blob: fe88ef67f93acb9bbcbf4fbee7c06747b3c8155e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110021#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110022#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_sb.h"
25#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
27#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110028#include "xfs_trans.h"
29#include "xfs_trans_priv.h"
30#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_log_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_log_recover.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_inode.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000034#include "xfs_trace.h"
Dave Chinnerf661f1e2012-10-08 21:56:02 +110035#include "xfs_fsops.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110036#include "xfs_cksum.h"
Brian Fosterbaff4e42014-07-15 08:07:29 +100037#include "xfs_sysfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
David Chinnereb01c9c2008-04-10 12:18:46 +100039kmem_zone_t *xfs_log_ticket_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/* Local miscellaneous function prototypes */
Mark Tinguelyad223e62012-06-14 09:22:15 -050042STATIC int
43xlog_commit_record(
44 struct xlog *log,
45 struct xlog_ticket *ticket,
46 struct xlog_in_core **iclog,
47 xfs_lsn_t *commitlsnp);
48
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050049STATIC struct xlog *
50xlog_alloc_log(
51 struct xfs_mount *mp,
52 struct xfs_buftarg *log_target,
53 xfs_daddr_t blk_offset,
54 int num_bblks);
Mark Tinguelyad223e62012-06-14 09:22:15 -050055STATIC int
56xlog_space_left(
57 struct xlog *log,
58 atomic64_t *head);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050059STATIC int
60xlog_sync(
61 struct xlog *log,
62 struct xlog_in_core *iclog);
63STATIC void
64xlog_dealloc_log(
65 struct xlog *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/* local state machine functions */
68STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050069STATIC void
70xlog_state_do_callback(
71 struct xlog *log,
72 int aborted,
73 struct xlog_in_core *iclog);
74STATIC int
75xlog_state_get_iclog_space(
76 struct xlog *log,
77 int len,
78 struct xlog_in_core **iclog,
79 struct xlog_ticket *ticket,
80 int *continued_write,
81 int *logoffsetp);
82STATIC int
83xlog_state_release_iclog(
84 struct xlog *log,
85 struct xlog_in_core *iclog);
86STATIC void
87xlog_state_switch_iclogs(
88 struct xlog *log,
89 struct xlog_in_core *iclog,
90 int eventual_size);
91STATIC void
92xlog_state_want_sync(
93 struct xlog *log,
94 struct xlog_in_core *iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
Mark Tinguelyad223e62012-06-14 09:22:15 -050096STATIC void
97xlog_grant_push_ail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050098 struct xlog *log,
99 int need_bytes);
100STATIC void
101xlog_regrant_reserve_log_space(
102 struct xlog *log,
103 struct xlog_ticket *ticket);
104STATIC void
105xlog_ungrant_log_space(
106 struct xlog *log,
107 struct xlog_ticket *ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100109#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500110STATIC void
111xlog_verify_dest_ptr(
112 struct xlog *log,
113 char *ptr);
Mark Tinguelyad223e62012-06-14 09:22:15 -0500114STATIC void
115xlog_verify_grant_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500116 struct xlog *log);
117STATIC void
118xlog_verify_iclog(
119 struct xlog *log,
120 struct xlog_in_core *iclog,
121 int count,
Thiago Farina667a9292012-11-12 21:32:59 -0200122 bool syncing);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500123STATIC void
124xlog_verify_tail_lsn(
125 struct xlog *log,
126 struct xlog_in_core *iclog,
127 xfs_lsn_t tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128#else
129#define xlog_verify_dest_ptr(a,b)
Dave Chinner3f336c62010-12-21 12:02:52 +1100130#define xlog_verify_grant_tail(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131#define xlog_verify_iclog(a,b,c,d)
132#define xlog_verify_tail_lsn(a,b,c)
133#endif
134
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500135STATIC int
136xlog_iclogs_empty(
137 struct xlog *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100139static void
Dave Chinner663e4962010-12-21 12:06:05 +1100140xlog_grant_sub_space(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500141 struct xlog *log,
142 atomic64_t *head,
143 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100144{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100145 int64_t head_val = atomic64_read(head);
146 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100147
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100148 do {
149 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100150
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100151 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100152
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100153 space -= bytes;
154 if (space < 0) {
155 space += log->l_logsize;
156 cycle--;
157 }
158
159 old = head_val;
160 new = xlog_assign_grant_head_val(cycle, space);
161 head_val = atomic64_cmpxchg(head, old, new);
162 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100163}
164
165static void
Dave Chinner663e4962010-12-21 12:06:05 +1100166xlog_grant_add_space(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500167 struct xlog *log,
168 atomic64_t *head,
169 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100170{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100171 int64_t head_val = atomic64_read(head);
172 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100173
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100174 do {
175 int tmp;
176 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100177
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100178 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100179
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100180 tmp = log->l_logsize - space;
181 if (tmp > bytes)
182 space += bytes;
183 else {
184 space = bytes - tmp;
185 cycle++;
186 }
187
188 old = head_val;
189 new = xlog_assign_grant_head_val(cycle, space);
190 head_val = atomic64_cmpxchg(head, old, new);
191 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100192}
Dave Chinnera69ed032010-12-21 12:08:20 +1100193
Christoph Hellwigc303c5b2012-02-20 02:31:26 +0000194STATIC void
195xlog_grant_head_init(
196 struct xlog_grant_head *head)
197{
198 xlog_assign_grant_head(&head->grant, 1, 0);
199 INIT_LIST_HEAD(&head->waiters);
200 spin_lock_init(&head->lock);
201}
202
Christoph Hellwiga79bf2d2012-02-20 02:31:27 +0000203STATIC void
204xlog_grant_head_wake_all(
205 struct xlog_grant_head *head)
206{
207 struct xlog_ticket *tic;
208
209 spin_lock(&head->lock);
210 list_for_each_entry(tic, &head->waiters, t_queue)
211 wake_up_process(tic->t_task);
212 spin_unlock(&head->lock);
213}
214
Christoph Hellwige179840d2012-02-20 02:31:29 +0000215static inline int
216xlog_ticket_reservation(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500217 struct xlog *log,
Christoph Hellwige179840d2012-02-20 02:31:29 +0000218 struct xlog_grant_head *head,
219 struct xlog_ticket *tic)
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000220{
Christoph Hellwige179840d2012-02-20 02:31:29 +0000221 if (head == &log->l_write_head) {
222 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
223 return tic->t_unit_res;
224 } else {
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000225 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
Christoph Hellwige179840d2012-02-20 02:31:29 +0000226 return tic->t_unit_res * tic->t_cnt;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000227 else
Christoph Hellwige179840d2012-02-20 02:31:29 +0000228 return tic->t_unit_res;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000229 }
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000230}
231
232STATIC bool
Christoph Hellwige179840d2012-02-20 02:31:29 +0000233xlog_grant_head_wake(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500234 struct xlog *log,
Christoph Hellwige179840d2012-02-20 02:31:29 +0000235 struct xlog_grant_head *head,
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000236 int *free_bytes)
237{
238 struct xlog_ticket *tic;
239 int need_bytes;
240
Christoph Hellwige179840d2012-02-20 02:31:29 +0000241 list_for_each_entry(tic, &head->waiters, t_queue) {
242 need_bytes = xlog_ticket_reservation(log, head, tic);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000243 if (*free_bytes < need_bytes)
244 return false;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000245
Christoph Hellwige179840d2012-02-20 02:31:29 +0000246 *free_bytes -= need_bytes;
247 trace_xfs_log_grant_wake_up(log, tic);
Christoph Hellwig14a72352012-02-20 02:31:24 +0000248 wake_up_process(tic->t_task);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000249 }
250
251 return true;
252}
253
254STATIC int
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000255xlog_grant_head_wait(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500256 struct xlog *log,
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000257 struct xlog_grant_head *head,
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000258 struct xlog_ticket *tic,
Dave Chinnera30b0362013-09-02 20:49:36 +1000259 int need_bytes) __releases(&head->lock)
260 __acquires(&head->lock)
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000261{
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000262 list_add_tail(&tic->t_queue, &head->waiters);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000263
264 do {
265 if (XLOG_FORCED_SHUTDOWN(log))
266 goto shutdown;
267 xlog_grant_push_ail(log, need_bytes);
268
Christoph Hellwig14a72352012-02-20 02:31:24 +0000269 __set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000270 spin_unlock(&head->lock);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000271
Christoph Hellwig14a72352012-02-20 02:31:24 +0000272 XFS_STATS_INC(xs_sleep_logspace);
273
274 trace_xfs_log_grant_sleep(log, tic);
275 schedule();
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000276 trace_xfs_log_grant_wake(log, tic);
277
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000278 spin_lock(&head->lock);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000279 if (XLOG_FORCED_SHUTDOWN(log))
280 goto shutdown;
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000281 } while (xlog_space_left(log, &head->grant) < need_bytes);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000282
283 list_del_init(&tic->t_queue);
284 return 0;
285shutdown:
286 list_del_init(&tic->t_queue);
Dave Chinner24513372014-06-25 14:58:08 +1000287 return -EIO;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000288}
289
Christoph Hellwig42ceedb2012-02-20 02:31:30 +0000290/*
291 * Atomically get the log space required for a log ticket.
292 *
293 * Once a ticket gets put onto head->waiters, it will only return after the
294 * needed reservation is satisfied.
295 *
296 * This function is structured so that it has a lock free fast path. This is
297 * necessary because every new transaction reservation will come through this
298 * path. Hence any lock will be globally hot if we take it unconditionally on
299 * every pass.
300 *
301 * As tickets are only ever moved on and off head->waiters under head->lock, we
302 * only need to take that lock if we are going to add the ticket to the queue
303 * and sleep. We can avoid taking the lock if the ticket was never added to
304 * head->waiters because the t_queue list head will be empty and we hold the
305 * only reference to it so it can safely be checked unlocked.
306 */
307STATIC int
308xlog_grant_head_check(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500309 struct xlog *log,
Christoph Hellwig42ceedb2012-02-20 02:31:30 +0000310 struct xlog_grant_head *head,
311 struct xlog_ticket *tic,
312 int *need_bytes)
313{
314 int free_bytes;
315 int error = 0;
316
317 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
318
319 /*
320 * If there are other waiters on the queue then give them a chance at
321 * logspace before us. Wake up the first waiters, if we do not wake
322 * up all the waiters then go to sleep waiting for more free space,
323 * otherwise try to get some space for this transaction.
324 */
325 *need_bytes = xlog_ticket_reservation(log, head, tic);
326 free_bytes = xlog_space_left(log, &head->grant);
327 if (!list_empty_careful(&head->waiters)) {
328 spin_lock(&head->lock);
329 if (!xlog_grant_head_wake(log, head, &free_bytes) ||
330 free_bytes < *need_bytes) {
331 error = xlog_grant_head_wait(log, head, tic,
332 *need_bytes);
333 }
334 spin_unlock(&head->lock);
335 } else if (free_bytes < *need_bytes) {
336 spin_lock(&head->lock);
337 error = xlog_grant_head_wait(log, head, tic, *need_bytes);
338 spin_unlock(&head->lock);
339 }
340
341 return error;
342}
343
Christoph Hellwig0adba532007-08-30 17:21:46 +1000344static void
345xlog_tic_reset_res(xlog_ticket_t *tic)
346{
347 tic->t_res_num = 0;
348 tic->t_res_arr_sum = 0;
349 tic->t_res_num_ophdrs = 0;
350}
351
352static void
353xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
354{
355 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
356 /* add to overflow and start again */
357 tic->t_res_o_flow += tic->t_res_arr_sum;
358 tic->t_res_num = 0;
359 tic->t_res_arr_sum = 0;
360 }
361
362 tic->t_res_arr[tic->t_res_num].r_len = len;
363 tic->t_res_arr[tic->t_res_num].r_type = type;
364 tic->t_res_arr_sum += len;
365 tic->t_res_num++;
366}
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368/*
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000369 * Replenish the byte reservation required by moving the grant write head.
370 */
371int
372xfs_log_regrant(
373 struct xfs_mount *mp,
374 struct xlog_ticket *tic)
375{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500376 struct xlog *log = mp->m_log;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000377 int need_bytes;
378 int error = 0;
379
380 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner24513372014-06-25 14:58:08 +1000381 return -EIO;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000382
383 XFS_STATS_INC(xs_try_logspace);
384
385 /*
386 * This is a new transaction on the ticket, so we need to change the
387 * transaction ID so that the next transaction has a different TID in
388 * the log. Just add one to the existing tid so that we can see chains
389 * of rolling transactions in the log easily.
390 */
391 tic->t_tid++;
392
393 xlog_grant_push_ail(log, tic->t_unit_res);
394
395 tic->t_curr_res = tic->t_unit_res;
396 xlog_tic_reset_res(tic);
397
398 if (tic->t_cnt > 0)
399 return 0;
400
401 trace_xfs_log_regrant(log, tic);
402
403 error = xlog_grant_head_check(log, &log->l_write_head, tic,
404 &need_bytes);
405 if (error)
406 goto out_error;
407
408 xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
409 trace_xfs_log_regrant_exit(log, tic);
410 xlog_verify_grant_tail(log);
411 return 0;
412
413out_error:
414 /*
415 * If we are failing, make sure the ticket doesn't have any current
416 * reservations. We don't want to add this back when the ticket/
417 * transaction gets cancelled.
418 */
419 tic->t_curr_res = 0;
420 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
421 return error;
422}
423
424/*
425 * Reserve log space and return a ticket corresponding the reservation.
426 *
427 * Each reservation is going to reserve extra space for a log record header.
428 * When writes happen to the on-disk log, we don't subtract the length of the
429 * log record header from any reservation. By wasting space in each
430 * reservation, we prevent over allocation problems.
431 */
432int
433xfs_log_reserve(
434 struct xfs_mount *mp,
435 int unit_bytes,
436 int cnt,
437 struct xlog_ticket **ticp,
438 __uint8_t client,
439 bool permanent,
440 uint t_type)
441{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500442 struct xlog *log = mp->m_log;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000443 struct xlog_ticket *tic;
444 int need_bytes;
445 int error = 0;
446
447 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
448
449 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner24513372014-06-25 14:58:08 +1000450 return -EIO;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000451
452 XFS_STATS_INC(xs_try_logspace);
453
454 ASSERT(*ticp == NULL);
455 tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
456 KM_SLEEP | KM_MAYFAIL);
457 if (!tic)
Dave Chinner24513372014-06-25 14:58:08 +1000458 return -ENOMEM;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000459
460 tic->t_trans_type = t_type;
461 *ticp = tic;
462
Dave Chinner437a2552012-11-28 13:01:00 +1100463 xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
464 : tic->t_unit_res);
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000465
466 trace_xfs_log_reserve(log, tic);
467
468 error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
469 &need_bytes);
470 if (error)
471 goto out_error;
472
473 xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
474 xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
475 trace_xfs_log_reserve_exit(log, tic);
476 xlog_verify_grant_tail(log);
477 return 0;
478
479out_error:
480 /*
481 * If we are failing, make sure the ticket doesn't have any current
482 * reservations. We don't want to add this back when the ticket/
483 * transaction gets cancelled.
484 */
485 tic->t_curr_res = 0;
486 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
487 return error;
488}
489
490
491/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 * NOTES:
493 *
494 * 1. currblock field gets updated at startup and after in-core logs
495 * marked as with WANT_SYNC.
496 */
497
498/*
499 * This routine is called when a user of a log manager ticket is done with
500 * the reservation. If the ticket was ever used, then a commit record for
501 * the associated transaction is written out as a log operation header with
502 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
503 * a given ticket. If the ticket was one with a permanent reservation, then
504 * a few operations are done differently. Permanent reservation tickets by
505 * default don't release the reservation. They just commit the current
506 * transaction with the belief that the reservation is still needed. A flag
507 * must be passed in before permanent reservations are actually released.
508 * When these type of tickets are not released, they need to be set into
509 * the inited state again. By doing this, a start record will be written
510 * out when the next write occurs.
511 */
512xfs_lsn_t
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000513xfs_log_done(
514 struct xfs_mount *mp,
515 struct xlog_ticket *ticket,
516 struct xlog_in_core **iclog,
517 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500519 struct xlog *log = mp->m_log;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000520 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (XLOG_FORCED_SHUTDOWN(log) ||
523 /*
524 * If nothing was ever written, don't write out commit record.
525 * If we get an error, just continue and give back the log ticket.
526 */
527 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
Dave Chinner55b66332010-03-23 11:43:17 +1100528 (xlog_commit_record(log, ticket, iclog, &lsn)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 lsn = (xfs_lsn_t) -1;
530 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
531 flags |= XFS_LOG_REL_PERM_RESERV;
532 }
533 }
534
535
536 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
537 (flags & XFS_LOG_REL_PERM_RESERV)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000538 trace_xfs_log_done_nonperm(log, ticket);
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000541 * Release ticket if not permanent reservation or a specific
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 * request has been made to release a permanent reservation.
543 */
544 xlog_ungrant_log_space(log, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100545 xfs_log_ticket_put(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000547 trace_xfs_log_done_perm(log, ticket);
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 xlog_regrant_reserve_log_space(log, ticket);
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000550 /* If this ticket was a permanent reservation and we aren't
551 * trying to release it, reset the inited flags; so next time
552 * we write, a start record will be written out.
553 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 ticket->t_flags |= XLOG_TIC_INITED;
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 return lsn;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000558}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560/*
561 * Attaches a new iclog I/O completion callback routine during
562 * transaction commit. If the log is in error state, a non-zero
563 * return code is handed back and the caller is responsible for
564 * executing the callback at an appropriate time.
565 */
566int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000567xfs_log_notify(
568 struct xfs_mount *mp,
569 struct xlog_in_core *iclog,
570 xfs_log_callback_t *cb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000572 int abortflg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
David Chinner114d23a2008-04-10 12:18:39 +1000574 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
576 if (!abortflg) {
577 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
578 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
579 cb->cb_next = NULL;
580 *(iclog->ic_callback_tail) = cb;
581 iclog->ic_callback_tail = &(cb->cb_next);
582 }
David Chinner114d23a2008-04-10 12:18:39 +1000583 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return abortflg;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000585}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000588xfs_log_release_iclog(
589 struct xfs_mount *mp,
590 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000592 if (xlog_state_release_iclog(mp->m_log, iclog)) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000593 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Dave Chinner24513372014-06-25 14:58:08 +1000594 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596
597 return 0;
598}
599
600/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 * Mount a log filesystem
602 *
603 * mp - ubiquitous xfs mount point structure
604 * log_target - buftarg of on-disk log device
605 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
606 * num_bblocks - Number of BBSIZE blocks in on-disk log
607 *
608 * Return error or zero.
609 */
610int
David Chinner249a8c12008-02-05 12:13:32 +1100611xfs_log_mount(
612 xfs_mount_t *mp,
613 xfs_buftarg_t *log_target,
614 xfs_daddr_t blk_offset,
615 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Jie Liu3e7b91c2013-08-12 20:50:03 +1000617 int error = 0;
618 int min_logfsbs;
David Chinner249a8c12008-02-05 12:13:32 +1100619
Dave Chinnerc99d6092014-05-05 16:18:37 +1000620 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
621 xfs_notice(mp, "Mounting V%d Filesystem",
622 XFS_SB_VERSION_NUM(&mp->m_sb));
623 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100624 xfs_notice(mp,
Dave Chinnerc99d6092014-05-05 16:18:37 +1000625"Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.",
626 XFS_SB_VERSION_NUM(&mp->m_sb));
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000627 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629
630 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200631 if (IS_ERR(mp->m_log)) {
Dave Chinner24513372014-06-25 14:58:08 +1000632 error = PTR_ERR(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100633 goto out;
634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 /*
Jie Liu3e7b91c2013-08-12 20:50:03 +1000637 * Validate the given log space and drop a critical message via syslog
638 * if the log size is too small that would lead to some unexpected
639 * situations in transaction log space reservation stage.
640 *
641 * Note: we can't just reject the mount if the validation fails. This
642 * would mean that people would have to downgrade their kernel just to
643 * remedy the situation as there is no way to grow the log (short of
644 * black magic surgery with xfs_db).
645 *
646 * We can, however, reject mounts for CRC format filesystems, as the
647 * mkfs binary being used to make the filesystem should never create a
648 * filesystem with a log that is too small.
649 */
650 min_logfsbs = xfs_log_calc_minimum_size(mp);
651
652 if (mp->m_sb.sb_logblocks < min_logfsbs) {
653 xfs_warn(mp,
654 "Log size %d blocks too small, minimum size is %d blocks",
655 mp->m_sb.sb_logblocks, min_logfsbs);
Dave Chinner24513372014-06-25 14:58:08 +1000656 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000657 } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) {
658 xfs_warn(mp,
659 "Log size %d blocks too large, maximum size is %lld blocks",
660 mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS);
Dave Chinner24513372014-06-25 14:58:08 +1000661 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000662 } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) {
663 xfs_warn(mp,
664 "log size %lld bytes too large, maximum size is %lld bytes",
665 XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks),
666 XFS_MAX_LOG_BYTES);
Dave Chinner24513372014-06-25 14:58:08 +1000667 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000668 }
669 if (error) {
670 if (xfs_sb_version_hascrc(&mp->m_sb)) {
671 xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
672 ASSERT(0);
673 goto out_free_log;
674 }
675 xfs_crit(mp,
676"Log size out of supported range. Continuing onwards, but if log hangs are\n"
677"experienced then please report this message in the bug report.");
678 }
679
680 /*
David Chinner249a8c12008-02-05 12:13:32 +1100681 * Initialize the AIL now we have a log.
682 */
David Chinner249a8c12008-02-05 12:13:32 +1100683 error = xfs_trans_ail_init(mp);
684 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100685 xfs_warn(mp, "AIL initialisation failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100686 goto out_free_log;
David Chinner249a8c12008-02-05 12:13:32 +1100687 }
David Chinnera9c21c12008-10-30 17:39:35 +1100688 mp->m_log->l_ailp = mp->m_ail;
David Chinner249a8c12008-02-05 12:13:32 +1100689
690 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 * skip log recovery on a norecovery mount. pretend it all
692 * just worked.
693 */
694 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
David Chinner249a8c12008-02-05 12:13:32 +1100695 int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000698 mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Eric Sandeen65be6052006-01-11 15:34:19 +1100700 error = xlog_recover(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000703 mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100705 xfs_warn(mp, "log mount/recovery failed: error %d",
706 error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100707 goto out_destroy_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
709 }
710
Brian Fosterbaff4e42014-07-15 08:07:29 +1000711 error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj,
712 "log");
713 if (error)
714 goto out_destroy_ail;
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 /* Normal transactions can now occur */
717 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
718
Dave Chinner71e330b2010-05-21 14:37:18 +1000719 /*
720 * Now the log has been fully initialised and we know were our
721 * space grant counters are, we can initialise the permanent ticket
722 * needed for delayed logging to work.
723 */
724 xlog_cil_init_post_recovery(mp->m_log);
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 return 0;
Christoph Hellwig26430752009-02-12 19:55:48 +0100727
728out_destroy_ail:
729 xfs_trans_ail_destroy(mp);
730out_free_log:
731 xlog_dealloc_log(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100732out:
David Chinner249a8c12008-02-05 12:13:32 +1100733 return error;
Christoph Hellwig26430752009-02-12 19:55:48 +0100734}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736/*
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100737 * Finish the recovery of the file system. This is separate from the
738 * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
739 * in the root and real-time bitmap inodes between calling xfs_log_mount() and
740 * here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 *
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100742 * If we finish recovery successfully, start the background log work. If we are
743 * not doing recovery, then we have a RO filesystem and we don't need to start
744 * it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 */
746int
Christoph Hellwig42490232008-08-13 16:49:32 +1000747xfs_log_mount_finish(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100749 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100751 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
Christoph Hellwig42490232008-08-13 16:49:32 +1000752 error = xlog_recover_finish(mp->m_log);
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100753 if (!error)
754 xfs_log_work_queue(mp);
755 } else {
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000756 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
758
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 return error;
761}
762
763/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 * Final log writes as part of unmount.
765 *
766 * Mark the filesystem clean as unmount happens. Note that during relocation
767 * this routine needs to be executed as part of source-bag while the
768 * deallocation must not be done until source-end.
769 */
770
771/*
772 * Unmount record used to have a string "Unmount filesystem--" in the
773 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
774 * We just write the magic number now since that particular field isn't
Zhi Yong Wu8e159e72013-08-12 03:15:00 +0000775 * currently architecture converted and "Unmount" is a bit foo.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 * As far as I know, there weren't any dependencies on the old behaviour.
777 */
778
779int
780xfs_log_unmount_write(xfs_mount_t *mp)
781{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500782 struct xlog *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 xlog_in_core_t *iclog;
784#ifdef DEBUG
785 xlog_in_core_t *first_iclog;
786#endif
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000787 xlog_ticket_t *tic = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 xfs_lsn_t lsn;
789 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /*
792 * Don't write out unmount record on read-only mounts.
793 * Or, if we are doing a forced umount (typically because of IO errors).
794 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000795 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return 0;
797
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000798 error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
David Chinnerb911ca02008-04-10 12:24:30 +1000799 ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801#ifdef DEBUG
802 first_iclog = iclog = log->l_iclog;
803 do {
804 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
805 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
806 ASSERT(iclog->ic_offset == 0);
807 }
808 iclog = iclog->ic_next;
809 } while (iclog != first_iclog);
810#endif
811 if (! (XLOG_FORCED_SHUTDOWN(log))) {
Tim Shimmin955e47a2006-09-28 11:04:16 +1000812 error = xfs_log_reserve(mp, 600, 1, &tic,
813 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 if (!error) {
Dave Chinner55b66332010-03-23 11:43:17 +1100815 /* the data section must be 32 bit size aligned */
816 struct {
817 __uint16_t magic;
818 __uint16_t pad1;
819 __uint32_t pad2; /* may as well make it 64 bits */
820 } magic = {
821 .magic = XLOG_UNMOUNT_TYPE,
822 };
823 struct xfs_log_iovec reg = {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000824 .i_addr = &magic,
Dave Chinner55b66332010-03-23 11:43:17 +1100825 .i_len = sizeof(magic),
826 .i_type = XLOG_REG_TYPE_UNMOUNT,
827 };
828 struct xfs_log_vec vec = {
829 .lv_niovecs = 1,
830 .lv_iovecp = &reg,
831 };
832
Dave Chinner39486592012-03-22 05:15:11 +0000833 /* remove inited flag, and account for space used */
Dave Chinner55b66332010-03-23 11:43:17 +1100834 tic->t_flags = 0;
Dave Chinner39486592012-03-22 05:15:11 +0000835 tic->t_curr_res -= sizeof(magic);
Dave Chinner55b66332010-03-23 11:43:17 +1100836 error = xlog_write(log, &vec, tic, &lsn,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 NULL, XLOG_UNMOUNT_TRANS);
838 /*
839 * At this point, we're umounting anyway,
840 * so there's no point in transitioning log state
841 * to IOERROR. Just continue...
842 */
843 }
844
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100845 if (error)
846 xfs_alert(mp, "%s: unmount record failed", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000849 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100851 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100853 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000854 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000856 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
858 iclog->ic_state == XLOG_STATE_DIRTY)) {
859 if (!XLOG_FORCED_SHUTDOWN(log)) {
Dave Chinnereb40a872010-12-21 12:09:01 +1100860 xlog_wait(&iclog->ic_force_wait,
861 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000863 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 }
865 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000866 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 }
Tim Shimmin955e47a2006-09-28 11:04:16 +1000868 if (tic) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000869 trace_xfs_log_umount_write(log, tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000870 xlog_ungrant_log_space(log, tic);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100871 xfs_log_ticket_put(tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 } else {
874 /*
875 * We're already in forced_shutdown mode, couldn't
876 * even attempt to write out the unmount transaction.
877 *
878 * Go through the motions of sync'ing and releasing
879 * the iclog, even though no I/O will actually happen,
Nathan Scottc41564b2006-03-29 08:55:14 +1000880 * we need to wait for other log I/Os that may already
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 * be in progress. Do this as a separate section of
882 * code so we'll know if we ever get stuck here that
883 * we're in this odd situation of trying to unmount
884 * a file system that went into forced_shutdown as
885 * the result of an unmount..
886 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000887 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100889 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
891 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100892 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000893 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000895 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
898 || iclog->ic_state == XLOG_STATE_DIRTY
899 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
900
Dave Chinnereb40a872010-12-21 12:09:01 +1100901 xlog_wait(&iclog->ic_force_wait,
902 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000904 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 }
906 }
907
David Chinner1bb7d6b2008-04-10 12:24:38 +1000908 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909} /* xfs_log_unmount_write */
910
911/*
Dave Chinnerc75921a2012-10-08 21:56:08 +1100912 * Empty the log for unmount/freeze.
Dave Chinnercf2931d2012-10-08 21:56:03 +1100913 *
914 * To do this, we first need to shut down the background log work so it is not
915 * trying to cover the log as we clean up. We then need to unpin all objects in
916 * the log so we can then flush them out. Once they have completed their IO and
917 * run the callbacks removing themselves from the AIL, we can write the unmount
Dave Chinnerc75921a2012-10-08 21:56:08 +1100918 * record.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 */
920void
Dave Chinnerc75921a2012-10-08 21:56:08 +1100921xfs_log_quiesce(
922 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923{
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100924 cancel_delayed_work_sync(&mp->m_log->l_work);
Dave Chinnercf2931d2012-10-08 21:56:03 +1100925 xfs_log_force(mp, XFS_LOG_SYNC);
926
927 /*
928 * The superblock buffer is uncached and while xfs_ail_push_all_sync()
929 * will push it, xfs_wait_buftarg() will not wait for it. Further,
930 * xfs_buf_iowait() cannot be used because it was pushed with the
931 * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
932 * the IO to complete.
933 */
934 xfs_ail_push_all_sync(mp->m_ail);
935 xfs_wait_buftarg(mp->m_ddev_targp);
936 xfs_buf_lock(mp->m_sb_bp);
937 xfs_buf_unlock(mp->m_sb_bp);
938
939 xfs_log_unmount_write(mp);
Dave Chinnerc75921a2012-10-08 21:56:08 +1100940}
941
942/*
943 * Shut down and release the AIL and Log.
944 *
945 * During unmount, we need to ensure we flush all the dirty metadata objects
946 * from the AIL so that the log is empty before we write the unmount record to
947 * the log. Once this is done, we can tear down the AIL and the log.
948 */
949void
950xfs_log_unmount(
951 struct xfs_mount *mp)
952{
953 xfs_log_quiesce(mp);
Dave Chinnercf2931d2012-10-08 21:56:03 +1100954
David Chinner249a8c12008-02-05 12:13:32 +1100955 xfs_trans_ail_destroy(mp);
Brian Fosterbaff4e42014-07-15 08:07:29 +1000956
957 xfs_sysfs_del(&mp->m_log->l_kobj);
958
Nathan Scottc41564b2006-03-29 08:55:14 +1000959 xlog_dealloc_log(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
961
Dave Chinner43f5efc2010-03-23 10:10:00 +1100962void
963xfs_log_item_init(
964 struct xfs_mount *mp,
965 struct xfs_log_item *item,
966 int type,
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000967 const struct xfs_item_ops *ops)
Dave Chinner43f5efc2010-03-23 10:10:00 +1100968{
969 item->li_mountp = mp;
970 item->li_ailp = mp->m_ail;
971 item->li_type = type;
972 item->li_ops = ops;
Dave Chinner71e330b2010-05-21 14:37:18 +1000973 item->li_lv = NULL;
974
975 INIT_LIST_HEAD(&item->li_ail);
976 INIT_LIST_HEAD(&item->li_cil);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100977}
978
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000979/*
980 * Wake up processes waiting for log space after we have moved the log tail.
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000981 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982void
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000983xfs_log_space_wake(
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000984 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500986 struct xlog *log = mp->m_log;
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000987 int free_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (XLOG_FORCED_SHUTDOWN(log))
990 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Christoph Hellwig28496962012-02-20 02:31:25 +0000992 if (!list_empty_careful(&log->l_write_head.waiters)) {
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000993 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
994
Christoph Hellwig28496962012-02-20 02:31:25 +0000995 spin_lock(&log->l_write_head.lock);
996 free_bytes = xlog_space_left(log, &log->l_write_head.grant);
Christoph Hellwige179840d2012-02-20 02:31:29 +0000997 xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
Christoph Hellwig28496962012-02-20 02:31:25 +0000998 spin_unlock(&log->l_write_head.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 }
Dave Chinner10547942010-12-21 12:02:25 +11001000
Christoph Hellwig28496962012-02-20 02:31:25 +00001001 if (!list_empty_careful(&log->l_reserve_head.waiters)) {
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001002 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
1003
Christoph Hellwig28496962012-02-20 02:31:25 +00001004 spin_lock(&log->l_reserve_head.lock);
1005 free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
Christoph Hellwige179840d2012-02-20 02:31:29 +00001006 xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
Christoph Hellwig28496962012-02-20 02:31:25 +00001007 spin_unlock(&log->l_reserve_head.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
Dave Chinner3f16b982010-12-21 12:29:01 +11001009}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011/*
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001012 * Determine if we have a transaction that has gone to disk that needs to be
1013 * covered. To begin the transition to the idle state firstly the log needs to
1014 * be idle. That means the CIL, the AIL and the iclogs needs to be empty before
1015 * we start attempting to cover the log.
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001016 *
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001017 * Only if we are then in a state where covering is needed, the caller is
1018 * informed that dummy transactions are required to move the log into the idle
1019 * state.
1020 *
1021 * If there are any items in the AIl or CIL, then we do not want to attempt to
1022 * cover the log as we may be in a situation where there isn't log space
1023 * available to run a dummy transaction and this can lead to deadlocks when the
1024 * tail of the log is pinned by an item that is modified in the CIL. Hence
1025 * there's no point in running a dummy transaction at this point because we
1026 * can't start trying to idle the log until both the CIL and AIL are empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 */
1028int
1029xfs_log_need_covered(xfs_mount_t *mp)
1030{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001031 struct xlog *log = mp->m_log;
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001032 int needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
David Chinner92821e22007-05-24 15:26:31 +10001034 if (!xfs_fs_writable(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 return 0;
1036
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001037 if (!xlog_cil_empty(log))
1038 return 0;
1039
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001040 spin_lock(&log->l_icloglock);
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001041 switch (log->l_covered_state) {
1042 case XLOG_STATE_COVER_DONE:
1043 case XLOG_STATE_COVER_DONE2:
1044 case XLOG_STATE_COVER_IDLE:
1045 break;
1046 case XLOG_STATE_COVER_NEED:
1047 case XLOG_STATE_COVER_NEED2:
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001048 if (xfs_ail_min_lsn(log->l_ailp))
1049 break;
1050 if (!xlog_iclogs_empty(log))
1051 break;
1052
1053 needed = 1;
1054 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
1055 log->l_covered_state = XLOG_STATE_COVER_DONE;
1056 else
1057 log->l_covered_state = XLOG_STATE_COVER_DONE2;
1058 break;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001059 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 needed = 1;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001061 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001063 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001064 return needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065}
1066
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001067/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 * We may be holding the log iclog lock upon entering this routine.
1069 */
1070xfs_lsn_t
Christoph Hellwig1c304622012-04-23 15:58:33 +10001071xlog_assign_tail_lsn_locked(
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001072 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Mark Tinguelyad223e62012-06-14 09:22:15 -05001074 struct xlog *log = mp->m_log;
Christoph Hellwig1c304622012-04-23 15:58:33 +10001075 struct xfs_log_item *lip;
1076 xfs_lsn_t tail_lsn;
1077
1078 assert_spin_locked(&mp->m_ail->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001080 /*
1081 * To make sure we always have a valid LSN for the log tail we keep
1082 * track of the last LSN which was committed in log->l_last_sync_lsn,
Christoph Hellwig1c304622012-04-23 15:58:33 +10001083 * and use that when the AIL was empty.
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001084 */
Christoph Hellwig1c304622012-04-23 15:58:33 +10001085 lip = xfs_ail_min(mp->m_ail);
1086 if (lip)
1087 tail_lsn = lip->li_lsn;
1088 else
Dave Chinner84f3c682010-12-03 22:11:29 +11001089 tail_lsn = atomic64_read(&log->l_last_sync_lsn);
Dave Chinner750b9c92013-11-01 15:27:18 +11001090 trace_xfs_log_assign_tail_lsn(log, tail_lsn);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001091 atomic64_set(&log->l_tail_lsn, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 return tail_lsn;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001093}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
Christoph Hellwig1c304622012-04-23 15:58:33 +10001095xfs_lsn_t
1096xlog_assign_tail_lsn(
1097 struct xfs_mount *mp)
1098{
1099 xfs_lsn_t tail_lsn;
1100
1101 spin_lock(&mp->m_ail->xa_lock);
1102 tail_lsn = xlog_assign_tail_lsn_locked(mp);
1103 spin_unlock(&mp->m_ail->xa_lock);
1104
1105 return tail_lsn;
1106}
1107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108/*
1109 * Return the space in the log between the tail and the head. The head
1110 * is passed in the cycle/bytes formal parms. In the special case where
1111 * the reserve head has wrapped passed the tail, this calculation is no
1112 * longer valid. In this case, just return 0 which means there is no space
1113 * in the log. This works for all places where this function is called
1114 * with the reserve head. Of course, if the write head were to ever
1115 * wrap the tail, we should blow up. Rather than catch this case here,
1116 * we depend on other ASSERTions in other parts of the code. XXXmiken
1117 *
1118 * This code also handles the case where the reservation head is behind
1119 * the tail. The details of this case are described below, but the end
1120 * result is that we return the size of the log as the amount of space left.
1121 */
David Chinnera8272ce2007-11-23 16:28:09 +11001122STATIC int
Dave Chinnera69ed032010-12-21 12:08:20 +11001123xlog_space_left(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001124 struct xlog *log,
Dave Chinnerc8a09ff2010-12-04 00:02:40 +11001125 atomic64_t *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126{
Dave Chinnera69ed032010-12-21 12:08:20 +11001127 int free_bytes;
1128 int tail_bytes;
1129 int tail_cycle;
1130 int head_cycle;
1131 int head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Dave Chinnera69ed032010-12-21 12:08:20 +11001133 xlog_crack_grant_head(head, &head_cycle, &head_bytes);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001134 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
1135 tail_bytes = BBTOB(tail_bytes);
Dave Chinnera69ed032010-12-21 12:08:20 +11001136 if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
1137 free_bytes = log->l_logsize - (head_bytes - tail_bytes);
1138 else if (tail_cycle + 1 < head_cycle)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 return 0;
Dave Chinnera69ed032010-12-21 12:08:20 +11001140 else if (tail_cycle < head_cycle) {
1141 ASSERT(tail_cycle == (head_cycle - 1));
1142 free_bytes = tail_bytes - head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 } else {
1144 /*
1145 * The reservation head is behind the tail.
1146 * In this case we just want to return the size of the
1147 * log as the amount of space left.
1148 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001149 xfs_alert(log->l_mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 "xlog_space_left: head behind tail\n"
1151 " tail_cycle = %d, tail_bytes = %d\n"
1152 " GH cycle = %d, GH bytes = %d",
Dave Chinnera69ed032010-12-21 12:08:20 +11001153 tail_cycle, tail_bytes, head_cycle, head_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 ASSERT(0);
1155 free_bytes = log->l_logsize;
1156 }
1157 return free_bytes;
Dave Chinnera69ed032010-12-21 12:08:20 +11001158}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160
1161/*
1162 * Log function which is called when an io completes.
1163 *
1164 * The log manager needs its own routine, in order to control what
1165 * happens with the buffer after the write completes.
1166 */
1167void
1168xlog_iodone(xfs_buf_t *bp)
1169{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001170 struct xlog_in_core *iclog = bp->b_fspriv;
1171 struct xlog *l = iclog->ic_log;
1172 int aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 /*
1175 * Race to shutdown the filesystem if we see an error.
1176 */
Dave Chinner36de9552014-06-06 16:02:12 +10001177 if (XFS_TEST_ERROR(bp->b_error, l->l_mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00001179 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001180 xfs_buf_stale(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001181 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /*
1183 * This flag will be propagated to the trans-committed
1184 * callback routines to let them know that the log-commit
1185 * didn't succeed.
1186 */
1187 aborted = XFS_LI_ABORTED;
1188 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
1189 aborted = XFS_LI_ABORTED;
1190 }
David Chinner3db296f2007-05-14 18:24:16 +10001191
1192 /* log I/O is always issued ASYNC */
1193 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 xlog_state_done_syncing(iclog, aborted);
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001195
David Chinner3db296f2007-05-14 18:24:16 +10001196 /*
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001197 * drop the buffer lock now that we are done. Nothing references
1198 * the buffer after this, so an unmount waiting on this lock can now
1199 * tear it down safely. As such, it is unsafe to reference the buffer
1200 * (bp) after the unlock as we could race with it being freed.
David Chinner3db296f2007-05-14 18:24:16 +10001201 */
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001202 xfs_buf_unlock(bp);
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11001203}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
1205/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 * Return size of each in-core log record buffer.
1207 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001208 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 *
1210 * If the filesystem blocksize is too large, we may need to choose a
1211 * larger size since the directory code currently logs entire blocks.
1212 */
1213
1214STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001215xlog_get_iclog_buffer_size(
1216 struct xfs_mount *mp,
1217 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
1219 int size;
1220 int xhdrs;
1221
Eric Sandeen1cb51252007-08-16 16:24:43 +10001222 if (mp->m_logbufs <= 0)
1223 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
1224 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001225 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 /*
1228 * Buffer size passed in from mount system call.
1229 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001230 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 size = log->l_iclog_size = mp->m_logbsize;
1232 log->l_iclog_size_log = 0;
1233 while (size != 1) {
1234 log->l_iclog_size_log++;
1235 size >>= 1;
1236 }
1237
Eric Sandeen62118702008-03-06 13:44:28 +11001238 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001239 /* # headers = size / 32k
1240 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 */
1242
1243 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
1244 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1245 xhdrs++;
1246 log->l_iclog_hsize = xhdrs << BBSHIFT;
1247 log->l_iclog_heads = xhdrs;
1248 } else {
1249 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1250 log->l_iclog_hsize = BBSIZE;
1251 log->l_iclog_heads = 1;
1252 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001253 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
1255
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001256 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +10001257 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1258 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 /* the default log size is 16k or 32k which is one header sector */
1261 log->l_iclog_hsize = BBSIZE;
1262 log->l_iclog_heads = 1;
1263
Christoph Hellwig7153f8b2009-02-09 08:36:46 +01001264done:
1265 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001266 if (mp->m_logbufs == 0)
1267 mp->m_logbufs = log->l_iclog_bufs;
1268 if (mp->m_logbsize == 0)
1269 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270} /* xlog_get_iclog_buffer_size */
1271
1272
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001273void
1274xfs_log_work_queue(
1275 struct xfs_mount *mp)
1276{
Dave Chinner58896082012-10-08 21:56:05 +11001277 queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work,
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001278 msecs_to_jiffies(xfs_syncd_centisecs * 10));
1279}
1280
1281/*
1282 * Every sync period we need to unpin all items in the AIL and push them to
1283 * disk. If there is nothing dirty, then we might need to cover the log to
1284 * indicate that the filesystem is idle.
1285 */
1286void
1287xfs_log_worker(
1288 struct work_struct *work)
1289{
1290 struct xlog *log = container_of(to_delayed_work(work),
1291 struct xlog, l_work);
1292 struct xfs_mount *mp = log->l_mp;
1293
1294 /* dgc: errors ignored - not fatal and nowhere to report them */
1295 if (xfs_log_need_covered(mp))
1296 xfs_fs_log_dummy(mp);
1297 else
1298 xfs_log_force(mp, 0);
1299
1300 /* start pushing all the metadata that is currently dirty */
1301 xfs_ail_push_all(mp->m_ail);
1302
1303 /* queue us up again */
1304 xfs_log_work_queue(mp);
1305}
1306
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307/*
1308 * This routine initializes some of the log structure for a given mount point.
1309 * Its primary purpose is to fill in enough, so recovery can occur. However,
1310 * some other stuff may be filled in too.
1311 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001312STATIC struct xlog *
1313xlog_alloc_log(
1314 struct xfs_mount *mp,
1315 struct xfs_buftarg *log_target,
1316 xfs_daddr_t blk_offset,
1317 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001319 struct xlog *log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 xlog_rec_header_t *head;
1321 xlog_in_core_t **iclogp;
1322 xlog_in_core_t *iclog, *prev_iclog=NULL;
1323 xfs_buf_t *bp;
1324 int i;
Dave Chinner24513372014-06-25 14:58:08 +10001325 int error = -ENOMEM;
Alex Elder69ce58f2010-04-20 17:09:59 +10001326 uint log2_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001328 log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001329 if (!log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001330 xfs_warn(mp, "Log allocation failed: No memory!");
Dave Chinnera6cb7672009-04-06 18:39:27 +02001331 goto out;
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
1334 log->l_mp = mp;
1335 log->l_targ = log_target;
1336 log->l_logsize = BBTOB(num_bblks);
1337 log->l_logBBstart = blk_offset;
1338 log->l_logBBsize = num_bblks;
1339 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1340 log->l_flags |= XLOG_ACTIVE_RECOVERY;
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001341 INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 log->l_prev_block = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001345 xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1346 xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
Christoph Hellwigc303c5b2012-02-20 02:31:26 +00001348
1349 xlog_grant_head_init(&log->l_reserve_head);
1350 xlog_grant_head_init(&log->l_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
Dave Chinner24513372014-06-25 14:58:08 +10001352 error = -EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001353 if (xfs_sb_version_hassector(&mp->m_sb)) {
Alex Elder69ce58f2010-04-20 17:09:59 +10001354 log2_size = mp->m_sb.sb_logsectlog;
1355 if (log2_size < BBSHIFT) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001356 xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1357 log2_size, BBSHIFT);
Alex Elder69ce58f2010-04-20 17:09:59 +10001358 goto out_free_log;
1359 }
1360
1361 log2_size -= BBSHIFT;
1362 if (log2_size > mp->m_sectbb_log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001363 xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1364 log2_size, mp->m_sectbb_log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001365 goto out_free_log;
1366 }
1367
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 /* for larger sector sizes, must have v2 or external log */
Alex Elder69ce58f2010-04-20 17:09:59 +10001369 if (log2_size && log->l_logBBstart > 0 &&
1370 !xfs_sb_version_haslogv2(&mp->m_sb)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001371 xfs_warn(mp,
1372 "log sector size (0x%x) invalid for configuration.",
1373 log2_size);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001374 goto out_free_log;
1375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 }
Alex Elder69ce58f2010-04-20 17:09:59 +10001377 log->l_sectBBsize = 1 << log2_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379 xlog_get_iclog_buffer_size(mp, log);
1380
Dave Chinner400b9d82014-08-04 12:42:40 +10001381 /*
1382 * Use a NULL block for the extra log buffer used during splits so that
1383 * it will trigger errors if we ever try to do IO on it without first
1384 * having set it up properly.
1385 */
Dave Chinner24513372014-06-25 14:58:08 +10001386 error = -ENOMEM;
Dave Chinner400b9d82014-08-04 12:42:40 +10001387 bp = xfs_buf_alloc(mp->m_logdev_targp, XFS_BUF_DADDR_NULL,
1388 BTOBB(log->l_iclog_size), 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001389 if (!bp)
1390 goto out_free_log;
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001391
1392 /*
1393 * The iclogbuf buffer locks are held over IO but we are not going to do
1394 * IO yet. Hence unlock the buffer so that the log IO path can grab it
1395 * when appropriately.
1396 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001397 ASSERT(xfs_buf_islocked(bp));
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001398 xfs_buf_unlock(bp);
1399
1400 bp->b_iodone = xlog_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 log->l_xbuf = bp;
1402
Eric Sandeen007c61c2007-10-11 17:43:56 +10001403 spin_lock_init(&log->l_icloglock);
Dave Chinnereb40a872010-12-21 12:09:01 +11001404 init_waitqueue_head(&log->l_flush_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 iclogp = &log->l_iclog;
1407 /*
1408 * The amount of memory to allocate for the iclog structure is
1409 * rather funky due to the way the structure is defined. It is
1410 * done this way so that we can use different sizes for machines
1411 * with different amounts of memory. See the definition of
1412 * xlog_in_core_t in xfs_log_priv.h for details.
1413 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 ASSERT(log->l_iclog_size >= 4096);
1415 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001416 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1417 if (!*iclogp)
1418 goto out_free_iclog;
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 iclog->ic_prev = prev_iclog;
1422 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001423
Dave Chinner686865f2010-09-24 20:07:47 +10001424 bp = xfs_buf_get_uncached(mp->m_logdev_targp,
Dave Chinnere70b73f2012-04-23 15:58:49 +10001425 BTOBB(log->l_iclog_size), 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001426 if (!bp)
1427 goto out_free_iclog;
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +02001428
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001429 ASSERT(xfs_buf_islocked(bp));
1430 xfs_buf_unlock(bp);
1431
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001432 bp->b_iodone = xlog_iodone;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001433 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001434 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001435#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
David Chinner4679b2d2008-04-10 12:18:54 +10001437#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 head = &iclog->ic_header;
1439 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001440 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1441 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001442 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001443 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001445 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1447
Dave Chinner4e94b712012-04-23 15:58:51 +10001448 iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 iclog->ic_state = XLOG_STATE_ACTIVE;
1450 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001451 atomic_set(&iclog->ic_refcnt, 0);
1452 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001454 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Dave Chinnereb40a872010-12-21 12:09:01 +11001456 init_waitqueue_head(&iclog->ic_force_wait);
1457 init_waitqueue_head(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 iclogp = &iclog->ic_next;
1460 }
1461 *iclogp = log->l_iclog; /* complete ring */
1462 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1463
Dave Chinner71e330b2010-05-21 14:37:18 +10001464 error = xlog_cil_init(log);
1465 if (error)
1466 goto out_free_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001468
1469out_free_iclog:
1470 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1471 prev_iclog = iclog->ic_next;
Dave Chinnereb40a872010-12-21 12:09:01 +11001472 if (iclog->ic_bp)
Dave Chinner644c3562008-11-10 16:50:24 +11001473 xfs_buf_free(iclog->ic_bp);
Dave Chinner644c3562008-11-10 16:50:24 +11001474 kmem_free(iclog);
1475 }
1476 spinlock_destroy(&log->l_icloglock);
Dave Chinner644c3562008-11-10 16:50:24 +11001477 xfs_buf_free(log->l_xbuf);
1478out_free_log:
1479 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001480out:
Dave Chinner24513372014-06-25 14:58:08 +10001481 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482} /* xlog_alloc_log */
1483
1484
1485/*
1486 * Write out the commit record of a transaction associated with the given
1487 * ticket. Return the lsn of the commit record.
1488 */
1489STATIC int
Dave Chinner55b66332010-03-23 11:43:17 +11001490xlog_commit_record(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001491 struct xlog *log,
Dave Chinner55b66332010-03-23 11:43:17 +11001492 struct xlog_ticket *ticket,
1493 struct xlog_in_core **iclog,
1494 xfs_lsn_t *commitlsnp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495{
Dave Chinner55b66332010-03-23 11:43:17 +11001496 struct xfs_mount *mp = log->l_mp;
1497 int error;
1498 struct xfs_log_iovec reg = {
1499 .i_addr = NULL,
1500 .i_len = 0,
1501 .i_type = XLOG_REG_TYPE_COMMIT,
1502 };
1503 struct xfs_log_vec vec = {
1504 .lv_niovecs = 1,
1505 .lv_iovecp = &reg,
1506 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
1508 ASSERT_ALWAYS(iclog);
Dave Chinner55b66332010-03-23 11:43:17 +11001509 error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1510 XLOG_COMMIT_TRANS);
1511 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +10001512 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +01001513 return error;
Dave Chinner55b66332010-03-23 11:43:17 +11001514}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516/*
1517 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1518 * log space. This code pushes on the lsn which would supposedly free up
1519 * the 25% which we want to leave free. We may need to adopt a policy which
1520 * pushes on an lsn which is further along in the log once we reach the high
1521 * water mark. In this manner, we would be creating a low water mark.
1522 */
David Chinnera8272ce2007-11-23 16:28:09 +11001523STATIC void
Dave Chinner2ced19c2010-12-21 12:09:20 +11001524xlog_grant_push_ail(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001525 struct xlog *log,
Dave Chinner2ced19c2010-12-21 12:09:20 +11001526 int need_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527{
Dave Chinner2ced19c2010-12-21 12:09:20 +11001528 xfs_lsn_t threshold_lsn = 0;
Dave Chinner84f3c682010-12-03 22:11:29 +11001529 xfs_lsn_t last_sync_lsn;
Dave Chinner2ced19c2010-12-21 12:09:20 +11001530 int free_blocks;
1531 int free_bytes;
1532 int threshold_block;
1533 int threshold_cycle;
1534 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
Dave Chinner2ced19c2010-12-21 12:09:20 +11001536 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Christoph Hellwig28496962012-02-20 02:31:25 +00001538 free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
Dave Chinner2ced19c2010-12-21 12:09:20 +11001539 free_blocks = BTOBBT(free_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Dave Chinner2ced19c2010-12-21 12:09:20 +11001541 /*
1542 * Set the threshold for the minimum number of free blocks in the
1543 * log to the maximum of what the caller needs, one quarter of the
1544 * log, and 256 blocks.
1545 */
1546 free_threshold = BTOBB(need_bytes);
1547 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1548 free_threshold = MAX(free_threshold, 256);
1549 if (free_blocks >= free_threshold)
1550 return;
1551
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001552 xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1553 &threshold_block);
1554 threshold_block += free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (threshold_block >= log->l_logBBsize) {
Dave Chinner2ced19c2010-12-21 12:09:20 +11001556 threshold_block -= log->l_logBBsize;
1557 threshold_cycle += 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
Dave Chinner2ced19c2010-12-21 12:09:20 +11001559 threshold_lsn = xlog_assign_lsn(threshold_cycle,
1560 threshold_block);
1561 /*
1562 * Don't pass in an lsn greater than the lsn of the last
Dave Chinner84f3c682010-12-03 22:11:29 +11001563 * log record known to be on disk. Use a snapshot of the last sync lsn
1564 * so that it doesn't change between the compare and the set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 */
Dave Chinner84f3c682010-12-03 22:11:29 +11001566 last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1567 if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1568 threshold_lsn = last_sync_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Dave Chinner2ced19c2010-12-21 12:09:20 +11001570 /*
1571 * Get the transaction layer to kick the dirty buffers out to
1572 * disk asynchronously. No point in trying to do this if
1573 * the filesystem is shutting down.
1574 */
1575 if (!XLOG_FORCED_SHUTDOWN(log))
Dave Chinnerfd074842011-04-08 12:45:07 +10001576 xfs_ail_push(log->l_ailp, threshold_lsn);
Dave Chinner2ced19c2010-12-21 12:09:20 +11001577}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Christoph Hellwig873ff552010-01-13 22:17:57 +00001579/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001580 * Stamp cycle number in every block
1581 */
1582STATIC void
1583xlog_pack_data(
1584 struct xlog *log,
1585 struct xlog_in_core *iclog,
1586 int roundoff)
1587{
1588 int i, j, k;
1589 int size = iclog->ic_offset + roundoff;
1590 __be32 cycle_lsn;
1591 xfs_caddr_t dp;
1592
1593 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
1594
1595 dp = iclog->ic_datap;
1596 for (i = 0; i < BTOBB(size); i++) {
1597 if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
1598 break;
1599 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
1600 *(__be32 *)dp = cycle_lsn;
1601 dp += BBSIZE;
1602 }
1603
1604 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1605 xlog_in_core_2_t *xhdr = iclog->ic_data;
1606
1607 for ( ; i < BTOBB(size); i++) {
1608 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1609 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1610 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
1611 *(__be32 *)dp = cycle_lsn;
1612 dp += BBSIZE;
1613 }
1614
1615 for (i = 1; i < log->l_iclog_heads; i++)
1616 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
1617 }
1618}
1619
1620/*
1621 * Calculate the checksum for a log buffer.
1622 *
1623 * This is a little more complicated than it should be because the various
1624 * headers and the actual data are non-contiguous.
1625 */
Dave Chinnerf9668a02012-11-28 13:01:03 +11001626__le32
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001627xlog_cksum(
1628 struct xlog *log,
1629 struct xlog_rec_header *rhead,
1630 char *dp,
1631 int size)
1632{
1633 __uint32_t crc;
1634
1635 /* first generate the crc for the record header ... */
1636 crc = xfs_start_cksum((char *)rhead,
1637 sizeof(struct xlog_rec_header),
1638 offsetof(struct xlog_rec_header, h_crc));
1639
1640 /* ... then for additional cycle data for v2 logs ... */
1641 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1642 union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
1643 int i;
1644
1645 for (i = 1; i < log->l_iclog_heads; i++) {
1646 crc = crc32c(crc, &xhdr[i].hic_xheader,
1647 sizeof(struct xlog_rec_ext_header));
1648 }
1649 }
1650
1651 /* ... and finally for the payload */
1652 crc = crc32c(crc, dp, size);
1653
1654 return xfs_end_cksum(crc);
1655}
1656
1657/*
Christoph Hellwig873ff552010-01-13 22:17:57 +00001658 * The bdstrat callback function for log bufs. This gives us a central
1659 * place to trap bufs in case we get hit by a log I/O error and need to
1660 * shutdown. Actually, in practice, even when we didn't get a log error,
1661 * we transition the iclogs to IOERROR state *after* flushing all existing
1662 * iclogs to disk. This is because we don't want anymore new transactions to be
1663 * started or completed afterwards.
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001664 *
1665 * We lock the iclogbufs here so that we can serialise against IO completion
1666 * during unmount. We might be processing a shutdown triggered during unmount,
1667 * and that can occur asynchronously to the unmount thread, and hence we need to
1668 * ensure that completes before tearing down the iclogbufs. Hence we need to
1669 * hold the buffer lock across the log IO to acheive that.
Christoph Hellwig873ff552010-01-13 22:17:57 +00001670 */
1671STATIC int
1672xlog_bdstrat(
1673 struct xfs_buf *bp)
1674{
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001675 struct xlog_in_core *iclog = bp->b_fspriv;
Christoph Hellwig873ff552010-01-13 22:17:57 +00001676
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001677 xfs_buf_lock(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001678 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Dave Chinner24513372014-06-25 14:58:08 +10001679 xfs_buf_ioerror(bp, -EIO);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001680 xfs_buf_stale(bp);
Dave Chinnere8aaba92014-10-02 09:04:22 +10001681 xfs_buf_ioend(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001682 /*
1683 * It would seem logical to return EIO here, but we rely on
1684 * the log state machine to propagate I/O errors instead of
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001685 * doing it here. Similarly, IO completion will unlock the
1686 * buffer, so we don't do it here.
Christoph Hellwig873ff552010-01-13 22:17:57 +00001687 */
1688 return 0;
1689 }
1690
Dave Chinner595bff72014-10-02 09:05:14 +10001691 xfs_buf_submit(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001692 return 0;
1693}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695/*
1696 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1697 * fashion. Previously, we should have moved the current iclog
1698 * ptr in the log to point to the next available iclog. This allows further
1699 * write to continue while this code syncs out an iclog ready to go.
1700 * Before an in-core log can be written out, the data section must be scanned
1701 * to save away the 1st word of each BBSIZE block into the header. We replace
1702 * it with the current cycle count. Each BBSIZE block is tagged with the
1703 * cycle count because there in an implicit assumption that drives will
1704 * guarantee that entire 512 byte blocks get written at once. In other words,
1705 * we can't have part of a 512 byte block written and part not written. By
1706 * tagging each block, we will know which blocks are valid when recovering
1707 * after an unclean shutdown.
1708 *
1709 * This routine is single threaded on the iclog. No other thread can be in
1710 * this routine with the same iclog. Changing contents of iclog can there-
1711 * fore be done without grabbing the state machine lock. Updating the global
1712 * log will require grabbing the lock though.
1713 *
1714 * The entire log manager uses a logical block numbering scheme. Only
1715 * log_sync (and then only bwrite()) know about the fact that the log may
1716 * not start with block zero on a given device. The log block start offset
1717 * is added immediately before calling bwrite().
1718 */
1719
David Chinnera8272ce2007-11-23 16:28:09 +11001720STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001721xlog_sync(
1722 struct xlog *log,
1723 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001726 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 uint count; /* byte count of bwrite */
1728 uint count_init; /* initial count before roundup */
1729 int roundoff; /* roundoff to BB or stripe */
1730 int split = 0; /* split write into two regions */
1731 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001732 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001733 int size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734
1735 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001736 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 /* Add for LR header */
1739 count_init = log->l_iclog_hsize + iclog->ic_offset;
1740
1741 /* Round out the log write size */
1742 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1743 /* we have a v2 stripe unit to use */
1744 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1745 } else {
1746 count = BBTOB(BTOBB(count_init));
1747 }
1748 roundoff = count - count_init;
1749 ASSERT(roundoff >= 0);
1750 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1751 roundoff < log->l_mp->m_sb.sb_logsunit)
1752 ||
1753 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1754 roundoff < BBTOB(1)));
1755
1756 /* move grant heads by roundoff in sync */
Christoph Hellwig28496962012-02-20 02:31:25 +00001757 xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
1758 xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759
1760 /* put cycle number in every block */
1761 xlog_pack_data(log, iclog, roundoff);
1762
1763 /* real byte length */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001764 size = iclog->ic_offset;
1765 if (v2)
1766 size += roundoff;
1767 iclog->ic_header.h_len = cpu_to_be32(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Nathan Scottf5faad72006-07-28 17:04:44 +10001769 bp = iclog->ic_bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001770 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1773
1774 /* Do we need to split this write into 2 parts? */
1775 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001776 char *dptr;
1777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1779 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001780 iclog->ic_bwritecnt = 2;
1781
1782 /*
1783 * Bump the cycle numbers at the start of each block in the
1784 * part of the iclog that ends up in the buffer that gets
1785 * written to the start of the log.
1786 *
1787 * Watch out for the header magic number case, though.
1788 */
1789 dptr = (char *)&iclog->ic_header + count;
1790 for (i = 0; i < split; i += BBSIZE) {
1791 __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr);
1792 if (++cycle == XLOG_HEADER_MAGIC_NUM)
1793 cycle++;
1794 *(__be32 *)dptr = cpu_to_be32(cycle);
1795
1796 dptr += BBSIZE;
1797 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 } else {
1799 iclog->ic_bwritecnt = 1;
1800 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001801
1802 /* calculcate the checksum */
1803 iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
1804 iclog->ic_datap, size);
1805
Dave Chinneraa0e8832012-04-23 15:58:52 +10001806 bp->b_io_length = BTOBB(count);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001807 bp->b_fspriv = iclog;
Nathan Scottf5faad72006-07-28 17:04:44 +10001808 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001810 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwig651701d2010-06-28 10:34:34 -04001811
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001812 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001813 bp->b_flags |= XBF_FUA;
1814
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001815 /*
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001816 * Flush the data device before flushing the log to make
1817 * sure all meta data written back from the AIL actually made
1818 * it to disk before stamping the new log tail LSN into the
1819 * log buffer. For an external log we need to issue the
1820 * flush explicitly, and unfortunately synchronously here;
1821 * for an internal log we can simply use the block layer
1822 * state machine for preflushes.
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001823 */
1824 if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
1825 xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001826 else
1827 bp->b_flags |= XBF_FLUSH;
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1831 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1832
Ben Myers003fd6c2013-01-18 14:17:46 -06001833 xlog_verify_iclog(log, iclog, count, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 /* account for log which doesn't start at block #0 */
1836 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1837 /*
1838 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1839 * is shutting down.
1840 */
1841 XFS_BUF_WRITE(bp);
1842
Christoph Hellwig901796a2011-10-10 16:52:49 +00001843 error = xlog_bdstrat(bp);
1844 if (error) {
1845 xfs_buf_ioerror_alert(bp, "xlog_sync");
Jesper Juhl014c2542006-01-15 02:37:08 +01001846 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 }
1848 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001849 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +00001851 xfs_buf_associate_memory(bp,
1852 (char *)&iclog->ic_header + count, split);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001853 bp->b_fspriv = iclog;
Nathan Scottf5faad72006-07-28 17:04:44 +10001854 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001856 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001857 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001858 bp->b_flags |= XBF_FUA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1861 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1862
Nathan Scottc41564b2006-03-29 08:55:14 +10001863 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1865 XFS_BUF_WRITE(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +00001866 error = xlog_bdstrat(bp);
1867 if (error) {
1868 xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
Jesper Juhl014c2542006-01-15 02:37:08 +01001869 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 }
1871 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001872 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873} /* xlog_sync */
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001876 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 */
David Chinnera8272ce2007-11-23 16:28:09 +11001878STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001879xlog_dealloc_log(
1880 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881{
1882 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 int i;
1884
Dave Chinner71e330b2010-05-21 14:37:18 +10001885 xlog_cil_destroy(log);
1886
Dave Chinner44396472011-04-21 09:34:27 +00001887 /*
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001888 * Cycle all the iclogbuf locks to make sure all log IO completion
1889 * is done before we tear down these buffers.
Dave Chinner44396472011-04-21 09:34:27 +00001890 */
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001891 iclog = log->l_iclog;
1892 for (i = 0; i < log->l_iclog_bufs; i++) {
1893 xfs_buf_lock(iclog->ic_bp);
1894 xfs_buf_unlock(iclog->ic_bp);
1895 iclog = iclog->ic_next;
1896 }
1897
1898 /*
1899 * Always need to ensure that the extra buffer does not point to memory
1900 * owned by another log buffer before we free it. Also, cycle the lock
1901 * first to ensure we've completed IO on it.
1902 */
1903 xfs_buf_lock(log->l_xbuf);
1904 xfs_buf_unlock(log->l_xbuf);
Dave Chinnere70b73f2012-04-23 15:58:49 +10001905 xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
Dave Chinner44396472011-04-21 09:34:27 +00001906 xfs_buf_free(log->l_xbuf);
1907
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 iclog = log->l_iclog;
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001909 for (i = 0; i < log->l_iclog_bufs; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 xfs_buf_free(iclog->ic_bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001912 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 iclog = next_iclog;
1914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 spinlock_destroy(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001918 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001919} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921/*
1922 * Update counters atomically now that memcpy is done.
1923 */
1924/* ARGSUSED */
1925static inline void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001926xlog_state_finish_copy(
1927 struct xlog *log,
1928 struct xlog_in_core *iclog,
1929 int record_cnt,
1930 int copy_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001932 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001934 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 iclog->ic_offset += copy_bytes;
1936
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001937 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938} /* xlog_state_finish_copy */
1939
1940
1941
1942
1943/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001944 * print out info relating to regions written which consume
1945 * the reservation
1946 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001947void
1948xlog_print_tic_res(
1949 struct xfs_mount *mp,
1950 struct xlog_ticket *ticket)
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001951{
1952 uint i;
1953 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1954
1955 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1956 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1957 "bformat",
1958 "bchunk",
1959 "efi_format",
1960 "efd_format",
1961 "iformat",
1962 "icore",
1963 "iext",
1964 "ibroot",
1965 "ilocal",
1966 "iattr_ext",
1967 "iattr_broot",
1968 "iattr_local",
1969 "qformat",
1970 "dquot",
1971 "quotaoff",
1972 "LR header",
1973 "unmount",
1974 "commit",
1975 "trans header"
1976 };
1977 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1978 "SETATTR_NOT_SIZE",
1979 "SETATTR_SIZE",
1980 "INACTIVE",
1981 "CREATE",
1982 "CREATE_TRUNC",
1983 "TRUNCATE_FILE",
1984 "REMOVE",
1985 "LINK",
1986 "RENAME",
1987 "MKDIR",
1988 "RMDIR",
1989 "SYMLINK",
1990 "SET_DMATTRS",
1991 "GROWFS",
1992 "STRAT_WRITE",
1993 "DIOSTRAT",
1994 "WRITE_SYNC",
1995 "WRITEID",
1996 "ADDAFORK",
1997 "ATTRINVAL",
1998 "ATRUNCATE",
1999 "ATTR_SET",
2000 "ATTR_RM",
2001 "ATTR_FLAG",
2002 "CLEAR_AGI_BUCKET",
2003 "QM_SBCHANGE",
2004 "DUMMY1",
2005 "DUMMY2",
2006 "QM_QUOTAOFF",
2007 "QM_DQALLOC",
2008 "QM_SETQLIM",
2009 "QM_DQCLUSTER",
2010 "QM_QINOCREATE",
2011 "QM_QUOTAOFF_END",
2012 "SB_UNIT",
2013 "FSYNC_TS",
2014 "GROWFSRT_ALLOC",
2015 "GROWFSRT_ZERO",
2016 "GROWFSRT_FREE",
2017 "SWAPEXT"
2018 };
2019
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002020 xfs_warn(mp,
Christoph Hellwig93b8a582011-12-06 21:58:07 +00002021 "xlog_write: reservation summary:\n"
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002022 " trans type = %s (%u)\n"
2023 " unit res = %d bytes\n"
2024 " current res = %d bytes\n"
2025 " total reg = %u bytes (o/flow = %u bytes)\n"
2026 " ophdrs = %u (ophdr space = %u bytes)\n"
2027 " ophdr + reg = %u bytes\n"
2028 " num regions = %u\n",
2029 ((ticket->t_trans_type <= 0 ||
2030 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
2031 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
2032 ticket->t_trans_type,
2033 ticket->t_unit_res,
2034 ticket->t_curr_res,
2035 ticket->t_res_arr_sum, ticket->t_res_o_flow,
2036 ticket->t_res_num_ophdrs, ophdr_spc,
2037 ticket->t_res_arr_sum +
2038 ticket->t_res_o_flow + ophdr_spc,
2039 ticket->t_res_num);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002040
2041 for (i = 0; i < ticket->t_res_num; i++) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002042 uint r_type = ticket->t_res_arr[i].r_type;
Eric Sandeen08e96e12013-10-11 20:59:05 -05002043 xfs_warn(mp, "region[%u]: %s - %u bytes", i,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002044 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
2045 "bad-rtype" : res_type_str[r_type-1]),
2046 ticket->t_res_arr[i].r_len);
2047 }
Dave Chinner169a7b02010-05-07 11:05:31 +10002048
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002049 xfs_alert_tag(mp, XFS_PTAG_LOGRES,
Christoph Hellwig93b8a582011-12-06 21:58:07 +00002050 "xlog_write: reservation ran out. Need to up reservation");
Chandra Seetharaman297aa632013-07-19 17:31:38 -05002051 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002052}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002053
2054/*
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002055 * Calculate the potential space needed by the log vector. Each region gets
2056 * its own xlog_op_header_t and may need to be double word aligned.
2057 */
2058static int
2059xlog_write_calc_vec_length(
2060 struct xlog_ticket *ticket,
Dave Chinner55b66332010-03-23 11:43:17 +11002061 struct xfs_log_vec *log_vector)
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002062{
Dave Chinner55b66332010-03-23 11:43:17 +11002063 struct xfs_log_vec *lv;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002064 int headers = 0;
2065 int len = 0;
2066 int i;
2067
2068 /* acct for start rec of xact */
2069 if (ticket->t_flags & XLOG_TIC_INITED)
2070 headers++;
2071
Dave Chinner55b66332010-03-23 11:43:17 +11002072 for (lv = log_vector; lv; lv = lv->lv_next) {
Dave Chinnerfd638752013-06-27 16:04:51 +10002073 /* we don't write ordered log vectors */
2074 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED)
2075 continue;
2076
Dave Chinner55b66332010-03-23 11:43:17 +11002077 headers += lv->lv_niovecs;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002078
Dave Chinner55b66332010-03-23 11:43:17 +11002079 for (i = 0; i < lv->lv_niovecs; i++) {
2080 struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
2081
2082 len += vecp->i_len;
2083 xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
2084 }
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002085 }
2086
2087 ticket->t_res_num_ophdrs += headers;
2088 len += headers * sizeof(struct xlog_op_header);
2089
2090 return len;
2091}
2092
2093/*
2094 * If first write for transaction, insert start record We can't be trying to
2095 * commit if we are inited. We can't have any "partial_copy" if we are inited.
2096 */
2097static int
2098xlog_write_start_rec(
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002099 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002100 struct xlog_ticket *ticket)
2101{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002102 if (!(ticket->t_flags & XLOG_TIC_INITED))
2103 return 0;
2104
2105 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2106 ophdr->oh_clientid = ticket->t_clientid;
2107 ophdr->oh_len = 0;
2108 ophdr->oh_flags = XLOG_START_TRANS;
2109 ophdr->oh_res2 = 0;
2110
2111 ticket->t_flags &= ~XLOG_TIC_INITED;
2112
2113 return sizeof(struct xlog_op_header);
2114}
2115
2116static xlog_op_header_t *
2117xlog_write_setup_ophdr(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002118 struct xlog *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002119 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002120 struct xlog_ticket *ticket,
2121 uint flags)
2122{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002123 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2124 ophdr->oh_clientid = ticket->t_clientid;
2125 ophdr->oh_res2 = 0;
2126
2127 /* are we copying a commit or unmount record? */
2128 ophdr->oh_flags = flags;
2129
2130 /*
2131 * We've seen logs corrupted with bad transaction client ids. This
2132 * makes sure that XFS doesn't generate them on. Turn this into an EIO
2133 * and shut down the filesystem.
2134 */
2135 switch (ophdr->oh_clientid) {
2136 case XFS_TRANSACTION:
2137 case XFS_VOLUME:
2138 case XFS_LOG:
2139 break;
2140 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002141 xfs_warn(log->l_mp,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002142 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
2143 ophdr->oh_clientid, ticket);
2144 return NULL;
2145 }
2146
2147 return ophdr;
2148}
2149
2150/*
2151 * Set up the parameters of the region copy into the log. This has
2152 * to handle region write split across multiple log buffers - this
2153 * state is kept external to this function so that this code can
Zhi Yong Wuac0e3002013-08-07 10:11:02 +00002154 * be written in an obvious, self documenting manner.
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002155 */
2156static int
2157xlog_write_setup_copy(
2158 struct xlog_ticket *ticket,
2159 struct xlog_op_header *ophdr,
2160 int space_available,
2161 int space_required,
2162 int *copy_off,
2163 int *copy_len,
2164 int *last_was_partial_copy,
2165 int *bytes_consumed)
2166{
2167 int still_to_copy;
2168
2169 still_to_copy = space_required - *bytes_consumed;
2170 *copy_off = *bytes_consumed;
2171
2172 if (still_to_copy <= space_available) {
2173 /* write of region completes here */
2174 *copy_len = still_to_copy;
2175 ophdr->oh_len = cpu_to_be32(*copy_len);
2176 if (*last_was_partial_copy)
2177 ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
2178 *last_was_partial_copy = 0;
2179 *bytes_consumed = 0;
2180 return 0;
2181 }
2182
2183 /* partial write of region, needs extra log op header reservation */
2184 *copy_len = space_available;
2185 ophdr->oh_len = cpu_to_be32(*copy_len);
2186 ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
2187 if (*last_was_partial_copy)
2188 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
2189 *bytes_consumed += *copy_len;
2190 (*last_was_partial_copy)++;
2191
2192 /* account for new log op header */
2193 ticket->t_curr_res -= sizeof(struct xlog_op_header);
2194 ticket->t_res_num_ophdrs++;
2195
2196 return sizeof(struct xlog_op_header);
2197}
2198
2199static int
2200xlog_write_copy_finish(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002201 struct xlog *log,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002202 struct xlog_in_core *iclog,
2203 uint flags,
2204 int *record_cnt,
2205 int *data_cnt,
2206 int *partial_copy,
2207 int *partial_copy_len,
2208 int log_offset,
2209 struct xlog_in_core **commit_iclog)
2210{
2211 if (*partial_copy) {
2212 /*
2213 * This iclog has already been marked WANT_SYNC by
2214 * xlog_state_get_iclog_space.
2215 */
2216 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2217 *record_cnt = 0;
2218 *data_cnt = 0;
2219 return xlog_state_release_iclog(log, iclog);
2220 }
2221
2222 *partial_copy = 0;
2223 *partial_copy_len = 0;
2224
2225 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
2226 /* no more space in this iclog - push it. */
2227 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2228 *record_cnt = 0;
2229 *data_cnt = 0;
2230
2231 spin_lock(&log->l_icloglock);
2232 xlog_state_want_sync(log, iclog);
2233 spin_unlock(&log->l_icloglock);
2234
2235 if (!commit_iclog)
2236 return xlog_state_release_iclog(log, iclog);
2237 ASSERT(flags & XLOG_COMMIT_TRANS);
2238 *commit_iclog = iclog;
2239 }
2240
2241 return 0;
2242}
2243
2244/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 * Write some region out to in-core log
2246 *
2247 * This will be called when writing externally provided regions or when
2248 * writing out a commit record for a given transaction.
2249 *
2250 * General algorithm:
2251 * 1. Find total length of this write. This may include adding to the
2252 * lengths passed in.
2253 * 2. Check whether we violate the tickets reservation.
2254 * 3. While writing to this iclog
2255 * A. Reserve as much space in this iclog as can get
2256 * B. If this is first write, save away start lsn
2257 * C. While writing this region:
2258 * 1. If first write of transaction, write start record
2259 * 2. Write log operation header (header per region)
2260 * 3. Find out if we can fit entire region into this iclog
2261 * 4. Potentially, verify destination memcpy ptr
2262 * 5. Memcpy (partial) region
2263 * 6. If partial copy, release iclog; otherwise, continue
2264 * copying more regions into current iclog
2265 * 4. Mark want sync bit (in simulation mode)
2266 * 5. Release iclog for potential flush to on-disk log.
2267 *
2268 * ERRORS:
2269 * 1. Panic if reservation is overrun. This should never happen since
2270 * reservation amounts are generated internal to the filesystem.
2271 * NOTES:
2272 * 1. Tickets are single threaded data structures.
2273 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
2274 * syncing routine. When a single log_write region needs to span
2275 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
2276 * on all log operation writes which don't contain the end of the
2277 * region. The XLOG_END_TRANS bit is used for the in-core log
2278 * operation which contains the end of the continued log_write region.
2279 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
2280 * we don't really know exactly how much space will be used. As a result,
2281 * we don't update ic_offset until the end when we know exactly how many
2282 * bytes have been written out.
2283 */
Dave Chinner71e330b2010-05-21 14:37:18 +10002284int
Christoph Hellwig35a8a722010-02-15 23:34:54 +00002285xlog_write(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002286 struct xlog *log,
Dave Chinner55b66332010-03-23 11:43:17 +11002287 struct xfs_log_vec *log_vector,
Christoph Hellwig35a8a722010-02-15 23:34:54 +00002288 struct xlog_ticket *ticket,
2289 xfs_lsn_t *start_lsn,
2290 struct xlog_in_core **commit_iclog,
2291 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292{
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002293 struct xlog_in_core *iclog = NULL;
Dave Chinner55b66332010-03-23 11:43:17 +11002294 struct xfs_log_iovec *vecp;
2295 struct xfs_log_vec *lv;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002296 int len;
2297 int index;
2298 int partial_copy = 0;
2299 int partial_copy_len = 0;
2300 int contwr = 0;
2301 int record_cnt = 0;
2302 int data_cnt = 0;
2303 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002305 *start_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Dave Chinner55b66332010-03-23 11:43:17 +11002307 len = xlog_write_calc_vec_length(ticket, log_vector);
Dave Chinner71e330b2010-05-21 14:37:18 +10002308
Christoph Hellwig93b8a582011-12-06 21:58:07 +00002309 /*
2310 * Region headers and bytes are already accounted for.
2311 * We only need to take into account start records and
2312 * split regions in this function.
2313 */
2314 if (ticket->t_flags & XLOG_TIC_INITED)
2315 ticket->t_curr_res -= sizeof(xlog_op_header_t);
2316
2317 /*
2318 * Commit record headers need to be accounted for. These
2319 * come in as separate writes so are easy to detect.
2320 */
2321 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
2322 ticket->t_curr_res -= sizeof(xlog_op_header_t);
Dave Chinner71e330b2010-05-21 14:37:18 +10002323
2324 if (ticket->t_curr_res < 0)
Dave Chinner55b66332010-03-23 11:43:17 +11002325 xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
Dave Chinner55b66332010-03-23 11:43:17 +11002327 index = 0;
2328 lv = log_vector;
2329 vecp = lv->lv_iovecp;
Dave Chinnerfd638752013-06-27 16:04:51 +10002330 while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002331 void *ptr;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002332 int log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002334 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
2335 &contwr, &log_offset);
2336 if (error)
2337 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002339 ASSERT(log_offset <= iclog->ic_size - 1);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002340 ptr = iclog->ic_datap + log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002342 /* start_lsn is the first lsn written to. That's all we need. */
2343 if (!*start_lsn)
2344 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002346 /*
2347 * This loop writes out as many regions as can fit in the amount
2348 * of space which was allocated by xlog_state_get_iclog_space().
2349 */
Dave Chinnerfd638752013-06-27 16:04:51 +10002350 while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
2351 struct xfs_log_iovec *reg;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002352 struct xlog_op_header *ophdr;
2353 int start_rec_copy;
2354 int copy_len;
2355 int copy_off;
Dave Chinnerfd638752013-06-27 16:04:51 +10002356 bool ordered = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
Dave Chinnerfd638752013-06-27 16:04:51 +10002358 /* ordered log vectors have no regions to write */
2359 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
2360 ASSERT(lv->lv_niovecs == 0);
2361 ordered = true;
2362 goto next_lv;
2363 }
2364
2365 reg = &vecp[index];
Dave Chinner55b66332010-03-23 11:43:17 +11002366 ASSERT(reg->i_len % sizeof(__int32_t) == 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002367 ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002369 start_rec_copy = xlog_write_start_rec(ptr, ticket);
2370 if (start_rec_copy) {
2371 record_cnt++;
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002372 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002373 start_rec_copy);
2374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002376 ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
2377 if (!ophdr)
Dave Chinner24513372014-06-25 14:58:08 +10002378 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002380 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002381 sizeof(struct xlog_op_header));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002383 len += xlog_write_setup_copy(ticket, ophdr,
2384 iclog->ic_size-log_offset,
Dave Chinner55b66332010-03-23 11:43:17 +11002385 reg->i_len,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002386 &copy_off, &copy_len,
2387 &partial_copy,
2388 &partial_copy_len);
2389 xlog_verify_dest_ptr(log, ptr);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002390
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002391 /* copy region */
2392 ASSERT(copy_len >= 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002393 memcpy(ptr, reg->i_addr + copy_off, copy_len);
2394 xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002395
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002396 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
2397 record_cnt++;
2398 data_cnt += contwr ? copy_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002400 error = xlog_write_copy_finish(log, iclog, flags,
2401 &record_cnt, &data_cnt,
2402 &partial_copy,
2403 &partial_copy_len,
2404 log_offset,
2405 commit_iclog);
2406 if (error)
2407 return error;
2408
2409 /*
2410 * if we had a partial copy, we need to get more iclog
2411 * space but we don't want to increment the region
2412 * index because there is still more is this region to
2413 * write.
2414 *
2415 * If we completed writing this region, and we flushed
2416 * the iclog (indicated by resetting of the record
2417 * count), then we also need to get more log space. If
2418 * this was the last record, though, we are done and
2419 * can just return.
2420 */
2421 if (partial_copy)
2422 break;
2423
Dave Chinner55b66332010-03-23 11:43:17 +11002424 if (++index == lv->lv_niovecs) {
Dave Chinnerfd638752013-06-27 16:04:51 +10002425next_lv:
Dave Chinner55b66332010-03-23 11:43:17 +11002426 lv = lv->lv_next;
2427 index = 0;
2428 if (lv)
2429 vecp = lv->lv_iovecp;
2430 }
Dave Chinnerfd638752013-06-27 16:04:51 +10002431 if (record_cnt == 0 && ordered == false) {
Dave Chinner55b66332010-03-23 11:43:17 +11002432 if (!lv)
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002433 return 0;
2434 break;
2435 }
2436 }
2437 }
2438
2439 ASSERT(len == 0);
2440
2441 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
2442 if (!commit_iclog)
2443 return xlog_state_release_iclog(log, iclog);
2444
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 ASSERT(flags & XLOG_COMMIT_TRANS);
2446 *commit_iclog = iclog;
2447 return 0;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002448}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450
2451/*****************************************************************************
2452 *
2453 * State Machine functions
2454 *
2455 *****************************************************************************
2456 */
2457
2458/* Clean iclogs starting from the head. This ordering must be
2459 * maintained, so an iclog doesn't become ACTIVE beyond one that
2460 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10002461 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 * iclog is trying to sync to disk.
2463 *
2464 * State Change: DIRTY -> ACTIVE
2465 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002466STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002467xlog_state_clean_log(
2468 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469{
2470 xlog_in_core_t *iclog;
2471 int changed = 0;
2472
2473 iclog = log->l_iclog;
2474 do {
2475 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2476 iclog->ic_state = XLOG_STATE_ACTIVE;
2477 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002478 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 /*
2480 * If the number of ops in this iclog indicate it just
2481 * contains the dummy transaction, we can
2482 * change state into IDLE (the second time around).
2483 * Otherwise we should change the state into
2484 * NEED a dummy.
2485 * We don't need to cover the dummy.
2486 */
2487 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002488 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2489 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 changed = 1;
2491 } else {
2492 /*
2493 * We have two dirty iclogs so start over
2494 * This could also be num of ops indicates
2495 * this is not the dummy going out.
2496 */
2497 changed = 2;
2498 }
2499 iclog->ic_header.h_num_logops = 0;
2500 memset(iclog->ic_header.h_cycle_data, 0,
2501 sizeof(iclog->ic_header.h_cycle_data));
2502 iclog->ic_header.h_lsn = 0;
2503 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2504 /* do nothing */;
2505 else
2506 break; /* stop cleaning */
2507 iclog = iclog->ic_next;
2508 } while (iclog != log->l_iclog);
2509
2510 /* log is locked when we are called */
2511 /*
2512 * Change state for the dummy log recording.
2513 * We usually go to NEED. But we go to NEED2 if the changed indicates
2514 * we are done writing the dummy record.
2515 * If we are done with the second dummy recored (DONE2), then
2516 * we go to IDLE.
2517 */
2518 if (changed) {
2519 switch (log->l_covered_state) {
2520 case XLOG_STATE_COVER_IDLE:
2521 case XLOG_STATE_COVER_NEED:
2522 case XLOG_STATE_COVER_NEED2:
2523 log->l_covered_state = XLOG_STATE_COVER_NEED;
2524 break;
2525
2526 case XLOG_STATE_COVER_DONE:
2527 if (changed == 1)
2528 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2529 else
2530 log->l_covered_state = XLOG_STATE_COVER_NEED;
2531 break;
2532
2533 case XLOG_STATE_COVER_DONE2:
2534 if (changed == 1)
2535 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2536 else
2537 log->l_covered_state = XLOG_STATE_COVER_NEED;
2538 break;
2539
2540 default:
2541 ASSERT(0);
2542 }
2543 }
2544} /* xlog_state_clean_log */
2545
2546STATIC xfs_lsn_t
2547xlog_get_lowest_lsn(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002548 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549{
2550 xlog_in_core_t *lsn_log;
2551 xfs_lsn_t lowest_lsn, lsn;
2552
2553 lsn_log = log->l_iclog;
2554 lowest_lsn = 0;
2555 do {
2556 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002557 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if ((lsn && !lowest_lsn) ||
2559 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2560 lowest_lsn = lsn;
2561 }
2562 }
2563 lsn_log = lsn_log->ic_next;
2564 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002565 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566}
2567
2568
2569STATIC void
2570xlog_state_do_callback(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002571 struct xlog *log,
2572 int aborted,
2573 struct xlog_in_core *ciclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574{
2575 xlog_in_core_t *iclog;
2576 xlog_in_core_t *first_iclog; /* used to know when we've
2577 * processed all iclogs once */
2578 xfs_log_callback_t *cb, *cb_next;
2579 int flushcnt = 0;
2580 xfs_lsn_t lowest_lsn;
2581 int ioerrors; /* counter: iclogs with errors */
2582 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2583 int funcdidcallbacks; /* flag: function did callbacks */
2584 int repeats; /* for issuing console warnings if
2585 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002586 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002588 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 first_iclog = iclog = log->l_iclog;
2590 ioerrors = 0;
2591 funcdidcallbacks = 0;
2592 repeats = 0;
2593
2594 do {
2595 /*
2596 * Scan all iclogs starting with the one pointed to by the
2597 * log. Reset this starting point each time the log is
2598 * unlocked (during callbacks).
2599 *
2600 * Keep looping through iclogs until one full pass is made
2601 * without running any callbacks.
2602 */
2603 first_iclog = log->l_iclog;
2604 iclog = log->l_iclog;
2605 loopdidcallbacks = 0;
2606 repeats++;
2607
2608 do {
2609
2610 /* skip all iclogs in the ACTIVE & DIRTY states */
2611 if (iclog->ic_state &
2612 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2613 iclog = iclog->ic_next;
2614 continue;
2615 }
2616
2617 /*
2618 * Between marking a filesystem SHUTDOWN and stopping
2619 * the log, we do flush all iclogs to disk (if there
2620 * wasn't a log I/O error). So, we do want things to
2621 * go smoothly in case of just a SHUTDOWN w/o a
2622 * LOG_IO_ERROR.
2623 */
2624 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2625 /*
2626 * Can only perform callbacks in order. Since
2627 * this iclog is not in the DONE_SYNC/
2628 * DO_CALLBACK state, we skip the rest and
2629 * just try to clean up. If we set our iclog
2630 * to DO_CALLBACK, we will not process it when
2631 * we retry since a previous iclog is in the
2632 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002633 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 */
2635 if (!(iclog->ic_state &
2636 (XLOG_STATE_DONE_SYNC |
2637 XLOG_STATE_DO_CALLBACK))) {
2638 if (ciclog && (ciclog->ic_state ==
2639 XLOG_STATE_DONE_SYNC)) {
2640 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2641 }
2642 break;
2643 }
2644 /*
2645 * We now have an iclog that is in either the
2646 * DO_CALLBACK or DONE_SYNC states. The other
2647 * states (WANT_SYNC, SYNCING, or CALLBACK were
2648 * caught by the above if and are going to
2649 * clean (i.e. we aren't doing their callbacks)
2650 * see the above if.
2651 */
2652
2653 /*
2654 * We will do one more check here to see if we
2655 * have chased our tail around.
2656 */
2657
2658 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002659 if (lowest_lsn &&
2660 XFS_LSN_CMP(lowest_lsn,
Dave Chinner84f3c682010-12-03 22:11:29 +11002661 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 iclog = iclog->ic_next;
2663 continue; /* Leave this iclog for
2664 * another thread */
2665 }
2666
2667 iclog->ic_state = XLOG_STATE_CALLBACK;
2668
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Dave Chinner84f3c682010-12-03 22:11:29 +11002670 /*
Dave Chinnerd35e88f2012-10-08 21:56:12 +11002671 * Completion of a iclog IO does not imply that
2672 * a transaction has completed, as transactions
2673 * can be large enough to span many iclogs. We
2674 * cannot change the tail of the log half way
2675 * through a transaction as this may be the only
2676 * transaction in the log and moving th etail to
2677 * point to the middle of it will prevent
2678 * recovery from finding the start of the
2679 * transaction. Hence we should only update the
2680 * last_sync_lsn if this iclog contains
2681 * transaction completion callbacks on it.
2682 *
2683 * We have to do this before we drop the
Dave Chinner84f3c682010-12-03 22:11:29 +11002684 * icloglock to ensure we are the only one that
2685 * can update it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 */
Dave Chinner84f3c682010-12-03 22:11:29 +11002687 ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2688 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
Dave Chinnerd35e88f2012-10-08 21:56:12 +11002689 if (iclog->ic_callback)
2690 atomic64_set(&log->l_last_sync_lsn,
2691 be64_to_cpu(iclog->ic_header.h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692
Dave Chinner84f3c682010-12-03 22:11:29 +11002693 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 ioerrors++;
Dave Chinner84f3c682010-12-03 22:11:29 +11002695
2696 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
David Chinner114d23a2008-04-10 12:18:39 +10002698 /*
2699 * Keep processing entries in the callback list until
2700 * we come around and it is empty. We need to
2701 * atomically see that the list is empty and change the
2702 * state to DIRTY so that we don't miss any more
2703 * callbacks being added.
2704 */
2705 spin_lock(&iclog->ic_callback_lock);
2706 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002707 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 iclog->ic_callback_tail = &(iclog->ic_callback);
2709 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002710 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
2712 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002713 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 cb_next = cb->cb_next;
2715 cb->cb_func(cb->cb_arg, aborted);
2716 }
David Chinner114d23a2008-04-10 12:18:39 +10002717 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 cb = iclog->ic_callback;
2719 }
2720
2721 loopdidcallbacks++;
2722 funcdidcallbacks++;
2723
David Chinner114d23a2008-04-10 12:18:39 +10002724 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002725 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002726 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2728 iclog->ic_state = XLOG_STATE_DIRTY;
2729
2730 /*
2731 * Transition from DIRTY to ACTIVE if applicable.
2732 * NOP if STATE_IOERROR.
2733 */
2734 xlog_state_clean_log(log);
2735
2736 /* wake up threads waiting in xfs_log_force() */
Dave Chinnereb40a872010-12-21 12:09:01 +11002737 wake_up_all(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738
2739 iclog = iclog->ic_next;
2740 } while (first_iclog != iclog);
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002741
2742 if (repeats > 5000) {
2743 flushcnt += repeats;
2744 repeats = 0;
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002745 xfs_warn(log->l_mp,
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002746 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002747 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 }
2749 } while (!ioerrors && loopdidcallbacks);
2750
2751 /*
2752 * make one last gasp attempt to see if iclogs are being left in
2753 * limbo..
2754 */
2755#ifdef DEBUG
2756 if (funcdidcallbacks) {
2757 first_iclog = iclog = log->l_iclog;
2758 do {
2759 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2760 /*
2761 * Terminate the loop if iclogs are found in states
2762 * which will cause other threads to clean up iclogs.
2763 *
2764 * SYNCING - i/o completion will go through logs
2765 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002766 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 * IOERROR - give up hope all ye who enter here
2768 */
2769 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2770 iclog->ic_state == XLOG_STATE_SYNCING ||
2771 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2772 iclog->ic_state == XLOG_STATE_IOERROR )
2773 break;
2774 iclog = iclog->ic_next;
2775 } while (first_iclog != iclog);
2776 }
2777#endif
2778
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002779 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2780 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002781 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002782
2783 if (wake)
Dave Chinnereb40a872010-12-21 12:09:01 +11002784 wake_up_all(&log->l_flush_wait);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002785}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786
2787
2788/*
2789 * Finish transitioning this iclog to the dirty state.
2790 *
2791 * Make sure that we completely execute this routine only when this is
2792 * the last call to the iclog. There is a good chance that iclog flushes,
2793 * when we reach the end of the physical log, get turned into 2 separate
2794 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2795 * routine. By using the reference count bwritecnt, we guarantee that only
2796 * the second completion goes through.
2797 *
2798 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002799 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 */
David Chinnera8272ce2007-11-23 16:28:09 +11002801STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802xlog_state_done_syncing(
2803 xlog_in_core_t *iclog,
2804 int aborted)
2805{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002806 struct xlog *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002808 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
2810 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2811 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002812 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2814
2815
2816 /*
2817 * If we got an error, either on the first buffer, or in the case of
2818 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2819 * and none should ever be attempted to be written to disk
2820 * again.
2821 */
2822 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2823 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002824 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 return;
2826 }
2827 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2828 }
2829
2830 /*
2831 * Someone could be sleeping prior to writing out the next
2832 * iclog buffer, we wake them all, one will get to do the
2833 * I/O, the others get to wait for the result.
2834 */
Dave Chinnereb40a872010-12-21 12:09:01 +11002835 wake_up_all(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002836 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2838} /* xlog_state_done_syncing */
2839
2840
2841/*
2842 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002843 * sleep. We wait on the flush queue on the head iclog as that should be
2844 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2845 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 *
2847 * The in-core logs are used in a circular fashion. They are not used
2848 * out-of-order even when an iclog past the head is free.
2849 *
2850 * return:
2851 * * log_offset where xlog_write() can start writing into the in-core
2852 * log's data space.
2853 * * in-core log pointer to which xlog_write() should write.
2854 * * boolean indicating this is a continued write to an in-core log.
2855 * If this is the last write, then the in-core log's offset field
2856 * needs to be incremented, depending on the amount of data which
2857 * is copied.
2858 */
David Chinnera8272ce2007-11-23 16:28:09 +11002859STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002860xlog_state_get_iclog_space(
2861 struct xlog *log,
2862 int len,
2863 struct xlog_in_core **iclogp,
2864 struct xlog_ticket *ticket,
2865 int *continued_write,
2866 int *logoffsetp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 int log_offset;
2869 xlog_rec_header_t *head;
2870 xlog_in_core_t *iclog;
2871 int error;
2872
2873restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002874 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002876 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10002877 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 }
2879
2880 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002881 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002883
2884 /* Wait for log writes to have flushed */
Dave Chinnereb40a872010-12-21 12:09:01 +11002885 xlog_wait(&log->l_flush_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 goto restart;
2887 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002888
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 head = &iclog->ic_header;
2890
David Chinner155cc6b2008-03-06 13:44:14 +11002891 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 log_offset = iclog->ic_offset;
2893
2894 /* On the 1st write to an iclog, figure out lsn. This works
2895 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2896 * committing to. If the offset is set, that's how many blocks
2897 * must be written.
2898 */
2899 if (log_offset == 0) {
2900 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002901 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002902 log->l_iclog_hsize,
2903 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002904 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2905 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002906 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 ASSERT(log->l_curr_block >= 0);
2908 }
2909
2910 /* If there is enough room to write everything, then do it. Otherwise,
2911 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2912 * bit is on, so this will get flushed out. Don't update ic_offset
2913 * until you know exactly how many bytes get copied. Therefore, wait
2914 * until later to update ic_offset.
2915 *
2916 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2917 * can fit into remaining data section.
2918 */
2919 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2920 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2921
Dave Chinner49641f12008-07-11 17:43:55 +10002922 /*
2923 * If I'm the only one writing to this iclog, sync it to disk.
2924 * We need to do an atomic compare and decrement here to avoid
2925 * racing with concurrent atomic_dec_and_lock() calls in
2926 * xlog_state_release_iclog() when there is more than one
2927 * reference to the iclog.
2928 */
2929 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2930 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002931 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002932 error = xlog_state_release_iclog(log, iclog);
2933 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002934 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002936 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 }
2938 goto restart;
2939 }
2940
2941 /* Do we have enough room to write the full amount in the remainder
2942 * of this iclog? Or must we continue a write on the next iclog and
2943 * mark this iclog as completely taken? In the case where we switch
2944 * iclogs (to mark it taken), this particular iclog will release/sync
2945 * to disk in xlog_write().
2946 */
2947 if (len <= iclog->ic_size - iclog->ic_offset) {
2948 *continued_write = 0;
2949 iclog->ic_offset += len;
2950 } else {
2951 *continued_write = 1;
2952 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2953 }
2954 *iclogp = iclog;
2955
2956 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002957 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
2959 *logoffsetp = log_offset;
2960 return 0;
2961} /* xlog_state_get_iclog_space */
2962
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963/* The first cnt-1 times through here we don't need to
2964 * move the grant write head because the permanent
2965 * reservation has reserved cnt times the unit amount.
2966 * Release part of current permanent unit reservation and
2967 * reset current reservation to be one units worth. Also
2968 * move grant reservation head forward.
2969 */
2970STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002971xlog_regrant_reserve_log_space(
2972 struct xlog *log,
2973 struct xlog_ticket *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002975 trace_xfs_log_regrant_reserve_enter(log, ticket);
2976
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 if (ticket->t_cnt > 0)
2978 ticket->t_cnt--;
2979
Christoph Hellwig28496962012-02-20 02:31:25 +00002980 xlog_grant_sub_space(log, &log->l_reserve_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11002981 ticket->t_curr_res);
Christoph Hellwig28496962012-02-20 02:31:25 +00002982 xlog_grant_sub_space(log, &log->l_write_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11002983 ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002985 xlog_tic_reset_res(ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002986
2987 trace_xfs_log_regrant_reserve_sub(log, ticket);
2988
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 /* just return if we still have some of the pre-reserved space */
Dave Chinnerd0eb2f32010-12-21 12:29:14 +11002990 if (ticket->t_cnt > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
Christoph Hellwig28496962012-02-20 02:31:25 +00002993 xlog_grant_add_space(log, &log->l_reserve_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11002994 ticket->t_unit_res);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002995
2996 trace_xfs_log_regrant_reserve_exit(log, ticket);
2997
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002999 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000} /* xlog_regrant_reserve_log_space */
3001
3002
3003/*
3004 * Give back the space left from a reservation.
3005 *
3006 * All the information we need to make a correct determination of space left
3007 * is present. For non-permanent reservations, things are quite easy. The
3008 * count should have been decremented to zero. We only need to deal with the
3009 * space remaining in the current reservation part of the ticket. If the
3010 * ticket contains a permanent reservation, there may be left over space which
3011 * needs to be released. A count of N means that N-1 refills of the current
3012 * reservation can be done before we need to ask for more space. The first
3013 * one goes to fill up the first current reservation. Once we run out of
3014 * space, the count will stay at zero and the only space remaining will be
3015 * in the current reservation field.
3016 */
3017STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003018xlog_ungrant_log_space(
3019 struct xlog *log,
3020 struct xlog_ticket *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021{
Dave Chinner663e4962010-12-21 12:06:05 +11003022 int bytes;
3023
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 if (ticket->t_cnt > 0)
3025 ticket->t_cnt--;
3026
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003027 trace_xfs_log_ungrant_enter(log, ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003028 trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029
Dave Chinner663e4962010-12-21 12:06:05 +11003030 /*
3031 * If this is a permanent reservation ticket, we may be able to free
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 * up more space based on the remaining count.
3033 */
Dave Chinner663e4962010-12-21 12:06:05 +11003034 bytes = ticket->t_curr_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 if (ticket->t_cnt > 0) {
3036 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Dave Chinner663e4962010-12-21 12:06:05 +11003037 bytes += ticket->t_unit_res*ticket->t_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 }
3039
Christoph Hellwig28496962012-02-20 02:31:25 +00003040 xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
3041 xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
Dave Chinner663e4962010-12-21 12:06:05 +11003042
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003043 trace_xfs_log_ungrant_exit(log, ticket);
3044
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +00003045 xfs_log_space_wake(log->l_mp);
Christoph Hellwig09a423a2012-02-20 02:31:20 +00003046}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
3048/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049 * Flush iclog to disk if this is the last reference to the given iclog and
3050 * the WANT_SYNC bit is set.
3051 *
3052 * When this function is entered, the iclog is not necessarily in the
3053 * WANT_SYNC state. It may be sitting around waiting to get filled.
3054 *
3055 *
3056 */
David Chinnera8272ce2007-11-23 16:28:09 +11003057STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11003058xlog_state_release_iclog(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003059 struct xlog *log,
3060 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 int sync = 0; /* do we sync? */
3063
David Chinner155cc6b2008-03-06 13:44:14 +11003064 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003065 return -EIO;
David Chinner155cc6b2008-03-06 13:44:14 +11003066
3067 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
3068 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
3069 return 0;
3070
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003072 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003073 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
3076 iclog->ic_state == XLOG_STATE_WANT_SYNC);
3077
David Chinner155cc6b2008-03-06 13:44:14 +11003078 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11003079 /* update tail before writing to iclog */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003080 xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081 sync++;
3082 iclog->ic_state = XLOG_STATE_SYNCING;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003083 iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
3084 xlog_verify_tail_lsn(log, iclog, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 /* cycle incremented when incrementing curr_block */
3086 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003087 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
3089 /*
3090 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10003091 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
3093 * this iclog has consistent data, so we ignore IOERROR
3094 * flags after this point.
3095 */
David Chinnerb5893342008-03-06 13:44:06 +11003096 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003098 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099} /* xlog_state_release_iclog */
3100
3101
3102/*
3103 * This routine will mark the current iclog in the ring as WANT_SYNC
3104 * and move the current iclog pointer to the next iclog in the ring.
3105 * When this routine is called from xlog_state_get_iclog_space(), the
3106 * exact size of the iclog has not yet been determined. All we know is
3107 * that every data block. We have run out of space in this log record.
3108 */
3109STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003110xlog_state_switch_iclogs(
3111 struct xlog *log,
3112 struct xlog_in_core *iclog,
3113 int eventual_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114{
3115 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
3116 if (!eventual_size)
3117 eventual_size = iclog->ic_offset;
3118 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003119 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 log->l_prev_block = log->l_curr_block;
3121 log->l_prev_cycle = log->l_curr_cycle;
3122
3123 /* roll log?: ic_offset changed later */
3124 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
3125
3126 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11003127 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 log->l_mp->m_sb.sb_logsunit > 1) {
3129 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
3130 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
3131 }
3132
3133 if (log->l_curr_block >= log->l_logBBsize) {
3134 log->l_curr_cycle++;
3135 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
3136 log->l_curr_cycle++;
3137 log->l_curr_block -= log->l_logBBsize;
3138 ASSERT(log->l_curr_block >= 0);
3139 }
3140 ASSERT(iclog == log->l_iclog);
3141 log->l_iclog = iclog->ic_next;
3142} /* xlog_state_switch_iclogs */
3143
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144/*
3145 * Write out all data in the in-core log as of this exact moment in time.
3146 *
3147 * Data may be written to the in-core log during this call. However,
3148 * we don't guarantee this data will be written out. A change from past
3149 * implementation means this routine will *not* write out zero length LRs.
3150 *
3151 * Basically, we try and perform an intelligent scan of the in-core logs.
3152 * If we determine there is no flushable data, we just return. There is no
3153 * flushable data if:
3154 *
3155 * 1. the current iclog is active and has no data; the previous iclog
3156 * is in the active or dirty state.
3157 * 2. the current iclog is drity, and the previous iclog is in the
3158 * active or dirty state.
3159 *
David Chinner12017fa2008-08-13 16:34:31 +10003160 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 *
3162 * 1. the current iclog is not in the active nor dirty state.
3163 * 2. the current iclog dirty, and the previous iclog is not in the
3164 * active nor dirty state.
3165 * 3. the current iclog is active, and there is another thread writing
3166 * to this particular iclog.
3167 * 4. a) the current iclog is active and has no other writers
3168 * b) when we return from flushing out this iclog, it is still
3169 * not in the active nor dirty state.
3170 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003171int
3172_xfs_log_force(
3173 struct xfs_mount *mp,
3174 uint flags,
3175 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176{
Mark Tinguelyad223e62012-06-14 09:22:15 -05003177 struct xlog *log = mp->m_log;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003178 struct xlog_in_core *iclog;
3179 xfs_lsn_t lsn;
3180
3181 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003183 xlog_cil_force(log);
Dave Chinner71e330b2010-05-21 14:37:18 +10003184
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003185 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186
3187 iclog = log->l_iclog;
3188 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003189 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003190 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 }
3192
3193 /* If the head iclog is not active nor dirty, we just attach
3194 * ourselves to the head and go to sleep.
3195 */
3196 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3197 iclog->ic_state == XLOG_STATE_DIRTY) {
3198 /*
3199 * If the head is dirty or (active and empty), then
3200 * we need to look at the previous iclog. If the previous
3201 * iclog is active or dirty we are done. There is nothing
3202 * to sync out. Otherwise, we attach ourselves to the
3203 * previous iclog and go to sleep.
3204 */
3205 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11003206 (atomic_read(&iclog->ic_refcnt) == 0
3207 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 iclog = iclog->ic_prev;
3209 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3210 iclog->ic_state == XLOG_STATE_DIRTY)
3211 goto no_sleep;
3212 else
3213 goto maybe_sleep;
3214 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11003215 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 /* We are the only one with access to this
3217 * iclog. Flush it out now. There should
3218 * be a roundoff of zero to show that someone
3219 * has already taken care of the roundoff from
3220 * the previous sync.
3221 */
David Chinner155cc6b2008-03-06 13:44:14 +11003222 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003223 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003225 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226
3227 if (xlog_state_release_iclog(log, iclog))
Dave Chinner24513372014-06-25 14:58:08 +10003228 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003229
3230 if (log_flushed)
3231 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003232 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003233 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 iclog->ic_state != XLOG_STATE_DIRTY)
3235 goto maybe_sleep;
3236 else
3237 goto no_sleep;
3238 } else {
3239 /* Someone else is writing to this iclog.
3240 * Use its call to flush out the data. However,
3241 * the other thread may not force out this LR,
3242 * so we mark it WANT_SYNC.
3243 */
3244 xlog_state_switch_iclogs(log, iclog, 0);
3245 goto maybe_sleep;
3246 }
3247 }
3248 }
3249
3250 /* By the time we come around again, the iclog could've been filled
3251 * which would give it another lsn. If we have a new lsn, just
3252 * return because the relevant data has been flushed.
3253 */
3254maybe_sleep:
3255 if (flags & XFS_LOG_SYNC) {
3256 /*
3257 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003258 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259 * Then we check again after waking up, in case our
3260 * sleep was disturbed by a bad news.
3261 */
3262 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003263 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003264 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 }
3266 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003267 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 /*
3269 * No need to grab the log lock here since we're
3270 * only deciding whether or not to return EIO
3271 * and the memory read should be atomic.
3272 */
3273 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003274 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003275 if (log_flushed)
3276 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277 } else {
3278
3279no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003280 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 }
3282 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003283}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284
3285/*
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003286 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3287 * about errors or whether the log was flushed or not. This is the normal
3288 * interface to use when trying to unpin items or move the log forward.
3289 */
3290void
3291xfs_log_force(
3292 xfs_mount_t *mp,
3293 uint flags)
3294{
3295 int error;
3296
Dave Chinner14c26c62012-04-24 16:33:31 +10003297 trace_xfs_log_force(mp, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003298 error = _xfs_log_force(mp, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003299 if (error)
3300 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003301}
3302
3303/*
3304 * Force the in-core log to disk for a specific LSN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305 *
3306 * Find in-core log with lsn.
3307 * If it is in the DIRTY state, just return.
3308 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3309 * state and go to sleep or return.
3310 * If it is in any other state, go to sleep or return.
3311 *
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003312 * Synchronous forces are implemented with a signal variable. All callers
3313 * to force a given lsn to disk will wait on a the sv attached to the
3314 * specific in-core log. When given in-core log finally completes its
3315 * write to disk, that thread will wake up all threads waiting on the
3316 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003318int
3319_xfs_log_force_lsn(
3320 struct xfs_mount *mp,
3321 xfs_lsn_t lsn,
3322 uint flags,
3323 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324{
Mark Tinguelyad223e62012-06-14 09:22:15 -05003325 struct xlog *log = mp->m_log;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003326 struct xlog_in_core *iclog;
3327 int already_slept = 0;
3328
3329 ASSERT(lsn != 0);
3330
3331 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003333 lsn = xlog_cil_force_lsn(log, lsn);
3334 if (lsn == NULLCOMMITLSN)
3335 return 0;
Dave Chinner71e330b2010-05-21 14:37:18 +10003336
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337try_again:
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003338 spin_lock(&log->l_icloglock);
3339 iclog = log->l_iclog;
3340 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003341 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003342 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 }
3344
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003345 do {
3346 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3347 iclog = iclog->ic_next;
3348 continue;
3349 }
3350
3351 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3352 spin_unlock(&log->l_icloglock);
3353 return 0;
3354 }
3355
3356 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3357 /*
3358 * We sleep here if we haven't already slept (e.g.
3359 * this is the first time we've looked at the correct
3360 * iclog buf) and the buffer before us is going to
3361 * be sync'ed. The reason for this is that if we
3362 * are doing sync transactions here, by waiting for
3363 * the previous I/O to complete, we can allow a few
3364 * more transactions into this iclog before we close
3365 * it down.
3366 *
3367 * Otherwise, we mark the buffer WANT_SYNC, and bump
3368 * up the refcnt so we can release the log (which
3369 * drops the ref count). The state switch keeps new
3370 * transaction commits from using this buffer. When
3371 * the current commits finish writing into the buffer,
3372 * the refcount will drop to zero and the buffer will
3373 * go out then.
3374 */
3375 if (!already_slept &&
3376 (iclog->ic_prev->ic_state &
3377 (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3378 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3379
3380 XFS_STATS_INC(xs_log_force_sleep);
3381
Dave Chinnereb40a872010-12-21 12:09:01 +11003382 xlog_wait(&iclog->ic_prev->ic_write_wait,
3383 &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003384 if (log_flushed)
3385 *log_flushed = 1;
3386 already_slept = 1;
3387 goto try_again;
3388 }
David Chinner155cc6b2008-03-06 13:44:14 +11003389 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003391 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392 if (xlog_state_release_iclog(log, iclog))
Dave Chinner24513372014-06-25 14:58:08 +10003393 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003394 if (log_flushed)
3395 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003396 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003399 if ((flags & XFS_LOG_SYNC) && /* sleep */
3400 !(iclog->ic_state &
3401 (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3402 /*
3403 * Don't wait on completion if we know that we've
3404 * gotten a log write error.
3405 */
3406 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3407 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003408 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003409 }
3410 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003411 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003412 /*
3413 * No need to grab the log lock here since we're
3414 * only deciding whether or not to return EIO
3415 * and the memory read should be atomic.
3416 */
3417 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003418 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003420 if (log_flushed)
3421 *log_flushed = 1;
3422 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003423 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 }
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003425
3426 return 0;
3427 } while (iclog != log->l_iclog);
3428
3429 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003431}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003433/*
3434 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3435 * about errors or whether the log was flushed or not. This is the normal
3436 * interface to use when trying to unpin items or move the log forward.
3437 */
3438void
3439xfs_log_force_lsn(
3440 xfs_mount_t *mp,
3441 xfs_lsn_t lsn,
3442 uint flags)
3443{
3444 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445
Dave Chinner14c26c62012-04-24 16:33:31 +10003446 trace_xfs_log_force(mp, lsn);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003447 error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003448 if (error)
3449 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003450}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451
3452/*
3453 * Called when we want to mark the current iclog as being ready to sync to
3454 * disk.
3455 */
David Chinnera8272ce2007-11-23 16:28:09 +11003456STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003457xlog_state_want_sync(
3458 struct xlog *log,
3459 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003461 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462
3463 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3464 xlog_state_switch_iclogs(log, iclog, 0);
3465 } else {
3466 ASSERT(iclog->ic_state &
3467 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3468 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003469}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470
3471
3472/*****************************************************************************
3473 *
3474 * TICKET functions
3475 *
3476 *****************************************************************************
3477 */
3478
3479/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003480 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003482void
3483xfs_log_ticket_put(
3484 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003486 ASSERT(atomic_read(&ticket->t_ref) > 0);
Dave Chinnereb40a872010-12-21 12:09:01 +11003487 if (atomic_dec_and_test(&ticket->t_ref))
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003488 kmem_zone_free(xfs_log_ticket_zone, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003489}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003491xlog_ticket_t *
3492xfs_log_ticket_get(
3493 xlog_ticket_t *ticket)
3494{
3495 ASSERT(atomic_read(&ticket->t_ref) > 0);
3496 atomic_inc(&ticket->t_ref);
3497 return ticket;
3498}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
3500/*
Jie Liue773fc92013-08-12 20:50:01 +10003501 * Figure out the total log space unit (in bytes) that would be
3502 * required for a log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 */
Jie Liue773fc92013-08-12 20:50:01 +10003504int
3505xfs_log_calc_unit_res(
3506 struct xfs_mount *mp,
3507 int unit_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508{
Jie Liue773fc92013-08-12 20:50:01 +10003509 struct xlog *log = mp->m_log;
3510 int iclog_space;
3511 uint num_headers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512
3513 /*
3514 * Permanent reservations have up to 'cnt'-1 active log operations
3515 * in the log. A unit in this case is the amount of space for one
3516 * of these log operations. Normal reservations have a cnt of 1
3517 * and their unit amount is the total amount of space required.
3518 *
3519 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003520 * which occupy space in the on-disk log.
3521 *
3522 * Normal form of a transaction is:
3523 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3524 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3525 *
3526 * We need to account for all the leadup data and trailer data
3527 * around the transaction data.
3528 * And then we need to account for the worst case in terms of using
3529 * more space.
3530 * The worst case will happen if:
3531 * - the placement of the transaction happens to be such that the
3532 * roundoff is at its maximum
3533 * - the transaction data is synced before the commit record is synced
3534 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3535 * Therefore the commit record is in its own Log Record.
3536 * This can happen as the commit record is called with its
3537 * own region to xlog_write().
3538 * This then means that in the worst case, roundoff can happen for
3539 * the commit-rec as well.
3540 * The commit-rec is smaller than padding in this scenario and so it is
3541 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 */
3543
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003544 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003546 unit_bytes += sizeof(xfs_trans_header_t);
3547
3548 /* for start-rec */
3549 unit_bytes += sizeof(xlog_op_header_t);
3550
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003551 /*
3552 * for LR headers - the space for data in an iclog is the size minus
3553 * the space used for the headers. If we use the iclog size, then we
3554 * undercalculate the number of headers required.
3555 *
3556 * Furthermore - the addition of op headers for split-recs might
3557 * increase the space required enough to require more log and op
3558 * headers, so take that into account too.
3559 *
3560 * IMPORTANT: This reservation makes the assumption that if this
3561 * transaction is the first in an iclog and hence has the LR headers
3562 * accounted to it, then the remaining space in the iclog is
3563 * exclusively for this transaction. i.e. if the transaction is larger
3564 * than the iclog, it will be the only thing in that iclog.
3565 * Fundamentally, this means we must pass the entire log vector to
3566 * xlog_write to guarantee this.
3567 */
3568 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3569 num_headers = howmany(unit_bytes, iclog_space);
3570
3571 /* for split-recs - ophdrs added when data split over LRs */
3572 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3573
3574 /* add extra header reservations if we overrun */
3575 while (!num_headers ||
3576 howmany(unit_bytes, iclog_space) > num_headers) {
3577 unit_bytes += sizeof(xlog_op_header_t);
3578 num_headers++;
3579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580 unit_bytes += log->l_iclog_hsize * num_headers;
3581
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003582 /* for commit-rec LR header - note: padding will subsume the ophdr */
3583 unit_bytes += log->l_iclog_hsize;
3584
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003585 /* for roundoff padding for transaction data and one for commit record */
Jie Liue773fc92013-08-12 20:50:01 +10003586 if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) {
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003587 /* log su roundoff */
Jie Liue773fc92013-08-12 20:50:01 +10003588 unit_bytes += 2 * mp->m_sb.sb_logsunit;
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003589 } else {
3590 /* BB roundoff */
Jie Liue773fc92013-08-12 20:50:01 +10003591 unit_bytes += 2 * BBSIZE;
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003592 }
3593
Jie Liue773fc92013-08-12 20:50:01 +10003594 return unit_bytes;
3595}
3596
3597/*
3598 * Allocate and initialise a new log ticket.
3599 */
3600struct xlog_ticket *
3601xlog_ticket_alloc(
3602 struct xlog *log,
3603 int unit_bytes,
3604 int cnt,
3605 char client,
3606 bool permanent,
3607 xfs_km_flags_t alloc_flags)
3608{
3609 struct xlog_ticket *tic;
3610 int unit_res;
3611
3612 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
3613 if (!tic)
3614 return NULL;
3615
3616 unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);
3617
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003618 atomic_set(&tic->t_ref, 1);
Christoph Hellwig14a72352012-02-20 02:31:24 +00003619 tic->t_task = current;
Dave Chinner10547942010-12-21 12:02:25 +11003620 INIT_LIST_HEAD(&tic->t_queue);
Jie Liue773fc92013-08-12 20:50:01 +10003621 tic->t_unit_res = unit_res;
3622 tic->t_curr_res = unit_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 tic->t_cnt = cnt;
3624 tic->t_ocnt = cnt;
Akinobu Mitaecb34032013-03-04 21:58:20 +09003625 tic->t_tid = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 tic->t_clientid = client;
3627 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003628 tic->t_trans_type = 0;
Christoph Hellwig9006fb92012-02-20 02:31:31 +00003629 if (permanent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 tic->t_flags |= XLOG_TIC_PERM_RESERV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631
Christoph Hellwig0adba532007-08-30 17:21:46 +10003632 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003633
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003635}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636
3637
3638/******************************************************************************
3639 *
3640 * Log debug routines
3641 *
3642 ******************************************************************************
3643 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003644#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645/*
3646 * Make sure that the destination ptr is within the valid data region of
3647 * one of the iclogs. This uses backup pointers stored in a different
3648 * part of the log in case we trash the log structure.
3649 */
3650void
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003651xlog_verify_dest_ptr(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003652 struct xlog *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003653 char *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654{
3655 int i;
3656 int good_ptr = 0;
3657
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003658 for (i = 0; i < log->l_iclog_bufs; i++) {
3659 if (ptr >= log->l_iclog_bak[i] &&
3660 ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 good_ptr++;
3662 }
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003663
3664 if (!good_ptr)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003665 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003666}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003668/*
3669 * Check to make sure the grant write head didn't just over lap the tail. If
3670 * the cycles are the same, we can't be overlapping. Otherwise, make sure that
3671 * the cycles differ by exactly one and check the byte count.
3672 *
3673 * This check is run unlocked, so can give false positives. Rather than assert
3674 * on failures, use a warn-once flag and a panic tag to allow the admin to
3675 * determine if they want to panic the machine when such an error occurs. For
3676 * debug kernels this will have the same effect as using an assert but, unlinke
3677 * an assert, it can be turned off at runtime.
3678 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679STATIC void
Dave Chinner3f336c62010-12-21 12:02:52 +11003680xlog_verify_grant_tail(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003681 struct xlog *log)
Dave Chinner3f336c62010-12-21 12:02:52 +11003682{
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003683 int tail_cycle, tail_blocks;
Dave Chinnera69ed032010-12-21 12:08:20 +11003684 int cycle, space;
Dave Chinner3f336c62010-12-21 12:02:52 +11003685
Christoph Hellwig28496962012-02-20 02:31:25 +00003686 xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003687 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3688 if (tail_cycle != cycle) {
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003689 if (cycle - 1 != tail_cycle &&
3690 !(log->l_flags & XLOG_TAIL_WARN)) {
3691 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3692 "%s: cycle - 1 != tail_cycle", __func__);
3693 log->l_flags |= XLOG_TAIL_WARN;
3694 }
3695
3696 if (space > BBTOB(tail_blocks) &&
3697 !(log->l_flags & XLOG_TAIL_WARN)) {
3698 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3699 "%s: space > BBTOB(tail_blocks)", __func__);
3700 log->l_flags |= XLOG_TAIL_WARN;
3701 }
Dave Chinner3f336c62010-12-21 12:02:52 +11003702 }
3703}
3704
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705/* check if it will fit */
3706STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003707xlog_verify_tail_lsn(
3708 struct xlog *log,
3709 struct xlog_in_core *iclog,
3710 xfs_lsn_t tail_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711{
3712 int blocks;
3713
3714 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3715 blocks =
3716 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3717 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003718 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003719 } else {
3720 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3721
3722 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003723 xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724
3725 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3726 if (blocks < BTOBB(iclog->ic_offset) + 1)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003727 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728 }
3729} /* xlog_verify_tail_lsn */
3730
3731/*
3732 * Perform a number of checks on the iclog before writing to disk.
3733 *
3734 * 1. Make sure the iclogs are still circular
3735 * 2. Make sure we have a good magic number
3736 * 3. Make sure we don't have magic numbers in the data
3737 * 4. Check fields of each log operation header for:
3738 * A. Valid client identifier
3739 * B. tid ptr value falls in valid ptr space (user space code)
3740 * C. Length in log record header is correct according to the
3741 * individual operation headers within record.
3742 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3743 * log, check the preceding blocks of the physical log to make sure all
3744 * the cycle numbers agree with the current cycle number.
3745 */
3746STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003747xlog_verify_iclog(
3748 struct xlog *log,
3749 struct xlog_in_core *iclog,
3750 int count,
Thiago Farina667a9292012-11-12 21:32:59 -02003751 bool syncing)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003752{
3753 xlog_op_header_t *ophead;
3754 xlog_in_core_t *icptr;
3755 xlog_in_core_2_t *xhdr;
3756 xfs_caddr_t ptr;
3757 xfs_caddr_t base_ptr;
3758 __psint_t field_offset;
3759 __uint8_t clientid;
3760 int len, i, j, k, op_len;
3761 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762
3763 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003764 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 icptr = log->l_iclog;
Geyslan G. Bem643f7c42013-10-30 16:01:00 -05003766 for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
3767 ASSERT(icptr);
3768
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 if (icptr != log->l_iclog)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003770 xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003771 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772
3773 /* check log magic numbers */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003774 if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003775 xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003776
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003777 ptr = (xfs_caddr_t) &iclog->ic_header;
3778 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003779 ptr += BBSIZE) {
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003780 if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003781 xfs_emerg(log->l_mp, "%s: unexpected magic num",
3782 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783 }
3784
3785 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003786 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 ptr = iclog->ic_datap;
3788 base_ptr = ptr;
3789 ophead = (xlog_op_header_t *)ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003790 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791 for (i = 0; i < len; i++) {
3792 ophead = (xlog_op_header_t *)ptr;
3793
3794 /* clientid is only 1 byte */
3795 field_offset = (__psint_t)
3796 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
Thiago Farina667a9292012-11-12 21:32:59 -02003797 if (!syncing || (field_offset & 0x1ff)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 clientid = ophead->oh_clientid;
3799 } else {
3800 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3801 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3802 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3803 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003804 clientid = xlog_get_client_id(
3805 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003807 clientid = xlog_get_client_id(
3808 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 }
3810 }
3811 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003812 xfs_warn(log->l_mp,
3813 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3814 __func__, clientid, ophead,
3815 (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816
3817 /* check length */
3818 field_offset = (__psint_t)
3819 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
Thiago Farina667a9292012-11-12 21:32:59 -02003820 if (!syncing || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003821 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 } else {
3823 idx = BTOBBT((__psint_t)&ophead->oh_len -
3824 (__psint_t)iclog->ic_datap);
3825 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3826 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3827 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003828 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003830 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003831 }
3832 }
3833 ptr += sizeof(xlog_op_header_t) + op_len;
3834 }
3835} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003836#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003837
3838/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003839 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003840 */
3841STATIC int
3842xlog_state_ioerror(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003843 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844{
3845 xlog_in_core_t *iclog, *ic;
3846
3847 iclog = log->l_iclog;
3848 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3849 /*
3850 * Mark all the incore logs IOERROR.
3851 * From now on, no log flushes will result.
3852 */
3853 ic = iclog;
3854 do {
3855 ic->ic_state = XLOG_STATE_IOERROR;
3856 ic = ic->ic_next;
3857 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003858 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859 }
3860 /*
3861 * Return non-zero, if state transition has already happened.
3862 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003863 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864}
3865
3866/*
3867 * This is called from xfs_force_shutdown, when we're forcibly
3868 * shutting down the filesystem, typically because of an IO error.
3869 * Our main objectives here are to make sure that:
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003870 * a. if !logerror, flush the logs to disk. Anything modified
3871 * after this is ignored.
3872 * b. the filesystem gets marked 'SHUTDOWN' for all interested
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 * parties to find out, 'atomically'.
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003874 * c. those who're sleeping on log reservations, pinned objects and
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 * other resources get woken up, and be told the bad news.
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003876 * d. nothing new gets queued up after (b) and (c) are done.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003877 *
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003878 * Note: for the !logerror case we need to flush the regions held in memory out
3879 * to disk first. This needs to be done before the log is marked as shutdown,
3880 * otherwise the iclog writes will fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 */
3882int
3883xfs_log_force_umount(
3884 struct xfs_mount *mp,
3885 int logerror)
3886{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003887 struct xlog *log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889
3890 log = mp->m_log;
3891
3892 /*
3893 * If this happens during log recovery, don't worry about
3894 * locking; the log isn't open for business yet.
3895 */
3896 if (!log ||
3897 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3898 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003899 if (mp->m_sb_bp)
3900 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003901 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902 }
3903
3904 /*
3905 * Somebody could've already done the hard work for us.
3906 * No need to get locks for this.
3907 */
3908 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3909 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003910 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003911 }
Dave Chinner9da1ab12010-05-17 15:51:59 +10003912
3913 /*
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003914 * Flush all the completed transactions to disk before marking the log
3915 * being shut down. We need to do it in this order to ensure that
3916 * completed operations are safely on disk before we shut down, and that
3917 * we don't have to issue any buffer IO after the shutdown flags are set
3918 * to guarantee this.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003919 */
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003920 if (!logerror)
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003921 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
Dave Chinner9da1ab12010-05-17 15:51:59 +10003922
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923 /*
Dave Chinner3f16b982010-12-21 12:29:01 +11003924 * mark the filesystem and the as in a shutdown state and wake
3925 * everybody up to tell them the bad news.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003927 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003929 if (mp->m_sb_bp)
3930 XFS_BUF_DONE(mp->m_sb_bp);
3931
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 /*
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003933 * Mark the log and the iclogs with IO error flags to prevent any
3934 * further log IO from being issued or completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935 */
3936 log->l_flags |= XLOG_IO_ERROR;
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003937 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003938 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939
3940 /*
Dave Chinner10547942010-12-21 12:02:25 +11003941 * We don't want anybody waiting for log reservations after this. That
3942 * means we have to wake up everybody queued up on reserveq as well as
3943 * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3944 * we don't enqueue anything once the SHUTDOWN flag is set, and this
Dave Chinner3f16b982010-12-21 12:29:01 +11003945 * action is protected by the grant locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003946 */
Christoph Hellwiga79bf2d2012-02-20 02:31:27 +00003947 xlog_grant_head_wake_all(&log->l_reserve_head);
3948 xlog_grant_head_wake_all(&log->l_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 /*
Dave Chinnerac983512014-05-07 08:05:50 +10003951 * Wake up everybody waiting on xfs_log_force. Wake the CIL push first
3952 * as if the log writes were completed. The abort handling in the log
3953 * item committed callback functions will do this again under lock to
3954 * avoid races.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 */
Dave Chinnerac983512014-05-07 08:05:50 +10003956 wake_up_all(&log->l_cilp->xc_commit_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3958
3959#ifdef XFSERRORDEBUG
3960 {
3961 xlog_in_core_t *iclog;
3962
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003963 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 iclog = log->l_iclog;
3965 do {
3966 ASSERT(iclog->ic_callback == 0);
3967 iclog = iclog->ic_next;
3968 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003969 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 }
3971#endif
3972 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01003973 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003974}
3975
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003976STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003977xlog_iclogs_empty(
3978 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003979{
3980 xlog_in_core_t *iclog;
3981
3982 iclog = log->l_iclog;
3983 do {
3984 /* endianness does not matter here, zero is zero in
3985 * any language.
3986 */
3987 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01003988 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 iclog = iclog->ic_next;
3990 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003991 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992}
Dave Chinnerf661f1e2012-10-08 21:56:02 +11003993