14:37:18 +1000441
442 if (!cil)
443 return 0;
444
Dave Chinner71e330b2010-05-21 14:37:18 +1000445 new_ctx = kmem_zalloc(sizeof(*new_ctx), KM_SLEEP|KM_NOFS);
446 new_ctx->ticket = xlog_cil_ticket_alloc(log);
447
Dave Chinner4c2d5422012-04-23 17:54:32 +1000448 down_write(&cil->xc_ctx_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000449 ctx = cil->xc_ctx;
450
Dave Chinner4bb928c2013-08-12 20:50:08 +1000451 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000452 push_seq = cil->xc_push_seq;
453 ASSERT(push_seq <= ctx->sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000454
Dave Chinner4c2d5422012-04-23 17:54:32 +1000455 /*
456 * Check if we've anything to push. If there is nothing, then we don't
457 * move on to a new sequence number and so we have to be able to push
458 * this sequence again later.
459 */
460 if (list_empty(&cil->xc_cil)) {
461 cil->xc_push_seq = 0;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000462 spin_unlock(&cil->xc_push_lock);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000463 goto out_skip;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000464 }
Dave Chinner4c2d5422012-04-23 17:54:32 +1000465
Dave Chinnera44f13e2010-08-24 11:40:03 +1000466
467 /* check for a previously pushed seqeunce */
Dave Chinner8af3dcd2014-09-23 15:57:59 +1000468 if (push_seq < cil->xc_ctx->sequence) {
469 spin_unlock(&cil->xc_push_lock);
Dave Chinnerdf806152010-05-17 15:52:13 +1000470 goto out_skip;
Dave Chinner8af3dcd2014-09-23 15:57:59 +1000471 }
472
473 /*
474 * We are now going to push this context, so add it to the committing
475 * list before we do anything else. This ensures that anyone waiting on
476 * this push can easily detect the difference between a "push in
477 * progress" and "CIL is empty, nothing to do".
478 *
479 * IOWs, a wait loop can now check for:
480 * the current sequence not being found on the committing list;
481 * an empty CIL; and
482 * an unchanged sequence number
483 * to detect a push that had nothing to do and therefore does not need
484 * waiting on. If the CIL is not empty, we get put on the committing
485 * list before emptying the CIL and bumping the sequence number. Hence
486 * an empty CIL and an unchanged sequence number means we jumped out
487 * above after doing nothing.
488 *
489 * Hence the waiter will either find the commit sequence on the
490 * committing list or the sequence number will be unchanged and the CIL
491 * still dirty. In that latter case, the push has not yet started, and
492 * so the waiter will have to continue trying to check the CIL
493 * committing list until it is found. In extreme cases of delay, the
494 * sequence may fully commit between the attempts the wait makes to wait
495 * on the commit sequence.
496 */
497 list_add(&ctx->committing, &cil->xc_committing);
498 spin_unlock(&cil->xc_push_lock);
Dave Chinnerdf806152010-05-17 15:52:13 +1000499
Dave Chinner71e330b2010-05-21 14:37:18 +1000500 /*
501 * pull all the log vectors off the items in the CIL, and
502 * remove the items from the CIL. We don't need the CIL lock
503 * here because it's only needed on the transaction commit
504 * side which is currently locked out by the flush lock.
505 */
506 lv = NULL;
Dave Chinner71e330b2010-05-21 14:37:18 +1000507 num_iovecs = 0;
Dave Chinner71e330b2010-05-21 14:37:18 +1000508 while (!list_empty(&cil->xc_cil)) {
509 struct xfs_log_item *item;
Dave Chinner71e330b2010-05-21 14:37:18 +1000510
511 item = list_first_entry(&cil->xc_cil,
512 struct xfs_log_item, li_cil);
513 list_del_init(&item->li_cil);
514 if (!ctx->lv_chain)
515 ctx->lv_chain = item->li_lv;
516 else
517 lv->lv_next = item->li_lv;
518 lv = item->li_lv;
519 item->li_lv = NULL;
Dave Chinner71e330b2010-05-21 14:37:18 +1000520 num_iovecs += lv->lv_niovecs;
Dave Chinner71e330b2010-05-21 14:37:18 +1000521 }
522
523 /*
524 * initialise the new context and attach it to the CIL. Then attach
525 * the current context to the CIL committing lsit so it can be found
526 * during log forces to extract the commit lsn of the sequence that
527 * needs to be forced.
528 */
529 INIT_LIST_HEAD(&new_ctx->committing);
530 INIT_LIST_HEAD(&new_ctx->busy_extents);
531 new_ctx->sequence = ctx->sequence + 1;
532 new_ctx->cil = cil;
533 cil->xc_ctx = new_ctx;
534
535 /*
536 * The switch is now done, so we can drop the context lock and move out
537 * of a shared context. We can't just go straight to the commit record,
538 * though - we need to synchronise with previous and future commits so
539 * that the commit records are correctly ordered in the log to ensure
540 * that we process items during log IO completion in the correct order.
541 *
542 * For example, if we get an EFI in one checkpoint and the EFD in the
543 * next (e.g. due to log forces), we do not want the checkpoint with
544 * the EFD to be committed before the checkpoint with the EFI. Hence
545 * we must strictly order the commit records of the checkpoints so
546 * that: a) the checkpoint callbacks are attached to the iclogs in the
547 * correct order; and b) the checkpoints are replayed in correct order
548 * in log recovery.
549 *
550 * Hence we need to add this context to the committing context list so
551 * that higher sequences will wait for us to write out a commit record
552 * before they do.
Dave Chinnerf876e442014-02-27 16:40:42 +1100553 *
554 * xfs_log_force_lsn requires us to mirror the new sequence into the cil
555 * structure atomically with the addition of this sequence to the
556 * committing list. This also ensures that we can do unlocked checks
557 * against the current sequence in log forces without risking
558 * deferencing a freed context pointer.
Dave Chinner71e330b2010-05-21 14:37:18 +1000559 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000560 spin_lock(&cil->xc_push_lock);
Dave Chinnerf876e442014-02-27 16:40:42 +1100561 cil->xc_current_sequence = new_ctx->sequence;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000562 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000563 up_write(&cil->xc_ctx_lock);
564
565 /*
566 * Build a checkpoint transaction header and write it to the log to
567 * begin the transaction. We need to account for the space used by the
568 * transaction header here as it is not accounted for in xlog_write().
569 *
570 * The LSN we need to pass to the log items on transaction commit is
571 * the LSN reported by the first log vector write. If we use the commit
572 * record lsn then we can move the tail beyond the grant write head.
573 */
574 tic = ctx->ticket;
575 thdr.th_magic = XFS_TRANS_HEADER_MAGIC;
576 thdr.th_type = XFS_TRANS_CHECKPOINT;
577 thdr.th_tid = tic->t_tid;
578 thdr.th_num_items = num_iovecs;
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000579 lhdr.i_addr = &thdr;
Dave Chinner71e330b2010-05-21 14:37:18 +1000580 lhdr.i_len = sizeof(xfs_trans_header_t);
581 lhdr.i_type = XLOG_REG_TYPE_TRANSHDR;
582 tic->t_curr_res -= lhdr.i_len + sizeof(xlog_op_header_t);
583
584 lvhdr.lv_niovecs = 1;
585 lvhdr.lv_iovecp = &lhdr;
586 lvhdr.lv_next = ctx->lv_chain;
587
588 error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0);
589 if (error)
Dave Chinner7db37c52011-01-27 12:02:00 +1100590 goto out_abort_free_ticket;
Dave Chinner71e330b2010-05-21 14:37:18 +1000591
592 /*
593 * now that we've written the checkpoint into the log, strictly
594 * order the commit records so replay will get them in the right order.
595 */
596restart:
Dave Chinner4bb928c2013-08-12 20:50:08 +1000597 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000598 list_for_each_entry(new_ctx, &cil->xc_committing, committing) {
599 /*
Dave Chinnerac983512014-05-07 08:05:50 +1000600 * Avoid getting stuck in this loop because we were woken by the
601 * shutdown, but then went back to sleep once already in the
602 * shutdown state.
603 */
604 if (XLOG_FORCED_SHUTDOWN(log)) {
605 spin_unlock(&cil->xc_push_lock);
606 goto out_abort_free_ticket;
607 }
608
609 /*
Dave Chinner71e330b2010-05-21 14:37:18 +1000610 * Higher sequences will wait for this one so skip them.
Dave Chinnerac983512014-05-07 08:05:50 +1000611 * Don't wait for our own sequence, either.
Dave Chinner71e330b2010-05-21 14:37:18 +1000612 */
613 if (new_ctx->sequence >= ctx->sequence)
614 continue;
615 if (!new_ctx->commit_lsn) {
616 /*
617 * It is still being pushed! Wait for the push to
618 * complete, then start again from the beginning.
619 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000620 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000621 goto restart;
622 }
623 }
Dave Chinner4bb928c2013-08-12 20:50:08 +1000624 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000625
Dave Chinner7db37c52011-01-27 12:02:00 +1100626 /* xfs_log_done always frees the ticket on error. */
Dave Chinner71e330b2010-05-21 14:37:18 +1000627 commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0);
Dave Chinner7db37c52011-01-27 12:02:00 +1100628 if (commit_lsn == -1)
Dave Chinner71e330b2010-05-21 14:37:18 +1000629 goto out_abort;
630
631 /* attach all the transactions w/ busy extents to iclog */
632 ctx->log_cb.cb_func = xlog_cil_committed;
633 ctx->log_cb.cb_arg = ctx;
634 error = xfs_log_notify(log->l_mp, commit_iclog, &ctx->log_cb);
635 if (error)
636 goto out_abort;
637
638 /*
639 * now the checkpoint commit is complete and we've attached the
640 * callbacks to the iclog we can assign the commit LSN to the context
641 * and wake up anyone who is waiting for the commit to complete.
642 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000643 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000644 ctx->commit_lsn = commit_lsn;
Dave Chinnereb40a872010-12-21 12:09:01 +1100645 wake_up_all(&cil->xc_commit_wait);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000646 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000647
648 /* release the hounds! */
649 return xfs_log_release_iclog(log->l_mp, commit_iclog);
650
651out_skip:
652 up_write(&cil->xc_ctx_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000653 xfs_log_ticket_put(new_ctx->ticket);
654 kmem_free(new_ctx);
655 return 0;
656
Dave Chinner7db37c52011-01-27 12:02:00 +1100657out_abort_free_ticket:
658 xfs_log_ticket_put(tic);
Dave Chinner71e330b2010-05-21 14:37:18 +1000659out_abort:
660 xlog_cil_committed(ctx, XFS_LI_ABORTED);
Dave Chinner24513372014-06-25 14:58:08 +1000661 return -EIO;
Dave Chinner71e330b2010-05-21 14:37:18 +1000662}
663
Dave Chinner4c2d5422012-04-23 17:54:32 +1000664static void
665xlog_cil_push_work(
666 struct work_struct *work)
667{
668 struct xfs_cil *cil = container_of(work, struct xfs_cil,
669 xc_push_work);
670 xlog_cil_push(cil->xc_log);
671}
672
673/*
674 * We need to push CIL every so often so we don't cache more than we can fit in
675 * the log. The limit really is that a checkpoint can't be more than half the
676 * log (the current checkpoint is not allowed to overwrite the previous
677 * checkpoint), but commit latency and memory usage limit this to a smaller
678 * size.
679 */
680static void
681xlog_cil_push_background(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500682 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000683{
684 struct xfs_cil *cil = log->l_cilp;
685
686 /*
687 * The cil won't be empty because we are called while holding the
688 * context lock so whatever we added to the CIL will still be there
689 */
690 ASSERT(!list_empty(&cil->xc_cil));
691
692 /*
693 * don't do a background push if we haven't used up all the
694 * space available yet.
695 */
696 if (cil->xc_ctx->space_used < XLOG_CIL_SPACE_LIMIT(log))
697 return;
698
Dave Chinner4bb928c2013-08-12 20:50:08 +1000699 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000700 if (cil->xc_push_seq < cil->xc_current_sequence) {
701 cil->xc_push_seq = cil->xc_current_sequence;
702 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
703 }
Dave Chinner4bb928c2013-08-12 20:50:08 +1000704 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000705
706}
707
Dave Chinnerf876e442014-02-27 16:40:42 +1100708/*
709 * xlog_cil_push_now() is used to trigger an immediate CIL push to the sequence
710 * number that is passed. When it returns, the work will be queued for
711 * @push_seq, but it won't be completed. The caller is expected to do any
712 * waiting for push_seq to complete if it is required.
713 */
Dave Chinner4c2d5422012-04-23 17:54:32 +1000714static void
Dave Chinnerf876e442014-02-27 16:40:42 +1100715xlog_cil_push_now(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500716 struct xlog *log,
Dave Chinner4c2d5422012-04-23 17:54:32 +1000717 xfs_lsn_t push_seq)
718{
719 struct xfs_cil *cil = log->l_cilp;
720
721 if (!cil)
722 return;
723
724 ASSERT(push_seq && push_seq <= cil->xc_current_sequence);
725
726 /* start on any pending background push to minimise wait time on it */
727 flush_work(&cil->xc_push_work);
728
729 /*
730 * If the CIL is empty or we've already pushed the sequence then
731 * there's no work we need to do.
732 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000733 spin_lock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000734 if (list_empty(&cil->xc_cil) || push_seq <= cil->xc_push_seq) {
Dave Chinner4bb928c2013-08-12 20:50:08 +1000735 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000736 return;
737 }
738
739 cil->xc_push_seq = push_seq;
Dave Chinnerf876e442014-02-27 16:40:42 +1100740 queue_work(log->l_mp->m_cil_workqueue, &cil->xc_push_work);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000741 spin_unlock(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000742}
743
Dave Chinner2c6e24c2013-10-15 09:17:49 +1100744bool
745xlog_cil_empty(
746 struct xlog *log)
747{
748 struct xfs_cil *cil = log->l_cilp;
749 bool empty = false;
750
751 spin_lock(&cil->xc_push_lock);
752 if (list_empty(&cil->xc_cil))
753 empty = true;
754 spin_unlock(&cil->xc_push_lock);
755 return empty;
756}
757
Dave Chinner71e330b2010-05-21 14:37:18 +1000758/*
Dave Chinnera44f13e2010-08-24 11:40:03 +1000759 * Commit a transaction with the given vector to the Committed Item List.
760 *
761 * To do this, we need to format the item, pin it in memory if required and
762 * account for the space used by the transaction. Once we have done that we
763 * need to release the unused reservation for the transaction, attach the
764 * transaction to the checkpoint context so we carry the busy extents through
765 * to checkpoint completion, and then unlock all the items in the transaction.
766 *
Dave Chinnera44f13e2010-08-24 11:40:03 +1000767 * Called with the context lock already held in read mode to lock out
768 * background commit, returns without it held once background commits are
769 * allowed again.
770 */
Jie Liuc6f97262014-02-07 15:26:07 +1100771void
Dave Chinnera44f13e2010-08-24 11:40:03 +1000772xfs_log_commit_cil(
773 struct xfs_mount *mp,
774 struct xfs_trans *tp,
Dave Chinnera44f13e2010-08-24 11:40:03 +1000775 xfs_lsn_t *commit_lsn,
776 int flags)
777{
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500778 struct xlog *log = mp->m_log;
Dave Chinner991aaf62013-08-12 20:50:07 +1000779 struct xfs_cil *cil = log->l_cilp;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000780 int log_flags = 0;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000781
782 if (flags & XFS_TRANS_RELEASE_LOG_RES)
783 log_flags = XFS_LOG_REL_PERM_RESERV;
784
Dave Chinnerf5baac32013-08-12 20:50:06 +1000785 /* lock out background commit */
Dave Chinner991aaf62013-08-12 20:50:07 +1000786 down_read(&cil->xc_ctx_lock);
Dave Chinnerf5baac32013-08-12 20:50:06 +1000787
Dave Chinner991aaf62013-08-12 20:50:07 +1000788 xlog_cil_insert_items(log, tp);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000789
790 /* check we didn't blow the reservation */
791 if (tp->t_ticket->t_curr_res < 0)
Dave Chinner991aaf62013-08-12 20:50:07 +1000792 xlog_print_tic_res(mp, tp->t_ticket);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000793
Dave Chinner991aaf62013-08-12 20:50:07 +1000794 tp->t_commit_lsn = cil->xc_ctx->sequence;
795 if (commit_lsn)
796 *commit_lsn = tp->t_commit_lsn;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000797
Dave Chinnera44f13e2010-08-24 11:40:03 +1000798 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
799 xfs_trans_unreserve_and_mod_sb(tp);
800
801 /*
802 * Once all the items of the transaction have been copied to the CIL,
803 * the items can be unlocked and freed.
804 *
805 * This needs to be done before we drop the CIL context lock because we
806 * have to update state in the log items and unlock them before they go
807 * to disk. If we don't, then the CIL checkpoint can race with us and
808 * we can run checkpoint completion before we've updated and unlocked
809 * the log items. This affects (at least) processing of stale buffers,
810 * inodes and EFIs.
811 */
Dave Chinner991aaf62013-08-12 20:50:07 +1000812 xfs_trans_free_items(tp, tp->t_commit_lsn, 0);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000813
Dave Chinner4c2d5422012-04-23 17:54:32 +1000814 xlog_cil_push_background(log);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000815
Dave Chinner991aaf62013-08-12 20:50:07 +1000816 up_read(&cil->xc_ctx_lock);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000817}
818
819/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000820 * Conditionally push the CIL based on the sequence passed in.
821 *
822 * We only need to push if we haven't already pushed the sequence
823 * number given. Hence the only time we will trigger a push here is
824 * if the push sequence is the same as the current context.
825 *
826 * We return the current commit lsn to allow the callers to determine if a
827 * iclog flush is necessary following this call.
Dave Chinner71e330b2010-05-21 14:37:18 +1000828 */
829xfs_lsn_t
Dave Chinnera44f13e2010-08-24 11:40:03 +1000830xlog_cil_force_lsn(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500831 struct xlog *log,
Dave Chinnera44f13e2010-08-24 11:40:03 +1000832 xfs_lsn_t sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000833{
834 struct xfs_cil *cil = log->l_cilp;
835 struct xfs_cil_ctx *ctx;
836 xfs_lsn_t commit_lsn = NULLCOMMITLSN;
837
Dave Chinnera44f13e2010-08-24 11:40:03 +1000838 ASSERT(sequence <= cil->xc_current_sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000839
Dave Chinnera44f13e2010-08-24 11:40:03 +1000840 /*
841 * check to see if we need to force out the current context.
842 * xlog_cil_push() handles racing pushes for the same sequence,
843 * so no need to deal with it here.
844 */
Dave Chinnerf876e442014-02-27 16:40:42 +1100845restart:
846 xlog_cil_push_now(log, sequence);
Dave Chinner71e330b2010-05-21 14:37:18 +1000847
848 /*
849 * See if we can find a previous sequence still committing.
Dave Chinner71e330b2010-05-21 14:37:18 +1000850 * We need to wait for all previous sequence commits to complete
851 * before allowing the force of push_seq to go ahead. Hence block
852 * on commits for those as well.
853 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000854 spin_lock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000855 list_for_each_entry(ctx, &cil->xc_committing, committing) {
Dave Chinnerac983512014-05-07 08:05:50 +1000856 /*
857 * Avoid getting stuck in this loop because we were woken by the
858 * shutdown, but then went back to sleep once already in the
859 * shutdown state.
860 */
861 if (XLOG_FORCED_SHUTDOWN(log))
862 goto out_shutdown;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000863 if (ctx->sequence > sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000864 continue;
865 if (!ctx->commit_lsn) {
866 /*
867 * It is still being pushed! Wait for the push to
868 * complete, then start again from the beginning.
869 */
Dave Chinner4bb928c2013-08-12 20:50:08 +1000870 xlog_wait(&cil->xc_commit_wait, &cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000871 goto restart;
872 }
Dave Chinnera44f13e2010-08-24 11:40:03 +1000873 if (ctx->sequence != sequence)
Dave Chinner71e330b2010-05-21 14:37:18 +1000874 continue;
875 /* found it! */
876 commit_lsn = ctx->commit_lsn;
877 }
Dave Chinnerf876e442014-02-27 16:40:42 +1100878
879 /*
880 * The call to xlog_cil_push_now() executes the push in the background.
881 * Hence by the time we have got here it our sequence may not have been
882 * pushed yet. This is true if the current sequence still matches the
883 * push sequence after the above wait loop and the CIL still contains
Dave Chinner8af3dcd2014-09-23 15:57:59 +1000884 * dirty objects. This is guaranteed by the push code first adding the
885 * context to the committing list before emptying the CIL.
Dave Chinnerf876e442014-02-27 16:40:42 +1100886 *
Dave Chinner8af3dcd2014-09-23 15:57:59 +1000887 * Hence if we don't find the context in the committing list and the
888 * current sequence number is unchanged then the CIL contents are
889 * significant. If the CIL is empty, if means there was nothing to push
890 * and that means there is nothing to wait for. If the CIL is not empty,
891 * it means we haven't yet started the push, because if it had started
892 * we would have found the context on the committing list.
Dave Chinnerf876e442014-02-27 16:40:42 +1100893 */
Dave Chinnerf876e442014-02-27 16:40:42 +1100894 if (sequence == cil->xc_current_sequence &&
895 !list_empty(&cil->xc_cil)) {
896 spin_unlock(&cil->xc_push_lock);
897 goto restart;
898 }
899
Dave Chinner4bb928c2013-08-12 20:50:08 +1000900 spin_unlock(&cil->xc_push_lock);
Dave Chinner71e330b2010-05-21 14:37:18 +1000901 return commit_lsn;
Dave Chinnerac983512014-05-07 08:05:50 +1000902
903 /*
904 * We detected a shutdown in progress. We need to trigger the log force
905 * to pass through it's iclog state machine error handling, even though
906 * we are already in a shutdown state. Hence we can't return
907 * NULLCOMMITLSN here as that has special meaning to log forces (i.e.
908 * LSN is already stable), so we return a zero LSN instead.
909 */
910out_shutdown:
911 spin_unlock(&cil->xc_push_lock);
912 return 0;
Dave Chinner71e330b2010-05-21 14:37:18 +1000913}
Dave Chinnerccf7c232010-05-20 23:19:42 +1000914
915/*
916 * Check if the current log item was first committed in this sequence.
917 * We can't rely on just the log item being in the CIL, we have to check
918 * the recorded commit sequence number.
919 *
920 * Note: for this to be used in a non-racy manner, it has to be called with
921 * CIL flushing locked out. As a result, it should only be used during the
922 * transaction commit process when deciding what to format into the item.
923 */
924bool
925xfs_log_item_in_current_chkpt(
926 struct xfs_log_item *lip)
927{
928 struct xfs_cil_ctx *ctx;
929
Dave Chinnerccf7c232010-05-20 23:19:42 +1000930 if (list_empty(&lip->li_cil))
931 return false;
932
933 ctx = lip->li_mountp->m_log->l_cilp->xc_ctx;
934
935 /*
936 * li_seq is written on the first commit of a log item to record the
937 * first checkpoint it is written to. Hence if it is different to the
938 * current sequence, we're in a new checkpoint.
939 */
940 if (XFS_LSN_CMP(lip->li_seq, ctx->sequence) != 0)
941 return false;
942 return true;
943}
Dave Chinner4c2d5422012-04-23 17:54:32 +1000944
945/*
946 * Perform initial CIL structure initialisation.
947 */
948int
949xlog_cil_init(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500950 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000951{
952 struct xfs_cil *cil;
953 struct xfs_cil_ctx *ctx;
954
955 cil = kmem_zalloc(sizeof(*cil), KM_SLEEP|KM_MAYFAIL);
956 if (!cil)
Dave Chinner24513372014-06-25 14:58:08 +1000957 return -ENOMEM;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000958
959 ctx = kmem_zalloc(sizeof(*ctx), KM_SLEEP|KM_MAYFAIL);
960 if (!ctx) {
961 kmem_free(cil);
Dave Chinner24513372014-06-25 14:58:08 +1000962 return -ENOMEM;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000963 }
964
965 INIT_WORK(&cil->xc_push_work, xlog_cil_push_work);
966 INIT_LIST_HEAD(&cil->xc_cil);
967 INIT_LIST_HEAD(&cil->xc_committing);
968 spin_lock_init(&cil->xc_cil_lock);
Dave Chinner4bb928c2013-08-12 20:50:08 +1000969 spin_lock_init(&cil->xc_push_lock);
Dave Chinner4c2d5422012-04-23 17:54:32 +1000970 init_rwsem(&cil->xc_ctx_lock);
971 init_waitqueue_head(&cil->xc_commit_wait);
972
973 INIT_LIST_HEAD(&ctx->committing);
974 INIT_LIST_HEAD(&ctx->busy_extents);
975 ctx->sequence = 1;
976 ctx->cil = cil;
977 cil->xc_ctx = ctx;
978 cil->xc_current_sequence = ctx->sequence;
979
980 cil->xc_log = log;
981 log->l_cilp = cil;
982 return 0;
983}
984
985void
986xlog_cil_destroy(
Mark Tinguelyf7bdf032012-06-14 09:22:15 -0500987 struct xlog *log)
Dave Chinner4c2d5422012-04-23 17:54:32 +1000988{
989 if (log->l_cilp->xc_ctx) {
990 if (log->l_cilp->xc_ctx->ticket)
991 xfs_log_ticket_put(log->l_cilp->xc_ctx->ticket);
992 kmem_free(log->l_cilp->xc_ctx);
993 }
994
995 ASSERT(list_empty(&log->l_cilp->xc_cil));
996 kmem_free(log->l_cilp);
997}
998