blob: 33f718f92a4849df234880dfe190069791d7bcb6 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_log_priv.h"
30#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110031#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_alloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_log_recover.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include "xfs_trans_priv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110036#include "xfs_dinode.h"
37#include "xfs_inode.h"
38#include "xfs_rw.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000039#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
David Chinnereb01c9c2008-04-10 12:18:46 +100041kmem_zone_t *xfs_log_ticket_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* Local miscellaneous function prototypes */
Dave Chinner55b66332010-03-23 11:43:17 +110044STATIC int xlog_commit_record(struct log *log, struct xlog_ticket *ticket,
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 xlog_in_core_t **, xfs_lsn_t *);
46STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp,
47 xfs_buftarg_t *log_target,
48 xfs_daddr_t blk_offset,
49 int num_bblks);
50STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes);
51STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
Nathan Scottc41564b2006-03-29 08:55:14 +100052STATIC void xlog_dealloc_log(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/* local state machine functions */
55STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
56STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog);
57STATIC int xlog_state_get_iclog_space(xlog_t *log,
58 int len,
59 xlog_in_core_t **iclog,
60 xlog_ticket_t *ticket,
61 int *continued_write,
62 int *logoffsetp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063STATIC int xlog_state_release_iclog(xlog_t *log,
64 xlog_in_core_t *iclog);
65STATIC void xlog_state_switch_iclogs(xlog_t *log,
66 xlog_in_core_t *iclog,
67 int eventual_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068STATIC void xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog);
69
70/* local functions to manipulate grant head */
71STATIC int xlog_grant_log_space(xlog_t *log,
72 xlog_ticket_t *xtic);
73STATIC void xlog_grant_push_ail(xfs_mount_t *mp,
74 int need_bytes);
75STATIC void xlog_regrant_reserve_log_space(xlog_t *log,
76 xlog_ticket_t *ticket);
77STATIC int xlog_regrant_write_log_space(xlog_t *log,
78 xlog_ticket_t *ticket);
79STATIC void xlog_ungrant_log_space(xlog_t *log,
80 xlog_ticket_t *ticket);
81
Nathan Scottcfcbbbd2005-11-02 15:12:04 +110082#if defined(DEBUG)
Christoph Hellwige6b1f272010-03-23 11:47:38 +110083STATIC void xlog_verify_dest_ptr(xlog_t *log, char *ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084STATIC void xlog_verify_grant_head(xlog_t *log, int equals);
85STATIC void xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog,
86 int count, boolean_t syncing);
87STATIC void xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog,
88 xfs_lsn_t tail_lsn);
89#else
90#define xlog_verify_dest_ptr(a,b)
91#define xlog_verify_grant_head(a,b)
92#define xlog_verify_iclog(a,b,c,d)
93#define xlog_verify_tail_lsn(a,b,c)
94#endif
95
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100096STATIC int xlog_iclogs_empty(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Christoph Hellwigdd954c62006-01-11 15:34:50 +110098
99static void
100xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
101{
102 if (*qp) {
103 tic->t_next = (*qp);
104 tic->t_prev = (*qp)->t_prev;
105 (*qp)->t_prev->t_next = tic;
106 (*qp)->t_prev = tic;
107 } else {
108 tic->t_prev = tic->t_next = tic;
109 *qp = tic;
110 }
111
112 tic->t_flags |= XLOG_TIC_IN_Q;
113}
114
115static void
116xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
117{
118 if (tic == tic->t_next) {
119 *qp = NULL;
120 } else {
121 *qp = tic->t_next;
122 tic->t_next->t_prev = tic->t_prev;
123 tic->t_prev->t_next = tic->t_next;
124 }
125
126 tic->t_next = tic->t_prev = NULL;
127 tic->t_flags &= ~XLOG_TIC_IN_Q;
128}
129
130static void
131xlog_grant_sub_space(struct log *log, int bytes)
132{
133 log->l_grant_write_bytes -= bytes;
134 if (log->l_grant_write_bytes < 0) {
135 log->l_grant_write_bytes += log->l_logsize;
136 log->l_grant_write_cycle--;
137 }
138
139 log->l_grant_reserve_bytes -= bytes;
140 if ((log)->l_grant_reserve_bytes < 0) {
141 log->l_grant_reserve_bytes += log->l_logsize;
142 log->l_grant_reserve_cycle--;
143 }
144
145}
146
147static void
148xlog_grant_add_space_write(struct log *log, int bytes)
149{
Michael Nishimotod729eae2008-05-19 16:34:20 +1000150 int tmp = log->l_logsize - log->l_grant_write_bytes;
151 if (tmp > bytes)
152 log->l_grant_write_bytes += bytes;
153 else {
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100154 log->l_grant_write_cycle++;
Michael Nishimotod729eae2008-05-19 16:34:20 +1000155 log->l_grant_write_bytes = bytes - tmp;
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100156 }
157}
158
159static void
160xlog_grant_add_space_reserve(struct log *log, int bytes)
161{
Michael Nishimotod729eae2008-05-19 16:34:20 +1000162 int tmp = log->l_logsize - log->l_grant_reserve_bytes;
163 if (tmp > bytes)
164 log->l_grant_reserve_bytes += bytes;
165 else {
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100166 log->l_grant_reserve_cycle++;
Michael Nishimotod729eae2008-05-19 16:34:20 +1000167 log->l_grant_reserve_bytes = bytes - tmp;
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100168 }
169}
170
171static inline void
172xlog_grant_add_space(struct log *log, int bytes)
173{
174 xlog_grant_add_space_write(log, bytes);
175 xlog_grant_add_space_reserve(log, bytes);
176}
177
Christoph Hellwig0adba532007-08-30 17:21:46 +1000178static void
179xlog_tic_reset_res(xlog_ticket_t *tic)
180{
181 tic->t_res_num = 0;
182 tic->t_res_arr_sum = 0;
183 tic->t_res_num_ophdrs = 0;
184}
185
186static void
187xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
188{
189 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
190 /* add to overflow and start again */
191 tic->t_res_o_flow += tic->t_res_arr_sum;
192 tic->t_res_num = 0;
193 tic->t_res_arr_sum = 0;
194 }
195
196 tic->t_res_arr[tic->t_res_num].r_len = len;
197 tic->t_res_arr[tic->t_res_num].r_type = type;
198 tic->t_res_arr_sum += len;
199 tic->t_res_num++;
200}
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/*
203 * NOTES:
204 *
205 * 1. currblock field gets updated at startup and after in-core logs
206 * marked as with WANT_SYNC.
207 */
208
209/*
210 * This routine is called when a user of a log manager ticket is done with
211 * the reservation. If the ticket was ever used, then a commit record for
212 * the associated transaction is written out as a log operation header with
213 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
214 * a given ticket. If the ticket was one with a permanent reservation, then
215 * a few operations are done differently. Permanent reservation tickets by
216 * default don't release the reservation. They just commit the current
217 * transaction with the belief that the reservation is still needed. A flag
218 * must be passed in before permanent reservations are actually released.
219 * When these type of tickets are not released, they need to be set into
220 * the inited state again. By doing this, a start record will be written
221 * out when the next write occurs.
222 */
223xfs_lsn_t
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000224xfs_log_done(
225 struct xfs_mount *mp,
226 struct xlog_ticket *ticket,
227 struct xlog_in_core **iclog,
228 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000230 struct log *log = mp->m_log;
231 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 if (XLOG_FORCED_SHUTDOWN(log) ||
234 /*
235 * If nothing was ever written, don't write out commit record.
236 * If we get an error, just continue and give back the log ticket.
237 */
238 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
Dave Chinner55b66332010-03-23 11:43:17 +1100239 (xlog_commit_record(log, ticket, iclog, &lsn)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 lsn = (xfs_lsn_t) -1;
241 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
242 flags |= XFS_LOG_REL_PERM_RESERV;
243 }
244 }
245
246
247 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
248 (flags & XFS_LOG_REL_PERM_RESERV)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000249 trace_xfs_log_done_nonperm(log, ticket);
250
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000252 * Release ticket if not permanent reservation or a specific
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 * request has been made to release a permanent reservation.
254 */
255 xlog_ungrant_log_space(log, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100256 xfs_log_ticket_put(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000258 trace_xfs_log_done_perm(log, ticket);
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 xlog_regrant_reserve_log_space(log, ticket);
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000261 /* If this ticket was a permanent reservation and we aren't
262 * trying to release it, reset the inited flags; so next time
263 * we write, a start record will be written out.
264 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 ticket->t_flags |= XLOG_TIC_INITED;
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268 return lsn;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000269}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271/*
272 * Attaches a new iclog I/O completion callback routine during
273 * transaction commit. If the log is in error state, a non-zero
274 * return code is handed back and the caller is responsible for
275 * executing the callback at an appropriate time.
276 */
277int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000278xfs_log_notify(
279 struct xfs_mount *mp,
280 struct xlog_in_core *iclog,
281 xfs_log_callback_t *cb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000283 int abortflg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
David Chinner114d23a2008-04-10 12:18:39 +1000285 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
287 if (!abortflg) {
288 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
289 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
290 cb->cb_next = NULL;
291 *(iclog->ic_callback_tail) = cb;
292 iclog->ic_callback_tail = &(cb->cb_next);
293 }
David Chinner114d23a2008-04-10 12:18:39 +1000294 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return abortflg;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000296}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
298int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000299xfs_log_release_iclog(
300 struct xfs_mount *mp,
301 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000303 if (xlog_state_release_iclog(mp->m_log, iclog)) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000304 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +0100305 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 }
307
308 return 0;
309}
310
311/*
312 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
313 * to the reservation.
314 * 2. Potentially, push buffers at tail of log to disk.
315 *
316 * Each reservation is going to reserve extra space for a log record header.
317 * When writes happen to the on-disk log, we don't subtract the length of the
318 * log record header from any reservation. By wasting space in each
319 * reservation, we prevent over allocation problems.
320 */
321int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000322xfs_log_reserve(
323 struct xfs_mount *mp,
324 int unit_bytes,
325 int cnt,
326 struct xlog_ticket **ticket,
327 __uint8_t client,
328 uint flags,
329 uint t_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000331 struct log *log = mp->m_log;
332 struct xlog_ticket *internal_ticket;
333 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
337 if (XLOG_FORCED_SHUTDOWN(log))
338 return XFS_ERROR(EIO);
339
340 XFS_STATS_INC(xs_try_logspace);
341
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (*ticket != NULL) {
344 ASSERT(flags & XFS_LOG_PERM_RESERV);
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000345 internal_ticket = *ticket;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000346
Dave Chinner524ee362010-05-07 11:05:05 +1000347 /*
348 * this is a new transaction on the ticket, so we need to
349 * change the transaction ID so that the next transaction has a
350 * different TID in the log. Just add one to the existing tid
351 * so that we can see chains of rolling transactions in the log
352 * easily.
353 */
354 internal_ticket->t_tid++;
355
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000356 trace_xfs_log_reserve(log, internal_ticket);
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
359 retval = xlog_regrant_write_log_space(log, internal_ticket);
360 } else {
361 /* may sleep if need to allocate more tickets */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100362 internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
Dave Chinner3383ca52010-05-07 11:04:17 +1000363 client, flags,
364 KM_SLEEP|KM_MAYFAIL);
David Chinnereb01c9c2008-04-10 12:18:46 +1000365 if (!internal_ticket)
366 return XFS_ERROR(ENOMEM);
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000367 internal_ticket->t_trans_type = t_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 *ticket = internal_ticket;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000369
370 trace_xfs_log_reserve(log, internal_ticket);
371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 xlog_grant_push_ail(mp,
373 (internal_ticket->t_unit_res *
374 internal_ticket->t_cnt));
375 retval = xlog_grant_log_space(log, internal_ticket);
376 }
377
378 return retval;
379} /* xfs_log_reserve */
380
381
382/*
383 * Mount a log filesystem
384 *
385 * mp - ubiquitous xfs mount point structure
386 * log_target - buftarg of on-disk log device
387 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
388 * num_bblocks - Number of BBSIZE blocks in on-disk log
389 *
390 * Return error or zero.
391 */
392int
David Chinner249a8c12008-02-05 12:13:32 +1100393xfs_log_mount(
394 xfs_mount_t *mp,
395 xfs_buftarg_t *log_target,
396 xfs_daddr_t blk_offset,
397 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
David Chinner249a8c12008-02-05 12:13:32 +1100399 int error;
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
402 cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
403 else {
404 cmn_err(CE_NOTE,
405 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
406 mp->m_fsname);
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000407 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
409
410 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200411 if (IS_ERR(mp->m_log)) {
412 error = -PTR_ERR(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100413 goto out;
414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /*
David Chinner249a8c12008-02-05 12:13:32 +1100417 * Initialize the AIL now we have a log.
418 */
David Chinner249a8c12008-02-05 12:13:32 +1100419 error = xfs_trans_ail_init(mp);
420 if (error) {
421 cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100422 goto out_free_log;
David Chinner249a8c12008-02-05 12:13:32 +1100423 }
David Chinnera9c21c12008-10-30 17:39:35 +1100424 mp->m_log->l_ailp = mp->m_ail;
David Chinner249a8c12008-02-05 12:13:32 +1100425
426 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 * skip log recovery on a norecovery mount. pretend it all
428 * just worked.
429 */
430 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
David Chinner249a8c12008-02-05 12:13:32 +1100431 int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000434 mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Eric Sandeen65be6052006-01-11 15:34:19 +1100436 error = xlog_recover(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000439 mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (error) {
441 cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100442 goto out_destroy_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
444 }
445
446 /* Normal transactions can now occur */
447 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
448
Dave Chinner71e330b2010-05-21 14:37:18 +1000449 /*
450 * Now the log has been fully initialised and we know were our
451 * space grant counters are, we can initialise the permanent ticket
452 * needed for delayed logging to work.
453 */
454 xlog_cil_init_post_recovery(mp->m_log);
455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 return 0;
Christoph Hellwig26430752009-02-12 19:55:48 +0100457
458out_destroy_ail:
459 xfs_trans_ail_destroy(mp);
460out_free_log:
461 xlog_dealloc_log(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100462out:
David Chinner249a8c12008-02-05 12:13:32 +1100463 return error;
Christoph Hellwig26430752009-02-12 19:55:48 +0100464}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466/*
467 * Finish the recovery of the file system. This is separate from
468 * the xfs_log_mount() call, because it depends on the code in
469 * xfs_mountfs() to read in the root and real-time bitmap inodes
470 * between calling xfs_log_mount() and here.
471 *
472 * mp - ubiquitous xfs mount point structure
473 */
474int
Christoph Hellwig42490232008-08-13 16:49:32 +1000475xfs_log_mount_finish(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 int error;
478
479 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
Christoph Hellwig42490232008-08-13 16:49:32 +1000480 error = xlog_recover_finish(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 else {
482 error = 0;
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000483 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485
486 return error;
487}
488
489/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 * Final log writes as part of unmount.
491 *
492 * Mark the filesystem clean as unmount happens. Note that during relocation
493 * this routine needs to be executed as part of source-bag while the
494 * deallocation must not be done until source-end.
495 */
496
497/*
498 * Unmount record used to have a string "Unmount filesystem--" in the
499 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
500 * We just write the magic number now since that particular field isn't
501 * currently architecture converted and "nUmount" is a bit foo.
502 * As far as I know, there weren't any dependencies on the old behaviour.
503 */
504
505int
506xfs_log_unmount_write(xfs_mount_t *mp)
507{
508 xlog_t *log = mp->m_log;
509 xlog_in_core_t *iclog;
510#ifdef DEBUG
511 xlog_in_core_t *first_iclog;
512#endif
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000513 xlog_ticket_t *tic = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 xfs_lsn_t lsn;
515 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 /*
518 * Don't write out unmount record on read-only mounts.
519 * Or, if we are doing a forced umount (typically because of IO errors).
520 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000521 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 return 0;
523
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000524 error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
David Chinnerb911ca02008-04-10 12:24:30 +1000525 ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527#ifdef DEBUG
528 first_iclog = iclog = log->l_iclog;
529 do {
530 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
531 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
532 ASSERT(iclog->ic_offset == 0);
533 }
534 iclog = iclog->ic_next;
535 } while (iclog != first_iclog);
536#endif
537 if (! (XLOG_FORCED_SHUTDOWN(log))) {
Tim Shimmin955e47a2006-09-28 11:04:16 +1000538 error = xfs_log_reserve(mp, 600, 1, &tic,
539 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (!error) {
Dave Chinner55b66332010-03-23 11:43:17 +1100541 /* the data section must be 32 bit size aligned */
542 struct {
543 __uint16_t magic;
544 __uint16_t pad1;
545 __uint32_t pad2; /* may as well make it 64 bits */
546 } magic = {
547 .magic = XLOG_UNMOUNT_TYPE,
548 };
549 struct xfs_log_iovec reg = {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000550 .i_addr = &magic,
Dave Chinner55b66332010-03-23 11:43:17 +1100551 .i_len = sizeof(magic),
552 .i_type = XLOG_REG_TYPE_UNMOUNT,
553 };
554 struct xfs_log_vec vec = {
555 .lv_niovecs = 1,
556 .lv_iovecp = &reg,
557 };
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 /* remove inited flag */
Dave Chinner55b66332010-03-23 11:43:17 +1100560 tic->t_flags = 0;
561 error = xlog_write(log, &vec, tic, &lsn,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 NULL, XLOG_UNMOUNT_TRANS);
563 /*
564 * At this point, we're umounting anyway,
565 * so there's no point in transitioning log state
566 * to IOERROR. Just continue...
567 */
568 }
569
570 if (error) {
571 xfs_fs_cmn_err(CE_ALERT, mp,
572 "xfs_log_unmount: unmount record failed");
573 }
574
575
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000576 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100578 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100580 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000581 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000583 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
585 iclog->ic_state == XLOG_STATE_DIRTY)) {
586 if (!XLOG_FORCED_SHUTDOWN(log)) {
David Chinner12017fa2008-08-13 16:34:31 +1000587 sv_wait(&iclog->ic_force_wait, PMEM,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 &log->l_icloglock, s);
589 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000590 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 }
592 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000593 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Tim Shimmin955e47a2006-09-28 11:04:16 +1000595 if (tic) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000596 trace_xfs_log_umount_write(log, tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000597 xlog_ungrant_log_space(log, tic);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100598 xfs_log_ticket_put(tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 } else {
601 /*
602 * We're already in forced_shutdown mode, couldn't
603 * even attempt to write out the unmount transaction.
604 *
605 * Go through the motions of sync'ing and releasing
606 * the iclog, even though no I/O will actually happen,
Nathan Scottc41564b2006-03-29 08:55:14 +1000607 * we need to wait for other log I/Os that may already
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 * be in progress. Do this as a separate section of
609 * code so we'll know if we ever get stuck here that
610 * we're in this odd situation of trying to unmount
611 * a file system that went into forced_shutdown as
612 * the result of an unmount..
613 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000614 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100616 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100619 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000620 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000622 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
625 || iclog->ic_state == XLOG_STATE_DIRTY
626 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
627
David Chinner12017fa2008-08-13 16:34:31 +1000628 sv_wait(&iclog->ic_force_wait, PMEM,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 &log->l_icloglock, s);
630 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000631 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 }
633 }
634
David Chinner1bb7d6b2008-04-10 12:24:38 +1000635 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636} /* xfs_log_unmount_write */
637
638/*
639 * Deallocate log structures for unmount/relocation.
David Chinner249a8c12008-02-05 12:13:32 +1100640 *
641 * We need to stop the aild from running before we destroy
642 * and deallocate the log as the aild references the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 */
644void
Christoph Hellwig21b699c2009-03-16 08:19:29 +0100645xfs_log_unmount(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
David Chinner249a8c12008-02-05 12:13:32 +1100647 xfs_trans_ail_destroy(mp);
Nathan Scottc41564b2006-03-29 08:55:14 +1000648 xlog_dealloc_log(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Dave Chinner43f5efc2010-03-23 10:10:00 +1100651void
652xfs_log_item_init(
653 struct xfs_mount *mp,
654 struct xfs_log_item *item,
655 int type,
656 struct xfs_item_ops *ops)
657{
658 item->li_mountp = mp;
659 item->li_ailp = mp->m_ail;
660 item->li_type = type;
661 item->li_ops = ops;
Dave Chinner71e330b2010-05-21 14:37:18 +1000662 item->li_lv = NULL;
663
664 INIT_LIST_HEAD(&item->li_ail);
665 INIT_LIST_HEAD(&item->li_cil);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100666}
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668/*
669 * Write region vectors to log. The write happens using the space reservation
670 * of the ticket (tic). It is not a requirement that all writes for a given
Dave Chinner9b9fc2b72010-03-23 11:21:11 +1100671 * transaction occur with one call to xfs_log_write(). However, it is important
672 * to note that the transaction reservation code makes an assumption about the
673 * number of log headers a transaction requires that may be violated if you
674 * don't pass all the transaction vectors in one call....
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 */
676int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000677xfs_log_write(
678 struct xfs_mount *mp,
679 struct xfs_log_iovec reg[],
680 int nentries,
681 struct xlog_ticket *tic,
682 xfs_lsn_t *start_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000684 struct log *log = mp->m_log;
685 int error;
Dave Chinner55b66332010-03-23 11:43:17 +1100686 struct xfs_log_vec vec = {
687 .lv_niovecs = nentries,
688 .lv_iovecp = reg,
689 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 if (XLOG_FORCED_SHUTDOWN(log))
692 return XFS_ERROR(EIO);
693
Dave Chinner55b66332010-03-23 11:43:17 +1100694 error = xlog_write(log, &vec, tic, start_lsn, NULL, 0);
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000695 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +1000696 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +0100697 return error;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000698}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700void
701xfs_log_move_tail(xfs_mount_t *mp,
702 xfs_lsn_t tail_lsn)
703{
704 xlog_ticket_t *tic;
705 xlog_t *log = mp->m_log;
706 int need_bytes, free_bytes, cycle, bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (XLOG_FORCED_SHUTDOWN(log))
709 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
711 if (tail_lsn == 0) {
712 /* needed since sync_lsn is 64 bits */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000713 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 tail_lsn = log->l_last_sync_lsn;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000715 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000718 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
721 * tail_lsn.
722 */
723 if (tail_lsn != 1) {
724 log->l_tail_lsn = tail_lsn;
725 }
726
727 if ((tic = log->l_write_headq)) {
728#ifdef DEBUG
729 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
730 panic("Recovery problem");
731#endif
732 cycle = log->l_grant_write_cycle;
733 bytes = log->l_grant_write_bytes;
734 free_bytes = xlog_space_left(log, cycle, bytes);
735 do {
736 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
737
738 if (free_bytes < tic->t_unit_res && tail_lsn != 1)
739 break;
740 tail_lsn = 0;
741 free_bytes -= tic->t_unit_res;
David Chinner12017fa2008-08-13 16:34:31 +1000742 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 tic = tic->t_next;
744 } while (tic != log->l_write_headq);
745 }
746 if ((tic = log->l_reserve_headq)) {
747#ifdef DEBUG
748 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
749 panic("Recovery problem");
750#endif
751 cycle = log->l_grant_reserve_cycle;
752 bytes = log->l_grant_reserve_bytes;
753 free_bytes = xlog_space_left(log, cycle, bytes);
754 do {
755 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
756 need_bytes = tic->t_unit_res*tic->t_cnt;
757 else
758 need_bytes = tic->t_unit_res;
759 if (free_bytes < need_bytes && tail_lsn != 1)
760 break;
761 tail_lsn = 0;
762 free_bytes -= need_bytes;
David Chinner12017fa2008-08-13 16:34:31 +1000763 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 tic = tic->t_next;
765 } while (tic != log->l_reserve_headq);
766 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000767 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768} /* xfs_log_move_tail */
769
770/*
771 * Determine if we have a transaction that has gone to disk
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000772 * that needs to be covered. To begin the transition to the idle state
773 * firstly the log needs to be idle (no AIL and nothing in the iclogs).
774 * If we are then in a state where covering is needed, the caller is informed
775 * that dummy transactions are required to move the log into the idle state.
776 *
777 * Because this is called as part of the sync process, we should also indicate
778 * that dummy transactions should be issued in anything but the covered or
779 * idle states. This ensures that the log tail is accurately reflected in
780 * the log at the end of the sync, hence if a crash occurrs avoids replay
781 * of transactions where the metadata is already on disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
783int
784xfs_log_need_covered(xfs_mount_t *mp)
785{
David Chinner27d8d5f2008-10-30 17:38:39 +1100786 int needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 xlog_t *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
David Chinner92821e22007-05-24 15:26:31 +1000789 if (!xfs_fs_writable(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return 0;
791
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000792 spin_lock(&log->l_icloglock);
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000793 switch (log->l_covered_state) {
794 case XLOG_STATE_COVER_DONE:
795 case XLOG_STATE_COVER_DONE2:
796 case XLOG_STATE_COVER_IDLE:
797 break;
798 case XLOG_STATE_COVER_NEED:
799 case XLOG_STATE_COVER_NEED2:
800 if (!xfs_trans_ail_tail(log->l_ailp) &&
801 xlog_iclogs_empty(log)) {
802 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
803 log->l_covered_state = XLOG_STATE_COVER_DONE;
804 else
805 log->l_covered_state = XLOG_STATE_COVER_DONE2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 }
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000807 /* FALLTHRU */
808 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 needed = 1;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000810 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000812 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +0100813 return needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
816/******************************************************************************
817 *
818 * local routines
819 *
820 ******************************************************************************
821 */
822
823/* xfs_trans_tail_ail returns 0 when there is nothing in the list.
824 * The log manager must keep track of the last LR which was committed
825 * to disk. The lsn of this LR will become the new tail_lsn whenever
826 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
827 * the situation where stuff could be written into the log but nothing
828 * was ever in the AIL when asked. Eventually, we panic since the
829 * tail hits the head.
830 *
831 * We may be holding the log iclog lock upon entering this routine.
832 */
833xfs_lsn_t
834xlog_assign_tail_lsn(xfs_mount_t *mp)
835{
836 xfs_lsn_t tail_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 xlog_t *log = mp->m_log;
838
David Chinner5b00f142008-10-30 17:39:00 +1100839 tail_lsn = xfs_trans_ail_tail(mp->m_ail);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000840 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 if (tail_lsn != 0) {
842 log->l_tail_lsn = tail_lsn;
843 } else {
844 tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn;
845 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000846 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 return tail_lsn;
849} /* xlog_assign_tail_lsn */
850
851
852/*
853 * Return the space in the log between the tail and the head. The head
854 * is passed in the cycle/bytes formal parms. In the special case where
855 * the reserve head has wrapped passed the tail, this calculation is no
856 * longer valid. In this case, just return 0 which means there is no space
857 * in the log. This works for all places where this function is called
858 * with the reserve head. Of course, if the write head were to ever
859 * wrap the tail, we should blow up. Rather than catch this case here,
860 * we depend on other ASSERTions in other parts of the code. XXXmiken
861 *
862 * This code also handles the case where the reservation head is behind
863 * the tail. The details of this case are described below, but the end
864 * result is that we return the size of the log as the amount of space left.
865 */
David Chinnera8272ce2007-11-23 16:28:09 +1100866STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867xlog_space_left(xlog_t *log, int cycle, int bytes)
868{
869 int free_bytes;
870 int tail_bytes;
871 int tail_cycle;
872
873 tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn));
874 tail_cycle = CYCLE_LSN(log->l_tail_lsn);
875 if ((tail_cycle == cycle) && (bytes >= tail_bytes)) {
876 free_bytes = log->l_logsize - (bytes - tail_bytes);
877 } else if ((tail_cycle + 1) < cycle) {
878 return 0;
879 } else if (tail_cycle < cycle) {
880 ASSERT(tail_cycle == (cycle - 1));
881 free_bytes = tail_bytes - bytes;
882 } else {
883 /*
884 * The reservation head is behind the tail.
885 * In this case we just want to return the size of the
886 * log as the amount of space left.
887 */
888 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
889 "xlog_space_left: head behind tail\n"
890 " tail_cycle = %d, tail_bytes = %d\n"
891 " GH cycle = %d, GH bytes = %d",
892 tail_cycle, tail_bytes, cycle, bytes);
893 ASSERT(0);
894 free_bytes = log->l_logsize;
895 }
896 return free_bytes;
897} /* xlog_space_left */
898
899
900/*
901 * Log function which is called when an io completes.
902 *
903 * The log manager needs its own routine, in order to control what
904 * happens with the buffer after the write completes.
905 */
906void
907xlog_iodone(xfs_buf_t *bp)
908{
909 xlog_in_core_t *iclog;
910 xlog_t *l;
911 int aborted;
912
913 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
914 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
915 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
916 aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 l = iclog->ic_log;
918
919 /*
Christoph Hellwig73f6aa42008-10-10 17:28:29 +1100920 * If the _XFS_BARRIER_FAILED flag was set by a lower
921 * layer, it means the underlying device no longer supports
David Chinner0bfefc42007-05-14 18:24:23 +1000922 * barrier I/O. Warn loudly and turn off barriers.
923 */
Christoph Hellwig73f6aa42008-10-10 17:28:29 +1100924 if (bp->b_flags & _XFS_BARRIER_FAILED) {
925 bp->b_flags &= ~_XFS_BARRIER_FAILED;
David Chinner0bfefc42007-05-14 18:24:23 +1000926 l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
927 xfs_fs_cmn_err(CE_WARN, l->l_mp,
928 "xlog_iodone: Barriers are no longer supported"
929 " by device. Disabling barriers\n");
David Chinner0bfefc42007-05-14 18:24:23 +1000930 }
931
932 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 * Race to shutdown the filesystem if we see an error.
934 */
935 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
936 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
937 xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
938 XFS_BUF_STALE(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +1000939 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 /*
941 * This flag will be propagated to the trans-committed
942 * callback routines to let them know that the log-commit
943 * didn't succeed.
944 */
945 aborted = XFS_LI_ABORTED;
946 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
947 aborted = XFS_LI_ABORTED;
948 }
David Chinner3db296f2007-05-14 18:24:16 +1000949
950 /* log I/O is always issued ASYNC */
951 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 xlog_state_done_syncing(iclog, aborted);
David Chinner3db296f2007-05-14 18:24:16 +1000953 /*
954 * do not reference the buffer (bp) here as we could race
955 * with it being freed after writing the unmount record to the
956 * log.
957 */
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959} /* xlog_iodone */
960
961/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * Return size of each in-core log record buffer.
963 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200964 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 *
966 * If the filesystem blocksize is too large, we may need to choose a
967 * larger size since the directory code currently logs entire blocks.
968 */
969
970STATIC void
971xlog_get_iclog_buffer_size(xfs_mount_t *mp,
972 xlog_t *log)
973{
974 int size;
975 int xhdrs;
976
Eric Sandeen1cb51252007-08-16 16:24:43 +1000977 if (mp->m_logbufs <= 0)
978 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
979 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100980 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982 /*
983 * Buffer size passed in from mount system call.
984 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100985 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 size = log->l_iclog_size = mp->m_logbsize;
987 log->l_iclog_size_log = 0;
988 while (size != 1) {
989 log->l_iclog_size_log++;
990 size >>= 1;
991 }
992
Eric Sandeen62118702008-03-06 13:44:28 +1100993 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200994 /* # headers = size / 32k
995 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 */
997
998 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
999 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1000 xhdrs++;
1001 log->l_iclog_hsize = xhdrs << BBSHIFT;
1002 log->l_iclog_heads = xhdrs;
1003 } else {
1004 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1005 log->l_iclog_hsize = BBSIZE;
1006 log->l_iclog_heads = 1;
1007 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001008 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001011 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +10001012 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1013 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
1015 /* the default log size is 16k or 32k which is one header sector */
1016 log->l_iclog_hsize = BBSIZE;
1017 log->l_iclog_heads = 1;
1018
Christoph Hellwig7153f8b2009-02-09 08:36:46 +01001019done:
1020 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001021 if (mp->m_logbufs == 0)
1022 mp->m_logbufs = log->l_iclog_bufs;
1023 if (mp->m_logbsize == 0)
1024 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025} /* xlog_get_iclog_buffer_size */
1026
1027
1028/*
1029 * This routine initializes some of the log structure for a given mount point.
1030 * Its primary purpose is to fill in enough, so recovery can occur. However,
1031 * some other stuff may be filled in too.
1032 */
1033STATIC xlog_t *
1034xlog_alloc_log(xfs_mount_t *mp,
1035 xfs_buftarg_t *log_target,
1036 xfs_daddr_t blk_offset,
1037 int num_bblks)
1038{
1039 xlog_t *log;
1040 xlog_rec_header_t *head;
1041 xlog_in_core_t **iclogp;
1042 xlog_in_core_t *iclog, *prev_iclog=NULL;
1043 xfs_buf_t *bp;
1044 int i;
Dave Chinnera6cb7672009-04-06 18:39:27 +02001045 int error = ENOMEM;
Alex Elder69ce58f2010-04-20 17:09:59 +10001046 uint log2_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Dave Chinner644c3562008-11-10 16:50:24 +11001048 log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001049 if (!log) {
1050 xlog_warn("XFS: Log allocation failed: No memory!");
1051 goto out;
1052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
1054 log->l_mp = mp;
1055 log->l_targ = log_target;
1056 log->l_logsize = BBTOB(num_bblks);
1057 log->l_logBBstart = blk_offset;
1058 log->l_logBBsize = num_bblks;
1059 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1060 log->l_flags |= XLOG_ACTIVE_RECOVERY;
1061
1062 log->l_prev_block = -1;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001063 log->l_tail_lsn = xlog_assign_lsn(1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1065 log->l_last_sync_lsn = log->l_tail_lsn;
1066 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
1067 log->l_grant_reserve_cycle = 1;
1068 log->l_grant_write_cycle = 1;
1069
Dave Chinnera6cb7672009-04-06 18:39:27 +02001070 error = EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001071 if (xfs_sb_version_hassector(&mp->m_sb)) {
Alex Elder69ce58f2010-04-20 17:09:59 +10001072 log2_size = mp->m_sb.sb_logsectlog;
1073 if (log2_size < BBSHIFT) {
1074 xlog_warn("XFS: Log sector size too small "
1075 "(0x%x < 0x%x)", log2_size, BBSHIFT);
1076 goto out_free_log;
1077 }
1078
1079 log2_size -= BBSHIFT;
1080 if (log2_size > mp->m_sectbb_log) {
1081 xlog_warn("XFS: Log sector size too large "
1082 "(0x%x > 0x%x)", log2_size, mp->m_sectbb_log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001083 goto out_free_log;
1084 }
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 /* for larger sector sizes, must have v2 or external log */
Alex Elder69ce58f2010-04-20 17:09:59 +10001087 if (log2_size && log->l_logBBstart > 0 &&
1088 !xfs_sb_version_haslogv2(&mp->m_sb)) {
1089
Dave Chinnera6cb7672009-04-06 18:39:27 +02001090 xlog_warn("XFS: log sector size (0x%x) invalid "
Alex Elder69ce58f2010-04-20 17:09:59 +10001091 "for configuration.", log2_size);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001092 goto out_free_log;
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
Alex Elder69ce58f2010-04-20 17:09:59 +10001095 log->l_sectBBsize = 1 << log2_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
1097 xlog_get_iclog_buffer_size(mp, log);
1098
Dave Chinnera6cb7672009-04-06 18:39:27 +02001099 error = ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001101 if (!bp)
1102 goto out_free_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1105 ASSERT(XFS_BUF_ISBUSY(bp));
1106 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
1107 log->l_xbuf = bp;
1108
Eric Sandeen007c61c2007-10-11 17:43:56 +10001109 spin_lock_init(&log->l_icloglock);
1110 spin_lock_init(&log->l_grant_lock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10001111 sv_init(&log->l_flush_wait, 0, "flush_wait");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1114 ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
1115
1116 iclogp = &log->l_iclog;
1117 /*
1118 * The amount of memory to allocate for the iclog structure is
1119 * rather funky due to the way the structure is defined. It is
1120 * done this way so that we can use different sizes for machines
1121 * with different amounts of memory. See the definition of
1122 * xlog_in_core_t in xfs_log_priv.h for details.
1123 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 ASSERT(log->l_iclog_size >= 4096);
1125 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001126 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1127 if (!*iclogp)
1128 goto out_free_iclog;
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 iclog->ic_prev = prev_iclog;
1132 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001133
1134 bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001135 if (!bp)
1136 goto out_free_iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001137 if (!XFS_BUF_CPSEMA(bp))
1138 ASSERT(0);
1139 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001140 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1141 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001142 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001143#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
David Chinner4679b2d2008-04-10 12:18:54 +10001145#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 head = &iclog->ic_header;
1147 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001148 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1149 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001150 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001151 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001153 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
1157 iclog->ic_state = XLOG_STATE_ACTIVE;
1158 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001159 atomic_set(&iclog->ic_refcnt, 0);
1160 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001162 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164 ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
1165 ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
David Chinner12017fa2008-08-13 16:34:31 +10001166 sv_init(&iclog->ic_force_wait, SV_DEFAULT, "iclog-force");
1167 sv_init(&iclog->ic_write_wait, SV_DEFAULT, "iclog-write");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 iclogp = &iclog->ic_next;
1170 }
1171 *iclogp = log->l_iclog; /* complete ring */
1172 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1173
Dave Chinner71e330b2010-05-21 14:37:18 +10001174 error = xlog_cil_init(log);
1175 if (error)
1176 goto out_free_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001178
1179out_free_iclog:
1180 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1181 prev_iclog = iclog->ic_next;
1182 if (iclog->ic_bp) {
1183 sv_destroy(&iclog->ic_force_wait);
1184 sv_destroy(&iclog->ic_write_wait);
1185 xfs_buf_free(iclog->ic_bp);
Dave Chinner644c3562008-11-10 16:50:24 +11001186 }
1187 kmem_free(iclog);
1188 }
1189 spinlock_destroy(&log->l_icloglock);
1190 spinlock_destroy(&log->l_grant_lock);
Dave Chinner644c3562008-11-10 16:50:24 +11001191 xfs_buf_free(log->l_xbuf);
1192out_free_log:
1193 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001194out:
1195 return ERR_PTR(-error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196} /* xlog_alloc_log */
1197
1198
1199/*
1200 * Write out the commit record of a transaction associated with the given
1201 * ticket. Return the lsn of the commit record.
1202 */
1203STATIC int
Dave Chinner55b66332010-03-23 11:43:17 +11001204xlog_commit_record(
1205 struct log *log,
1206 struct xlog_ticket *ticket,
1207 struct xlog_in_core **iclog,
1208 xfs_lsn_t *commitlsnp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
Dave Chinner55b66332010-03-23 11:43:17 +11001210 struct xfs_mount *mp = log->l_mp;
1211 int error;
1212 struct xfs_log_iovec reg = {
1213 .i_addr = NULL,
1214 .i_len = 0,
1215 .i_type = XLOG_REG_TYPE_COMMIT,
1216 };
1217 struct xfs_log_vec vec = {
1218 .lv_niovecs = 1,
1219 .lv_iovecp = &reg,
1220 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 ASSERT_ALWAYS(iclog);
Dave Chinner55b66332010-03-23 11:43:17 +11001223 error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1224 XLOG_COMMIT_TRANS);
1225 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +10001226 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +01001227 return error;
Dave Chinner55b66332010-03-23 11:43:17 +11001228}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230/*
1231 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1232 * log space. This code pushes on the lsn which would supposedly free up
1233 * the 25% which we want to leave free. We may need to adopt a policy which
1234 * pushes on an lsn which is further along in the log once we reach the high
1235 * water mark. In this manner, we would be creating a low water mark.
1236 */
David Chinnera8272ce2007-11-23 16:28:09 +11001237STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238xlog_grant_push_ail(xfs_mount_t *mp,
1239 int need_bytes)
1240{
1241 xlog_t *log = mp->m_log; /* pointer to the log */
1242 xfs_lsn_t tail_lsn; /* lsn of the log tail */
1243 xfs_lsn_t threshold_lsn = 0; /* lsn we'd like to be at */
1244 int free_blocks; /* free blocks left to write to */
1245 int free_bytes; /* free bytes left to write to */
1246 int threshold_block; /* block in lsn we'd like to be at */
1247 int threshold_cycle; /* lsn cycle we'd like to be at */
1248 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
1251
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001252 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 free_bytes = xlog_space_left(log,
1254 log->l_grant_reserve_cycle,
1255 log->l_grant_reserve_bytes);
1256 tail_lsn = log->l_tail_lsn;
1257 free_blocks = BTOBBT(free_bytes);
1258
1259 /*
1260 * Set the threshold for the minimum number of free blocks in the
1261 * log to the maximum of what the caller needs, one quarter of the
1262 * log, and 256 blocks.
1263 */
1264 free_threshold = BTOBB(need_bytes);
1265 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1266 free_threshold = MAX(free_threshold, 256);
1267 if (free_blocks < free_threshold) {
1268 threshold_block = BLOCK_LSN(tail_lsn) + free_threshold;
1269 threshold_cycle = CYCLE_LSN(tail_lsn);
1270 if (threshold_block >= log->l_logBBsize) {
1271 threshold_block -= log->l_logBBsize;
1272 threshold_cycle += 1;
1273 }
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001274 threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276 /* Don't pass in an lsn greater than the lsn of the last
1277 * log record known to be on disk.
1278 */
1279 if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0)
1280 threshold_lsn = log->l_last_sync_lsn;
1281 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001282 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284 /*
1285 * Get the transaction layer to kick the dirty buffers out to
1286 * disk asynchronously. No point in trying to do this if
1287 * the filesystem is shutting down.
1288 */
1289 if (threshold_lsn &&
1290 !XLOG_FORCED_SHUTDOWN(log))
David Chinner783a2f62008-10-30 17:39:58 +11001291 xfs_trans_ail_push(log->l_ailp, threshold_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292} /* xlog_grant_push_ail */
1293
Christoph Hellwig873ff552010-01-13 22:17:57 +00001294/*
1295 * The bdstrat callback function for log bufs. This gives us a central
1296 * place to trap bufs in case we get hit by a log I/O error and need to
1297 * shutdown. Actually, in practice, even when we didn't get a log error,
1298 * we transition the iclogs to IOERROR state *after* flushing all existing
1299 * iclogs to disk. This is because we don't want anymore new transactions to be
1300 * started or completed afterwards.
1301 */
1302STATIC int
1303xlog_bdstrat(
1304 struct xfs_buf *bp)
1305{
1306 struct xlog_in_core *iclog;
1307
1308 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1309 if (iclog->ic_state & XLOG_STATE_IOERROR) {
1310 XFS_BUF_ERROR(bp, EIO);
1311 XFS_BUF_STALE(bp);
1312 xfs_biodone(bp);
1313 /*
1314 * It would seem logical to return EIO here, but we rely on
1315 * the log state machine to propagate I/O errors instead of
1316 * doing it here.
1317 */
1318 return 0;
1319 }
1320
1321 bp->b_flags |= _XBF_RUN_QUEUES;
1322 xfs_buf_iorequest(bp);
1323 return 0;
1324}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
1326/*
1327 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1328 * fashion. Previously, we should have moved the current iclog
1329 * ptr in the log to point to the next available iclog. This allows further
1330 * write to continue while this code syncs out an iclog ready to go.
1331 * Before an in-core log can be written out, the data section must be scanned
1332 * to save away the 1st word of each BBSIZE block into the header. We replace
1333 * it with the current cycle count. Each BBSIZE block is tagged with the
1334 * cycle count because there in an implicit assumption that drives will
1335 * guarantee that entire 512 byte blocks get written at once. In other words,
1336 * we can't have part of a 512 byte block written and part not written. By
1337 * tagging each block, we will know which blocks are valid when recovering
1338 * after an unclean shutdown.
1339 *
1340 * This routine is single threaded on the iclog. No other thread can be in
1341 * this routine with the same iclog. Changing contents of iclog can there-
1342 * fore be done without grabbing the state machine lock. Updating the global
1343 * log will require grabbing the lock though.
1344 *
1345 * The entire log manager uses a logical block numbering scheme. Only
1346 * log_sync (and then only bwrite()) know about the fact that the log may
1347 * not start with block zero on a given device. The log block start offset
1348 * is added immediately before calling bwrite().
1349 */
1350
David Chinnera8272ce2007-11-23 16:28:09 +11001351STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352xlog_sync(xlog_t *log,
1353 xlog_in_core_t *iclog)
1354{
1355 xfs_caddr_t dptr; /* pointer to byte sized element */
1356 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001357 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 uint count; /* byte count of bwrite */
1359 uint count_init; /* initial count before roundup */
1360 int roundoff; /* roundoff to BB or stripe */
1361 int split = 0; /* split write into two regions */
1362 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001363 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001366 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
1368 /* Add for LR header */
1369 count_init = log->l_iclog_hsize + iclog->ic_offset;
1370
1371 /* Round out the log write size */
1372 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1373 /* we have a v2 stripe unit to use */
1374 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1375 } else {
1376 count = BBTOB(BTOBB(count_init));
1377 }
1378 roundoff = count - count_init;
1379 ASSERT(roundoff >= 0);
1380 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1381 roundoff < log->l_mp->m_sb.sb_logsunit)
1382 ||
1383 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1384 roundoff < BBTOB(1)));
1385
1386 /* move grant heads by roundoff in sync */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001387 spin_lock(&log->l_grant_lock);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11001388 xlog_grant_add_space(log, roundoff);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001389 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391 /* put cycle number in every block */
1392 xlog_pack_data(log, iclog, roundoff);
1393
1394 /* real byte length */
1395 if (v2) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001396 iclog->ic_header.h_len =
1397 cpu_to_be32(iclog->ic_offset + roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001399 iclog->ic_header.h_len =
1400 cpu_to_be32(iclog->ic_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402
Nathan Scottf5faad72006-07-28 17:04:44 +10001403 bp = iclog->ic_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
1405 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001406 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
1408 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1409
1410 /* Do we need to split this write into 2 parts? */
1411 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1412 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1413 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1414 iclog->ic_bwritecnt = 2; /* split into 2 writes */
1415 } else {
1416 iclog->ic_bwritecnt = 1;
1417 }
David Chinner511105b2007-05-24 15:21:57 +10001418 XFS_BUF_SET_COUNT(bp, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */
Nathan Scottf5faad72006-07-28 17:04:44 +10001420 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 XFS_BUF_BUSY(bp);
1422 XFS_BUF_ASYNC(bp);
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001423 bp->b_flags |= XBF_LOG_BUFFER;
Christoph Hellwig651701d2010-06-28 10:34:34 -04001424
1425 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001426 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1429 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1430
1431 xlog_verify_iclog(log, iclog, count, B_TRUE);
1432
1433 /* account for log which doesn't start at block #0 */
1434 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1435 /*
1436 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1437 * is shutting down.
1438 */
1439 XFS_BUF_WRITE(bp);
1440
Christoph Hellwig873ff552010-01-13 22:17:57 +00001441 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
1443 XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001444 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
1446 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001447 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
1449 (unsigned long)1);
1450 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1451 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
1452 XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
1453 (__psint_t)count), split);
1454 XFS_BUF_SET_FSPRIVATE(bp, iclog);
Nathan Scottf5faad72006-07-28 17:04:44 +10001455 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 XFS_BUF_BUSY(bp);
1457 XFS_BUF_ASYNC(bp);
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001458 bp->b_flags |= XBF_LOG_BUFFER;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001459 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1460 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 dptr = XFS_BUF_PTR(bp);
1462 /*
1463 * Bump the cycle numbers at the start of each block
1464 * since this part of the buffer is at the start of
1465 * a new cycle. Watch out for the header magic number
1466 * case, though.
1467 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001468 for (i = 0; i < split; i += BBSIZE) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001469 be32_add_cpu((__be32 *)dptr, 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001470 if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001471 be32_add_cpu((__be32 *)dptr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 dptr += BBSIZE;
1473 }
1474
1475 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1476 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1477
Nathan Scottc41564b2006-03-29 08:55:14 +10001478 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1480 XFS_BUF_WRITE(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001481 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
1483 bp, XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001484 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
1486 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001487 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488} /* xlog_sync */
1489
1490
1491/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001492 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 */
David Chinnera8272ce2007-11-23 16:28:09 +11001494STATIC void
Nathan Scottc41564b2006-03-29 08:55:14 +10001495xlog_dealloc_log(xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
1497 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 int i;
1499
Dave Chinner71e330b2010-05-21 14:37:18 +10001500 xlog_cil_destroy(log);
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 iclog = log->l_iclog;
1503 for (i=0; i<log->l_iclog_bufs; i++) {
David Chinner12017fa2008-08-13 16:34:31 +10001504 sv_destroy(&iclog->ic_force_wait);
1505 sv_destroy(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 xfs_buf_free(iclog->ic_bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001508 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 iclog = next_iclog;
1510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 spinlock_destroy(&log->l_icloglock);
1512 spinlock_destroy(&log->l_grant_lock);
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 xfs_buf_free(log->l_xbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001516 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001517} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
1519/*
1520 * Update counters atomically now that memcpy is done.
1521 */
1522/* ARGSUSED */
1523static inline void
1524xlog_state_finish_copy(xlog_t *log,
1525 xlog_in_core_t *iclog,
1526 int record_cnt,
1527 int copy_bytes)
1528{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001529 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001531 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 iclog->ic_offset += copy_bytes;
1533
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001534 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535} /* xlog_state_finish_copy */
1536
1537
1538
1539
1540/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001541 * print out info relating to regions written which consume
1542 * the reservation
1543 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001544void
1545xlog_print_tic_res(
1546 struct xfs_mount *mp,
1547 struct xlog_ticket *ticket)
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001548{
1549 uint i;
1550 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1551
1552 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1553 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1554 "bformat",
1555 "bchunk",
1556 "efi_format",
1557 "efd_format",
1558 "iformat",
1559 "icore",
1560 "iext",
1561 "ibroot",
1562 "ilocal",
1563 "iattr_ext",
1564 "iattr_broot",
1565 "iattr_local",
1566 "qformat",
1567 "dquot",
1568 "quotaoff",
1569 "LR header",
1570 "unmount",
1571 "commit",
1572 "trans header"
1573 };
1574 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1575 "SETATTR_NOT_SIZE",
1576 "SETATTR_SIZE",
1577 "INACTIVE",
1578 "CREATE",
1579 "CREATE_TRUNC",
1580 "TRUNCATE_FILE",
1581 "REMOVE",
1582 "LINK",
1583 "RENAME",
1584 "MKDIR",
1585 "RMDIR",
1586 "SYMLINK",
1587 "SET_DMATTRS",
1588 "GROWFS",
1589 "STRAT_WRITE",
1590 "DIOSTRAT",
1591 "WRITE_SYNC",
1592 "WRITEID",
1593 "ADDAFORK",
1594 "ATTRINVAL",
1595 "ATRUNCATE",
1596 "ATTR_SET",
1597 "ATTR_RM",
1598 "ATTR_FLAG",
1599 "CLEAR_AGI_BUCKET",
1600 "QM_SBCHANGE",
1601 "DUMMY1",
1602 "DUMMY2",
1603 "QM_QUOTAOFF",
1604 "QM_DQALLOC",
1605 "QM_SETQLIM",
1606 "QM_DQCLUSTER",
1607 "QM_QINOCREATE",
1608 "QM_QUOTAOFF_END",
1609 "SB_UNIT",
1610 "FSYNC_TS",
1611 "GROWFSRT_ALLOC",
1612 "GROWFSRT_ZERO",
1613 "GROWFSRT_FREE",
1614 "SWAPEXT"
1615 };
1616
1617 xfs_fs_cmn_err(CE_WARN, mp,
1618 "xfs_log_write: reservation summary:\n"
1619 " trans type = %s (%u)\n"
1620 " unit res = %d bytes\n"
1621 " current res = %d bytes\n"
1622 " total reg = %u bytes (o/flow = %u bytes)\n"
1623 " ophdrs = %u (ophdr space = %u bytes)\n"
1624 " ophdr + reg = %u bytes\n"
1625 " num regions = %u\n",
1626 ((ticket->t_trans_type <= 0 ||
1627 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1628 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1629 ticket->t_trans_type,
1630 ticket->t_unit_res,
1631 ticket->t_curr_res,
1632 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1633 ticket->t_res_num_ophdrs, ophdr_spc,
1634 ticket->t_res_arr_sum +
Tim Shimmin12598452006-01-11 21:02:47 +11001635 ticket->t_res_o_flow + ophdr_spc,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001636 ticket->t_res_num);
1637
1638 for (i = 0; i < ticket->t_res_num; i++) {
Tim Shimmin12598452006-01-11 21:02:47 +11001639 uint r_type = ticket->t_res_arr[i].r_type;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001640 cmn_err(CE_WARN,
1641 "region[%u]: %s - %u bytes\n",
1642 i,
1643 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1644 "bad-rtype" : res_type_str[r_type-1]),
1645 ticket->t_res_arr[i].r_len);
1646 }
Dave Chinner169a7b02010-05-07 11:05:31 +10001647
1648 xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
1649 "xfs_log_write: reservation ran out. Need to up reservation");
1650 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001651}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001652
1653/*
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001654 * Calculate the potential space needed by the log vector. Each region gets
1655 * its own xlog_op_header_t and may need to be double word aligned.
1656 */
1657static int
1658xlog_write_calc_vec_length(
1659 struct xlog_ticket *ticket,
Dave Chinner55b66332010-03-23 11:43:17 +11001660 struct xfs_log_vec *log_vector)
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001661{
Dave Chinner55b66332010-03-23 11:43:17 +11001662 struct xfs_log_vec *lv;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001663 int headers = 0;
1664 int len = 0;
1665 int i;
1666
1667 /* acct for start rec of xact */
1668 if (ticket->t_flags & XLOG_TIC_INITED)
1669 headers++;
1670
Dave Chinner55b66332010-03-23 11:43:17 +11001671 for (lv = log_vector; lv; lv = lv->lv_next) {
1672 headers += lv->lv_niovecs;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001673
Dave Chinner55b66332010-03-23 11:43:17 +11001674 for (i = 0; i < lv->lv_niovecs; i++) {
1675 struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
1676
1677 len += vecp->i_len;
1678 xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
1679 }
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001680 }
1681
1682 ticket->t_res_num_ophdrs += headers;
1683 len += headers * sizeof(struct xlog_op_header);
1684
1685 return len;
1686}
1687
1688/*
1689 * If first write for transaction, insert start record We can't be trying to
1690 * commit if we are inited. We can't have any "partial_copy" if we are inited.
1691 */
1692static int
1693xlog_write_start_rec(
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001694 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001695 struct xlog_ticket *ticket)
1696{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001697 if (!(ticket->t_flags & XLOG_TIC_INITED))
1698 return 0;
1699
1700 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1701 ophdr->oh_clientid = ticket->t_clientid;
1702 ophdr->oh_len = 0;
1703 ophdr->oh_flags = XLOG_START_TRANS;
1704 ophdr->oh_res2 = 0;
1705
1706 ticket->t_flags &= ~XLOG_TIC_INITED;
1707
1708 return sizeof(struct xlog_op_header);
1709}
1710
1711static xlog_op_header_t *
1712xlog_write_setup_ophdr(
1713 struct log *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001714 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001715 struct xlog_ticket *ticket,
1716 uint flags)
1717{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001718 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1719 ophdr->oh_clientid = ticket->t_clientid;
1720 ophdr->oh_res2 = 0;
1721
1722 /* are we copying a commit or unmount record? */
1723 ophdr->oh_flags = flags;
1724
1725 /*
1726 * We've seen logs corrupted with bad transaction client ids. This
1727 * makes sure that XFS doesn't generate them on. Turn this into an EIO
1728 * and shut down the filesystem.
1729 */
1730 switch (ophdr->oh_clientid) {
1731 case XFS_TRANSACTION:
1732 case XFS_VOLUME:
1733 case XFS_LOG:
1734 break;
1735 default:
1736 xfs_fs_cmn_err(CE_WARN, log->l_mp,
1737 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1738 ophdr->oh_clientid, ticket);
1739 return NULL;
1740 }
1741
1742 return ophdr;
1743}
1744
1745/*
1746 * Set up the parameters of the region copy into the log. This has
1747 * to handle region write split across multiple log buffers - this
1748 * state is kept external to this function so that this code can
1749 * can be written in an obvious, self documenting manner.
1750 */
1751static int
1752xlog_write_setup_copy(
1753 struct xlog_ticket *ticket,
1754 struct xlog_op_header *ophdr,
1755 int space_available,
1756 int space_required,
1757 int *copy_off,
1758 int *copy_len,
1759 int *last_was_partial_copy,
1760 int *bytes_consumed)
1761{
1762 int still_to_copy;
1763
1764 still_to_copy = space_required - *bytes_consumed;
1765 *copy_off = *bytes_consumed;
1766
1767 if (still_to_copy <= space_available) {
1768 /* write of region completes here */
1769 *copy_len = still_to_copy;
1770 ophdr->oh_len = cpu_to_be32(*copy_len);
1771 if (*last_was_partial_copy)
1772 ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
1773 *last_was_partial_copy = 0;
1774 *bytes_consumed = 0;
1775 return 0;
1776 }
1777
1778 /* partial write of region, needs extra log op header reservation */
1779 *copy_len = space_available;
1780 ophdr->oh_len = cpu_to_be32(*copy_len);
1781 ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
1782 if (*last_was_partial_copy)
1783 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
1784 *bytes_consumed += *copy_len;
1785 (*last_was_partial_copy)++;
1786
1787 /* account for new log op header */
1788 ticket->t_curr_res -= sizeof(struct xlog_op_header);
1789 ticket->t_res_num_ophdrs++;
1790
1791 return sizeof(struct xlog_op_header);
1792}
1793
1794static int
1795xlog_write_copy_finish(
1796 struct log *log,
1797 struct xlog_in_core *iclog,
1798 uint flags,
1799 int *record_cnt,
1800 int *data_cnt,
1801 int *partial_copy,
1802 int *partial_copy_len,
1803 int log_offset,
1804 struct xlog_in_core **commit_iclog)
1805{
1806 if (*partial_copy) {
1807 /*
1808 * This iclog has already been marked WANT_SYNC by
1809 * xlog_state_get_iclog_space.
1810 */
1811 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1812 *record_cnt = 0;
1813 *data_cnt = 0;
1814 return xlog_state_release_iclog(log, iclog);
1815 }
1816
1817 *partial_copy = 0;
1818 *partial_copy_len = 0;
1819
1820 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
1821 /* no more space in this iclog - push it. */
1822 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1823 *record_cnt = 0;
1824 *data_cnt = 0;
1825
1826 spin_lock(&log->l_icloglock);
1827 xlog_state_want_sync(log, iclog);
1828 spin_unlock(&log->l_icloglock);
1829
1830 if (!commit_iclog)
1831 return xlog_state_release_iclog(log, iclog);
1832 ASSERT(flags & XLOG_COMMIT_TRANS);
1833 *commit_iclog = iclog;
1834 }
1835
1836 return 0;
1837}
1838
1839/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 * Write some region out to in-core log
1841 *
1842 * This will be called when writing externally provided regions or when
1843 * writing out a commit record for a given transaction.
1844 *
1845 * General algorithm:
1846 * 1. Find total length of this write. This may include adding to the
1847 * lengths passed in.
1848 * 2. Check whether we violate the tickets reservation.
1849 * 3. While writing to this iclog
1850 * A. Reserve as much space in this iclog as can get
1851 * B. If this is first write, save away start lsn
1852 * C. While writing this region:
1853 * 1. If first write of transaction, write start record
1854 * 2. Write log operation header (header per region)
1855 * 3. Find out if we can fit entire region into this iclog
1856 * 4. Potentially, verify destination memcpy ptr
1857 * 5. Memcpy (partial) region
1858 * 6. If partial copy, release iclog; otherwise, continue
1859 * copying more regions into current iclog
1860 * 4. Mark want sync bit (in simulation mode)
1861 * 5. Release iclog for potential flush to on-disk log.
1862 *
1863 * ERRORS:
1864 * 1. Panic if reservation is overrun. This should never happen since
1865 * reservation amounts are generated internal to the filesystem.
1866 * NOTES:
1867 * 1. Tickets are single threaded data structures.
1868 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1869 * syncing routine. When a single log_write region needs to span
1870 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1871 * on all log operation writes which don't contain the end of the
1872 * region. The XLOG_END_TRANS bit is used for the in-core log
1873 * operation which contains the end of the continued log_write region.
1874 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1875 * we don't really know exactly how much space will be used. As a result,
1876 * we don't update ic_offset until the end when we know exactly how many
1877 * bytes have been written out.
1878 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001879int
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001880xlog_write(
Dave Chinner55b66332010-03-23 11:43:17 +11001881 struct log *log,
1882 struct xfs_log_vec *log_vector,
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001883 struct xlog_ticket *ticket,
1884 xfs_lsn_t *start_lsn,
1885 struct xlog_in_core **commit_iclog,
1886 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001888 struct xlog_in_core *iclog = NULL;
Dave Chinner55b66332010-03-23 11:43:17 +11001889 struct xfs_log_iovec *vecp;
1890 struct xfs_log_vec *lv;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001891 int len;
1892 int index;
1893 int partial_copy = 0;
1894 int partial_copy_len = 0;
1895 int contwr = 0;
1896 int record_cnt = 0;
1897 int data_cnt = 0;
1898 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001900 *start_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Dave Chinner55b66332010-03-23 11:43:17 +11001902 len = xlog_write_calc_vec_length(ticket, log_vector);
Dave Chinner71e330b2010-05-21 14:37:18 +10001903 if (log->l_cilp) {
1904 /*
1905 * Region headers and bytes are already accounted for.
1906 * We only need to take into account start records and
1907 * split regions in this function.
1908 */
1909 if (ticket->t_flags & XLOG_TIC_INITED)
1910 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1911
1912 /*
1913 * Commit record headers need to be accounted for. These
1914 * come in as separate writes so are easy to detect.
1915 */
1916 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
1917 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1918 } else
1919 ticket->t_curr_res -= len;
1920
1921 if (ticket->t_curr_res < 0)
Dave Chinner55b66332010-03-23 11:43:17 +11001922 xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Dave Chinner55b66332010-03-23 11:43:17 +11001924 index = 0;
1925 lv = log_vector;
1926 vecp = lv->lv_iovecp;
1927 while (lv && index < lv->lv_niovecs) {
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001928 void *ptr;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001929 int log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001931 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
1932 &contwr, &log_offset);
1933 if (error)
1934 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001936 ASSERT(log_offset <= iclog->ic_size - 1);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001937 ptr = iclog->ic_datap + log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001939 /* start_lsn is the first lsn written to. That's all we need. */
1940 if (!*start_lsn)
1941 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001943 /*
1944 * This loop writes out as many regions as can fit in the amount
1945 * of space which was allocated by xlog_state_get_iclog_space().
1946 */
Dave Chinner55b66332010-03-23 11:43:17 +11001947 while (lv && index < lv->lv_niovecs) {
1948 struct xfs_log_iovec *reg = &vecp[index];
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001949 struct xlog_op_header *ophdr;
1950 int start_rec_copy;
1951 int copy_len;
1952 int copy_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Dave Chinner55b66332010-03-23 11:43:17 +11001954 ASSERT(reg->i_len % sizeof(__int32_t) == 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001955 ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001957 start_rec_copy = xlog_write_start_rec(ptr, ticket);
1958 if (start_rec_copy) {
1959 record_cnt++;
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001960 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001961 start_rec_copy);
1962 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001964 ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
1965 if (!ophdr)
1966 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001968 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001969 sizeof(struct xlog_op_header));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001971 len += xlog_write_setup_copy(ticket, ophdr,
1972 iclog->ic_size-log_offset,
Dave Chinner55b66332010-03-23 11:43:17 +11001973 reg->i_len,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001974 &copy_off, &copy_len,
1975 &partial_copy,
1976 &partial_copy_len);
1977 xlog_verify_dest_ptr(log, ptr);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001978
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001979 /* copy region */
1980 ASSERT(copy_len >= 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001981 memcpy(ptr, reg->i_addr + copy_off, copy_len);
1982 xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001983
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001984 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
1985 record_cnt++;
1986 data_cnt += contwr ? copy_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001988 error = xlog_write_copy_finish(log, iclog, flags,
1989 &record_cnt, &data_cnt,
1990 &partial_copy,
1991 &partial_copy_len,
1992 log_offset,
1993 commit_iclog);
1994 if (error)
1995 return error;
1996
1997 /*
1998 * if we had a partial copy, we need to get more iclog
1999 * space but we don't want to increment the region
2000 * index because there is still more is this region to
2001 * write.
2002 *
2003 * If we completed writing this region, and we flushed
2004 * the iclog (indicated by resetting of the record
2005 * count), then we also need to get more log space. If
2006 * this was the last record, though, we are done and
2007 * can just return.
2008 */
2009 if (partial_copy)
2010 break;
2011
Dave Chinner55b66332010-03-23 11:43:17 +11002012 if (++index == lv->lv_niovecs) {
2013 lv = lv->lv_next;
2014 index = 0;
2015 if (lv)
2016 vecp = lv->lv_iovecp;
2017 }
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002018 if (record_cnt == 0) {
Dave Chinner55b66332010-03-23 11:43:17 +11002019 if (!lv)
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002020 return 0;
2021 break;
2022 }
2023 }
2024 }
2025
2026 ASSERT(len == 0);
2027
2028 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
2029 if (!commit_iclog)
2030 return xlog_state_release_iclog(log, iclog);
2031
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 ASSERT(flags & XLOG_COMMIT_TRANS);
2033 *commit_iclog = iclog;
2034 return 0;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11002035}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
2037
2038/*****************************************************************************
2039 *
2040 * State Machine functions
2041 *
2042 *****************************************************************************
2043 */
2044
2045/* Clean iclogs starting from the head. This ordering must be
2046 * maintained, so an iclog doesn't become ACTIVE beyond one that
2047 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10002048 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 * iclog is trying to sync to disk.
2050 *
2051 * State Change: DIRTY -> ACTIVE
2052 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002053STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054xlog_state_clean_log(xlog_t *log)
2055{
2056 xlog_in_core_t *iclog;
2057 int changed = 0;
2058
2059 iclog = log->l_iclog;
2060 do {
2061 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2062 iclog->ic_state = XLOG_STATE_ACTIVE;
2063 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002064 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 /*
2066 * If the number of ops in this iclog indicate it just
2067 * contains the dummy transaction, we can
2068 * change state into IDLE (the second time around).
2069 * Otherwise we should change the state into
2070 * NEED a dummy.
2071 * We don't need to cover the dummy.
2072 */
2073 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002074 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2075 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 changed = 1;
2077 } else {
2078 /*
2079 * We have two dirty iclogs so start over
2080 * This could also be num of ops indicates
2081 * this is not the dummy going out.
2082 */
2083 changed = 2;
2084 }
2085 iclog->ic_header.h_num_logops = 0;
2086 memset(iclog->ic_header.h_cycle_data, 0,
2087 sizeof(iclog->ic_header.h_cycle_data));
2088 iclog->ic_header.h_lsn = 0;
2089 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2090 /* do nothing */;
2091 else
2092 break; /* stop cleaning */
2093 iclog = iclog->ic_next;
2094 } while (iclog != log->l_iclog);
2095
2096 /* log is locked when we are called */
2097 /*
2098 * Change state for the dummy log recording.
2099 * We usually go to NEED. But we go to NEED2 if the changed indicates
2100 * we are done writing the dummy record.
2101 * If we are done with the second dummy recored (DONE2), then
2102 * we go to IDLE.
2103 */
2104 if (changed) {
2105 switch (log->l_covered_state) {
2106 case XLOG_STATE_COVER_IDLE:
2107 case XLOG_STATE_COVER_NEED:
2108 case XLOG_STATE_COVER_NEED2:
2109 log->l_covered_state = XLOG_STATE_COVER_NEED;
2110 break;
2111
2112 case XLOG_STATE_COVER_DONE:
2113 if (changed == 1)
2114 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2115 else
2116 log->l_covered_state = XLOG_STATE_COVER_NEED;
2117 break;
2118
2119 case XLOG_STATE_COVER_DONE2:
2120 if (changed == 1)
2121 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2122 else
2123 log->l_covered_state = XLOG_STATE_COVER_NEED;
2124 break;
2125
2126 default:
2127 ASSERT(0);
2128 }
2129 }
2130} /* xlog_state_clean_log */
2131
2132STATIC xfs_lsn_t
2133xlog_get_lowest_lsn(
2134 xlog_t *log)
2135{
2136 xlog_in_core_t *lsn_log;
2137 xfs_lsn_t lowest_lsn, lsn;
2138
2139 lsn_log = log->l_iclog;
2140 lowest_lsn = 0;
2141 do {
2142 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002143 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if ((lsn && !lowest_lsn) ||
2145 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2146 lowest_lsn = lsn;
2147 }
2148 }
2149 lsn_log = lsn_log->ic_next;
2150 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002151 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152}
2153
2154
2155STATIC void
2156xlog_state_do_callback(
2157 xlog_t *log,
2158 int aborted,
2159 xlog_in_core_t *ciclog)
2160{
2161 xlog_in_core_t *iclog;
2162 xlog_in_core_t *first_iclog; /* used to know when we've
2163 * processed all iclogs once */
2164 xfs_log_callback_t *cb, *cb_next;
2165 int flushcnt = 0;
2166 xfs_lsn_t lowest_lsn;
2167 int ioerrors; /* counter: iclogs with errors */
2168 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2169 int funcdidcallbacks; /* flag: function did callbacks */
2170 int repeats; /* for issuing console warnings if
2171 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002172 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002174 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 first_iclog = iclog = log->l_iclog;
2176 ioerrors = 0;
2177 funcdidcallbacks = 0;
2178 repeats = 0;
2179
2180 do {
2181 /*
2182 * Scan all iclogs starting with the one pointed to by the
2183 * log. Reset this starting point each time the log is
2184 * unlocked (during callbacks).
2185 *
2186 * Keep looping through iclogs until one full pass is made
2187 * without running any callbacks.
2188 */
2189 first_iclog = log->l_iclog;
2190 iclog = log->l_iclog;
2191 loopdidcallbacks = 0;
2192 repeats++;
2193
2194 do {
2195
2196 /* skip all iclogs in the ACTIVE & DIRTY states */
2197 if (iclog->ic_state &
2198 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2199 iclog = iclog->ic_next;
2200 continue;
2201 }
2202
2203 /*
2204 * Between marking a filesystem SHUTDOWN and stopping
2205 * the log, we do flush all iclogs to disk (if there
2206 * wasn't a log I/O error). So, we do want things to
2207 * go smoothly in case of just a SHUTDOWN w/o a
2208 * LOG_IO_ERROR.
2209 */
2210 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2211 /*
2212 * Can only perform callbacks in order. Since
2213 * this iclog is not in the DONE_SYNC/
2214 * DO_CALLBACK state, we skip the rest and
2215 * just try to clean up. If we set our iclog
2216 * to DO_CALLBACK, we will not process it when
2217 * we retry since a previous iclog is in the
2218 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002219 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 */
2221 if (!(iclog->ic_state &
2222 (XLOG_STATE_DONE_SYNC |
2223 XLOG_STATE_DO_CALLBACK))) {
2224 if (ciclog && (ciclog->ic_state ==
2225 XLOG_STATE_DONE_SYNC)) {
2226 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2227 }
2228 break;
2229 }
2230 /*
2231 * We now have an iclog that is in either the
2232 * DO_CALLBACK or DONE_SYNC states. The other
2233 * states (WANT_SYNC, SYNCING, or CALLBACK were
2234 * caught by the above if and are going to
2235 * clean (i.e. we aren't doing their callbacks)
2236 * see the above if.
2237 */
2238
2239 /*
2240 * We will do one more check here to see if we
2241 * have chased our tail around.
2242 */
2243
2244 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002245 if (lowest_lsn &&
2246 XFS_LSN_CMP(lowest_lsn,
2247 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 iclog = iclog->ic_next;
2249 continue; /* Leave this iclog for
2250 * another thread */
2251 }
2252
2253 iclog->ic_state = XLOG_STATE_CALLBACK;
2254
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002255 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
2257 /* l_last_sync_lsn field protected by
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002258 * l_grant_lock. Don't worry about iclog's lsn.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 * No one else can be here except us.
2260 */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002261 spin_lock(&log->l_grant_lock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002262 ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn,
2263 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2264 log->l_last_sync_lsn =
2265 be64_to_cpu(iclog->ic_header.h_lsn);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002266 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 } else {
David Chinner114d23a2008-04-10 12:18:39 +10002269 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 ioerrors++;
2271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
David Chinner114d23a2008-04-10 12:18:39 +10002273 /*
2274 * Keep processing entries in the callback list until
2275 * we come around and it is empty. We need to
2276 * atomically see that the list is empty and change the
2277 * state to DIRTY so that we don't miss any more
2278 * callbacks being added.
2279 */
2280 spin_lock(&iclog->ic_callback_lock);
2281 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002282 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283 iclog->ic_callback_tail = &(iclog->ic_callback);
2284 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002285 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002288 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 cb_next = cb->cb_next;
2290 cb->cb_func(cb->cb_arg, aborted);
2291 }
David Chinner114d23a2008-04-10 12:18:39 +10002292 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 cb = iclog->ic_callback;
2294 }
2295
2296 loopdidcallbacks++;
2297 funcdidcallbacks++;
2298
David Chinner114d23a2008-04-10 12:18:39 +10002299 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002300 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002301 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2303 iclog->ic_state = XLOG_STATE_DIRTY;
2304
2305 /*
2306 * Transition from DIRTY to ACTIVE if applicable.
2307 * NOP if STATE_IOERROR.
2308 */
2309 xlog_state_clean_log(log);
2310
2311 /* wake up threads waiting in xfs_log_force() */
David Chinner12017fa2008-08-13 16:34:31 +10002312 sv_broadcast(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
2314 iclog = iclog->ic_next;
2315 } while (first_iclog != iclog);
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002316
2317 if (repeats > 5000) {
2318 flushcnt += repeats;
2319 repeats = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 xfs_fs_cmn_err(CE_WARN, log->l_mp,
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002321 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002322 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 }
2324 } while (!ioerrors && loopdidcallbacks);
2325
2326 /*
2327 * make one last gasp attempt to see if iclogs are being left in
2328 * limbo..
2329 */
2330#ifdef DEBUG
2331 if (funcdidcallbacks) {
2332 first_iclog = iclog = log->l_iclog;
2333 do {
2334 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2335 /*
2336 * Terminate the loop if iclogs are found in states
2337 * which will cause other threads to clean up iclogs.
2338 *
2339 * SYNCING - i/o completion will go through logs
2340 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002341 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 * IOERROR - give up hope all ye who enter here
2343 */
2344 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2345 iclog->ic_state == XLOG_STATE_SYNCING ||
2346 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2347 iclog->ic_state == XLOG_STATE_IOERROR )
2348 break;
2349 iclog = iclog->ic_next;
2350 } while (first_iclog != iclog);
2351 }
2352#endif
2353
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002354 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2355 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002356 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002357
2358 if (wake)
2359 sv_broadcast(&log->l_flush_wait);
2360}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
2362
2363/*
2364 * Finish transitioning this iclog to the dirty state.
2365 *
2366 * Make sure that we completely execute this routine only when this is
2367 * the last call to the iclog. There is a good chance that iclog flushes,
2368 * when we reach the end of the physical log, get turned into 2 separate
2369 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2370 * routine. By using the reference count bwritecnt, we guarantee that only
2371 * the second completion goes through.
2372 *
2373 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002374 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 */
David Chinnera8272ce2007-11-23 16:28:09 +11002376STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377xlog_state_done_syncing(
2378 xlog_in_core_t *iclog,
2379 int aborted)
2380{
2381 xlog_t *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002383 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2386 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002387 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2389
2390
2391 /*
2392 * If we got an error, either on the first buffer, or in the case of
2393 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2394 * and none should ever be attempted to be written to disk
2395 * again.
2396 */
2397 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2398 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002399 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 return;
2401 }
2402 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2403 }
2404
2405 /*
2406 * Someone could be sleeping prior to writing out the next
2407 * iclog buffer, we wake them all, one will get to do the
2408 * I/O, the others get to wait for the result.
2409 */
David Chinner12017fa2008-08-13 16:34:31 +10002410 sv_broadcast(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002411 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2413} /* xlog_state_done_syncing */
2414
2415
2416/*
2417 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002418 * sleep. We wait on the flush queue on the head iclog as that should be
2419 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2420 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 *
2422 * The in-core logs are used in a circular fashion. They are not used
2423 * out-of-order even when an iclog past the head is free.
2424 *
2425 * return:
2426 * * log_offset where xlog_write() can start writing into the in-core
2427 * log's data space.
2428 * * in-core log pointer to which xlog_write() should write.
2429 * * boolean indicating this is a continued write to an in-core log.
2430 * If this is the last write, then the in-core log's offset field
2431 * needs to be incremented, depending on the amount of data which
2432 * is copied.
2433 */
David Chinnera8272ce2007-11-23 16:28:09 +11002434STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435xlog_state_get_iclog_space(xlog_t *log,
2436 int len,
2437 xlog_in_core_t **iclogp,
2438 xlog_ticket_t *ticket,
2439 int *continued_write,
2440 int *logoffsetp)
2441{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 int log_offset;
2443 xlog_rec_header_t *head;
2444 xlog_in_core_t *iclog;
2445 int error;
2446
2447restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002448 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002450 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 return XFS_ERROR(EIO);
2452 }
2453
2454 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002455 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002457
2458 /* Wait for log writes to have flushed */
2459 sv_wait(&log->l_flush_wait, 0, &log->l_icloglock, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 goto restart;
2461 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002462
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 head = &iclog->ic_header;
2464
David Chinner155cc6b2008-03-06 13:44:14 +11002465 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 log_offset = iclog->ic_offset;
2467
2468 /* On the 1st write to an iclog, figure out lsn. This works
2469 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2470 * committing to. If the offset is set, that's how many blocks
2471 * must be written.
2472 */
2473 if (log_offset == 0) {
2474 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002475 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002476 log->l_iclog_hsize,
2477 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002478 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2479 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002480 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 ASSERT(log->l_curr_block >= 0);
2482 }
2483
2484 /* If there is enough room to write everything, then do it. Otherwise,
2485 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2486 * bit is on, so this will get flushed out. Don't update ic_offset
2487 * until you know exactly how many bytes get copied. Therefore, wait
2488 * until later to update ic_offset.
2489 *
2490 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2491 * can fit into remaining data section.
2492 */
2493 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2494 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2495
Dave Chinner49641f12008-07-11 17:43:55 +10002496 /*
2497 * If I'm the only one writing to this iclog, sync it to disk.
2498 * We need to do an atomic compare and decrement here to avoid
2499 * racing with concurrent atomic_dec_and_lock() calls in
2500 * xlog_state_release_iclog() when there is more than one
2501 * reference to the iclog.
2502 */
2503 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2504 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002505 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002506 error = xlog_state_release_iclog(log, iclog);
2507 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002508 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002510 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 }
2512 goto restart;
2513 }
2514
2515 /* Do we have enough room to write the full amount in the remainder
2516 * of this iclog? Or must we continue a write on the next iclog and
2517 * mark this iclog as completely taken? In the case where we switch
2518 * iclogs (to mark it taken), this particular iclog will release/sync
2519 * to disk in xlog_write().
2520 */
2521 if (len <= iclog->ic_size - iclog->ic_offset) {
2522 *continued_write = 0;
2523 iclog->ic_offset += len;
2524 } else {
2525 *continued_write = 1;
2526 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2527 }
2528 *iclogp = iclog;
2529
2530 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002531 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
2533 *logoffsetp = log_offset;
2534 return 0;
2535} /* xlog_state_get_iclog_space */
2536
2537/*
2538 * Atomically get the log space required for a log ticket.
2539 *
2540 * Once a ticket gets put onto the reserveq, it will only return after
2541 * the needed reservation is satisfied.
2542 */
2543STATIC int
2544xlog_grant_log_space(xlog_t *log,
2545 xlog_ticket_t *tic)
2546{
2547 int free_bytes;
2548 int need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549#ifdef DEBUG
2550 xfs_lsn_t tail_lsn;
2551#endif
2552
2553
2554#ifdef DEBUG
2555 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2556 panic("grant Recovery problem");
2557#endif
2558
2559 /* Is there space or do we need to sleep? */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002560 spin_lock(&log->l_grant_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002561
2562 trace_xfs_log_grant_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564 /* something is already sleeping; insert new transaction at end */
2565 if (log->l_reserve_headq) {
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002566 xlog_ins_ticketq(&log->l_reserve_headq, tic);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002567
2568 trace_xfs_log_grant_sleep1(log, tic);
2569
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 /*
2571 * Gotta check this before going to sleep, while we're
2572 * holding the grant lock.
2573 */
2574 if (XLOG_FORCED_SHUTDOWN(log))
2575 goto error_return;
2576
2577 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002578 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 /*
2580 * If we got an error, and the filesystem is shutting down,
2581 * we'll catch it down below. So just continue...
2582 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002583 trace_xfs_log_grant_wake1(log, tic);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002584 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 }
2586 if (tic->t_flags & XFS_LOG_PERM_RESERV)
2587 need_bytes = tic->t_unit_res*tic->t_ocnt;
2588 else
2589 need_bytes = tic->t_unit_res;
2590
2591redo:
2592 if (XLOG_FORCED_SHUTDOWN(log))
2593 goto error_return;
2594
2595 free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle,
2596 log->l_grant_reserve_bytes);
2597 if (free_bytes < need_bytes) {
2598 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002599 xlog_ins_ticketq(&log->l_reserve_headq, tic);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002600
2601 trace_xfs_log_grant_sleep2(log, tic);
2602
Dave Chinner9d7fef72009-04-06 18:42:59 +02002603 spin_unlock(&log->l_grant_lock);
2604 xlog_grant_push_ail(log->l_mp, need_bytes);
2605 spin_lock(&log->l_grant_lock);
2606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002608 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Dave Chinner9d7fef72009-04-06 18:42:59 +02002610 spin_lock(&log->l_grant_lock);
2611 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002614 trace_xfs_log_grant_wake2(log, tic);
2615
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 goto redo;
2617 } else if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002618 xlog_del_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
2620 /* we've got enough space */
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002621 xlog_grant_add_space(log, need_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622#ifdef DEBUG
2623 tail_lsn = log->l_tail_lsn;
2624 /*
2625 * Check to make sure the grant write head didn't just over lap the
2626 * tail. If the cycles are the same, we can't be overlapping.
2627 * Otherwise, make sure that the cycles differ by exactly one and
2628 * check the byte count.
2629 */
2630 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2631 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2632 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2633 }
2634#endif
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002635 trace_xfs_log_grant_exit(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002637 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 return 0;
2639
2640 error_return:
2641 if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002642 xlog_del_ticketq(&log->l_reserve_headq, tic);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002643
2644 trace_xfs_log_grant_error(log, tic);
2645
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 /*
2647 * If we are failing, make sure the ticket doesn't have any
2648 * current reservations. We don't want to add this back when
2649 * the ticket/transaction gets cancelled.
2650 */
2651 tic->t_curr_res = 0;
2652 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002653 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 return XFS_ERROR(EIO);
2655} /* xlog_grant_log_space */
2656
2657
2658/*
2659 * Replenish the byte reservation required by moving the grant write head.
2660 *
2661 *
2662 */
2663STATIC int
2664xlog_regrant_write_log_space(xlog_t *log,
2665 xlog_ticket_t *tic)
2666{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 int free_bytes, need_bytes;
2668 xlog_ticket_t *ntic;
2669#ifdef DEBUG
2670 xfs_lsn_t tail_lsn;
2671#endif
2672
2673 tic->t_curr_res = tic->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002674 xlog_tic_reset_res(tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
2676 if (tic->t_cnt > 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01002677 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678
2679#ifdef DEBUG
2680 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2681 panic("regrant Recovery problem");
2682#endif
2683
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002684 spin_lock(&log->l_grant_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002685
2686 trace_xfs_log_regrant_write_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
2688 if (XLOG_FORCED_SHUTDOWN(log))
2689 goto error_return;
2690
2691 /* If there are other waiters on the queue then give them a
2692 * chance at logspace before us. Wake up the first waiters,
2693 * if we do not wake up all the waiters then go to sleep waiting
2694 * for more free space, otherwise try to get some space for
2695 * this transaction.
2696 */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002697 need_bytes = tic->t_unit_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 if ((ntic = log->l_write_headq)) {
2699 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2700 log->l_grant_write_bytes);
2701 do {
2702 ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);
2703
2704 if (free_bytes < ntic->t_unit_res)
2705 break;
2706 free_bytes -= ntic->t_unit_res;
David Chinner12017fa2008-08-13 16:34:31 +10002707 sv_signal(&ntic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 ntic = ntic->t_next;
2709 } while (ntic != log->l_write_headq);
2710
2711 if (ntic != log->l_write_headq) {
2712 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002713 xlog_ins_ticketq(&log->l_write_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002715 trace_xfs_log_regrant_write_sleep1(log, tic);
2716
Dave Chinner9d7fef72009-04-06 18:42:59 +02002717 spin_unlock(&log->l_grant_lock);
2718 xlog_grant_push_ail(log->l_mp, need_bytes);
2719 spin_lock(&log->l_grant_lock);
2720
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002722 sv_wait(&tic->t_wait, PINOD|PLTWAIT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 &log->l_grant_lock, s);
2724
2725 /* If we're shutting down, this tic is already
2726 * off the queue */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002727 spin_lock(&log->l_grant_lock);
2728 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002731 trace_xfs_log_regrant_write_wake1(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
2733 }
2734
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735redo:
2736 if (XLOG_FORCED_SHUTDOWN(log))
2737 goto error_return;
2738
2739 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2740 log->l_grant_write_bytes);
2741 if (free_bytes < need_bytes) {
2742 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002743 xlog_ins_ticketq(&log->l_write_headq, tic);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002744 spin_unlock(&log->l_grant_lock);
2745 xlog_grant_push_ail(log->l_mp, need_bytes);
2746 spin_lock(&log->l_grant_lock);
2747
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 XFS_STATS_INC(xs_sleep_logspace);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002749 trace_xfs_log_regrant_write_sleep2(log, tic);
2750
David Chinner12017fa2008-08-13 16:34:31 +10002751 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752
2753 /* If we're shutting down, this tic is already off the queue */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002754 spin_lock(&log->l_grant_lock);
2755 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002758 trace_xfs_log_regrant_write_wake2(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 goto redo;
2760 } else if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002761 xlog_del_ticketq(&log->l_write_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002763 /* we've got enough space */
2764 xlog_grant_add_space_write(log, need_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765#ifdef DEBUG
2766 tail_lsn = log->l_tail_lsn;
2767 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2768 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2769 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2770 }
2771#endif
2772
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002773 trace_xfs_log_regrant_write_exit(log, tic);
2774
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002776 spin_unlock(&log->l_grant_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01002777 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778
2779
2780 error_return:
2781 if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002782 xlog_del_ticketq(&log->l_reserve_headq, tic);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002783
2784 trace_xfs_log_regrant_write_error(log, tic);
2785
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 /*
2787 * If we are failing, make sure the ticket doesn't have any
2788 * current reservations. We don't want to add this back when
2789 * the ticket/transaction gets cancelled.
2790 */
2791 tic->t_curr_res = 0;
2792 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002793 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 return XFS_ERROR(EIO);
2795} /* xlog_regrant_write_log_space */
2796
2797
2798/* The first cnt-1 times through here we don't need to
2799 * move the grant write head because the permanent
2800 * reservation has reserved cnt times the unit amount.
2801 * Release part of current permanent unit reservation and
2802 * reset current reservation to be one units worth. Also
2803 * move grant reservation head forward.
2804 */
2805STATIC void
2806xlog_regrant_reserve_log_space(xlog_t *log,
2807 xlog_ticket_t *ticket)
2808{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002809 trace_xfs_log_regrant_reserve_enter(log, ticket);
2810
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 if (ticket->t_cnt > 0)
2812 ticket->t_cnt--;
2813
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002814 spin_lock(&log->l_grant_lock);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002815 xlog_grant_sub_space(log, ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002817 xlog_tic_reset_res(ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002818
2819 trace_xfs_log_regrant_reserve_sub(log, ticket);
2820
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 xlog_verify_grant_head(log, 1);
2822
2823 /* just return if we still have some of the pre-reserved space */
2824 if (ticket->t_cnt > 0) {
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002825 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 return;
2827 }
2828
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002829 xlog_grant_add_space_reserve(log, ticket->t_unit_res);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002830
2831 trace_xfs_log_regrant_reserve_exit(log, ticket);
2832
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 xlog_verify_grant_head(log, 0);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002834 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002836 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837} /* xlog_regrant_reserve_log_space */
2838
2839
2840/*
2841 * Give back the space left from a reservation.
2842 *
2843 * All the information we need to make a correct determination of space left
2844 * is present. For non-permanent reservations, things are quite easy. The
2845 * count should have been decremented to zero. We only need to deal with the
2846 * space remaining in the current reservation part of the ticket. If the
2847 * ticket contains a permanent reservation, there may be left over space which
2848 * needs to be released. A count of N means that N-1 refills of the current
2849 * reservation can be done before we need to ask for more space. The first
2850 * one goes to fill up the first current reservation. Once we run out of
2851 * space, the count will stay at zero and the only space remaining will be
2852 * in the current reservation field.
2853 */
2854STATIC void
2855xlog_ungrant_log_space(xlog_t *log,
2856 xlog_ticket_t *ticket)
2857{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 if (ticket->t_cnt > 0)
2859 ticket->t_cnt--;
2860
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002861 spin_lock(&log->l_grant_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002862 trace_xfs_log_ungrant_enter(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002864 xlog_grant_sub_space(log, ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002866 trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
2868 /* If this is a permanent reservation ticket, we may be able to free
2869 * up more space based on the remaining count.
2870 */
2871 if (ticket->t_cnt > 0) {
2872 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002873 xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 }
2875
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002876 trace_xfs_log_ungrant_exit(log, ticket);
2877
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002879 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 xfs_log_move_tail(log->l_mp, 1);
2881} /* xlog_ungrant_log_space */
2882
2883
2884/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 * Flush iclog to disk if this is the last reference to the given iclog and
2886 * the WANT_SYNC bit is set.
2887 *
2888 * When this function is entered, the iclog is not necessarily in the
2889 * WANT_SYNC state. It may be sitting around waiting to get filled.
2890 *
2891 *
2892 */
David Chinnera8272ce2007-11-23 16:28:09 +11002893STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11002894xlog_state_release_iclog(
2895 xlog_t *log,
2896 xlog_in_core_t *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 int sync = 0; /* do we sync? */
2899
David Chinner155cc6b2008-03-06 13:44:14 +11002900 if (iclog->ic_state & XLOG_STATE_IOERROR)
2901 return XFS_ERROR(EIO);
2902
2903 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
2904 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
2905 return 0;
2906
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002908 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 return XFS_ERROR(EIO);
2910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
2912 iclog->ic_state == XLOG_STATE_WANT_SYNC);
2913
David Chinner155cc6b2008-03-06 13:44:14 +11002914 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11002915 /* update tail before writing to iclog */
2916 xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 sync++;
2918 iclog->ic_state = XLOG_STATE_SYNCING;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002919 iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn);
2921 /* cycle incremented when incrementing curr_block */
2922 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002923 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924
2925 /*
2926 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10002927 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2929 * this iclog has consistent data, so we ignore IOERROR
2930 * flags after this point.
2931 */
David Chinnerb5893342008-03-06 13:44:06 +11002932 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002934 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935} /* xlog_state_release_iclog */
2936
2937
2938/*
2939 * This routine will mark the current iclog in the ring as WANT_SYNC
2940 * and move the current iclog pointer to the next iclog in the ring.
2941 * When this routine is called from xlog_state_get_iclog_space(), the
2942 * exact size of the iclog has not yet been determined. All we know is
2943 * that every data block. We have run out of space in this log record.
2944 */
2945STATIC void
2946xlog_state_switch_iclogs(xlog_t *log,
2947 xlog_in_core_t *iclog,
2948 int eventual_size)
2949{
2950 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2951 if (!eventual_size)
2952 eventual_size = iclog->ic_offset;
2953 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002954 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 log->l_prev_block = log->l_curr_block;
2956 log->l_prev_cycle = log->l_curr_cycle;
2957
2958 /* roll log?: ic_offset changed later */
2959 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
2960
2961 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11002962 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 log->l_mp->m_sb.sb_logsunit > 1) {
2964 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
2965 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
2966 }
2967
2968 if (log->l_curr_block >= log->l_logBBsize) {
2969 log->l_curr_cycle++;
2970 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
2971 log->l_curr_cycle++;
2972 log->l_curr_block -= log->l_logBBsize;
2973 ASSERT(log->l_curr_block >= 0);
2974 }
2975 ASSERT(iclog == log->l_iclog);
2976 log->l_iclog = iclog->ic_next;
2977} /* xlog_state_switch_iclogs */
2978
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979/*
2980 * Write out all data in the in-core log as of this exact moment in time.
2981 *
2982 * Data may be written to the in-core log during this call. However,
2983 * we don't guarantee this data will be written out. A change from past
2984 * implementation means this routine will *not* write out zero length LRs.
2985 *
2986 * Basically, we try and perform an intelligent scan of the in-core logs.
2987 * If we determine there is no flushable data, we just return. There is no
2988 * flushable data if:
2989 *
2990 * 1. the current iclog is active and has no data; the previous iclog
2991 * is in the active or dirty state.
2992 * 2. the current iclog is drity, and the previous iclog is in the
2993 * active or dirty state.
2994 *
David Chinner12017fa2008-08-13 16:34:31 +10002995 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 *
2997 * 1. the current iclog is not in the active nor dirty state.
2998 * 2. the current iclog dirty, and the previous iclog is not in the
2999 * active nor dirty state.
3000 * 3. the current iclog is active, and there is another thread writing
3001 * to this particular iclog.
3002 * 4. a) the current iclog is active and has no other writers
3003 * b) when we return from flushing out this iclog, it is still
3004 * not in the active nor dirty state.
3005 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003006int
3007_xfs_log_force(
3008 struct xfs_mount *mp,
3009 uint flags,
3010 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003012 struct log *log = mp->m_log;
3013 struct xlog_in_core *iclog;
3014 xfs_lsn_t lsn;
3015
3016 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Dave Chinnera44f13e2010-08-24 11:40:03 +10003018 if (log->l_cilp)
3019 xlog_cil_force(log);
Dave Chinner71e330b2010-05-21 14:37:18 +10003020
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003021 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
3023 iclog = log->l_iclog;
3024 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003025 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 return XFS_ERROR(EIO);
3027 }
3028
3029 /* If the head iclog is not active nor dirty, we just attach
3030 * ourselves to the head and go to sleep.
3031 */
3032 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3033 iclog->ic_state == XLOG_STATE_DIRTY) {
3034 /*
3035 * If the head is dirty or (active and empty), then
3036 * we need to look at the previous iclog. If the previous
3037 * iclog is active or dirty we are done. There is nothing
3038 * to sync out. Otherwise, we attach ourselves to the
3039 * previous iclog and go to sleep.
3040 */
3041 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11003042 (atomic_read(&iclog->ic_refcnt) == 0
3043 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 iclog = iclog->ic_prev;
3045 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
3046 iclog->ic_state == XLOG_STATE_DIRTY)
3047 goto no_sleep;
3048 else
3049 goto maybe_sleep;
3050 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11003051 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052 /* We are the only one with access to this
3053 * iclog. Flush it out now. There should
3054 * be a roundoff of zero to show that someone
3055 * has already taken care of the roundoff from
3056 * the previous sync.
3057 */
David Chinner155cc6b2008-03-06 13:44:14 +11003058 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003059 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003061 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062
3063 if (xlog_state_release_iclog(log, iclog))
3064 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003065
3066 if (log_flushed)
3067 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003068 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003069 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 iclog->ic_state != XLOG_STATE_DIRTY)
3071 goto maybe_sleep;
3072 else
3073 goto no_sleep;
3074 } else {
3075 /* Someone else is writing to this iclog.
3076 * Use its call to flush out the data. However,
3077 * the other thread may not force out this LR,
3078 * so we mark it WANT_SYNC.
3079 */
3080 xlog_state_switch_iclogs(log, iclog, 0);
3081 goto maybe_sleep;
3082 }
3083 }
3084 }
3085
3086 /* By the time we come around again, the iclog could've been filled
3087 * which would give it another lsn. If we have a new lsn, just
3088 * return because the relevant data has been flushed.
3089 */
3090maybe_sleep:
3091 if (flags & XFS_LOG_SYNC) {
3092 /*
3093 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003094 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 * Then we check again after waking up, in case our
3096 * sleep was disturbed by a bad news.
3097 */
3098 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003099 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 return XFS_ERROR(EIO);
3101 }
3102 XFS_STATS_INC(xs_log_force_sleep);
David Chinner12017fa2008-08-13 16:34:31 +10003103 sv_wait(&iclog->ic_force_wait, PINOD, &log->l_icloglock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 /*
3105 * No need to grab the log lock here since we're
3106 * only deciding whether or not to return EIO
3107 * and the memory read should be atomic.
3108 */
3109 if (iclog->ic_state & XLOG_STATE_IOERROR)
3110 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003111 if (log_flushed)
3112 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 } else {
3114
3115no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003116 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 }
3118 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003119}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120
3121/*
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003122 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3123 * about errors or whether the log was flushed or not. This is the normal
3124 * interface to use when trying to unpin items or move the log forward.
3125 */
3126void
3127xfs_log_force(
3128 xfs_mount_t *mp,
3129 uint flags)
3130{
3131 int error;
3132
3133 error = _xfs_log_force(mp, flags, NULL);
3134 if (error) {
3135 xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
3136 "error %d returned.", error);
3137 }
3138}
3139
3140/*
3141 * Force the in-core log to disk for a specific LSN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 *
3143 * Find in-core log with lsn.
3144 * If it is in the DIRTY state, just return.
3145 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3146 * state and go to sleep or return.
3147 * If it is in any other state, go to sleep or return.
3148 *
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003149 * Synchronous forces are implemented with a signal variable. All callers
3150 * to force a given lsn to disk will wait on a the sv attached to the
3151 * specific in-core log. When given in-core log finally completes its
3152 * write to disk, that thread will wake up all threads waiting on the
3153 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003155int
3156_xfs_log_force_lsn(
3157 struct xfs_mount *mp,
3158 xfs_lsn_t lsn,
3159 uint flags,
3160 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003162 struct log *log = mp->m_log;
3163 struct xlog_in_core *iclog;
3164 int already_slept = 0;
3165
3166 ASSERT(lsn != 0);
3167
3168 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169
Dave Chinner71e330b2010-05-21 14:37:18 +10003170 if (log->l_cilp) {
Dave Chinnera44f13e2010-08-24 11:40:03 +10003171 lsn = xlog_cil_force_lsn(log, lsn);
Dave Chinner71e330b2010-05-21 14:37:18 +10003172 if (lsn == NULLCOMMITLSN)
3173 return 0;
3174 }
3175
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176try_again:
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003177 spin_lock(&log->l_icloglock);
3178 iclog = log->l_iclog;
3179 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003180 spin_unlock(&log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003181 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 }
3183
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003184 do {
3185 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3186 iclog = iclog->ic_next;
3187 continue;
3188 }
3189
3190 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3191 spin_unlock(&log->l_icloglock);
3192 return 0;
3193 }
3194
3195 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3196 /*
3197 * We sleep here if we haven't already slept (e.g.
3198 * this is the first time we've looked at the correct
3199 * iclog buf) and the buffer before us is going to
3200 * be sync'ed. The reason for this is that if we
3201 * are doing sync transactions here, by waiting for
3202 * the previous I/O to complete, we can allow a few
3203 * more transactions into this iclog before we close
3204 * it down.
3205 *
3206 * Otherwise, we mark the buffer WANT_SYNC, and bump
3207 * up the refcnt so we can release the log (which
3208 * drops the ref count). The state switch keeps new
3209 * transaction commits from using this buffer. When
3210 * the current commits finish writing into the buffer,
3211 * the refcount will drop to zero and the buffer will
3212 * go out then.
3213 */
3214 if (!already_slept &&
3215 (iclog->ic_prev->ic_state &
3216 (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3217 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3218
3219 XFS_STATS_INC(xs_log_force_sleep);
3220
3221 sv_wait(&iclog->ic_prev->ic_write_wait,
3222 PSWP, &log->l_icloglock, s);
3223 if (log_flushed)
3224 *log_flushed = 1;
3225 already_slept = 1;
3226 goto try_again;
3227 }
David Chinner155cc6b2008-03-06 13:44:14 +11003228 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003230 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 if (xlog_state_release_iclog(log, iclog))
3232 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003233 if (log_flushed)
3234 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003235 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003238 if ((flags & XFS_LOG_SYNC) && /* sleep */
3239 !(iclog->ic_state &
3240 (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3241 /*
3242 * Don't wait on completion if we know that we've
3243 * gotten a log write error.
3244 */
3245 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3246 spin_unlock(&log->l_icloglock);
3247 return XFS_ERROR(EIO);
3248 }
3249 XFS_STATS_INC(xs_log_force_sleep);
3250 sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
3251 /*
3252 * No need to grab the log lock here since we're
3253 * only deciding whether or not to return EIO
3254 * and the memory read should be atomic.
3255 */
3256 if (iclog->ic_state & XLOG_STATE_IOERROR)
3257 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003259 if (log_flushed)
3260 *log_flushed = 1;
3261 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003262 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 }
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003264
3265 return 0;
3266 } while (iclog != log->l_iclog);
3267
3268 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003270}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003272/*
3273 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3274 * about errors or whether the log was flushed or not. This is the normal
3275 * interface to use when trying to unpin items or move the log forward.
3276 */
3277void
3278xfs_log_force_lsn(
3279 xfs_mount_t *mp,
3280 xfs_lsn_t lsn,
3281 uint flags)
3282{
3283 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003285 error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
3286 if (error) {
3287 xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
3288 "error %d returned.", error);
3289 }
3290}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291
3292/*
3293 * Called when we want to mark the current iclog as being ready to sync to
3294 * disk.
3295 */
David Chinnera8272ce2007-11-23 16:28:09 +11003296STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3298{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003299 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300
3301 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3302 xlog_state_switch_iclogs(log, iclog, 0);
3303 } else {
3304 ASSERT(iclog->ic_state &
3305 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3306 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003307}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308
3309
3310/*****************************************************************************
3311 *
3312 * TICKET functions
3313 *
3314 *****************************************************************************
3315 */
3316
3317/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003318 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003320void
3321xfs_log_ticket_put(
3322 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003324 ASSERT(atomic_read(&ticket->t_ref) > 0);
3325 if (atomic_dec_and_test(&ticket->t_ref)) {
3326 sv_destroy(&ticket->t_wait);
3327 kmem_zone_free(xfs_log_ticket_zone, ticket);
3328 }
3329}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003331xlog_ticket_t *
3332xfs_log_ticket_get(
3333 xlog_ticket_t *ticket)
3334{
3335 ASSERT(atomic_read(&ticket->t_ref) > 0);
3336 atomic_inc(&ticket->t_ref);
3337 return ticket;
3338}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
Dave Chinner955833c2010-05-14 21:41:46 +10003340xlog_tid_t
3341xfs_log_get_trans_ident(
3342 struct xfs_trans *tp)
3343{
3344 return tp->t_ticket->t_tid;
3345}
3346
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347/*
David Chinnereb01c9c2008-04-10 12:18:46 +10003348 * Allocate and initialise a new log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349 */
Dave Chinner71e330b2010-05-21 14:37:18 +10003350xlog_ticket_t *
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003351xlog_ticket_alloc(
3352 struct log *log,
3353 int unit_bytes,
3354 int cnt,
3355 char client,
Dave Chinner3383ca52010-05-07 11:04:17 +10003356 uint xflags,
3357 int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003359 struct xlog_ticket *tic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 uint num_headers;
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003361 int iclog_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362
Dave Chinner3383ca52010-05-07 11:04:17 +10003363 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
David Chinnereb01c9c2008-04-10 12:18:46 +10003364 if (!tic)
3365 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366
3367 /*
3368 * Permanent reservations have up to 'cnt'-1 active log operations
3369 * in the log. A unit in this case is the amount of space for one
3370 * of these log operations. Normal reservations have a cnt of 1
3371 * and their unit amount is the total amount of space required.
3372 *
3373 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003374 * which occupy space in the on-disk log.
3375 *
3376 * Normal form of a transaction is:
3377 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3378 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3379 *
3380 * We need to account for all the leadup data and trailer data
3381 * around the transaction data.
3382 * And then we need to account for the worst case in terms of using
3383 * more space.
3384 * The worst case will happen if:
3385 * - the placement of the transaction happens to be such that the
3386 * roundoff is at its maximum
3387 * - the transaction data is synced before the commit record is synced
3388 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3389 * Therefore the commit record is in its own Log Record.
3390 * This can happen as the commit record is called with its
3391 * own region to xlog_write().
3392 * This then means that in the worst case, roundoff can happen for
3393 * the commit-rec as well.
3394 * The commit-rec is smaller than padding in this scenario and so it is
3395 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 */
3397
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003398 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003400 unit_bytes += sizeof(xfs_trans_header_t);
3401
3402 /* for start-rec */
3403 unit_bytes += sizeof(xlog_op_header_t);
3404
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003405 /*
3406 * for LR headers - the space for data in an iclog is the size minus
3407 * the space used for the headers. If we use the iclog size, then we
3408 * undercalculate the number of headers required.
3409 *
3410 * Furthermore - the addition of op headers for split-recs might
3411 * increase the space required enough to require more log and op
3412 * headers, so take that into account too.
3413 *
3414 * IMPORTANT: This reservation makes the assumption that if this
3415 * transaction is the first in an iclog and hence has the LR headers
3416 * accounted to it, then the remaining space in the iclog is
3417 * exclusively for this transaction. i.e. if the transaction is larger
3418 * than the iclog, it will be the only thing in that iclog.
3419 * Fundamentally, this means we must pass the entire log vector to
3420 * xlog_write to guarantee this.
3421 */
3422 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3423 num_headers = howmany(unit_bytes, iclog_space);
3424
3425 /* for split-recs - ophdrs added when data split over LRs */
3426 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3427
3428 /* add extra header reservations if we overrun */
3429 while (!num_headers ||
3430 howmany(unit_bytes, iclog_space) > num_headers) {
3431 unit_bytes += sizeof(xlog_op_header_t);
3432 num_headers++;
3433 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 unit_bytes += log->l_iclog_hsize * num_headers;
3435
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003436 /* for commit-rec LR header - note: padding will subsume the ophdr */
3437 unit_bytes += log->l_iclog_hsize;
3438
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003439 /* for roundoff padding for transaction data and one for commit record */
Eric Sandeen62118702008-03-06 13:44:28 +11003440 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003441 log->l_mp->m_sb.sb_logsunit > 1) {
3442 /* log su roundoff */
3443 unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
3444 } else {
3445 /* BB roundoff */
3446 unit_bytes += 2*BBSIZE;
3447 }
3448
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003449 atomic_set(&tic->t_ref, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450 tic->t_unit_res = unit_bytes;
3451 tic->t_curr_res = unit_bytes;
3452 tic->t_cnt = cnt;
3453 tic->t_ocnt = cnt;
Dave Chinnerf9837102010-04-14 15:47:55 +10003454 tic->t_tid = random32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455 tic->t_clientid = client;
3456 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003457 tic->t_trans_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 if (xflags & XFS_LOG_PERM_RESERV)
3459 tic->t_flags |= XLOG_TIC_PERM_RESERV;
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003460 sv_init(&tic->t_wait, SV_DEFAULT, "logtick");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461
Christoph Hellwig0adba532007-08-30 17:21:46 +10003462 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003463
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003465}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466
3467
3468/******************************************************************************
3469 *
3470 * Log debug routines
3471 *
3472 ******************************************************************************
3473 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003474#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475/*
3476 * Make sure that the destination ptr is within the valid data region of
3477 * one of the iclogs. This uses backup pointers stored in a different
3478 * part of the log in case we trash the log structure.
3479 */
3480void
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003481xlog_verify_dest_ptr(
3482 struct log *log,
3483 char *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484{
3485 int i;
3486 int good_ptr = 0;
3487
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003488 for (i = 0; i < log->l_iclog_bufs; i++) {
3489 if (ptr >= log->l_iclog_bak[i] &&
3490 ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491 good_ptr++;
3492 }
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003493
3494 if (!good_ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003496}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497
3498STATIC void
3499xlog_verify_grant_head(xlog_t *log, int equals)
3500{
3501 if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) {
3502 if (equals)
3503 ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes);
3504 else
3505 ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes);
3506 } else {
3507 ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle);
3508 ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes);
3509 }
3510} /* xlog_verify_grant_head */
3511
3512/* check if it will fit */
3513STATIC void
3514xlog_verify_tail_lsn(xlog_t *log,
3515 xlog_in_core_t *iclog,
3516 xfs_lsn_t tail_lsn)
3517{
3518 int blocks;
3519
3520 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3521 blocks =
3522 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3523 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3524 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3525 } else {
3526 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3527
3528 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3529 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3530
3531 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3532 if (blocks < BTOBB(iclog->ic_offset) + 1)
3533 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3534 }
3535} /* xlog_verify_tail_lsn */
3536
3537/*
3538 * Perform a number of checks on the iclog before writing to disk.
3539 *
3540 * 1. Make sure the iclogs are still circular
3541 * 2. Make sure we have a good magic number
3542 * 3. Make sure we don't have magic numbers in the data
3543 * 4. Check fields of each log operation header for:
3544 * A. Valid client identifier
3545 * B. tid ptr value falls in valid ptr space (user space code)
3546 * C. Length in log record header is correct according to the
3547 * individual operation headers within record.
3548 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3549 * log, check the preceding blocks of the physical log to make sure all
3550 * the cycle numbers agree with the current cycle number.
3551 */
3552STATIC void
3553xlog_verify_iclog(xlog_t *log,
3554 xlog_in_core_t *iclog,
3555 int count,
3556 boolean_t syncing)
3557{
3558 xlog_op_header_t *ophead;
3559 xlog_in_core_t *icptr;
3560 xlog_in_core_2_t *xhdr;
3561 xfs_caddr_t ptr;
3562 xfs_caddr_t base_ptr;
3563 __psint_t field_offset;
3564 __uint8_t clientid;
3565 int len, i, j, k, op_len;
3566 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567
3568 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003569 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570 icptr = log->l_iclog;
3571 for (i=0; i < log->l_iclog_bufs; i++) {
Christoph Hellwig4b809162007-08-16 15:37:36 +10003572 if (icptr == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 xlog_panic("xlog_verify_iclog: invalid ptr");
3574 icptr = icptr->ic_next;
3575 }
3576 if (icptr != log->l_iclog)
3577 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003578 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579
3580 /* check log magic numbers */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003581 if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582 xlog_panic("xlog_verify_iclog: invalid magic num");
3583
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003584 ptr = (xfs_caddr_t) &iclog->ic_header;
3585 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003586 ptr += BBSIZE) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003587 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588 xlog_panic("xlog_verify_iclog: unexpected magic num");
3589 }
3590
3591 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003592 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 ptr = iclog->ic_datap;
3594 base_ptr = ptr;
3595 ophead = (xlog_op_header_t *)ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003596 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597 for (i = 0; i < len; i++) {
3598 ophead = (xlog_op_header_t *)ptr;
3599
3600 /* clientid is only 1 byte */
3601 field_offset = (__psint_t)
3602 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3603 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3604 clientid = ophead->oh_clientid;
3605 } else {
3606 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3607 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3608 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3609 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003610 clientid = xlog_get_client_id(
3611 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003613 clientid = xlog_get_client_id(
3614 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 }
3616 }
3617 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Christoph Hellwigda1650a2005-11-02 10:21:35 +11003618 cmn_err(CE_WARN, "xlog_verify_iclog: "
3619 "invalid clientid %d op 0x%p offset 0x%lx",
3620 clientid, ophead, (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621
3622 /* check length */
3623 field_offset = (__psint_t)
3624 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3625 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003626 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 } else {
3628 idx = BTOBBT((__psint_t)&ophead->oh_len -
3629 (__psint_t)iclog->ic_datap);
3630 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3631 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3632 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003633 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003635 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 }
3637 }
3638 ptr += sizeof(xlog_op_header_t) + op_len;
3639 }
3640} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003641#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642
3643/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003644 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645 */
3646STATIC int
3647xlog_state_ioerror(
3648 xlog_t *log)
3649{
3650 xlog_in_core_t *iclog, *ic;
3651
3652 iclog = log->l_iclog;
3653 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3654 /*
3655 * Mark all the incore logs IOERROR.
3656 * From now on, no log flushes will result.
3657 */
3658 ic = iclog;
3659 do {
3660 ic->ic_state = XLOG_STATE_IOERROR;
3661 ic = ic->ic_next;
3662 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 }
3665 /*
3666 * Return non-zero, if state transition has already happened.
3667 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003668 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669}
3670
3671/*
3672 * This is called from xfs_force_shutdown, when we're forcibly
3673 * shutting down the filesystem, typically because of an IO error.
3674 * Our main objectives here are to make sure that:
3675 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3676 * parties to find out, 'atomically'.
3677 * b. those who're sleeping on log reservations, pinned objects and
3678 * other resources get woken up, and be told the bad news.
3679 * c. nothing new gets queued up after (a) and (b) are done.
3680 * d. if !logerror, flush the iclogs to disk, then seal them off
3681 * for business.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003682 *
3683 * Note: for delayed logging the !logerror case needs to flush the regions
3684 * held in memory out to the iclogs before flushing them to disk. This needs
3685 * to be done before the log is marked as shutdown, otherwise the flush to the
3686 * iclogs will fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 */
3688int
3689xfs_log_force_umount(
3690 struct xfs_mount *mp,
3691 int logerror)
3692{
3693 xlog_ticket_t *tic;
3694 xlog_t *log;
3695 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697 log = mp->m_log;
3698
3699 /*
3700 * If this happens during log recovery, don't worry about
3701 * locking; the log isn't open for business yet.
3702 */
3703 if (!log ||
3704 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3705 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003706 if (mp->m_sb_bp)
3707 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003708 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 }
3710
3711 /*
3712 * Somebody could've already done the hard work for us.
3713 * No need to get locks for this.
3714 */
3715 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3716 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003717 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 }
3719 retval = 0;
Dave Chinner9da1ab12010-05-17 15:51:59 +10003720
3721 /*
3722 * Flush the in memory commit item list before marking the log as
3723 * being shut down. We need to do it in this order to ensure all the
3724 * completed transactions are flushed to disk with the xfs_log_force()
3725 * call below.
3726 */
3727 if (!logerror && (mp->m_flags & XFS_MOUNT_DELAYLOG))
Dave Chinnera44f13e2010-08-24 11:40:03 +10003728 xlog_cil_force(log);
Dave Chinner9da1ab12010-05-17 15:51:59 +10003729
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 /*
3731 * We must hold both the GRANT lock and the LOG lock,
3732 * before we mark the filesystem SHUTDOWN and wake
3733 * everybody up to tell the bad news.
3734 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003735 spin_lock(&log->l_icloglock);
David Chinner6b1d1a72008-04-10 12:19:02 +10003736 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003738 if (mp->m_sb_bp)
3739 XFS_BUF_DONE(mp->m_sb_bp);
3740
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 /*
3742 * This flag is sort of redundant because of the mount flag, but
3743 * it's good to maintain the separation between the log and the rest
3744 * of XFS.
3745 */
3746 log->l_flags |= XLOG_IO_ERROR;
3747
3748 /*
3749 * If we hit a log error, we want to mark all the iclogs IOERROR
3750 * while we're still holding the loglock.
3751 */
3752 if (logerror)
3753 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003754 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755
3756 /*
3757 * We don't want anybody waiting for log reservations
3758 * after this. That means we have to wake up everybody
3759 * queued up on reserve_headq as well as write_headq.
3760 * In addition, we make sure in xlog_{re}grant_log_space
3761 * that we don't enqueue anything once the SHUTDOWN flag
3762 * is set, and this action is protected by the GRANTLOCK.
3763 */
3764 if ((tic = log->l_reserve_headq)) {
3765 do {
David Chinner12017fa2008-08-13 16:34:31 +10003766 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 tic = tic->t_next;
3768 } while (tic != log->l_reserve_headq);
3769 }
3770
3771 if ((tic = log->l_write_headq)) {
3772 do {
David Chinner12017fa2008-08-13 16:34:31 +10003773 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774 tic = tic->t_next;
3775 } while (tic != log->l_write_headq);
3776 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10003777 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003779 if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780 ASSERT(!logerror);
3781 /*
3782 * Force the incore logs to disk before shutting the
3783 * log down completely.
3784 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003785 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
3786
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003787 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003789 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790 }
3791 /*
3792 * Wake up everybody waiting on xfs_log_force.
3793 * Callback all log item committed functions as if the
3794 * log writes were completed.
3795 */
3796 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3797
3798#ifdef XFSERRORDEBUG
3799 {
3800 xlog_in_core_t *iclog;
3801
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003802 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003803 iclog = log->l_iclog;
3804 do {
3805 ASSERT(iclog->ic_callback == 0);
3806 iclog = iclog->ic_next;
3807 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003808 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 }
3810#endif
3811 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01003812 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813}
3814
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003815STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816xlog_iclogs_empty(xlog_t *log)
3817{
3818 xlog_in_core_t *iclog;
3819
3820 iclog = log->l_iclog;
3821 do {
3822 /* endianness does not matter here, zero is zero in
3823 * any language.
3824 */
3825 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01003826 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 iclog = iclog->ic_next;
3828 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003829 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830}