blob: aaadee0969c929022725b11dcc2cd3a66995bd29 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_mount.h"
25#include "xfs_error.h"
Dave Chinner239880e2013-10-23 10:50:10 +110026#include "xfs_trans.h"
27#include "xfs_trans_priv.h"
28#include "xfs_log.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_log_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_log_recover.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_inode.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000032#include "xfs_trace.h"
Dave Chinnerf661f1e2012-10-08 21:56:02 +110033#include "xfs_fsops.h"
Christoph Hellwig0e446be2012-11-12 22:54:24 +110034#include "xfs_cksum.h"
Brian Fosterbaff4e42014-07-15 08:07:29 +100035#include "xfs_sysfs.h"
Dave Chinner61e63ec2015-01-22 09:10:31 +110036#include "xfs_sb.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
David Chinnereb01c9c2008-04-10 12:18:46 +100038kmem_zone_t *xfs_log_ticket_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Linus Torvalds1da177e2005-04-16 15:20:36 -070040/* Local miscellaneous function prototypes */
Mark Tinguelyad223e62012-06-14 09:22:15 -050041STATIC int
42xlog_commit_record(
43 struct xlog *log,
44 struct xlog_ticket *ticket,
45 struct xlog_in_core **iclog,
46 xfs_lsn_t *commitlsnp);
47
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050048STATIC struct xlog *
49xlog_alloc_log(
50 struct xfs_mount *mp,
51 struct xfs_buftarg *log_target,
52 xfs_daddr_t blk_offset,
53 int num_bblks);
Mark Tinguelyad223e62012-06-14 09:22:15 -050054STATIC int
55xlog_space_left(
56 struct xlog *log,
57 atomic64_t *head);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050058STATIC int
59xlog_sync(
60 struct xlog *log,
61 struct xlog_in_core *iclog);
62STATIC void
63xlog_dealloc_log(
64 struct xlog *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66/* local state machine functions */
67STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050068STATIC void
69xlog_state_do_callback(
70 struct xlog *log,
71 int aborted,
72 struct xlog_in_core *iclog);
73STATIC int
74xlog_state_get_iclog_space(
75 struct xlog *log,
76 int len,
77 struct xlog_in_core **iclog,
78 struct xlog_ticket *ticket,
79 int *continued_write,
80 int *logoffsetp);
81STATIC int
82xlog_state_release_iclog(
83 struct xlog *log,
84 struct xlog_in_core *iclog);
85STATIC void
86xlog_state_switch_iclogs(
87 struct xlog *log,
88 struct xlog_in_core *iclog,
89 int eventual_size);
90STATIC void
91xlog_state_want_sync(
92 struct xlog *log,
93 struct xlog_in_core *iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Mark Tinguelyad223e62012-06-14 09:22:15 -050095STATIC void
96xlog_grant_push_ail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -050097 struct xlog *log,
98 int need_bytes);
99STATIC void
100xlog_regrant_reserve_log_space(
101 struct xlog *log,
102 struct xlog_ticket *ticket);
103STATIC void
104xlog_ungrant_log_space(
105 struct xlog *log,
106 struct xlog_ticket *ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100108#if defined(DEBUG)
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500109STATIC void
110xlog_verify_dest_ptr(
111 struct xlog *log,
Christoph Hellwig5809d5e2015-06-22 09:44:47 +1000112 void *ptr);
Mark Tinguelyad223e62012-06-14 09:22:15 -0500113STATIC void
114xlog_verify_grant_tail(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500115 struct xlog *log);
116STATIC void
117xlog_verify_iclog(
118 struct xlog *log,
119 struct xlog_in_core *iclog,
120 int count,
Thiago Farina667a9292012-11-12 21:32:59 -0200121 bool syncing);
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500122STATIC void
123xlog_verify_tail_lsn(
124 struct xlog *log,
125 struct xlog_in_core *iclog,
126 xfs_lsn_t tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127#else
128#define xlog_verify_dest_ptr(a,b)
Dave Chinner3f336c62010-12-21 12:02:52 +1100129#define xlog_verify_grant_tail(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130#define xlog_verify_iclog(a,b,c,d)
131#define xlog_verify_tail_lsn(a,b,c)
132#endif
133
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500134STATIC int
135xlog_iclogs_empty(
136 struct xlog *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100138static void
Dave Chinner663e4962010-12-21 12:06:05 +1100139xlog_grant_sub_space(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500140 struct xlog *log,
141 atomic64_t *head,
142 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100143{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100144 int64_t head_val = atomic64_read(head);
145 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100146
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100147 do {
148 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100149
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100150 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100151
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100152 space -= bytes;
153 if (space < 0) {
154 space += log->l_logsize;
155 cycle--;
156 }
157
158 old = head_val;
159 new = xlog_assign_grant_head_val(cycle, space);
160 head_val = atomic64_cmpxchg(head, old, new);
161 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100162}
163
164static void
Dave Chinner663e4962010-12-21 12:06:05 +1100165xlog_grant_add_space(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500166 struct xlog *log,
167 atomic64_t *head,
168 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100169{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100170 int64_t head_val = atomic64_read(head);
171 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100172
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100173 do {
174 int tmp;
175 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100176
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100177 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100178
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100179 tmp = log->l_logsize - space;
180 if (tmp > bytes)
181 space += bytes;
182 else {
183 space = bytes - tmp;
184 cycle++;
185 }
186
187 old = head_val;
188 new = xlog_assign_grant_head_val(cycle, space);
189 head_val = atomic64_cmpxchg(head, old, new);
190 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100191}
Dave Chinnera69ed032010-12-21 12:08:20 +1100192
Christoph Hellwigc303c5b2012-02-20 02:31:26 +0000193STATIC void
194xlog_grant_head_init(
195 struct xlog_grant_head *head)
196{
197 xlog_assign_grant_head(&head->grant, 1, 0);
198 INIT_LIST_HEAD(&head->waiters);
199 spin_lock_init(&head->lock);
200}
201
Christoph Hellwiga79bf2d2012-02-20 02:31:27 +0000202STATIC void
203xlog_grant_head_wake_all(
204 struct xlog_grant_head *head)
205{
206 struct xlog_ticket *tic;
207
208 spin_lock(&head->lock);
209 list_for_each_entry(tic, &head->waiters, t_queue)
210 wake_up_process(tic->t_task);
211 spin_unlock(&head->lock);
212}
213
Christoph Hellwige179840d2012-02-20 02:31:29 +0000214static inline int
215xlog_ticket_reservation(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500216 struct xlog *log,
Christoph Hellwige179840d2012-02-20 02:31:29 +0000217 struct xlog_grant_head *head,
218 struct xlog_ticket *tic)
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000219{
Christoph Hellwige179840d2012-02-20 02:31:29 +0000220 if (head == &log->l_write_head) {
221 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
222 return tic->t_unit_res;
223 } else {
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000224 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
Christoph Hellwige179840d2012-02-20 02:31:29 +0000225 return tic->t_unit_res * tic->t_cnt;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000226 else
Christoph Hellwige179840d2012-02-20 02:31:29 +0000227 return tic->t_unit_res;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000228 }
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000229}
230
231STATIC bool
Christoph Hellwige179840d2012-02-20 02:31:29 +0000232xlog_grant_head_wake(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500233 struct xlog *log,
Christoph Hellwige179840d2012-02-20 02:31:29 +0000234 struct xlog_grant_head *head,
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000235 int *free_bytes)
236{
237 struct xlog_ticket *tic;
238 int need_bytes;
239
Christoph Hellwige179840d2012-02-20 02:31:29 +0000240 list_for_each_entry(tic, &head->waiters, t_queue) {
241 need_bytes = xlog_ticket_reservation(log, head, tic);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000242 if (*free_bytes < need_bytes)
243 return false;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000244
Christoph Hellwige179840d2012-02-20 02:31:29 +0000245 *free_bytes -= need_bytes;
246 trace_xfs_log_grant_wake_up(log, tic);
Christoph Hellwig14a72352012-02-20 02:31:24 +0000247 wake_up_process(tic->t_task);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000248 }
249
250 return true;
251}
252
253STATIC int
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000254xlog_grant_head_wait(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500255 struct xlog *log,
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000256 struct xlog_grant_head *head,
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000257 struct xlog_ticket *tic,
Dave Chinnera30b0362013-09-02 20:49:36 +1000258 int need_bytes) __releases(&head->lock)
259 __acquires(&head->lock)
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000260{
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000261 list_add_tail(&tic->t_queue, &head->waiters);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000262
263 do {
264 if (XLOG_FORCED_SHUTDOWN(log))
265 goto shutdown;
266 xlog_grant_push_ail(log, need_bytes);
267
Christoph Hellwig14a72352012-02-20 02:31:24 +0000268 __set_current_state(TASK_UNINTERRUPTIBLE);
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000269 spin_unlock(&head->lock);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000270
Christoph Hellwig14a72352012-02-20 02:31:24 +0000271 XFS_STATS_INC(xs_sleep_logspace);
272
273 trace_xfs_log_grant_sleep(log, tic);
274 schedule();
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000275 trace_xfs_log_grant_wake(log, tic);
276
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000277 spin_lock(&head->lock);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000278 if (XLOG_FORCED_SHUTDOWN(log))
279 goto shutdown;
Christoph Hellwig23ee3df2012-02-20 02:31:28 +0000280 } while (xlog_space_left(log, &head->grant) < need_bytes);
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000281
282 list_del_init(&tic->t_queue);
283 return 0;
284shutdown:
285 list_del_init(&tic->t_queue);
Dave Chinner24513372014-06-25 14:58:08 +1000286 return -EIO;
Christoph Hellwig9f9c19e2011-11-28 08:17:36 +0000287}
288
Christoph Hellwig42ceedb2012-02-20 02:31:30 +0000289/*
290 * Atomically get the log space required for a log ticket.
291 *
292 * Once a ticket gets put onto head->waiters, it will only return after the
293 * needed reservation is satisfied.
294 *
295 * This function is structured so that it has a lock free fast path. This is
296 * necessary because every new transaction reservation will come through this
297 * path. Hence any lock will be globally hot if we take it unconditionally on
298 * every pass.
299 *
300 * As tickets are only ever moved on and off head->waiters under head->lock, we
301 * only need to take that lock if we are going to add the ticket to the queue
302 * and sleep. We can avoid taking the lock if the ticket was never added to
303 * head->waiters because the t_queue list head will be empty and we hold the
304 * only reference to it so it can safely be checked unlocked.
305 */
306STATIC int
307xlog_grant_head_check(
Mark Tinguelyad223e62012-06-14 09:22:15 -0500308 struct xlog *log,
Christoph Hellwig42ceedb2012-02-20 02:31:30 +0000309 struct xlog_grant_head *head,
310 struct xlog_ticket *tic,
311 int *need_bytes)
312{
313 int free_bytes;
314 int error = 0;
315
316 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
317
318 /*
319 * If there are other waiters on the queue then give them a chance at
320 * logspace before us. Wake up the first waiters, if we do not wake
321 * up all the waiters then go to sleep waiting for more free space,
322 * otherwise try to get some space for this transaction.
323 */
324 *need_bytes = xlog_ticket_reservation(log, head, tic);
325 free_bytes = xlog_space_left(log, &head->grant);
326 if (!list_empty_careful(&head->waiters)) {
327 spin_lock(&head->lock);
328 if (!xlog_grant_head_wake(log, head, &free_bytes) ||
329 free_bytes < *need_bytes) {
330 error = xlog_grant_head_wait(log, head, tic,
331 *need_bytes);
332 }
333 spin_unlock(&head->lock);
334 } else if (free_bytes < *need_bytes) {
335 spin_lock(&head->lock);
336 error = xlog_grant_head_wait(log, head, tic, *need_bytes);
337 spin_unlock(&head->lock);
338 }
339
340 return error;
341}
342
Christoph Hellwig0adba532007-08-30 17:21:46 +1000343static void
344xlog_tic_reset_res(xlog_ticket_t *tic)
345{
346 tic->t_res_num = 0;
347 tic->t_res_arr_sum = 0;
348 tic->t_res_num_ophdrs = 0;
349}
350
351static void
352xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
353{
354 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
355 /* add to overflow and start again */
356 tic->t_res_o_flow += tic->t_res_arr_sum;
357 tic->t_res_num = 0;
358 tic->t_res_arr_sum = 0;
359 }
360
361 tic->t_res_arr[tic->t_res_num].r_len = len;
362 tic->t_res_arr[tic->t_res_num].r_type = type;
363 tic->t_res_arr_sum += len;
364 tic->t_res_num++;
365}
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367/*
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000368 * Replenish the byte reservation required by moving the grant write head.
369 */
370int
371xfs_log_regrant(
372 struct xfs_mount *mp,
373 struct xlog_ticket *tic)
374{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500375 struct xlog *log = mp->m_log;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000376 int need_bytes;
377 int error = 0;
378
379 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner24513372014-06-25 14:58:08 +1000380 return -EIO;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000381
382 XFS_STATS_INC(xs_try_logspace);
383
384 /*
385 * This is a new transaction on the ticket, so we need to change the
386 * transaction ID so that the next transaction has a different TID in
387 * the log. Just add one to the existing tid so that we can see chains
388 * of rolling transactions in the log easily.
389 */
390 tic->t_tid++;
391
392 xlog_grant_push_ail(log, tic->t_unit_res);
393
394 tic->t_curr_res = tic->t_unit_res;
395 xlog_tic_reset_res(tic);
396
397 if (tic->t_cnt > 0)
398 return 0;
399
400 trace_xfs_log_regrant(log, tic);
401
402 error = xlog_grant_head_check(log, &log->l_write_head, tic,
403 &need_bytes);
404 if (error)
405 goto out_error;
406
407 xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
408 trace_xfs_log_regrant_exit(log, tic);
409 xlog_verify_grant_tail(log);
410 return 0;
411
412out_error:
413 /*
414 * If we are failing, make sure the ticket doesn't have any current
415 * reservations. We don't want to add this back when the ticket/
416 * transaction gets cancelled.
417 */
418 tic->t_curr_res = 0;
419 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
420 return error;
421}
422
423/*
424 * Reserve log space and return a ticket corresponding the reservation.
425 *
426 * Each reservation is going to reserve extra space for a log record header.
427 * When writes happen to the on-disk log, we don't subtract the length of the
428 * log record header from any reservation. By wasting space in each
429 * reservation, we prevent over allocation problems.
430 */
431int
432xfs_log_reserve(
433 struct xfs_mount *mp,
434 int unit_bytes,
435 int cnt,
436 struct xlog_ticket **ticp,
437 __uint8_t client,
438 bool permanent,
439 uint t_type)
440{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500441 struct xlog *log = mp->m_log;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000442 struct xlog_ticket *tic;
443 int need_bytes;
444 int error = 0;
445
446 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
447
448 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner24513372014-06-25 14:58:08 +1000449 return -EIO;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000450
451 XFS_STATS_INC(xs_try_logspace);
452
453 ASSERT(*ticp == NULL);
454 tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
455 KM_SLEEP | KM_MAYFAIL);
456 if (!tic)
Dave Chinner24513372014-06-25 14:58:08 +1000457 return -ENOMEM;
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000458
459 tic->t_trans_type = t_type;
460 *ticp = tic;
461
Dave Chinner437a2552012-11-28 13:01:00 +1100462 xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
463 : tic->t_unit_res);
Christoph Hellwig9006fb92012-02-20 02:31:31 +0000464
465 trace_xfs_log_reserve(log, tic);
466
467 error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
468 &need_bytes);
469 if (error)
470 goto out_error;
471
472 xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
473 xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
474 trace_xfs_log_reserve_exit(log, tic);
475 xlog_verify_grant_tail(log);
476 return 0;
477
478out_error:
479 /*
480 * If we are failing, make sure the ticket doesn't have any current
481 * reservations. We don't want to add this back when the ticket/
482 * transaction gets cancelled.
483 */
484 tic->t_curr_res = 0;
485 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
486 return error;
487}
488
489
490/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 * NOTES:
492 *
493 * 1. currblock field gets updated at startup and after in-core logs
494 * marked as with WANT_SYNC.
495 */
496
497/*
498 * This routine is called when a user of a log manager ticket is done with
499 * the reservation. If the ticket was ever used, then a commit record for
500 * the associated transaction is written out as a log operation header with
501 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
502 * a given ticket. If the ticket was one with a permanent reservation, then
503 * a few operations are done differently. Permanent reservation tickets by
504 * default don't release the reservation. They just commit the current
505 * transaction with the belief that the reservation is still needed. A flag
506 * must be passed in before permanent reservations are actually released.
507 * When these type of tickets are not released, they need to be set into
508 * the inited state again. By doing this, a start record will be written
509 * out when the next write occurs.
510 */
511xfs_lsn_t
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000512xfs_log_done(
513 struct xfs_mount *mp,
514 struct xlog_ticket *ticket,
515 struct xlog_in_core **iclog,
Christoph Hellwigf78c3902015-06-04 13:48:20 +1000516 bool regrant)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
Mark Tinguelyad223e62012-06-14 09:22:15 -0500518 struct xlog *log = mp->m_log;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000519 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (XLOG_FORCED_SHUTDOWN(log) ||
522 /*
523 * If nothing was ever written, don't write out commit record.
524 * If we get an error, just continue and give back the log ticket.
525 */
526 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
Dave Chinner55b66332010-03-23 11:43:17 +1100527 (xlog_commit_record(log, ticket, iclog, &lsn)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 lsn = (xfs_lsn_t) -1;
Christoph Hellwigf78c3902015-06-04 13:48:20 +1000529 regrant = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 }
531
532
Christoph Hellwigf78c3902015-06-04 13:48:20 +1000533 if (!regrant) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000534 trace_xfs_log_done_nonperm(log, ticket);
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000537 * Release ticket if not permanent reservation or a specific
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * request has been made to release a permanent reservation.
539 */
540 xlog_ungrant_log_space(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000542 trace_xfs_log_done_perm(log, ticket);
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 xlog_regrant_reserve_log_space(log, ticket);
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000545 /* If this ticket was a permanent reservation and we aren't
546 * trying to release it, reset the inited flags; so next time
547 * we write, a start record will be written out.
548 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 ticket->t_flags |= XLOG_TIC_INITED;
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Christoph Hellwigf78c3902015-06-04 13:48:20 +1000552 xfs_log_ticket_put(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 return lsn;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000554}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556/*
557 * Attaches a new iclog I/O completion callback routine during
558 * transaction commit. If the log is in error state, a non-zero
559 * return code is handed back and the caller is responsible for
560 * executing the callback at an appropriate time.
561 */
562int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000563xfs_log_notify(
564 struct xfs_mount *mp,
565 struct xlog_in_core *iclog,
566 xfs_log_callback_t *cb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000568 int abortflg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
David Chinner114d23a2008-04-10 12:18:39 +1000570 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
572 if (!abortflg) {
573 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
574 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
575 cb->cb_next = NULL;
576 *(iclog->ic_callback_tail) = cb;
577 iclog->ic_callback_tail = &(cb->cb_next);
578 }
David Chinner114d23a2008-04-10 12:18:39 +1000579 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 return abortflg;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000581}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
583int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000584xfs_log_release_iclog(
585 struct xfs_mount *mp,
586 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000588 if (xlog_state_release_iclog(mp->m_log, iclog)) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000589 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Dave Chinner24513372014-06-25 14:58:08 +1000590 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
592
593 return 0;
594}
595
596/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 * Mount a log filesystem
598 *
599 * mp - ubiquitous xfs mount point structure
600 * log_target - buftarg of on-disk log device
601 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
602 * num_bblocks - Number of BBSIZE blocks in on-disk log
603 *
604 * Return error or zero.
605 */
606int
David Chinner249a8c12008-02-05 12:13:32 +1100607xfs_log_mount(
608 xfs_mount_t *mp,
609 xfs_buftarg_t *log_target,
610 xfs_daddr_t blk_offset,
611 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Jie Liu3e7b91c2013-08-12 20:50:03 +1000613 int error = 0;
614 int min_logfsbs;
David Chinner249a8c12008-02-05 12:13:32 +1100615
Dave Chinnerc99d6092014-05-05 16:18:37 +1000616 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
617 xfs_notice(mp, "Mounting V%d Filesystem",
618 XFS_SB_VERSION_NUM(&mp->m_sb));
619 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100620 xfs_notice(mp,
Dave Chinnerc99d6092014-05-05 16:18:37 +1000621"Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.",
622 XFS_SB_VERSION_NUM(&mp->m_sb));
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000623 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625
626 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200627 if (IS_ERR(mp->m_log)) {
Dave Chinner24513372014-06-25 14:58:08 +1000628 error = PTR_ERR(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100629 goto out;
630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 /*
Jie Liu3e7b91c2013-08-12 20:50:03 +1000633 * Validate the given log space and drop a critical message via syslog
634 * if the log size is too small that would lead to some unexpected
635 * situations in transaction log space reservation stage.
636 *
637 * Note: we can't just reject the mount if the validation fails. This
638 * would mean that people would have to downgrade their kernel just to
639 * remedy the situation as there is no way to grow the log (short of
640 * black magic surgery with xfs_db).
641 *
642 * We can, however, reject mounts for CRC format filesystems, as the
643 * mkfs binary being used to make the filesystem should never create a
644 * filesystem with a log that is too small.
645 */
646 min_logfsbs = xfs_log_calc_minimum_size(mp);
647
648 if (mp->m_sb.sb_logblocks < min_logfsbs) {
649 xfs_warn(mp,
650 "Log size %d blocks too small, minimum size is %d blocks",
651 mp->m_sb.sb_logblocks, min_logfsbs);
Dave Chinner24513372014-06-25 14:58:08 +1000652 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000653 } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) {
654 xfs_warn(mp,
655 "Log size %d blocks too large, maximum size is %lld blocks",
656 mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS);
Dave Chinner24513372014-06-25 14:58:08 +1000657 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000658 } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) {
659 xfs_warn(mp,
660 "log size %lld bytes too large, maximum size is %lld bytes",
661 XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks),
662 XFS_MAX_LOG_BYTES);
Dave Chinner24513372014-06-25 14:58:08 +1000663 error = -EINVAL;
Jie Liu3e7b91c2013-08-12 20:50:03 +1000664 }
665 if (error) {
666 if (xfs_sb_version_hascrc(&mp->m_sb)) {
667 xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
668 ASSERT(0);
669 goto out_free_log;
670 }
Joe Perchesf41febd2015-07-29 11:52:04 +1000671 xfs_crit(mp, "Log size out of supported range.");
Jie Liu3e7b91c2013-08-12 20:50:03 +1000672 xfs_crit(mp,
Joe Perchesf41febd2015-07-29 11:52:04 +1000673"Continuing onwards, but if log hangs are experienced then please report this message in the bug report.");
Jie Liu3e7b91c2013-08-12 20:50:03 +1000674 }
675
676 /*
David Chinner249a8c12008-02-05 12:13:32 +1100677 * Initialize the AIL now we have a log.
678 */
David Chinner249a8c12008-02-05 12:13:32 +1100679 error = xfs_trans_ail_init(mp);
680 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100681 xfs_warn(mp, "AIL initialisation failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100682 goto out_free_log;
David Chinner249a8c12008-02-05 12:13:32 +1100683 }
David Chinnera9c21c12008-10-30 17:39:35 +1100684 mp->m_log->l_ailp = mp->m_ail;
David Chinner249a8c12008-02-05 12:13:32 +1100685
686 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 * skip log recovery on a norecovery mount. pretend it all
688 * just worked.
689 */
690 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
David Chinner249a8c12008-02-05 12:13:32 +1100691 int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000694 mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Eric Sandeen65be6052006-01-11 15:34:19 +1100696 error = xlog_recover(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000699 mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100701 xfs_warn(mp, "log mount/recovery failed: error %d",
702 error);
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000703 xlog_recover_cancel(mp->m_log);
Christoph Hellwig26430752009-02-12 19:55:48 +0100704 goto out_destroy_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
706 }
707
Brian Fosterbaff4e42014-07-15 08:07:29 +1000708 error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj,
709 "log");
710 if (error)
711 goto out_destroy_ail;
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 /* Normal transactions can now occur */
714 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
715
Dave Chinner71e330b2010-05-21 14:37:18 +1000716 /*
717 * Now the log has been fully initialised and we know were our
718 * space grant counters are, we can initialise the permanent ticket
719 * needed for delayed logging to work.
720 */
721 xlog_cil_init_post_recovery(mp->m_log);
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 return 0;
Christoph Hellwig26430752009-02-12 19:55:48 +0100724
725out_destroy_ail:
726 xfs_trans_ail_destroy(mp);
727out_free_log:
728 xlog_dealloc_log(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100729out:
David Chinner249a8c12008-02-05 12:13:32 +1100730 return error;
Christoph Hellwig26430752009-02-12 19:55:48 +0100731}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733/*
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100734 * Finish the recovery of the file system. This is separate from the
735 * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
736 * in the root and real-time bitmap inodes between calling xfs_log_mount() and
737 * here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 *
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100739 * If we finish recovery successfully, start the background log work. If we are
740 * not doing recovery, then we have a RO filesystem and we don't need to start
741 * it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 */
743int
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000744xfs_log_mount_finish(
745 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746{
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100747 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000749 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000750 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000751 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 }
753
Brian Fosterf0b2efa2015-08-19 09:58:36 +1000754 error = xlog_recover_finish(mp->m_log);
755 if (!error)
756 xfs_log_work_queue(mp);
757
758 return error;
759}
760
761/*
762 * The mount has failed. Cancel the recovery if it hasn't completed and destroy
763 * the log.
764 */
765int
766xfs_log_mount_cancel(
767 struct xfs_mount *mp)
768{
769 int error;
770
771 error = xlog_recover_cancel(mp->m_log);
772 xfs_log_unmount(mp);
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 return error;
775}
776
777/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 * Final log writes as part of unmount.
779 *
780 * Mark the filesystem clean as unmount happens. Note that during relocation
781 * this routine needs to be executed as part of source-bag while the
782 * deallocation must not be done until source-end.
783 */
784
785/*
786 * Unmount record used to have a string "Unmount filesystem--" in the
787 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
788 * We just write the magic number now since that particular field isn't
Zhi Yong Wu8e159e72013-08-12 03:15:00 +0000789 * currently architecture converted and "Unmount" is a bit foo.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 * As far as I know, there weren't any dependencies on the old behaviour.
791 */
792
793int
794xfs_log_unmount_write(xfs_mount_t *mp)
795{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500796 struct xlog *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 xlog_in_core_t *iclog;
798#ifdef DEBUG
799 xlog_in_core_t *first_iclog;
800#endif
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000801 xlog_ticket_t *tic = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 xfs_lsn_t lsn;
803 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 /*
806 * Don't write out unmount record on read-only mounts.
807 * Or, if we are doing a forced umount (typically because of IO errors).
808 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000809 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return 0;
811
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000812 error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
David Chinnerb911ca02008-04-10 12:24:30 +1000813 ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815#ifdef DEBUG
816 first_iclog = iclog = log->l_iclog;
817 do {
818 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
819 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
820 ASSERT(iclog->ic_offset == 0);
821 }
822 iclog = iclog->ic_next;
823 } while (iclog != first_iclog);
824#endif
825 if (! (XLOG_FORCED_SHUTDOWN(log))) {
Tim Shimmin955e47a2006-09-28 11:04:16 +1000826 error = xfs_log_reserve(mp, 600, 1, &tic,
827 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 if (!error) {
Dave Chinner55b66332010-03-23 11:43:17 +1100829 /* the data section must be 32 bit size aligned */
830 struct {
831 __uint16_t magic;
832 __uint16_t pad1;
833 __uint32_t pad2; /* may as well make it 64 bits */
834 } magic = {
835 .magic = XLOG_UNMOUNT_TYPE,
836 };
837 struct xfs_log_iovec reg = {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000838 .i_addr = &magic,
Dave Chinner55b66332010-03-23 11:43:17 +1100839 .i_len = sizeof(magic),
840 .i_type = XLOG_REG_TYPE_UNMOUNT,
841 };
842 struct xfs_log_vec vec = {
843 .lv_niovecs = 1,
844 .lv_iovecp = &reg,
845 };
846
Dave Chinner39486592012-03-22 05:15:11 +0000847 /* remove inited flag, and account for space used */
Dave Chinner55b66332010-03-23 11:43:17 +1100848 tic->t_flags = 0;
Dave Chinner39486592012-03-22 05:15:11 +0000849 tic->t_curr_res -= sizeof(magic);
Dave Chinner55b66332010-03-23 11:43:17 +1100850 error = xlog_write(log, &vec, tic, &lsn,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 NULL, XLOG_UNMOUNT_TRANS);
852 /*
853 * At this point, we're umounting anyway,
854 * so there's no point in transitioning log state
855 * to IOERROR. Just continue...
856 */
857 }
858
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100859 if (error)
860 xfs_alert(mp, "%s: unmount record failed", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
862
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000863 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100865 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100867 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000868 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000870 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
872 iclog->ic_state == XLOG_STATE_DIRTY)) {
873 if (!XLOG_FORCED_SHUTDOWN(log)) {
Dave Chinnereb40a872010-12-21 12:09:01 +1100874 xlog_wait(&iclog->ic_force_wait,
875 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000877 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
879 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000880 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
Tim Shimmin955e47a2006-09-28 11:04:16 +1000882 if (tic) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000883 trace_xfs_log_umount_write(log, tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000884 xlog_ungrant_log_space(log, tic);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100885 xfs_log_ticket_put(tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 } else {
888 /*
889 * We're already in forced_shutdown mode, couldn't
890 * even attempt to write out the unmount transaction.
891 *
892 * Go through the motions of sync'ing and releasing
893 * the iclog, even though no I/O will actually happen,
Nathan Scottc41564b2006-03-29 08:55:14 +1000894 * we need to wait for other log I/Os that may already
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 * be in progress. Do this as a separate section of
896 * code so we'll know if we ever get stuck here that
897 * we're in this odd situation of trying to unmount
898 * a file system that went into forced_shutdown as
899 * the result of an unmount..
900 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000901 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100903 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100906 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000907 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000909 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
911 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
912 || iclog->ic_state == XLOG_STATE_DIRTY
913 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
914
Dave Chinnereb40a872010-12-21 12:09:01 +1100915 xlog_wait(&iclog->ic_force_wait,
916 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000918 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 }
920 }
921
David Chinner1bb7d6b2008-04-10 12:24:38 +1000922 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923} /* xfs_log_unmount_write */
924
925/*
Dave Chinnerc75921a2012-10-08 21:56:08 +1100926 * Empty the log for unmount/freeze.
Dave Chinnercf2931d2012-10-08 21:56:03 +1100927 *
928 * To do this, we first need to shut down the background log work so it is not
929 * trying to cover the log as we clean up. We then need to unpin all objects in
930 * the log so we can then flush them out. Once they have completed their IO and
931 * run the callbacks removing themselves from the AIL, we can write the unmount
Dave Chinnerc75921a2012-10-08 21:56:08 +1100932 * record.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 */
934void
Dave Chinnerc75921a2012-10-08 21:56:08 +1100935xfs_log_quiesce(
936 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100938 cancel_delayed_work_sync(&mp->m_log->l_work);
Dave Chinnercf2931d2012-10-08 21:56:03 +1100939 xfs_log_force(mp, XFS_LOG_SYNC);
940
941 /*
942 * The superblock buffer is uncached and while xfs_ail_push_all_sync()
943 * will push it, xfs_wait_buftarg() will not wait for it. Further,
944 * xfs_buf_iowait() cannot be used because it was pushed with the
945 * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
946 * the IO to complete.
947 */
948 xfs_ail_push_all_sync(mp->m_ail);
949 xfs_wait_buftarg(mp->m_ddev_targp);
950 xfs_buf_lock(mp->m_sb_bp);
951 xfs_buf_unlock(mp->m_sb_bp);
952
953 xfs_log_unmount_write(mp);
Dave Chinnerc75921a2012-10-08 21:56:08 +1100954}
955
956/*
957 * Shut down and release the AIL and Log.
958 *
959 * During unmount, we need to ensure we flush all the dirty metadata objects
960 * from the AIL so that the log is empty before we write the unmount record to
961 * the log. Once this is done, we can tear down the AIL and the log.
962 */
963void
964xfs_log_unmount(
965 struct xfs_mount *mp)
966{
967 xfs_log_quiesce(mp);
Dave Chinnercf2931d2012-10-08 21:56:03 +1100968
David Chinner249a8c12008-02-05 12:13:32 +1100969 xfs_trans_ail_destroy(mp);
Brian Fosterbaff4e42014-07-15 08:07:29 +1000970
971 xfs_sysfs_del(&mp->m_log->l_kobj);
972
Nathan Scottc41564b2006-03-29 08:55:14 +1000973 xlog_dealloc_log(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975
Dave Chinner43f5efc2010-03-23 10:10:00 +1100976void
977xfs_log_item_init(
978 struct xfs_mount *mp,
979 struct xfs_log_item *item,
980 int type,
Christoph Hellwig272e42b2011-10-28 09:54:24 +0000981 const struct xfs_item_ops *ops)
Dave Chinner43f5efc2010-03-23 10:10:00 +1100982{
983 item->li_mountp = mp;
984 item->li_ailp = mp->m_ail;
985 item->li_type = type;
986 item->li_ops = ops;
Dave Chinner71e330b2010-05-21 14:37:18 +1000987 item->li_lv = NULL;
988
989 INIT_LIST_HEAD(&item->li_ail);
990 INIT_LIST_HEAD(&item->li_cil);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100991}
992
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000993/*
994 * Wake up processes waiting for log space after we have moved the log tail.
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000995 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996void
Christoph Hellwig09a423a2012-02-20 02:31:20 +0000997xfs_log_space_wake(
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +0000998 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Mark Tinguelyad223e62012-06-14 09:22:15 -05001000 struct xlog *log = mp->m_log;
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +00001001 int free_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 if (XLOG_FORCED_SHUTDOWN(log))
1004 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Christoph Hellwig28496962012-02-20 02:31:25 +00001006 if (!list_empty_careful(&log->l_write_head.waiters)) {
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001007 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
1008
Christoph Hellwig28496962012-02-20 02:31:25 +00001009 spin_lock(&log->l_write_head.lock);
1010 free_bytes = xlog_space_left(log, &log->l_write_head.grant);
Christoph Hellwige179840d2012-02-20 02:31:29 +00001011 xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
Christoph Hellwig28496962012-02-20 02:31:25 +00001012 spin_unlock(&log->l_write_head.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 }
Dave Chinner10547942010-12-21 12:02:25 +11001014
Christoph Hellwig28496962012-02-20 02:31:25 +00001015 if (!list_empty_careful(&log->l_reserve_head.waiters)) {
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001016 ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));
1017
Christoph Hellwig28496962012-02-20 02:31:25 +00001018 spin_lock(&log->l_reserve_head.lock);
1019 free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
Christoph Hellwige179840d2012-02-20 02:31:29 +00001020 xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
Christoph Hellwig28496962012-02-20 02:31:25 +00001021 spin_unlock(&log->l_reserve_head.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
Dave Chinner3f16b982010-12-21 12:29:01 +11001023}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025/*
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001026 * Determine if we have a transaction that has gone to disk that needs to be
1027 * covered. To begin the transition to the idle state firstly the log needs to
1028 * be idle. That means the CIL, the AIL and the iclogs needs to be empty before
1029 * we start attempting to cover the log.
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001030 *
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001031 * Only if we are then in a state where covering is needed, the caller is
1032 * informed that dummy transactions are required to move the log into the idle
1033 * state.
1034 *
1035 * If there are any items in the AIl or CIL, then we do not want to attempt to
1036 * cover the log as we may be in a situation where there isn't log space
1037 * available to run a dummy transaction and this can lead to deadlocks when the
1038 * tail of the log is pinned by an item that is modified in the CIL. Hence
1039 * there's no point in running a dummy transaction at this point because we
1040 * can't start trying to idle the log until both the CIL and AIL are empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 */
1042int
1043xfs_log_need_covered(xfs_mount_t *mp)
1044{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001045 struct xlog *log = mp->m_log;
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001046 int needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Brian Foster91ee5752014-11-28 14:02:59 +11001048 if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 return 0;
1050
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001051 if (!xlog_cil_empty(log))
1052 return 0;
1053
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001054 spin_lock(&log->l_icloglock);
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001055 switch (log->l_covered_state) {
1056 case XLOG_STATE_COVER_DONE:
1057 case XLOG_STATE_COVER_DONE2:
1058 case XLOG_STATE_COVER_IDLE:
1059 break;
1060 case XLOG_STATE_COVER_NEED:
1061 case XLOG_STATE_COVER_NEED2:
Dave Chinner2c6e24c2013-10-15 09:17:49 +11001062 if (xfs_ail_min_lsn(log->l_ailp))
1063 break;
1064 if (!xlog_iclogs_empty(log))
1065 break;
1066
1067 needed = 1;
1068 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
1069 log->l_covered_state = XLOG_STATE_COVER_DONE;
1070 else
1071 log->l_covered_state = XLOG_STATE_COVER_DONE2;
1072 break;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001073 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 needed = 1;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +10001075 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001077 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +01001078 return needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079}
1080
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001081/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 * We may be holding the log iclog lock upon entering this routine.
1083 */
1084xfs_lsn_t
Christoph Hellwig1c304622012-04-23 15:58:33 +10001085xlog_assign_tail_lsn_locked(
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001086 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087{
Mark Tinguelyad223e62012-06-14 09:22:15 -05001088 struct xlog *log = mp->m_log;
Christoph Hellwig1c304622012-04-23 15:58:33 +10001089 struct xfs_log_item *lip;
1090 xfs_lsn_t tail_lsn;
1091
1092 assert_spin_locked(&mp->m_ail->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001094 /*
1095 * To make sure we always have a valid LSN for the log tail we keep
1096 * track of the last LSN which was committed in log->l_last_sync_lsn,
Christoph Hellwig1c304622012-04-23 15:58:33 +10001097 * and use that when the AIL was empty.
Christoph Hellwig09a423a2012-02-20 02:31:20 +00001098 */
Christoph Hellwig1c304622012-04-23 15:58:33 +10001099 lip = xfs_ail_min(mp->m_ail);
1100 if (lip)
1101 tail_lsn = lip->li_lsn;
1102 else
Dave Chinner84f3c682010-12-03 22:11:29 +11001103 tail_lsn = atomic64_read(&log->l_last_sync_lsn);
Dave Chinner750b9c92013-11-01 15:27:18 +11001104 trace_xfs_log_assign_tail_lsn(log, tail_lsn);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001105 atomic64_set(&log->l_tail_lsn, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 return tail_lsn;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001107}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Christoph Hellwig1c304622012-04-23 15:58:33 +10001109xfs_lsn_t
1110xlog_assign_tail_lsn(
1111 struct xfs_mount *mp)
1112{
1113 xfs_lsn_t tail_lsn;
1114
1115 spin_lock(&mp->m_ail->xa_lock);
1116 tail_lsn = xlog_assign_tail_lsn_locked(mp);
1117 spin_unlock(&mp->m_ail->xa_lock);
1118
1119 return tail_lsn;
1120}
1121
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122/*
1123 * Return the space in the log between the tail and the head. The head
1124 * is passed in the cycle/bytes formal parms. In the special case where
1125 * the reserve head has wrapped passed the tail, this calculation is no
1126 * longer valid. In this case, just return 0 which means there is no space
1127 * in the log. This works for all places where this function is called
1128 * with the reserve head. Of course, if the write head were to ever
1129 * wrap the tail, we should blow up. Rather than catch this case here,
1130 * we depend on other ASSERTions in other parts of the code. XXXmiken
1131 *
1132 * This code also handles the case where the reservation head is behind
1133 * the tail. The details of this case are described below, but the end
1134 * result is that we return the size of the log as the amount of space left.
1135 */
David Chinnera8272ce2007-11-23 16:28:09 +11001136STATIC int
Dave Chinnera69ed032010-12-21 12:08:20 +11001137xlog_space_left(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001138 struct xlog *log,
Dave Chinnerc8a09ff2010-12-04 00:02:40 +11001139 atomic64_t *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140{
Dave Chinnera69ed032010-12-21 12:08:20 +11001141 int free_bytes;
1142 int tail_bytes;
1143 int tail_cycle;
1144 int head_cycle;
1145 int head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Dave Chinnera69ed032010-12-21 12:08:20 +11001147 xlog_crack_grant_head(head, &head_cycle, &head_bytes);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001148 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
1149 tail_bytes = BBTOB(tail_bytes);
Dave Chinnera69ed032010-12-21 12:08:20 +11001150 if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
1151 free_bytes = log->l_logsize - (head_bytes - tail_bytes);
1152 else if (tail_cycle + 1 < head_cycle)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 return 0;
Dave Chinnera69ed032010-12-21 12:08:20 +11001154 else if (tail_cycle < head_cycle) {
1155 ASSERT(tail_cycle == (head_cycle - 1));
1156 free_bytes = tail_bytes - head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 } else {
1158 /*
1159 * The reservation head is behind the tail.
1160 * In this case we just want to return the size of the
1161 * log as the amount of space left.
1162 */
Joe Perchesf41febd2015-07-29 11:52:04 +10001163 xfs_alert(log->l_mp, "xlog_space_left: head behind tail");
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001164 xfs_alert(log->l_mp,
Joe Perchesf41febd2015-07-29 11:52:04 +10001165 " tail_cycle = %d, tail_bytes = %d",
1166 tail_cycle, tail_bytes);
1167 xfs_alert(log->l_mp,
1168 " GH cycle = %d, GH bytes = %d",
1169 head_cycle, head_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 ASSERT(0);
1171 free_bytes = log->l_logsize;
1172 }
1173 return free_bytes;
Dave Chinnera69ed032010-12-21 12:08:20 +11001174}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
1176
1177/*
1178 * Log function which is called when an io completes.
1179 *
1180 * The log manager needs its own routine, in order to control what
1181 * happens with the buffer after the write completes.
1182 */
1183void
1184xlog_iodone(xfs_buf_t *bp)
1185{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001186 struct xlog_in_core *iclog = bp->b_fspriv;
1187 struct xlog *l = iclog->ic_log;
1188 int aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 /*
1191 * Race to shutdown the filesystem if we see an error.
1192 */
Dave Chinner36de9552014-06-06 16:02:12 +10001193 if (XFS_TEST_ERROR(bp->b_error, l->l_mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
Christoph Hellwig901796a2011-10-10 16:52:49 +00001195 xfs_buf_ioerror_alert(bp, __func__);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001196 xfs_buf_stale(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001197 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 /*
1199 * This flag will be propagated to the trans-committed
1200 * callback routines to let them know that the log-commit
1201 * didn't succeed.
1202 */
1203 aborted = XFS_LI_ABORTED;
1204 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
1205 aborted = XFS_LI_ABORTED;
1206 }
David Chinner3db296f2007-05-14 18:24:16 +10001207
1208 /* log I/O is always issued ASYNC */
1209 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 xlog_state_done_syncing(iclog, aborted);
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001211
David Chinner3db296f2007-05-14 18:24:16 +10001212 /*
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001213 * drop the buffer lock now that we are done. Nothing references
1214 * the buffer after this, so an unmount waiting on this lock can now
1215 * tear it down safely. As such, it is unsafe to reference the buffer
1216 * (bp) after the unlock as we could race with it being freed.
David Chinner3db296f2007-05-14 18:24:16 +10001217 */
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001218 xfs_buf_unlock(bp);
Dave Chinnerc3f8fc72012-11-12 22:54:01 +11001219}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
1221/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 * Return size of each in-core log record buffer.
1223 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001224 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 *
1226 * If the filesystem blocksize is too large, we may need to choose a
1227 * larger size since the directory code currently logs entire blocks.
1228 */
1229
1230STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001231xlog_get_iclog_buffer_size(
1232 struct xfs_mount *mp,
1233 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
1235 int size;
1236 int xhdrs;
1237
Eric Sandeen1cb51252007-08-16 16:24:43 +10001238 if (mp->m_logbufs <= 0)
1239 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
1240 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001241 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
1243 /*
1244 * Buffer size passed in from mount system call.
1245 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001246 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 size = log->l_iclog_size = mp->m_logbsize;
1248 log->l_iclog_size_log = 0;
1249 while (size != 1) {
1250 log->l_iclog_size_log++;
1251 size >>= 1;
1252 }
1253
Eric Sandeen62118702008-03-06 13:44:28 +11001254 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001255 /* # headers = size / 32k
1256 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
1258
1259 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
1260 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1261 xhdrs++;
1262 log->l_iclog_hsize = xhdrs << BBSHIFT;
1263 log->l_iclog_heads = xhdrs;
1264 } else {
1265 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1266 log->l_iclog_hsize = BBSIZE;
1267 log->l_iclog_heads = 1;
1268 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001269 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 }
1271
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001272 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +10001273 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1274 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276 /* the default log size is 16k or 32k which is one header sector */
1277 log->l_iclog_hsize = BBSIZE;
1278 log->l_iclog_heads = 1;
1279
Christoph Hellwig7153f8b2009-02-09 08:36:46 +01001280done:
1281 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001282 if (mp->m_logbufs == 0)
1283 mp->m_logbufs = log->l_iclog_bufs;
1284 if (mp->m_logbsize == 0)
1285 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286} /* xlog_get_iclog_buffer_size */
1287
1288
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001289void
1290xfs_log_work_queue(
1291 struct xfs_mount *mp)
1292{
Dave Chinner58896082012-10-08 21:56:05 +11001293 queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work,
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001294 msecs_to_jiffies(xfs_syncd_centisecs * 10));
1295}
1296
1297/*
1298 * Every sync period we need to unpin all items in the AIL and push them to
1299 * disk. If there is nothing dirty, then we might need to cover the log to
1300 * indicate that the filesystem is idle.
1301 */
1302void
1303xfs_log_worker(
1304 struct work_struct *work)
1305{
1306 struct xlog *log = container_of(to_delayed_work(work),
1307 struct xlog, l_work);
1308 struct xfs_mount *mp = log->l_mp;
1309
1310 /* dgc: errors ignored - not fatal and nowhere to report them */
Dave Chinner61e63ec2015-01-22 09:10:31 +11001311 if (xfs_log_need_covered(mp)) {
1312 /*
1313 * Dump a transaction into the log that contains no real change.
1314 * This is needed to stamp the current tail LSN into the log
1315 * during the covering operation.
1316 *
1317 * We cannot use an inode here for this - that will push dirty
1318 * state back up into the VFS and then periodic inode flushing
1319 * will prevent log covering from making progress. Hence we
1320 * synchronously log the superblock instead to ensure the
1321 * superblock is immediately unpinned and can be written back.
1322 */
1323 xfs_sync_sb(mp, true);
1324 } else
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001325 xfs_log_force(mp, 0);
1326
1327 /* start pushing all the metadata that is currently dirty */
1328 xfs_ail_push_all(mp->m_ail);
1329
1330 /* queue us up again */
1331 xfs_log_work_queue(mp);
1332}
1333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334/*
1335 * This routine initializes some of the log structure for a given mount point.
1336 * Its primary purpose is to fill in enough, so recovery can occur. However,
1337 * some other stuff may be filled in too.
1338 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001339STATIC struct xlog *
1340xlog_alloc_log(
1341 struct xfs_mount *mp,
1342 struct xfs_buftarg *log_target,
1343 xfs_daddr_t blk_offset,
1344 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001346 struct xlog *log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 xlog_rec_header_t *head;
1348 xlog_in_core_t **iclogp;
1349 xlog_in_core_t *iclog, *prev_iclog=NULL;
1350 xfs_buf_t *bp;
1351 int i;
Dave Chinner24513372014-06-25 14:58:08 +10001352 int error = -ENOMEM;
Alex Elder69ce58f2010-04-20 17:09:59 +10001353 uint log2_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001355 log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001356 if (!log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001357 xfs_warn(mp, "Log allocation failed: No memory!");
Dave Chinnera6cb7672009-04-06 18:39:27 +02001358 goto out;
1359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
1361 log->l_mp = mp;
1362 log->l_targ = log_target;
1363 log->l_logsize = BBTOB(num_bblks);
1364 log->l_logBBstart = blk_offset;
1365 log->l_logBBsize = num_bblks;
1366 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1367 log->l_flags |= XLOG_ACTIVE_RECOVERY;
Dave Chinnerf661f1e2012-10-08 21:56:02 +11001368 INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 log->l_prev_block = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001372 xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1373 xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
Christoph Hellwigc303c5b2012-02-20 02:31:26 +00001375
1376 xlog_grant_head_init(&log->l_reserve_head);
1377 xlog_grant_head_init(&log->l_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
Dave Chinner24513372014-06-25 14:58:08 +10001379 error = -EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001380 if (xfs_sb_version_hassector(&mp->m_sb)) {
Alex Elder69ce58f2010-04-20 17:09:59 +10001381 log2_size = mp->m_sb.sb_logsectlog;
1382 if (log2_size < BBSHIFT) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001383 xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1384 log2_size, BBSHIFT);
Alex Elder69ce58f2010-04-20 17:09:59 +10001385 goto out_free_log;
1386 }
1387
1388 log2_size -= BBSHIFT;
1389 if (log2_size > mp->m_sectbb_log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001390 xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1391 log2_size, mp->m_sectbb_log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001392 goto out_free_log;
1393 }
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /* for larger sector sizes, must have v2 or external log */
Alex Elder69ce58f2010-04-20 17:09:59 +10001396 if (log2_size && log->l_logBBstart > 0 &&
1397 !xfs_sb_version_haslogv2(&mp->m_sb)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001398 xfs_warn(mp,
1399 "log sector size (0x%x) invalid for configuration.",
1400 log2_size);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001401 goto out_free_log;
1402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 }
Alex Elder69ce58f2010-04-20 17:09:59 +10001404 log->l_sectBBsize = 1 << log2_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
1406 xlog_get_iclog_buffer_size(mp, log);
1407
Dave Chinner400b9d82014-08-04 12:42:40 +10001408 /*
1409 * Use a NULL block for the extra log buffer used during splits so that
1410 * it will trigger errors if we ever try to do IO on it without first
1411 * having set it up properly.
1412 */
Dave Chinner24513372014-06-25 14:58:08 +10001413 error = -ENOMEM;
Dave Chinner400b9d82014-08-04 12:42:40 +10001414 bp = xfs_buf_alloc(mp->m_logdev_targp, XFS_BUF_DADDR_NULL,
1415 BTOBB(log->l_iclog_size), 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001416 if (!bp)
1417 goto out_free_log;
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001418
1419 /*
1420 * The iclogbuf buffer locks are held over IO but we are not going to do
1421 * IO yet. Hence unlock the buffer so that the log IO path can grab it
1422 * when appropriately.
1423 */
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001424 ASSERT(xfs_buf_islocked(bp));
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001425 xfs_buf_unlock(bp);
1426
Brian Foster96ab7952014-12-24 09:46:23 +11001427 /* use high priority wq for log I/O completion */
1428 bp->b_ioend_wq = mp->m_log_workqueue;
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001429 bp->b_iodone = xlog_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 log->l_xbuf = bp;
1431
Eric Sandeen007c61c2007-10-11 17:43:56 +10001432 spin_lock_init(&log->l_icloglock);
Dave Chinnereb40a872010-12-21 12:09:01 +11001433 init_waitqueue_head(&log->l_flush_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 iclogp = &log->l_iclog;
1436 /*
1437 * The amount of memory to allocate for the iclog structure is
1438 * rather funky due to the way the structure is defined. It is
1439 * done this way so that we can use different sizes for machines
1440 * with different amounts of memory. See the definition of
1441 * xlog_in_core_t in xfs_log_priv.h for details.
1442 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 ASSERT(log->l_iclog_size >= 4096);
1444 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001445 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1446 if (!*iclogp)
1447 goto out_free_iclog;
1448
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 iclog->ic_prev = prev_iclog;
1451 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001452
Dave Chinner686865f2010-09-24 20:07:47 +10001453 bp = xfs_buf_get_uncached(mp->m_logdev_targp,
Dave Chinnere70b73f2012-04-23 15:58:49 +10001454 BTOBB(log->l_iclog_size), 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001455 if (!bp)
1456 goto out_free_iclog;
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +02001457
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001458 ASSERT(xfs_buf_islocked(bp));
1459 xfs_buf_unlock(bp);
1460
Brian Foster96ab7952014-12-24 09:46:23 +11001461 /* use high priority wq for log I/O completion */
1462 bp->b_ioend_wq = mp->m_log_workqueue;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001463 bp->b_iodone = xlog_iodone;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001464 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001465 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001466#ifdef DEBUG
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10001467 log->l_iclog_bak[i] = &iclog->ic_header;
David Chinner4679b2d2008-04-10 12:18:54 +10001468#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 head = &iclog->ic_header;
1470 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001471 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1472 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001473 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001474 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001476 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1478
Dave Chinner4e94b712012-04-23 15:58:51 +10001479 iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 iclog->ic_state = XLOG_STATE_ACTIVE;
1481 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001482 atomic_set(&iclog->ic_refcnt, 0);
1483 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001485 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Dave Chinnereb40a872010-12-21 12:09:01 +11001487 init_waitqueue_head(&iclog->ic_force_wait);
1488 init_waitqueue_head(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
1490 iclogp = &iclog->ic_next;
1491 }
1492 *iclogp = log->l_iclog; /* complete ring */
1493 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1494
Dave Chinner71e330b2010-05-21 14:37:18 +10001495 error = xlog_cil_init(log);
1496 if (error)
1497 goto out_free_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001499
1500out_free_iclog:
1501 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1502 prev_iclog = iclog->ic_next;
Dave Chinnereb40a872010-12-21 12:09:01 +11001503 if (iclog->ic_bp)
Dave Chinner644c3562008-11-10 16:50:24 +11001504 xfs_buf_free(iclog->ic_bp);
Dave Chinner644c3562008-11-10 16:50:24 +11001505 kmem_free(iclog);
1506 }
1507 spinlock_destroy(&log->l_icloglock);
Dave Chinner644c3562008-11-10 16:50:24 +11001508 xfs_buf_free(log->l_xbuf);
1509out_free_log:
1510 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001511out:
Dave Chinner24513372014-06-25 14:58:08 +10001512 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513} /* xlog_alloc_log */
1514
1515
1516/*
1517 * Write out the commit record of a transaction associated with the given
1518 * ticket. Return the lsn of the commit record.
1519 */
1520STATIC int
Dave Chinner55b66332010-03-23 11:43:17 +11001521xlog_commit_record(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001522 struct xlog *log,
Dave Chinner55b66332010-03-23 11:43:17 +11001523 struct xlog_ticket *ticket,
1524 struct xlog_in_core **iclog,
1525 xfs_lsn_t *commitlsnp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
Dave Chinner55b66332010-03-23 11:43:17 +11001527 struct xfs_mount *mp = log->l_mp;
1528 int error;
1529 struct xfs_log_iovec reg = {
1530 .i_addr = NULL,
1531 .i_len = 0,
1532 .i_type = XLOG_REG_TYPE_COMMIT,
1533 };
1534 struct xfs_log_vec vec = {
1535 .lv_niovecs = 1,
1536 .lv_iovecp = &reg,
1537 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539 ASSERT_ALWAYS(iclog);
Dave Chinner55b66332010-03-23 11:43:17 +11001540 error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1541 XLOG_COMMIT_TRANS);
1542 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +10001543 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +01001544 return error;
Dave Chinner55b66332010-03-23 11:43:17 +11001545}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
1547/*
1548 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1549 * log space. This code pushes on the lsn which would supposedly free up
1550 * the 25% which we want to leave free. We may need to adopt a policy which
1551 * pushes on an lsn which is further along in the log once we reach the high
1552 * water mark. In this manner, we would be creating a low water mark.
1553 */
David Chinnera8272ce2007-11-23 16:28:09 +11001554STATIC void
Dave Chinner2ced19c2010-12-21 12:09:20 +11001555xlog_grant_push_ail(
Mark Tinguelyad223e62012-06-14 09:22:15 -05001556 struct xlog *log,
Dave Chinner2ced19c2010-12-21 12:09:20 +11001557 int need_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558{
Dave Chinner2ced19c2010-12-21 12:09:20 +11001559 xfs_lsn_t threshold_lsn = 0;
Dave Chinner84f3c682010-12-03 22:11:29 +11001560 xfs_lsn_t last_sync_lsn;
Dave Chinner2ced19c2010-12-21 12:09:20 +11001561 int free_blocks;
1562 int free_bytes;
1563 int threshold_block;
1564 int threshold_cycle;
1565 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Dave Chinner2ced19c2010-12-21 12:09:20 +11001567 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Christoph Hellwig28496962012-02-20 02:31:25 +00001569 free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
Dave Chinner2ced19c2010-12-21 12:09:20 +11001570 free_blocks = BTOBBT(free_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
Dave Chinner2ced19c2010-12-21 12:09:20 +11001572 /*
1573 * Set the threshold for the minimum number of free blocks in the
1574 * log to the maximum of what the caller needs, one quarter of the
1575 * log, and 256 blocks.
1576 */
1577 free_threshold = BTOBB(need_bytes);
1578 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1579 free_threshold = MAX(free_threshold, 256);
1580 if (free_blocks >= free_threshold)
1581 return;
1582
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001583 xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1584 &threshold_block);
1585 threshold_block += free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (threshold_block >= log->l_logBBsize) {
Dave Chinner2ced19c2010-12-21 12:09:20 +11001587 threshold_block -= log->l_logBBsize;
1588 threshold_cycle += 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 }
Dave Chinner2ced19c2010-12-21 12:09:20 +11001590 threshold_lsn = xlog_assign_lsn(threshold_cycle,
1591 threshold_block);
1592 /*
1593 * Don't pass in an lsn greater than the lsn of the last
Dave Chinner84f3c682010-12-03 22:11:29 +11001594 * log record known to be on disk. Use a snapshot of the last sync lsn
1595 * so that it doesn't change between the compare and the set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 */
Dave Chinner84f3c682010-12-03 22:11:29 +11001597 last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1598 if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1599 threshold_lsn = last_sync_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Dave Chinner2ced19c2010-12-21 12:09:20 +11001601 /*
1602 * Get the transaction layer to kick the dirty buffers out to
1603 * disk asynchronously. No point in trying to do this if
1604 * the filesystem is shutting down.
1605 */
1606 if (!XLOG_FORCED_SHUTDOWN(log))
Dave Chinnerfd074842011-04-08 12:45:07 +10001607 xfs_ail_push(log->l_ailp, threshold_lsn);
Dave Chinner2ced19c2010-12-21 12:09:20 +11001608}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Christoph Hellwig873ff552010-01-13 22:17:57 +00001610/*
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001611 * Stamp cycle number in every block
1612 */
1613STATIC void
1614xlog_pack_data(
1615 struct xlog *log,
1616 struct xlog_in_core *iclog,
1617 int roundoff)
1618{
1619 int i, j, k;
1620 int size = iclog->ic_offset + roundoff;
1621 __be32 cycle_lsn;
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10001622 char *dp;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001623
1624 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
1625
1626 dp = iclog->ic_datap;
1627 for (i = 0; i < BTOBB(size); i++) {
1628 if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
1629 break;
1630 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
1631 *(__be32 *)dp = cycle_lsn;
1632 dp += BBSIZE;
1633 }
1634
1635 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1636 xlog_in_core_2_t *xhdr = iclog->ic_data;
1637
1638 for ( ; i < BTOBB(size); i++) {
1639 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1640 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
1641 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
1642 *(__be32 *)dp = cycle_lsn;
1643 dp += BBSIZE;
1644 }
1645
1646 for (i = 1; i < log->l_iclog_heads; i++)
1647 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
1648 }
1649}
1650
1651/*
1652 * Calculate the checksum for a log buffer.
1653 *
1654 * This is a little more complicated than it should be because the various
1655 * headers and the actual data are non-contiguous.
1656 */
Dave Chinnerf9668a02012-11-28 13:01:03 +11001657__le32
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001658xlog_cksum(
1659 struct xlog *log,
1660 struct xlog_rec_header *rhead,
1661 char *dp,
1662 int size)
1663{
1664 __uint32_t crc;
1665
1666 /* first generate the crc for the record header ... */
1667 crc = xfs_start_cksum((char *)rhead,
1668 sizeof(struct xlog_rec_header),
1669 offsetof(struct xlog_rec_header, h_crc));
1670
1671 /* ... then for additional cycle data for v2 logs ... */
1672 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1673 union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
1674 int i;
Brian Fostera3f20012015-08-19 09:59:50 +10001675 int xheads;
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001676
Brian Fostera3f20012015-08-19 09:59:50 +10001677 xheads = size / XLOG_HEADER_CYCLE_SIZE;
1678 if (size % XLOG_HEADER_CYCLE_SIZE)
1679 xheads++;
1680
1681 for (i = 1; i < xheads; i++) {
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001682 crc = crc32c(crc, &xhdr[i].hic_xheader,
1683 sizeof(struct xlog_rec_ext_header));
1684 }
1685 }
1686
1687 /* ... and finally for the payload */
1688 crc = crc32c(crc, dp, size);
1689
1690 return xfs_end_cksum(crc);
1691}
1692
1693/*
Christoph Hellwig873ff552010-01-13 22:17:57 +00001694 * The bdstrat callback function for log bufs. This gives us a central
1695 * place to trap bufs in case we get hit by a log I/O error and need to
1696 * shutdown. Actually, in practice, even when we didn't get a log error,
1697 * we transition the iclogs to IOERROR state *after* flushing all existing
1698 * iclogs to disk. This is because we don't want anymore new transactions to be
1699 * started or completed afterwards.
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001700 *
1701 * We lock the iclogbufs here so that we can serialise against IO completion
1702 * during unmount. We might be processing a shutdown triggered during unmount,
1703 * and that can occur asynchronously to the unmount thread, and hence we need to
1704 * ensure that completes before tearing down the iclogbufs. Hence we need to
1705 * hold the buffer lock across the log IO to acheive that.
Christoph Hellwig873ff552010-01-13 22:17:57 +00001706 */
1707STATIC int
1708xlog_bdstrat(
1709 struct xfs_buf *bp)
1710{
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001711 struct xlog_in_core *iclog = bp->b_fspriv;
Christoph Hellwig873ff552010-01-13 22:17:57 +00001712
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001713 xfs_buf_lock(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001714 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Dave Chinner24513372014-06-25 14:58:08 +10001715 xfs_buf_ioerror(bp, -EIO);
Christoph Hellwigc867cb62011-10-10 16:52:46 +00001716 xfs_buf_stale(bp);
Dave Chinnere8aaba92014-10-02 09:04:22 +10001717 xfs_buf_ioend(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001718 /*
1719 * It would seem logical to return EIO here, but we rely on
1720 * the log state machine to propagate I/O errors instead of
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001721 * doing it here. Similarly, IO completion will unlock the
1722 * buffer, so we don't do it here.
Christoph Hellwig873ff552010-01-13 22:17:57 +00001723 */
1724 return 0;
1725 }
1726
Dave Chinner595bff72014-10-02 09:05:14 +10001727 xfs_buf_submit(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001728 return 0;
1729}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
1731/*
1732 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1733 * fashion. Previously, we should have moved the current iclog
1734 * ptr in the log to point to the next available iclog. This allows further
1735 * write to continue while this code syncs out an iclog ready to go.
1736 * Before an in-core log can be written out, the data section must be scanned
1737 * to save away the 1st word of each BBSIZE block into the header. We replace
1738 * it with the current cycle count. Each BBSIZE block is tagged with the
1739 * cycle count because there in an implicit assumption that drives will
1740 * guarantee that entire 512 byte blocks get written at once. In other words,
1741 * we can't have part of a 512 byte block written and part not written. By
1742 * tagging each block, we will know which blocks are valid when recovering
1743 * after an unclean shutdown.
1744 *
1745 * This routine is single threaded on the iclog. No other thread can be in
1746 * this routine with the same iclog. Changing contents of iclog can there-
1747 * fore be done without grabbing the state machine lock. Updating the global
1748 * log will require grabbing the lock though.
1749 *
1750 * The entire log manager uses a logical block numbering scheme. Only
1751 * log_sync (and then only bwrite()) know about the fact that the log may
1752 * not start with block zero on a given device. The log block start offset
1753 * is added immediately before calling bwrite().
1754 */
1755
David Chinnera8272ce2007-11-23 16:28:09 +11001756STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001757xlog_sync(
1758 struct xlog *log,
1759 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001762 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 uint count; /* byte count of bwrite */
1764 uint count_init; /* initial count before roundup */
1765 int roundoff; /* roundoff to BB or stripe */
1766 int split = 0; /* split write into two regions */
1767 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001768 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001769 int size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001772 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
1774 /* Add for LR header */
1775 count_init = log->l_iclog_hsize + iclog->ic_offset;
1776
1777 /* Round out the log write size */
1778 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1779 /* we have a v2 stripe unit to use */
1780 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1781 } else {
1782 count = BBTOB(BTOBB(count_init));
1783 }
1784 roundoff = count - count_init;
1785 ASSERT(roundoff >= 0);
1786 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1787 roundoff < log->l_mp->m_sb.sb_logsunit)
1788 ||
1789 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1790 roundoff < BBTOB(1)));
1791
1792 /* move grant heads by roundoff in sync */
Christoph Hellwig28496962012-02-20 02:31:25 +00001793 xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
1794 xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 /* put cycle number in every block */
1797 xlog_pack_data(log, iclog, roundoff);
1798
1799 /* real byte length */
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001800 size = iclog->ic_offset;
1801 if (v2)
1802 size += roundoff;
1803 iclog->ic_header.h_len = cpu_to_be32(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Nathan Scottf5faad72006-07-28 17:04:44 +10001805 bp = iclog->ic_bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001806 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
1808 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1809
1810 /* Do we need to split this write into 2 parts? */
1811 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001812 char *dptr;
1813
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1815 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001816 iclog->ic_bwritecnt = 2;
1817
1818 /*
1819 * Bump the cycle numbers at the start of each block in the
1820 * part of the iclog that ends up in the buffer that gets
1821 * written to the start of the log.
1822 *
1823 * Watch out for the header magic number case, though.
1824 */
1825 dptr = (char *)&iclog->ic_header + count;
1826 for (i = 0; i < split; i += BBSIZE) {
1827 __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr);
1828 if (++cycle == XLOG_HEADER_MAGIC_NUM)
1829 cycle++;
1830 *(__be32 *)dptr = cpu_to_be32(cycle);
1831
1832 dptr += BBSIZE;
1833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 } else {
1835 iclog->ic_bwritecnt = 1;
1836 }
Christoph Hellwig0e446be2012-11-12 22:54:24 +11001837
1838 /* calculcate the checksum */
1839 iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
1840 iclog->ic_datap, size);
1841
Dave Chinneraa0e8832012-04-23 15:58:52 +10001842 bp->b_io_length = BTOBB(count);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001843 bp->b_fspriv = iclog;
Nathan Scottf5faad72006-07-28 17:04:44 +10001844 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001846 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwig651701d2010-06-28 10:34:34 -04001847
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001848 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001849 bp->b_flags |= XBF_FUA;
1850
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001851 /*
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001852 * Flush the data device before flushing the log to make
1853 * sure all meta data written back from the AIL actually made
1854 * it to disk before stamping the new log tail LSN into the
1855 * log buffer. For an external log we need to issue the
1856 * flush explicitly, and unfortunately synchronously here;
1857 * for an internal log we can simply use the block layer
1858 * state machine for preflushes.
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001859 */
1860 if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
1861 xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001862 else
1863 bp->b_flags |= XBF_FLUSH;
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865
1866 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1867 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1868
Ben Myers003fd6c2013-01-18 14:17:46 -06001869 xlog_verify_iclog(log, iclog, count, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871 /* account for log which doesn't start at block #0 */
1872 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1873 /*
1874 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1875 * is shutting down.
1876 */
1877 XFS_BUF_WRITE(bp);
1878
Christoph Hellwig901796a2011-10-10 16:52:49 +00001879 error = xlog_bdstrat(bp);
1880 if (error) {
1881 xfs_buf_ioerror_alert(bp, "xlog_sync");
Jesper Juhl014c2542006-01-15 02:37:08 +01001882 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 }
1884 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001885 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
Chandra Seetharaman02fe03d2011-07-22 23:40:22 +00001887 xfs_buf_associate_memory(bp,
1888 (char *)&iclog->ic_header + count, split);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001889 bp->b_fspriv = iclog;
Nathan Scottf5faad72006-07-28 17:04:44 +10001890 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001892 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001893 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001894 bp->b_flags |= XBF_FUA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
1896 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1897 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1898
Nathan Scottc41564b2006-03-29 08:55:14 +10001899 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1901 XFS_BUF_WRITE(bp);
Christoph Hellwig901796a2011-10-10 16:52:49 +00001902 error = xlog_bdstrat(bp);
1903 if (error) {
1904 xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
Jesper Juhl014c2542006-01-15 02:37:08 +01001905 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 }
1907 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001908 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909} /* xlog_sync */
1910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001912 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 */
David Chinnera8272ce2007-11-23 16:28:09 +11001914STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001915xlog_dealloc_log(
1916 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917{
1918 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 int i;
1920
Dave Chinner71e330b2010-05-21 14:37:18 +10001921 xlog_cil_destroy(log);
1922
Dave Chinner44396472011-04-21 09:34:27 +00001923 /*
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001924 * Cycle all the iclogbuf locks to make sure all log IO completion
1925 * is done before we tear down these buffers.
Dave Chinner44396472011-04-21 09:34:27 +00001926 */
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001927 iclog = log->l_iclog;
1928 for (i = 0; i < log->l_iclog_bufs; i++) {
1929 xfs_buf_lock(iclog->ic_bp);
1930 xfs_buf_unlock(iclog->ic_bp);
1931 iclog = iclog->ic_next;
1932 }
1933
1934 /*
1935 * Always need to ensure that the extra buffer does not point to memory
1936 * owned by another log buffer before we free it. Also, cycle the lock
1937 * first to ensure we've completed IO on it.
1938 */
1939 xfs_buf_lock(log->l_xbuf);
1940 xfs_buf_unlock(log->l_xbuf);
Dave Chinnere70b73f2012-04-23 15:58:49 +10001941 xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
Dave Chinner44396472011-04-21 09:34:27 +00001942 xfs_buf_free(log->l_xbuf);
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 iclog = log->l_iclog;
Dave Chinner9c23ecc2014-04-17 08:15:26 +10001945 for (i = 0; i < log->l_iclog_bufs; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 xfs_buf_free(iclog->ic_bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001948 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 iclog = next_iclog;
1950 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 spinlock_destroy(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001954 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001955} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957/*
1958 * Update counters atomically now that memcpy is done.
1959 */
1960/* ARGSUSED */
1961static inline void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05001962xlog_state_finish_copy(
1963 struct xlog *log,
1964 struct xlog_in_core *iclog,
1965 int record_cnt,
1966 int copy_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001968 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001970 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 iclog->ic_offset += copy_bytes;
1972
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001973 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974} /* xlog_state_finish_copy */
1975
1976
1977
1978
1979/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001980 * print out info relating to regions written which consume
1981 * the reservation
1982 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001983void
1984xlog_print_tic_res(
1985 struct xfs_mount *mp,
1986 struct xlog_ticket *ticket)
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001987{
1988 uint i;
1989 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1990
1991 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1992 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1993 "bformat",
1994 "bchunk",
1995 "efi_format",
1996 "efd_format",
1997 "iformat",
1998 "icore",
1999 "iext",
2000 "ibroot",
2001 "ilocal",
2002 "iattr_ext",
2003 "iattr_broot",
2004 "iattr_local",
2005 "qformat",
2006 "dquot",
2007 "quotaoff",
2008 "LR header",
2009 "unmount",
2010 "commit",
2011 "trans header"
2012 };
2013 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
2014 "SETATTR_NOT_SIZE",
2015 "SETATTR_SIZE",
2016 "INACTIVE",
2017 "CREATE",
2018 "CREATE_TRUNC",
2019 "TRUNCATE_FILE",
2020 "REMOVE",
2021 "LINK",
2022 "RENAME",
2023 "MKDIR",
2024 "RMDIR",
2025 "SYMLINK",
2026 "SET_DMATTRS",
2027 "GROWFS",
2028 "STRAT_WRITE",
2029 "DIOSTRAT",
2030 "WRITE_SYNC",
2031 "WRITEID",
2032 "ADDAFORK",
2033 "ATTRINVAL",
2034 "ATRUNCATE",
2035 "ATTR_SET",
2036 "ATTR_RM",
2037 "ATTR_FLAG",
2038 "CLEAR_AGI_BUCKET",
2039 "QM_SBCHANGE",
2040 "DUMMY1",
2041 "DUMMY2",
2042 "QM_QUOTAOFF",
2043 "QM_DQALLOC",
2044 "QM_SETQLIM",
2045 "QM_DQCLUSTER",
2046 "QM_QINOCREATE",
2047 "QM_QUOTAOFF_END",
2048 "SB_UNIT",
2049 "FSYNC_TS",
2050 "GROWFSRT_ALLOC",
2051 "GROWFSRT_ZERO",
2052 "GROWFSRT_FREE",
2053 "SWAPEXT"
2054 };
2055
Joe Perchesf41febd2015-07-29 11:52:04 +10002056 xfs_warn(mp, "xlog_write: reservation summary:");
2057 xfs_warn(mp, " trans type = %s (%u)",
2058 ((ticket->t_trans_type <= 0 ||
2059 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002060 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
Joe Perchesf41febd2015-07-29 11:52:04 +10002061 ticket->t_trans_type);
2062 xfs_warn(mp, " unit res = %d bytes",
2063 ticket->t_unit_res);
2064 xfs_warn(mp, " current res = %d bytes",
2065 ticket->t_curr_res);
2066 xfs_warn(mp, " total reg = %u bytes (o/flow = %u bytes)",
2067 ticket->t_res_arr_sum, ticket->t_res_o_flow);
2068 xfs_warn(mp, " ophdrs = %u (ophdr space = %u bytes)",
2069 ticket->t_res_num_ophdrs, ophdr_spc);
2070 xfs_warn(mp, " ophdr + reg = %u bytes",
2071 ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc);
2072 xfs_warn(mp, " num regions = %u",
2073 ticket->t_res_num);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002074
2075 for (i = 0; i < ticket->t_res_num; i++) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002076 uint r_type = ticket->t_res_arr[i].r_type;
Eric Sandeen08e96e12013-10-11 20:59:05 -05002077 xfs_warn(mp, "region[%u]: %s - %u bytes", i,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002078 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
2079 "bad-rtype" : res_type_str[r_type-1]),
2080 ticket->t_res_arr[i].r_len);
2081 }
Dave Chinner169a7b02010-05-07 11:05:31 +10002082
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002083 xfs_alert_tag(mp, XFS_PTAG_LOGRES,
Christoph Hellwig93b8a582011-12-06 21:58:07 +00002084 "xlog_write: reservation ran out. Need to up reservation");
Chandra Seetharaman297aa632013-07-19 17:31:38 -05002085 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002086}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002087
2088/*
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002089 * Calculate the potential space needed by the log vector. Each region gets
2090 * its own xlog_op_header_t and may need to be double word aligned.
2091 */
2092static int
2093xlog_write_calc_vec_length(
2094 struct xlog_ticket *ticket,
Dave Chinner55b66332010-03-23 11:43:17 +11002095 struct xfs_log_vec *log_vector)
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002096{
Dave Chinner55b66332010-03-23 11:43:17 +11002097 struct xfs_log_vec *lv;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002098 int headers = 0;
2099 int len = 0;
2100 int i;
2101
2102 /* acct for start rec of xact */
2103 if (ticket->t_flags & XLOG_TIC_INITED)
2104 headers++;
2105
Dave Chinner55b66332010-03-23 11:43:17 +11002106 for (lv = log_vector; lv; lv = lv->lv_next) {
Dave Chinnerfd638752013-06-27 16:04:51 +10002107 /* we don't write ordered log vectors */
2108 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED)
2109 continue;
2110
Dave Chinner55b66332010-03-23 11:43:17 +11002111 headers += lv->lv_niovecs;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002112
Dave Chinner55b66332010-03-23 11:43:17 +11002113 for (i = 0; i < lv->lv_niovecs; i++) {
2114 struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
2115
2116 len += vecp->i_len;
2117 xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
2118 }
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002119 }
2120
2121 ticket->t_res_num_ophdrs += headers;
2122 len += headers * sizeof(struct xlog_op_header);
2123
2124 return len;
2125}
2126
2127/*
2128 * If first write for transaction, insert start record We can't be trying to
2129 * commit if we are inited. We can't have any "partial_copy" if we are inited.
2130 */
2131static int
2132xlog_write_start_rec(
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002133 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002134 struct xlog_ticket *ticket)
2135{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002136 if (!(ticket->t_flags & XLOG_TIC_INITED))
2137 return 0;
2138
2139 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2140 ophdr->oh_clientid = ticket->t_clientid;
2141 ophdr->oh_len = 0;
2142 ophdr->oh_flags = XLOG_START_TRANS;
2143 ophdr->oh_res2 = 0;
2144
2145 ticket->t_flags &= ~XLOG_TIC_INITED;
2146
2147 return sizeof(struct xlog_op_header);
2148}
2149
2150static xlog_op_header_t *
2151xlog_write_setup_ophdr(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002152 struct xlog *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002153 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002154 struct xlog_ticket *ticket,
2155 uint flags)
2156{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002157 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
2158 ophdr->oh_clientid = ticket->t_clientid;
2159 ophdr->oh_res2 = 0;
2160
2161 /* are we copying a commit or unmount record? */
2162 ophdr->oh_flags = flags;
2163
2164 /*
2165 * We've seen logs corrupted with bad transaction client ids. This
2166 * makes sure that XFS doesn't generate them on. Turn this into an EIO
2167 * and shut down the filesystem.
2168 */
2169 switch (ophdr->oh_clientid) {
2170 case XFS_TRANSACTION:
2171 case XFS_VOLUME:
2172 case XFS_LOG:
2173 break;
2174 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002175 xfs_warn(log->l_mp,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002176 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
2177 ophdr->oh_clientid, ticket);
2178 return NULL;
2179 }
2180
2181 return ophdr;
2182}
2183
2184/*
2185 * Set up the parameters of the region copy into the log. This has
2186 * to handle region write split across multiple log buffers - this
2187 * state is kept external to this function so that this code can
Zhi Yong Wuac0e3002013-08-07 10:11:02 +00002188 * be written in an obvious, self documenting manner.
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002189 */
2190static int
2191xlog_write_setup_copy(
2192 struct xlog_ticket *ticket,
2193 struct xlog_op_header *ophdr,
2194 int space_available,
2195 int space_required,
2196 int *copy_off,
2197 int *copy_len,
2198 int *last_was_partial_copy,
2199 int *bytes_consumed)
2200{
2201 int still_to_copy;
2202
2203 still_to_copy = space_required - *bytes_consumed;
2204 *copy_off = *bytes_consumed;
2205
2206 if (still_to_copy <= space_available) {
2207 /* write of region completes here */
2208 *copy_len = still_to_copy;
2209 ophdr->oh_len = cpu_to_be32(*copy_len);
2210 if (*last_was_partial_copy)
2211 ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
2212 *last_was_partial_copy = 0;
2213 *bytes_consumed = 0;
2214 return 0;
2215 }
2216
2217 /* partial write of region, needs extra log op header reservation */
2218 *copy_len = space_available;
2219 ophdr->oh_len = cpu_to_be32(*copy_len);
2220 ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
2221 if (*last_was_partial_copy)
2222 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
2223 *bytes_consumed += *copy_len;
2224 (*last_was_partial_copy)++;
2225
2226 /* account for new log op header */
2227 ticket->t_curr_res -= sizeof(struct xlog_op_header);
2228 ticket->t_res_num_ophdrs++;
2229
2230 return sizeof(struct xlog_op_header);
2231}
2232
2233static int
2234xlog_write_copy_finish(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002235 struct xlog *log,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002236 struct xlog_in_core *iclog,
2237 uint flags,
2238 int *record_cnt,
2239 int *data_cnt,
2240 int *partial_copy,
2241 int *partial_copy_len,
2242 int log_offset,
2243 struct xlog_in_core **commit_iclog)
2244{
2245 if (*partial_copy) {
2246 /*
2247 * This iclog has already been marked WANT_SYNC by
2248 * xlog_state_get_iclog_space.
2249 */
2250 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2251 *record_cnt = 0;
2252 *data_cnt = 0;
2253 return xlog_state_release_iclog(log, iclog);
2254 }
2255
2256 *partial_copy = 0;
2257 *partial_copy_len = 0;
2258
2259 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
2260 /* no more space in this iclog - push it. */
2261 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
2262 *record_cnt = 0;
2263 *data_cnt = 0;
2264
2265 spin_lock(&log->l_icloglock);
2266 xlog_state_want_sync(log, iclog);
2267 spin_unlock(&log->l_icloglock);
2268
2269 if (!commit_iclog)
2270 return xlog_state_release_iclog(log, iclog);
2271 ASSERT(flags & XLOG_COMMIT_TRANS);
2272 *commit_iclog = iclog;
2273 }
2274
2275 return 0;
2276}
2277
2278/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 * Write some region out to in-core log
2280 *
2281 * This will be called when writing externally provided regions or when
2282 * writing out a commit record for a given transaction.
2283 *
2284 * General algorithm:
2285 * 1. Find total length of this write. This may include adding to the
2286 * lengths passed in.
2287 * 2. Check whether we violate the tickets reservation.
2288 * 3. While writing to this iclog
2289 * A. Reserve as much space in this iclog as can get
2290 * B. If this is first write, save away start lsn
2291 * C. While writing this region:
2292 * 1. If first write of transaction, write start record
2293 * 2. Write log operation header (header per region)
2294 * 3. Find out if we can fit entire region into this iclog
2295 * 4. Potentially, verify destination memcpy ptr
2296 * 5. Memcpy (partial) region
2297 * 6. If partial copy, release iclog; otherwise, continue
2298 * copying more regions into current iclog
2299 * 4. Mark want sync bit (in simulation mode)
2300 * 5. Release iclog for potential flush to on-disk log.
2301 *
2302 * ERRORS:
2303 * 1. Panic if reservation is overrun. This should never happen since
2304 * reservation amounts are generated internal to the filesystem.
2305 * NOTES:
2306 * 1. Tickets are single threaded data structures.
2307 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
2308 * syncing routine. When a single log_write region needs to span
2309 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
2310 * on all log operation writes which don't contain the end of the
2311 * region. The XLOG_END_TRANS bit is used for the in-core log
2312 * operation which contains the end of the continued log_write region.
2313 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
2314 * we don't really know exactly how much space will be used. As a result,
2315 * we don't update ic_offset until the end when we know exactly how many
2316 * bytes have been written out.
2317 */
Dave Chinner71e330b2010-05-21 14:37:18 +10002318int
Christoph Hellwig35a8a722010-02-15 23:34:54 +00002319xlog_write(
Mark Tinguelyad223e62012-06-14 09:22:15 -05002320 struct xlog *log,
Dave Chinner55b66332010-03-23 11:43:17 +11002321 struct xfs_log_vec *log_vector,
Christoph Hellwig35a8a722010-02-15 23:34:54 +00002322 struct xlog_ticket *ticket,
2323 xfs_lsn_t *start_lsn,
2324 struct xlog_in_core **commit_iclog,
2325 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326{
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002327 struct xlog_in_core *iclog = NULL;
Dave Chinner55b66332010-03-23 11:43:17 +11002328 struct xfs_log_iovec *vecp;
2329 struct xfs_log_vec *lv;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002330 int len;
2331 int index;
2332 int partial_copy = 0;
2333 int partial_copy_len = 0;
2334 int contwr = 0;
2335 int record_cnt = 0;
2336 int data_cnt = 0;
2337 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002339 *start_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340
Dave Chinner55b66332010-03-23 11:43:17 +11002341 len = xlog_write_calc_vec_length(ticket, log_vector);
Dave Chinner71e330b2010-05-21 14:37:18 +10002342
Christoph Hellwig93b8a582011-12-06 21:58:07 +00002343 /*
2344 * Region headers and bytes are already accounted for.
2345 * We only need to take into account start records and
2346 * split regions in this function.
2347 */
2348 if (ticket->t_flags & XLOG_TIC_INITED)
2349 ticket->t_curr_res -= sizeof(xlog_op_header_t);
2350
2351 /*
2352 * Commit record headers need to be accounted for. These
2353 * come in as separate writes so are easy to detect.
2354 */
2355 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
2356 ticket->t_curr_res -= sizeof(xlog_op_header_t);
Dave Chinner71e330b2010-05-21 14:37:18 +10002357
2358 if (ticket->t_curr_res < 0)
Dave Chinner55b66332010-03-23 11:43:17 +11002359 xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Dave Chinner55b66332010-03-23 11:43:17 +11002361 index = 0;
2362 lv = log_vector;
2363 vecp = lv->lv_iovecp;
Dave Chinnerfd638752013-06-27 16:04:51 +10002364 while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002365 void *ptr;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002366 int log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002368 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
2369 &contwr, &log_offset);
2370 if (error)
2371 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002373 ASSERT(log_offset <= iclog->ic_size - 1);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002374 ptr = iclog->ic_datap + log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002376 /* start_lsn is the first lsn written to. That's all we need. */
2377 if (!*start_lsn)
2378 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002380 /*
2381 * This loop writes out as many regions as can fit in the amount
2382 * of space which was allocated by xlog_state_get_iclog_space().
2383 */
Dave Chinnerfd638752013-06-27 16:04:51 +10002384 while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
2385 struct xfs_log_iovec *reg;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002386 struct xlog_op_header *ophdr;
2387 int start_rec_copy;
2388 int copy_len;
2389 int copy_off;
Dave Chinnerfd638752013-06-27 16:04:51 +10002390 bool ordered = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
Dave Chinnerfd638752013-06-27 16:04:51 +10002392 /* ordered log vectors have no regions to write */
2393 if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
2394 ASSERT(lv->lv_niovecs == 0);
2395 ordered = true;
2396 goto next_lv;
2397 }
2398
2399 reg = &vecp[index];
Dave Chinner55b66332010-03-23 11:43:17 +11002400 ASSERT(reg->i_len % sizeof(__int32_t) == 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002401 ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002403 start_rec_copy = xlog_write_start_rec(ptr, ticket);
2404 if (start_rec_copy) {
2405 record_cnt++;
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002406 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002407 start_rec_copy);
2408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002410 ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
2411 if (!ophdr)
Dave Chinner24513372014-06-25 14:58:08 +10002412 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002414 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002415 sizeof(struct xlog_op_header));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002417 len += xlog_write_setup_copy(ticket, ophdr,
2418 iclog->ic_size-log_offset,
Dave Chinner55b66332010-03-23 11:43:17 +11002419 reg->i_len,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002420 &copy_off, &copy_len,
2421 &partial_copy,
2422 &partial_copy_len);
2423 xlog_verify_dest_ptr(log, ptr);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002424
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002425 /* copy region */
2426 ASSERT(copy_len >= 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11002427 memcpy(ptr, reg->i_addr + copy_off, copy_len);
2428 xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11002429
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002430 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
2431 record_cnt++;
2432 data_cnt += contwr ? copy_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002434 error = xlog_write_copy_finish(log, iclog, flags,
2435 &record_cnt, &data_cnt,
2436 &partial_copy,
2437 &partial_copy_len,
2438 log_offset,
2439 commit_iclog);
2440 if (error)
2441 return error;
2442
2443 /*
2444 * if we had a partial copy, we need to get more iclog
2445 * space but we don't want to increment the region
2446 * index because there is still more is this region to
2447 * write.
2448 *
2449 * If we completed writing this region, and we flushed
2450 * the iclog (indicated by resetting of the record
2451 * count), then we also need to get more log space. If
2452 * this was the last record, though, we are done and
2453 * can just return.
2454 */
2455 if (partial_copy)
2456 break;
2457
Dave Chinner55b66332010-03-23 11:43:17 +11002458 if (++index == lv->lv_niovecs) {
Dave Chinnerfd638752013-06-27 16:04:51 +10002459next_lv:
Dave Chinner55b66332010-03-23 11:43:17 +11002460 lv = lv->lv_next;
2461 index = 0;
2462 if (lv)
2463 vecp = lv->lv_iovecp;
2464 }
Dave Chinnerfd638752013-06-27 16:04:51 +10002465 if (record_cnt == 0 && ordered == false) {
Dave Chinner55b66332010-03-23 11:43:17 +11002466 if (!lv)
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002467 return 0;
2468 break;
2469 }
2470 }
2471 }
2472
2473 ASSERT(len == 0);
2474
2475 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
2476 if (!commit_iclog)
2477 return xlog_state_release_iclog(log, iclog);
2478
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 ASSERT(flags & XLOG_COMMIT_TRANS);
2480 *commit_iclog = iclog;
2481 return 0;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002482}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
2484
2485/*****************************************************************************
2486 *
2487 * State Machine functions
2488 *
2489 *****************************************************************************
2490 */
2491
2492/* Clean iclogs starting from the head. This ordering must be
2493 * maintained, so an iclog doesn't become ACTIVE beyond one that
2494 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10002495 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 * iclog is trying to sync to disk.
2497 *
2498 * State Change: DIRTY -> ACTIVE
2499 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002500STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002501xlog_state_clean_log(
2502 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503{
2504 xlog_in_core_t *iclog;
2505 int changed = 0;
2506
2507 iclog = log->l_iclog;
2508 do {
2509 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2510 iclog->ic_state = XLOG_STATE_ACTIVE;
2511 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002512 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 /*
2514 * If the number of ops in this iclog indicate it just
2515 * contains the dummy transaction, we can
2516 * change state into IDLE (the second time around).
2517 * Otherwise we should change the state into
2518 * NEED a dummy.
2519 * We don't need to cover the dummy.
2520 */
2521 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002522 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2523 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 changed = 1;
2525 } else {
2526 /*
2527 * We have two dirty iclogs so start over
2528 * This could also be num of ops indicates
2529 * this is not the dummy going out.
2530 */
2531 changed = 2;
2532 }
2533 iclog->ic_header.h_num_logops = 0;
2534 memset(iclog->ic_header.h_cycle_data, 0,
2535 sizeof(iclog->ic_header.h_cycle_data));
2536 iclog->ic_header.h_lsn = 0;
2537 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2538 /* do nothing */;
2539 else
2540 break; /* stop cleaning */
2541 iclog = iclog->ic_next;
2542 } while (iclog != log->l_iclog);
2543
2544 /* log is locked when we are called */
2545 /*
2546 * Change state for the dummy log recording.
2547 * We usually go to NEED. But we go to NEED2 if the changed indicates
2548 * we are done writing the dummy record.
2549 * If we are done with the second dummy recored (DONE2), then
2550 * we go to IDLE.
2551 */
2552 if (changed) {
2553 switch (log->l_covered_state) {
2554 case XLOG_STATE_COVER_IDLE:
2555 case XLOG_STATE_COVER_NEED:
2556 case XLOG_STATE_COVER_NEED2:
2557 log->l_covered_state = XLOG_STATE_COVER_NEED;
2558 break;
2559
2560 case XLOG_STATE_COVER_DONE:
2561 if (changed == 1)
2562 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2563 else
2564 log->l_covered_state = XLOG_STATE_COVER_NEED;
2565 break;
2566
2567 case XLOG_STATE_COVER_DONE2:
2568 if (changed == 1)
2569 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2570 else
2571 log->l_covered_state = XLOG_STATE_COVER_NEED;
2572 break;
2573
2574 default:
2575 ASSERT(0);
2576 }
2577 }
2578} /* xlog_state_clean_log */
2579
2580STATIC xfs_lsn_t
2581xlog_get_lowest_lsn(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002582 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583{
2584 xlog_in_core_t *lsn_log;
2585 xfs_lsn_t lowest_lsn, lsn;
2586
2587 lsn_log = log->l_iclog;
2588 lowest_lsn = 0;
2589 do {
2590 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002591 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 if ((lsn && !lowest_lsn) ||
2593 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2594 lowest_lsn = lsn;
2595 }
2596 }
2597 lsn_log = lsn_log->ic_next;
2598 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002599 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600}
2601
2602
2603STATIC void
2604xlog_state_do_callback(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002605 struct xlog *log,
2606 int aborted,
2607 struct xlog_in_core *ciclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608{
2609 xlog_in_core_t *iclog;
2610 xlog_in_core_t *first_iclog; /* used to know when we've
2611 * processed all iclogs once */
2612 xfs_log_callback_t *cb, *cb_next;
2613 int flushcnt = 0;
2614 xfs_lsn_t lowest_lsn;
2615 int ioerrors; /* counter: iclogs with errors */
2616 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2617 int funcdidcallbacks; /* flag: function did callbacks */
2618 int repeats; /* for issuing console warnings if
2619 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002620 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002622 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 first_iclog = iclog = log->l_iclog;
2624 ioerrors = 0;
2625 funcdidcallbacks = 0;
2626 repeats = 0;
2627
2628 do {
2629 /*
2630 * Scan all iclogs starting with the one pointed to by the
2631 * log. Reset this starting point each time the log is
2632 * unlocked (during callbacks).
2633 *
2634 * Keep looping through iclogs until one full pass is made
2635 * without running any callbacks.
2636 */
2637 first_iclog = log->l_iclog;
2638 iclog = log->l_iclog;
2639 loopdidcallbacks = 0;
2640 repeats++;
2641
2642 do {
2643
2644 /* skip all iclogs in the ACTIVE & DIRTY states */
2645 if (iclog->ic_state &
2646 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2647 iclog = iclog->ic_next;
2648 continue;
2649 }
2650
2651 /*
2652 * Between marking a filesystem SHUTDOWN and stopping
2653 * the log, we do flush all iclogs to disk (if there
2654 * wasn't a log I/O error). So, we do want things to
2655 * go smoothly in case of just a SHUTDOWN w/o a
2656 * LOG_IO_ERROR.
2657 */
2658 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2659 /*
2660 * Can only perform callbacks in order. Since
2661 * this iclog is not in the DONE_SYNC/
2662 * DO_CALLBACK state, we skip the rest and
2663 * just try to clean up. If we set our iclog
2664 * to DO_CALLBACK, we will not process it when
2665 * we retry since a previous iclog is in the
2666 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002667 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 */
2669 if (!(iclog->ic_state &
2670 (XLOG_STATE_DONE_SYNC |
2671 XLOG_STATE_DO_CALLBACK))) {
2672 if (ciclog && (ciclog->ic_state ==
2673 XLOG_STATE_DONE_SYNC)) {
2674 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2675 }
2676 break;
2677 }
2678 /*
2679 * We now have an iclog that is in either the
2680 * DO_CALLBACK or DONE_SYNC states. The other
2681 * states (WANT_SYNC, SYNCING, or CALLBACK were
2682 * caught by the above if and are going to
2683 * clean (i.e. we aren't doing their callbacks)
2684 * see the above if.
2685 */
2686
2687 /*
2688 * We will do one more check here to see if we
2689 * have chased our tail around.
2690 */
2691
2692 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002693 if (lowest_lsn &&
2694 XFS_LSN_CMP(lowest_lsn,
Dave Chinner84f3c682010-12-03 22:11:29 +11002695 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 iclog = iclog->ic_next;
2697 continue; /* Leave this iclog for
2698 * another thread */
2699 }
2700
2701 iclog->ic_state = XLOG_STATE_CALLBACK;
2702
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
Dave Chinner84f3c682010-12-03 22:11:29 +11002704 /*
Dave Chinnerd35e88f2012-10-08 21:56:12 +11002705 * Completion of a iclog IO does not imply that
2706 * a transaction has completed, as transactions
2707 * can be large enough to span many iclogs. We
2708 * cannot change the tail of the log half way
2709 * through a transaction as this may be the only
2710 * transaction in the log and moving th etail to
2711 * point to the middle of it will prevent
2712 * recovery from finding the start of the
2713 * transaction. Hence we should only update the
2714 * last_sync_lsn if this iclog contains
2715 * transaction completion callbacks on it.
2716 *
2717 * We have to do this before we drop the
Dave Chinner84f3c682010-12-03 22:11:29 +11002718 * icloglock to ensure we are the only one that
2719 * can update it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 */
Dave Chinner84f3c682010-12-03 22:11:29 +11002721 ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2722 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
Dave Chinnerd35e88f2012-10-08 21:56:12 +11002723 if (iclog->ic_callback)
2724 atomic64_set(&log->l_last_sync_lsn,
2725 be64_to_cpu(iclog->ic_header.h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726
Dave Chinner84f3c682010-12-03 22:11:29 +11002727 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 ioerrors++;
Dave Chinner84f3c682010-12-03 22:11:29 +11002729
2730 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731
David Chinner114d23a2008-04-10 12:18:39 +10002732 /*
2733 * Keep processing entries in the callback list until
2734 * we come around and it is empty. We need to
2735 * atomically see that the list is empty and change the
2736 * state to DIRTY so that we don't miss any more
2737 * callbacks being added.
2738 */
2739 spin_lock(&iclog->ic_callback_lock);
2740 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002741 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 iclog->ic_callback_tail = &(iclog->ic_callback);
2743 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002744 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745
2746 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002747 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 cb_next = cb->cb_next;
2749 cb->cb_func(cb->cb_arg, aborted);
2750 }
David Chinner114d23a2008-04-10 12:18:39 +10002751 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 cb = iclog->ic_callback;
2753 }
2754
2755 loopdidcallbacks++;
2756 funcdidcallbacks++;
2757
David Chinner114d23a2008-04-10 12:18:39 +10002758 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002759 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002760 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2762 iclog->ic_state = XLOG_STATE_DIRTY;
2763
2764 /*
2765 * Transition from DIRTY to ACTIVE if applicable.
2766 * NOP if STATE_IOERROR.
2767 */
2768 xlog_state_clean_log(log);
2769
2770 /* wake up threads waiting in xfs_log_force() */
Dave Chinnereb40a872010-12-21 12:09:01 +11002771 wake_up_all(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
2773 iclog = iclog->ic_next;
2774 } while (first_iclog != iclog);
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002775
2776 if (repeats > 5000) {
2777 flushcnt += repeats;
2778 repeats = 0;
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002779 xfs_warn(log->l_mp,
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002780 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002781 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 }
2783 } while (!ioerrors && loopdidcallbacks);
2784
2785 /*
2786 * make one last gasp attempt to see if iclogs are being left in
2787 * limbo..
2788 */
2789#ifdef DEBUG
2790 if (funcdidcallbacks) {
2791 first_iclog = iclog = log->l_iclog;
2792 do {
2793 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2794 /*
2795 * Terminate the loop if iclogs are found in states
2796 * which will cause other threads to clean up iclogs.
2797 *
2798 * SYNCING - i/o completion will go through logs
2799 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002800 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 * IOERROR - give up hope all ye who enter here
2802 */
2803 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2804 iclog->ic_state == XLOG_STATE_SYNCING ||
2805 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2806 iclog->ic_state == XLOG_STATE_IOERROR )
2807 break;
2808 iclog = iclog->ic_next;
2809 } while (first_iclog != iclog);
2810 }
2811#endif
2812
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002813 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2814 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002815 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002816
2817 if (wake)
Dave Chinnereb40a872010-12-21 12:09:01 +11002818 wake_up_all(&log->l_flush_wait);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002819}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820
2821
2822/*
2823 * Finish transitioning this iclog to the dirty state.
2824 *
2825 * Make sure that we completely execute this routine only when this is
2826 * the last call to the iclog. There is a good chance that iclog flushes,
2827 * when we reach the end of the physical log, get turned into 2 separate
2828 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2829 * routine. By using the reference count bwritecnt, we guarantee that only
2830 * the second completion goes through.
2831 *
2832 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002833 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 */
David Chinnera8272ce2007-11-23 16:28:09 +11002835STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836xlog_state_done_syncing(
2837 xlog_in_core_t *iclog,
2838 int aborted)
2839{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002840 struct xlog *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002842 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
2844 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2845 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002846 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2848
2849
2850 /*
2851 * If we got an error, either on the first buffer, or in the case of
2852 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2853 * and none should ever be attempted to be written to disk
2854 * again.
2855 */
2856 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2857 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002858 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 return;
2860 }
2861 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2862 }
2863
2864 /*
2865 * Someone could be sleeping prior to writing out the next
2866 * iclog buffer, we wake them all, one will get to do the
2867 * I/O, the others get to wait for the result.
2868 */
Dave Chinnereb40a872010-12-21 12:09:01 +11002869 wake_up_all(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002870 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2872} /* xlog_state_done_syncing */
2873
2874
2875/*
2876 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002877 * sleep. We wait on the flush queue on the head iclog as that should be
2878 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2879 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 *
2881 * The in-core logs are used in a circular fashion. They are not used
2882 * out-of-order even when an iclog past the head is free.
2883 *
2884 * return:
2885 * * log_offset where xlog_write() can start writing into the in-core
2886 * log's data space.
2887 * * in-core log pointer to which xlog_write() should write.
2888 * * boolean indicating this is a continued write to an in-core log.
2889 * If this is the last write, then the in-core log's offset field
2890 * needs to be incremented, depending on the amount of data which
2891 * is copied.
2892 */
David Chinnera8272ce2007-11-23 16:28:09 +11002893STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05002894xlog_state_get_iclog_space(
2895 struct xlog *log,
2896 int len,
2897 struct xlog_in_core **iclogp,
2898 struct xlog_ticket *ticket,
2899 int *continued_write,
2900 int *logoffsetp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 int log_offset;
2903 xlog_rec_header_t *head;
2904 xlog_in_core_t *iclog;
2905 int error;
2906
2907restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002908 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002910 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10002911 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 }
2913
2914 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002915 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002917
2918 /* Wait for log writes to have flushed */
Dave Chinnereb40a872010-12-21 12:09:01 +11002919 xlog_wait(&log->l_flush_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 goto restart;
2921 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002922
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 head = &iclog->ic_header;
2924
David Chinner155cc6b2008-03-06 13:44:14 +11002925 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 log_offset = iclog->ic_offset;
2927
2928 /* On the 1st write to an iclog, figure out lsn. This works
2929 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2930 * committing to. If the offset is set, that's how many blocks
2931 * must be written.
2932 */
2933 if (log_offset == 0) {
2934 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002935 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002936 log->l_iclog_hsize,
2937 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002938 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2939 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002940 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 ASSERT(log->l_curr_block >= 0);
2942 }
2943
2944 /* If there is enough room to write everything, then do it. Otherwise,
2945 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2946 * bit is on, so this will get flushed out. Don't update ic_offset
2947 * until you know exactly how many bytes get copied. Therefore, wait
2948 * until later to update ic_offset.
2949 *
2950 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2951 * can fit into remaining data section.
2952 */
2953 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2954 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2955
Dave Chinner49641f12008-07-11 17:43:55 +10002956 /*
2957 * If I'm the only one writing to this iclog, sync it to disk.
2958 * We need to do an atomic compare and decrement here to avoid
2959 * racing with concurrent atomic_dec_and_lock() calls in
2960 * xlog_state_release_iclog() when there is more than one
2961 * reference to the iclog.
2962 */
2963 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2964 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002965 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002966 error = xlog_state_release_iclog(log, iclog);
2967 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002968 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002970 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 }
2972 goto restart;
2973 }
2974
2975 /* Do we have enough room to write the full amount in the remainder
2976 * of this iclog? Or must we continue a write on the next iclog and
2977 * mark this iclog as completely taken? In the case where we switch
2978 * iclogs (to mark it taken), this particular iclog will release/sync
2979 * to disk in xlog_write().
2980 */
2981 if (len <= iclog->ic_size - iclog->ic_offset) {
2982 *continued_write = 0;
2983 iclog->ic_offset += len;
2984 } else {
2985 *continued_write = 1;
2986 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2987 }
2988 *iclogp = iclog;
2989
2990 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002991 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993 *logoffsetp = log_offset;
2994 return 0;
2995} /* xlog_state_get_iclog_space */
2996
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997/* The first cnt-1 times through here we don't need to
2998 * move the grant write head because the permanent
2999 * reservation has reserved cnt times the unit amount.
3000 * Release part of current permanent unit reservation and
3001 * reset current reservation to be one units worth. Also
3002 * move grant reservation head forward.
3003 */
3004STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003005xlog_regrant_reserve_log_space(
3006 struct xlog *log,
3007 struct xlog_ticket *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003009 trace_xfs_log_regrant_reserve_enter(log, ticket);
3010
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 if (ticket->t_cnt > 0)
3012 ticket->t_cnt--;
3013
Christoph Hellwig28496962012-02-20 02:31:25 +00003014 xlog_grant_sub_space(log, &log->l_reserve_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11003015 ticket->t_curr_res);
Christoph Hellwig28496962012-02-20 02:31:25 +00003016 xlog_grant_sub_space(log, &log->l_write_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11003017 ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10003019 xlog_tic_reset_res(ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003020
3021 trace_xfs_log_regrant_reserve_sub(log, ticket);
3022
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 /* just return if we still have some of the pre-reserved space */
Dave Chinnerd0eb2f32010-12-21 12:29:14 +11003024 if (ticket->t_cnt > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
Christoph Hellwig28496962012-02-20 02:31:25 +00003027 xlog_grant_add_space(log, &log->l_reserve_head.grant,
Dave Chinnera69ed032010-12-21 12:08:20 +11003028 ticket->t_unit_res);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003029
3030 trace_xfs_log_regrant_reserve_exit(log, ticket);
3031
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10003033 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034} /* xlog_regrant_reserve_log_space */
3035
3036
3037/*
3038 * Give back the space left from a reservation.
3039 *
3040 * All the information we need to make a correct determination of space left
3041 * is present. For non-permanent reservations, things are quite easy. The
3042 * count should have been decremented to zero. We only need to deal with the
3043 * space remaining in the current reservation part of the ticket. If the
3044 * ticket contains a permanent reservation, there may be left over space which
3045 * needs to be released. A count of N means that N-1 refills of the current
3046 * reservation can be done before we need to ask for more space. The first
3047 * one goes to fill up the first current reservation. Once we run out of
3048 * space, the count will stay at zero and the only space remaining will be
3049 * in the current reservation field.
3050 */
3051STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003052xlog_ungrant_log_space(
3053 struct xlog *log,
3054 struct xlog_ticket *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055{
Dave Chinner663e4962010-12-21 12:06:05 +11003056 int bytes;
3057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 if (ticket->t_cnt > 0)
3059 ticket->t_cnt--;
3060
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003061 trace_xfs_log_ungrant_enter(log, ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003062 trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063
Dave Chinner663e4962010-12-21 12:06:05 +11003064 /*
3065 * If this is a permanent reservation ticket, we may be able to free
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 * up more space based on the remaining count.
3067 */
Dave Chinner663e4962010-12-21 12:06:05 +11003068 bytes = ticket->t_curr_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 if (ticket->t_cnt > 0) {
3070 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Dave Chinner663e4962010-12-21 12:06:05 +11003071 bytes += ticket->t_unit_res*ticket->t_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072 }
3073
Christoph Hellwig28496962012-02-20 02:31:25 +00003074 xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
3075 xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
Dave Chinner663e4962010-12-21 12:06:05 +11003076
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00003077 trace_xfs_log_ungrant_exit(log, ticket);
3078
Christoph Hellwigcfb7cdc2012-02-20 02:31:23 +00003079 xfs_log_space_wake(log->l_mp);
Christoph Hellwig09a423a2012-02-20 02:31:20 +00003080}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081
3082/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 * Flush iclog to disk if this is the last reference to the given iclog and
3084 * the WANT_SYNC bit is set.
3085 *
3086 * When this function is entered, the iclog is not necessarily in the
3087 * WANT_SYNC state. It may be sitting around waiting to get filled.
3088 *
3089 *
3090 */
David Chinnera8272ce2007-11-23 16:28:09 +11003091STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11003092xlog_state_release_iclog(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003093 struct xlog *log,
3094 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 int sync = 0; /* do we sync? */
3097
David Chinner155cc6b2008-03-06 13:44:14 +11003098 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003099 return -EIO;
David Chinner155cc6b2008-03-06 13:44:14 +11003100
3101 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
3102 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
3103 return 0;
3104
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003106 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003107 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
3110 iclog->ic_state == XLOG_STATE_WANT_SYNC);
3111
David Chinner155cc6b2008-03-06 13:44:14 +11003112 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11003113 /* update tail before writing to iclog */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003114 xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 sync++;
3116 iclog->ic_state = XLOG_STATE_SYNCING;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003117 iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
3118 xlog_verify_tail_lsn(log, iclog, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 /* cycle incremented when incrementing curr_block */
3120 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003121 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122
3123 /*
3124 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10003125 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
3127 * this iclog has consistent data, so we ignore IOERROR
3128 * flags after this point.
3129 */
David Chinnerb5893342008-03-06 13:44:06 +11003130 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003132 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133} /* xlog_state_release_iclog */
3134
3135
3136/*
3137 * This routine will mark the current iclog in the ring as WANT_SYNC
3138 * and move the current iclog pointer to the next iclog in the ring.
3139 * When this routine is called from xlog_state_get_iclog_space(), the
3140 * exact size of the iclog has not yet been determined. All we know is
3141 * that every data block. We have run out of space in this log record.
3142 */
3143STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003144xlog_state_switch_iclogs(
3145 struct xlog *log,
3146 struct xlog_in_core *iclog,
3147 int eventual_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148{
3149 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
3150 if (!eventual_size)
3151 eventual_size = iclog->ic_offset;
3152 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003153 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 log->l_prev_block = log->l_curr_block;
3155 log->l_prev_cycle = log->l_curr_cycle;
3156
3157 /* roll log?: ic_offset changed later */
3158 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
3159
3160 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11003161 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 log->l_mp->m_sb.sb_logsunit > 1) {
3163 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
3164 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
3165 }
3166
3167 if (log->l_curr_block >= log->l_logBBsize) {
3168 log->l_curr_cycle++;
3169 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
3170 log->l_curr_cycle++;
3171 log->l_curr_block -= log->l_logBBsize;
3172 ASSERT(log->l_curr_block >= 0);
3173 }
3174 ASSERT(iclog == log->l_iclog);
3175 log->l_iclog = iclog->ic_next;
3176} /* xlog_state_switch_iclogs */
3177
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178/*
3179 * Write out all data in the in-core log as of this exact moment in time.
3180 *
3181 * Data may be written to the in-core log during this call. However,
3182 * we don't guarantee this data will be written out. A change from past
3183 * implementation means this routine will *not* write out zero length LRs.
3184 *
3185 * Basically, we try and perform an intelligent scan of the in-core logs.
3186 * If we determine there is no flushable data, we just return. There is no
3187 * flushable data if:
3188 *
3189 * 1. the current iclog is active and has no data; the previous iclog
3190 * is in the active or dirty state.
3191 * 2. the current iclog is drity, and the previous iclog is in the
3192 * active or dirty state.
3193 *
David Chinner12017fa2008-08-13 16:34:31 +10003194 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195 *
3196 * 1. the current iclog is not in the active nor dirty state.
3197 * 2. the current iclog dirty, and the previous iclog is not in the
3198 * active nor dirty state.
3199 * 3. the current iclog is active, and there is another thread writing
3200 * to this particular iclog.
3201 * 4. a) the current iclog is active and has no other writers
3202 * b) when we return from flushing out this iclog, it is still
3203 * not in the active nor dirty state.
3204 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003205int
3206_xfs_log_force(
3207 struct xfs_mount *mp,
3208 uint flags,
3209 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
Mark Tinguelyad223e62012-06-14 09:22:15 -05003211 struct xlog *log = mp->m_log;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003212 struct xlog_in_core *iclog;
3213 xfs_lsn_t lsn;
3214
3215 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003217 xlog_cil_force(log);
Dave Chinner71e330b2010-05-21 14:37:18 +10003218
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003219 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
3221 iclog = log->l_iclog;
3222 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003223 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003224 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225 }
3226
3227 /* If the head iclog is not active nor dirty, we just attach
3228 * ourselves to the head and go to sleep.
3229 */
3230 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3231 iclog->ic_state == XLOG_STATE_DIRTY) {
3232 /*
3233 * If the head is dirty or (active and empty), then
3234 * we need to look at the previous iclog. If the previous
3235 * iclog is active or dirty we are done. There is nothing
3236 * to sync out. Otherwise, we attach ourselves to the
3237 * previous iclog and go to sleep.
3238 */
3239 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11003240 (atomic_read(&iclog->ic_refcnt) == 0
3241 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 iclog = iclog->ic_prev;
3243 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3244 iclog->ic_state == XLOG_STATE_DIRTY)
3245 goto no_sleep;
3246 else
3247 goto maybe_sleep;
3248 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11003249 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250 /* We are the only one with access to this
3251 * iclog. Flush it out now. There should
3252 * be a roundoff of zero to show that someone
3253 * has already taken care of the roundoff from
3254 * the previous sync.
3255 */
David Chinner155cc6b2008-03-06 13:44:14 +11003256 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003257 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003259 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
3261 if (xlog_state_release_iclog(log, iclog))
Dave Chinner24513372014-06-25 14:58:08 +10003262 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003263
3264 if (log_flushed)
3265 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003266 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003267 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 iclog->ic_state != XLOG_STATE_DIRTY)
3269 goto maybe_sleep;
3270 else
3271 goto no_sleep;
3272 } else {
3273 /* Someone else is writing to this iclog.
3274 * Use its call to flush out the data. However,
3275 * the other thread may not force out this LR,
3276 * so we mark it WANT_SYNC.
3277 */
3278 xlog_state_switch_iclogs(log, iclog, 0);
3279 goto maybe_sleep;
3280 }
3281 }
3282 }
3283
3284 /* By the time we come around again, the iclog could've been filled
3285 * which would give it another lsn. If we have a new lsn, just
3286 * return because the relevant data has been flushed.
3287 */
3288maybe_sleep:
3289 if (flags & XFS_LOG_SYNC) {
3290 /*
3291 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003292 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 * Then we check again after waking up, in case our
3294 * sleep was disturbed by a bad news.
3295 */
3296 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003297 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003298 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 }
3300 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003301 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 /*
3303 * No need to grab the log lock here since we're
3304 * only deciding whether or not to return EIO
3305 * and the memory read should be atomic.
3306 */
3307 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003308 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003309 if (log_flushed)
3310 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311 } else {
3312
3313no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003314 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315 }
3316 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003317}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318
3319/*
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003320 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3321 * about errors or whether the log was flushed or not. This is the normal
3322 * interface to use when trying to unpin items or move the log forward.
3323 */
3324void
3325xfs_log_force(
3326 xfs_mount_t *mp,
3327 uint flags)
3328{
3329 int error;
3330
Dave Chinner14c26c62012-04-24 16:33:31 +10003331 trace_xfs_log_force(mp, 0);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003332 error = _xfs_log_force(mp, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003333 if (error)
3334 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003335}
3336
3337/*
3338 * Force the in-core log to disk for a specific LSN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 *
3340 * Find in-core log with lsn.
3341 * If it is in the DIRTY state, just return.
3342 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3343 * state and go to sleep or return.
3344 * If it is in any other state, go to sleep or return.
3345 *
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003346 * Synchronous forces are implemented with a signal variable. All callers
3347 * to force a given lsn to disk will wait on a the sv attached to the
3348 * specific in-core log. When given in-core log finally completes its
3349 * write to disk, that thread will wake up all threads waiting on the
3350 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003352int
3353_xfs_log_force_lsn(
3354 struct xfs_mount *mp,
3355 xfs_lsn_t lsn,
3356 uint flags,
3357 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
Mark Tinguelyad223e62012-06-14 09:22:15 -05003359 struct xlog *log = mp->m_log;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003360 struct xlog_in_core *iclog;
3361 int already_slept = 0;
3362
3363 ASSERT(lsn != 0);
3364
3365 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003367 lsn = xlog_cil_force_lsn(log, lsn);
3368 if (lsn == NULLCOMMITLSN)
3369 return 0;
Dave Chinner71e330b2010-05-21 14:37:18 +10003370
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371try_again:
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003372 spin_lock(&log->l_icloglock);
3373 iclog = log->l_iclog;
3374 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003375 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003376 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377 }
3378
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003379 do {
3380 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3381 iclog = iclog->ic_next;
3382 continue;
3383 }
3384
3385 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3386 spin_unlock(&log->l_icloglock);
3387 return 0;
3388 }
3389
3390 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3391 /*
3392 * We sleep here if we haven't already slept (e.g.
3393 * this is the first time we've looked at the correct
3394 * iclog buf) and the buffer before us is going to
3395 * be sync'ed. The reason for this is that if we
3396 * are doing sync transactions here, by waiting for
3397 * the previous I/O to complete, we can allow a few
3398 * more transactions into this iclog before we close
3399 * it down.
3400 *
3401 * Otherwise, we mark the buffer WANT_SYNC, and bump
3402 * up the refcnt so we can release the log (which
3403 * drops the ref count). The state switch keeps new
3404 * transaction commits from using this buffer. When
3405 * the current commits finish writing into the buffer,
3406 * the refcount will drop to zero and the buffer will
3407 * go out then.
3408 */
3409 if (!already_slept &&
3410 (iclog->ic_prev->ic_state &
3411 (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3412 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3413
3414 XFS_STATS_INC(xs_log_force_sleep);
3415
Dave Chinnereb40a872010-12-21 12:09:01 +11003416 xlog_wait(&iclog->ic_prev->ic_write_wait,
3417 &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003418 if (log_flushed)
3419 *log_flushed = 1;
3420 already_slept = 1;
3421 goto try_again;
3422 }
David Chinner155cc6b2008-03-06 13:44:14 +11003423 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003425 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 if (xlog_state_release_iclog(log, iclog))
Dave Chinner24513372014-06-25 14:58:08 +10003427 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003428 if (log_flushed)
3429 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003430 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003433 if ((flags & XFS_LOG_SYNC) && /* sleep */
3434 !(iclog->ic_state &
3435 (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3436 /*
3437 * Don't wait on completion if we know that we've
3438 * gotten a log write error.
3439 */
3440 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3441 spin_unlock(&log->l_icloglock);
Dave Chinner24513372014-06-25 14:58:08 +10003442 return -EIO;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003443 }
3444 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003445 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003446 /*
3447 * No need to grab the log lock here since we're
3448 * only deciding whether or not to return EIO
3449 * and the memory read should be atomic.
3450 */
3451 if (iclog->ic_state & XLOG_STATE_IOERROR)
Dave Chinner24513372014-06-25 14:58:08 +10003452 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003454 if (log_flushed)
3455 *log_flushed = 1;
3456 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003457 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 }
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003459
3460 return 0;
3461 } while (iclog != log->l_iclog);
3462
3463 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003465}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003467/*
3468 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3469 * about errors or whether the log was flushed or not. This is the normal
3470 * interface to use when trying to unpin items or move the log forward.
3471 */
3472void
3473xfs_log_force_lsn(
3474 xfs_mount_t *mp,
3475 xfs_lsn_t lsn,
3476 uint flags)
3477{
3478 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479
Dave Chinner14c26c62012-04-24 16:33:31 +10003480 trace_xfs_log_force(mp, lsn);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003481 error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003482 if (error)
3483 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003484}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486/*
3487 * Called when we want to mark the current iclog as being ready to sync to
3488 * disk.
3489 */
David Chinnera8272ce2007-11-23 16:28:09 +11003490STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003491xlog_state_want_sync(
3492 struct xlog *log,
3493 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003495 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496
3497 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3498 xlog_state_switch_iclogs(log, iclog, 0);
3499 } else {
3500 ASSERT(iclog->ic_state &
3501 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3502 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003503}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504
3505
3506/*****************************************************************************
3507 *
3508 * TICKET functions
3509 *
3510 *****************************************************************************
3511 */
3512
3513/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003514 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003516void
3517xfs_log_ticket_put(
3518 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003520 ASSERT(atomic_read(&ticket->t_ref) > 0);
Dave Chinnereb40a872010-12-21 12:09:01 +11003521 if (atomic_dec_and_test(&ticket->t_ref))
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003522 kmem_zone_free(xfs_log_ticket_zone, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003523}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003525xlog_ticket_t *
3526xfs_log_ticket_get(
3527 xlog_ticket_t *ticket)
3528{
3529 ASSERT(atomic_read(&ticket->t_ref) > 0);
3530 atomic_inc(&ticket->t_ref);
3531 return ticket;
3532}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533
3534/*
Jie Liue773fc92013-08-12 20:50:01 +10003535 * Figure out the total log space unit (in bytes) that would be
3536 * required for a log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 */
Jie Liue773fc92013-08-12 20:50:01 +10003538int
3539xfs_log_calc_unit_res(
3540 struct xfs_mount *mp,
3541 int unit_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542{
Jie Liue773fc92013-08-12 20:50:01 +10003543 struct xlog *log = mp->m_log;
3544 int iclog_space;
3545 uint num_headers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546
3547 /*
3548 * Permanent reservations have up to 'cnt'-1 active log operations
3549 * in the log. A unit in this case is the amount of space for one
3550 * of these log operations. Normal reservations have a cnt of 1
3551 * and their unit amount is the total amount of space required.
3552 *
3553 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003554 * which occupy space in the on-disk log.
3555 *
3556 * Normal form of a transaction is:
3557 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3558 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3559 *
3560 * We need to account for all the leadup data and trailer data
3561 * around the transaction data.
3562 * And then we need to account for the worst case in terms of using
3563 * more space.
3564 * The worst case will happen if:
3565 * - the placement of the transaction happens to be such that the
3566 * roundoff is at its maximum
3567 * - the transaction data is synced before the commit record is synced
3568 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3569 * Therefore the commit record is in its own Log Record.
3570 * This can happen as the commit record is called with its
3571 * own region to xlog_write().
3572 * This then means that in the worst case, roundoff can happen for
3573 * the commit-rec as well.
3574 * The commit-rec is smaller than padding in this scenario and so it is
3575 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 */
3577
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003578 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003580 unit_bytes += sizeof(xfs_trans_header_t);
3581
3582 /* for start-rec */
3583 unit_bytes += sizeof(xlog_op_header_t);
3584
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003585 /*
3586 * for LR headers - the space for data in an iclog is the size minus
3587 * the space used for the headers. If we use the iclog size, then we
3588 * undercalculate the number of headers required.
3589 *
3590 * Furthermore - the addition of op headers for split-recs might
3591 * increase the space required enough to require more log and op
3592 * headers, so take that into account too.
3593 *
3594 * IMPORTANT: This reservation makes the assumption that if this
3595 * transaction is the first in an iclog and hence has the LR headers
3596 * accounted to it, then the remaining space in the iclog is
3597 * exclusively for this transaction. i.e. if the transaction is larger
3598 * than the iclog, it will be the only thing in that iclog.
3599 * Fundamentally, this means we must pass the entire log vector to
3600 * xlog_write to guarantee this.
3601 */
3602 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3603 num_headers = howmany(unit_bytes, iclog_space);
3604
3605 /* for split-recs - ophdrs added when data split over LRs */
3606 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3607
3608 /* add extra header reservations if we overrun */
3609 while (!num_headers ||
3610 howmany(unit_bytes, iclog_space) > num_headers) {
3611 unit_bytes += sizeof(xlog_op_header_t);
3612 num_headers++;
3613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 unit_bytes += log->l_iclog_hsize * num_headers;
3615
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003616 /* for commit-rec LR header - note: padding will subsume the ophdr */
3617 unit_bytes += log->l_iclog_hsize;
3618
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003619 /* for roundoff padding for transaction data and one for commit record */
Jie Liue773fc92013-08-12 20:50:01 +10003620 if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) {
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003621 /* log su roundoff */
Jie Liue773fc92013-08-12 20:50:01 +10003622 unit_bytes += 2 * mp->m_sb.sb_logsunit;
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003623 } else {
3624 /* BB roundoff */
Jie Liue773fc92013-08-12 20:50:01 +10003625 unit_bytes += 2 * BBSIZE;
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003626 }
3627
Jie Liue773fc92013-08-12 20:50:01 +10003628 return unit_bytes;
3629}
3630
3631/*
3632 * Allocate and initialise a new log ticket.
3633 */
3634struct xlog_ticket *
3635xlog_ticket_alloc(
3636 struct xlog *log,
3637 int unit_bytes,
3638 int cnt,
3639 char client,
3640 bool permanent,
3641 xfs_km_flags_t alloc_flags)
3642{
3643 struct xlog_ticket *tic;
3644 int unit_res;
3645
3646 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
3647 if (!tic)
3648 return NULL;
3649
3650 unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);
3651
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003652 atomic_set(&tic->t_ref, 1);
Christoph Hellwig14a72352012-02-20 02:31:24 +00003653 tic->t_task = current;
Dave Chinner10547942010-12-21 12:02:25 +11003654 INIT_LIST_HEAD(&tic->t_queue);
Jie Liue773fc92013-08-12 20:50:01 +10003655 tic->t_unit_res = unit_res;
3656 tic->t_curr_res = unit_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 tic->t_cnt = cnt;
3658 tic->t_ocnt = cnt;
Akinobu Mitaecb34032013-03-04 21:58:20 +09003659 tic->t_tid = prandom_u32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 tic->t_clientid = client;
3661 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003662 tic->t_trans_type = 0;
Christoph Hellwig9006fb92012-02-20 02:31:31 +00003663 if (permanent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 tic->t_flags |= XLOG_TIC_PERM_RESERV;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665
Christoph Hellwig0adba532007-08-30 17:21:46 +10003666 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003667
Linus Torvalds1da177e2005-04-16 15:20:36 -07003668 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003669}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670
3671
3672/******************************************************************************
3673 *
3674 * Log debug routines
3675 *
3676 ******************************************************************************
3677 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003678#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679/*
3680 * Make sure that the destination ptr is within the valid data region of
3681 * one of the iclogs. This uses backup pointers stored in a different
3682 * part of the log in case we trash the log structure.
3683 */
3684void
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003685xlog_verify_dest_ptr(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003686 struct xlog *log,
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003687 void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688{
3689 int i;
3690 int good_ptr = 0;
3691
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003692 for (i = 0; i < log->l_iclog_bufs; i++) {
3693 if (ptr >= log->l_iclog_bak[i] &&
3694 ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003695 good_ptr++;
3696 }
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003697
3698 if (!good_ptr)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003699 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003702/*
3703 * Check to make sure the grant write head didn't just over lap the tail. If
3704 * the cycles are the same, we can't be overlapping. Otherwise, make sure that
3705 * the cycles differ by exactly one and check the byte count.
3706 *
3707 * This check is run unlocked, so can give false positives. Rather than assert
3708 * on failures, use a warn-once flag and a panic tag to allow the admin to
3709 * determine if they want to panic the machine when such an error occurs. For
3710 * debug kernels this will have the same effect as using an assert but, unlinke
3711 * an assert, it can be turned off at runtime.
3712 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713STATIC void
Dave Chinner3f336c62010-12-21 12:02:52 +11003714xlog_verify_grant_tail(
Mark Tinguelyad223e62012-06-14 09:22:15 -05003715 struct xlog *log)
Dave Chinner3f336c62010-12-21 12:02:52 +11003716{
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003717 int tail_cycle, tail_blocks;
Dave Chinnera69ed032010-12-21 12:08:20 +11003718 int cycle, space;
Dave Chinner3f336c62010-12-21 12:02:52 +11003719
Christoph Hellwig28496962012-02-20 02:31:25 +00003720 xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003721 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3722 if (tail_cycle != cycle) {
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003723 if (cycle - 1 != tail_cycle &&
3724 !(log->l_flags & XLOG_TAIL_WARN)) {
3725 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3726 "%s: cycle - 1 != tail_cycle", __func__);
3727 log->l_flags |= XLOG_TAIL_WARN;
3728 }
3729
3730 if (space > BBTOB(tail_blocks) &&
3731 !(log->l_flags & XLOG_TAIL_WARN)) {
3732 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3733 "%s: space > BBTOB(tail_blocks)", __func__);
3734 log->l_flags |= XLOG_TAIL_WARN;
3735 }
Dave Chinner3f336c62010-12-21 12:02:52 +11003736 }
3737}
3738
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739/* check if it will fit */
3740STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003741xlog_verify_tail_lsn(
3742 struct xlog *log,
3743 struct xlog_in_core *iclog,
3744 xfs_lsn_t tail_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745{
3746 int blocks;
3747
3748 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3749 blocks =
3750 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3751 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003752 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003753 } else {
3754 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3755
3756 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003757 xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003758
3759 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3760 if (blocks < BTOBB(iclog->ic_offset) + 1)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003761 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762 }
3763} /* xlog_verify_tail_lsn */
3764
3765/*
3766 * Perform a number of checks on the iclog before writing to disk.
3767 *
3768 * 1. Make sure the iclogs are still circular
3769 * 2. Make sure we have a good magic number
3770 * 3. Make sure we don't have magic numbers in the data
3771 * 4. Check fields of each log operation header for:
3772 * A. Valid client identifier
3773 * B. tid ptr value falls in valid ptr space (user space code)
3774 * C. Length in log record header is correct according to the
3775 * individual operation headers within record.
3776 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3777 * log, check the preceding blocks of the physical log to make sure all
3778 * the cycle numbers agree with the current cycle number.
3779 */
3780STATIC void
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003781xlog_verify_iclog(
3782 struct xlog *log,
3783 struct xlog_in_core *iclog,
3784 int count,
Thiago Farina667a9292012-11-12 21:32:59 -02003785 bool syncing)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786{
3787 xlog_op_header_t *ophead;
3788 xlog_in_core_t *icptr;
3789 xlog_in_core_2_t *xhdr;
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003790 void *base_ptr, *ptr, *p;
Christoph Hellwigdb9d67d2015-06-22 09:43:32 +10003791 ptrdiff_t field_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 __uint8_t clientid;
3793 int len, i, j, k, op_len;
3794 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795
3796 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003797 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798 icptr = log->l_iclog;
Geyslan G. Bem643f7c42013-10-30 16:01:00 -05003799 for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
3800 ASSERT(icptr);
3801
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802 if (icptr != log->l_iclog)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003803 xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003804 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805
3806 /* check log magic numbers */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003807 if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003808 xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003810 base_ptr = ptr = &iclog->ic_header;
3811 p = &iclog->ic_header;
3812 for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) {
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003813 if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003814 xfs_emerg(log->l_mp, "%s: unexpected magic num",
3815 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 }
3817
3818 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003819 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003820 base_ptr = ptr = iclog->ic_datap;
3821 ophead = ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003822 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 for (i = 0; i < len; i++) {
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003824 ophead = ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825
3826 /* clientid is only 1 byte */
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003827 p = &ophead->oh_clientid;
3828 field_offset = p - base_ptr;
Thiago Farina667a9292012-11-12 21:32:59 -02003829 if (!syncing || (field_offset & 0x1ff)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830 clientid = ophead->oh_clientid;
3831 } else {
Christoph Hellwigb2a922c2015-06-22 09:45:10 +10003832 idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3834 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3835 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003836 clientid = xlog_get_client_id(
3837 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003839 clientid = xlog_get_client_id(
3840 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 }
3842 }
3843 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003844 xfs_warn(log->l_mp,
3845 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3846 __func__, clientid, ophead,
3847 (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848
3849 /* check length */
Christoph Hellwig5809d5e2015-06-22 09:44:47 +10003850 p = &ophead->oh_len;
3851 field_offset = p - base_ptr;
Thiago Farina667a9292012-11-12 21:32:59 -02003852 if (!syncing || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003853 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 } else {
Christoph Hellwigdb9d67d2015-06-22 09:43:32 +10003855 idx = BTOBBT((uintptr_t)&ophead->oh_len -
3856 (uintptr_t)iclog->ic_datap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3858 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3859 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003860 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003862 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 }
3864 }
3865 ptr += sizeof(xlog_op_header_t) + op_len;
3866 }
3867} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003868#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869
3870/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003871 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 */
3873STATIC int
3874xlog_state_ioerror(
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003875 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876{
3877 xlog_in_core_t *iclog, *ic;
3878
3879 iclog = log->l_iclog;
3880 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3881 /*
3882 * Mark all the incore logs IOERROR.
3883 * From now on, no log flushes will result.
3884 */
3885 ic = iclog;
3886 do {
3887 ic->ic_state = XLOG_STATE_IOERROR;
3888 ic = ic->ic_next;
3889 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003890 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 }
3892 /*
3893 * Return non-zero, if state transition has already happened.
3894 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003895 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896}
3897
3898/*
3899 * This is called from xfs_force_shutdown, when we're forcibly
3900 * shutting down the filesystem, typically because of an IO error.
3901 * Our main objectives here are to make sure that:
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003902 * a. if !logerror, flush the logs to disk. Anything modified
3903 * after this is ignored.
3904 * b. the filesystem gets marked 'SHUTDOWN' for all interested
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 * parties to find out, 'atomically'.
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003906 * c. those who're sleeping on log reservations, pinned objects and
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 * other resources get woken up, and be told the bad news.
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003908 * d. nothing new gets queued up after (b) and (c) are done.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003909 *
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003910 * Note: for the !logerror case we need to flush the regions held in memory out
3911 * to disk first. This needs to be done before the log is marked as shutdown,
3912 * otherwise the iclog writes will fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003913 */
3914int
3915xfs_log_force_umount(
3916 struct xfs_mount *mp,
3917 int logerror)
3918{
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05003919 struct xlog *log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003920 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003921
3922 log = mp->m_log;
3923
3924 /*
3925 * If this happens during log recovery, don't worry about
3926 * locking; the log isn't open for business yet.
3927 */
3928 if (!log ||
3929 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3930 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003931 if (mp->m_sb_bp)
3932 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 }
3935
3936 /*
3937 * Somebody could've already done the hard work for us.
3938 * No need to get locks for this.
3939 */
3940 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3941 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003942 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003943 }
Dave Chinner9da1ab12010-05-17 15:51:59 +10003944
3945 /*
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003946 * Flush all the completed transactions to disk before marking the log
3947 * being shut down. We need to do it in this order to ensure that
3948 * completed operations are safely on disk before we shut down, and that
3949 * we don't have to issue any buffer IO after the shutdown flags are set
3950 * to guarantee this.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003951 */
Christoph Hellwig93b8a582011-12-06 21:58:07 +00003952 if (!logerror)
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003953 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
Dave Chinner9da1ab12010-05-17 15:51:59 +10003954
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 /*
Dave Chinner3f16b982010-12-21 12:29:01 +11003956 * mark the filesystem and the as in a shutdown state and wake
3957 * everybody up to tell them the bad news.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003959 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003960 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003961 if (mp->m_sb_bp)
3962 XFS_BUF_DONE(mp->m_sb_bp);
3963
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 /*
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003965 * Mark the log and the iclogs with IO error flags to prevent any
3966 * further log IO from being issued or completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967 */
3968 log->l_flags |= XLOG_IO_ERROR;
Dave Chinnera870fe6d2014-10-02 09:02:28 +10003969 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003970 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003971
3972 /*
Dave Chinner10547942010-12-21 12:02:25 +11003973 * We don't want anybody waiting for log reservations after this. That
3974 * means we have to wake up everybody queued up on reserveq as well as
3975 * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3976 * we don't enqueue anything once the SHUTDOWN flag is set, and this
Dave Chinner3f16b982010-12-21 12:29:01 +11003977 * action is protected by the grant locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 */
Christoph Hellwiga79bf2d2012-02-20 02:31:27 +00003979 xlog_grant_head_wake_all(&log->l_reserve_head);
3980 xlog_grant_head_wake_all(&log->l_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003981
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 /*
Dave Chinnerac983512014-05-07 08:05:50 +10003983 * Wake up everybody waiting on xfs_log_force. Wake the CIL push first
3984 * as if the log writes were completed. The abort handling in the log
3985 * item committed callback functions will do this again under lock to
3986 * avoid races.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 */
Dave Chinnerac983512014-05-07 08:05:50 +10003988 wake_up_all(&log->l_cilp->xc_commit_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3990
3991#ifdef XFSERRORDEBUG
3992 {
3993 xlog_in_core_t *iclog;
3994
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003995 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996 iclog = log->l_iclog;
3997 do {
3998 ASSERT(iclog->ic_callback == 0);
3999 iclog = iclog->ic_next;
4000 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10004001 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004002 }
4003#endif
4004 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01004005 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006}
4007
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10004008STATIC int
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -05004009xlog_iclogs_empty(
4010 struct xlog *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004011{
4012 xlog_in_core_t *iclog;
4013
4014 iclog = log->l_iclog;
4015 do {
4016 /* endianness does not matter here, zero is zero in
4017 * any language.
4018 */
4019 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01004020 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021 iclog = iclog->ic_next;
4022 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01004023 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024}
Dave Chinnerf661f1e2012-10-08 21:56:02 +11004025