blob: 25efa9b8a6029ac35df73942459132abd71dc228 [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);
Dave Chinnerc8a09ff2010-12-04 00:02:40 +110050STATIC int xlog_space_left(struct log *log, atomic64_t *head);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051STATIC 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);
Dave Chinner2ced19c2010-12-21 12:09:20 +110073STATIC void xlog_grant_push_ail(struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 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);
Dave Chinner3f336c62010-12-21 12:02:52 +110084STATIC void xlog_verify_grant_tail(struct log *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085STATIC 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)
Dave Chinner3f336c62010-12-21 12:02:52 +110091#define xlog_verify_grant_tail(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#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 +110098static void
Dave Chinner663e4962010-12-21 12:06:05 +110099xlog_grant_sub_space(
100 struct log *log,
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100101 atomic64_t *head,
Dave Chinner663e4962010-12-21 12:06:05 +1100102 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100103{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100104 int64_t head_val = atomic64_read(head);
105 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100106
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100107 do {
108 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100109
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100110 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100111
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100112 space -= bytes;
113 if (space < 0) {
114 space += log->l_logsize;
115 cycle--;
116 }
117
118 old = head_val;
119 new = xlog_assign_grant_head_val(cycle, space);
120 head_val = atomic64_cmpxchg(head, old, new);
121 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100122}
123
124static void
Dave Chinner663e4962010-12-21 12:06:05 +1100125xlog_grant_add_space(
126 struct log *log,
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100127 atomic64_t *head,
Dave Chinner663e4962010-12-21 12:06:05 +1100128 int bytes)
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100129{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100130 int64_t head_val = atomic64_read(head);
131 int64_t new, old;
Dave Chinnera69ed032010-12-21 12:08:20 +1100132
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100133 do {
134 int tmp;
135 int cycle, space;
Dave Chinnera69ed032010-12-21 12:08:20 +1100136
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100137 xlog_crack_grant_head_val(head_val, &cycle, &space);
Dave Chinnera69ed032010-12-21 12:08:20 +1100138
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100139 tmp = log->l_logsize - space;
140 if (tmp > bytes)
141 space += bytes;
142 else {
143 space = bytes - tmp;
144 cycle++;
145 }
146
147 old = head_val;
148 new = xlog_assign_grant_head_val(cycle, space);
149 head_val = atomic64_cmpxchg(head, old, new);
150 } while (head_val != old);
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100151}
Dave Chinnera69ed032010-12-21 12:08:20 +1100152
Christoph Hellwig0adba532007-08-30 17:21:46 +1000153static void
154xlog_tic_reset_res(xlog_ticket_t *tic)
155{
156 tic->t_res_num = 0;
157 tic->t_res_arr_sum = 0;
158 tic->t_res_num_ophdrs = 0;
159}
160
161static void
162xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
163{
164 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
165 /* add to overflow and start again */
166 tic->t_res_o_flow += tic->t_res_arr_sum;
167 tic->t_res_num = 0;
168 tic->t_res_arr_sum = 0;
169 }
170
171 tic->t_res_arr[tic->t_res_num].r_len = len;
172 tic->t_res_arr[tic->t_res_num].r_type = type;
173 tic->t_res_arr_sum += len;
174 tic->t_res_num++;
175}
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177/*
178 * NOTES:
179 *
180 * 1. currblock field gets updated at startup and after in-core logs
181 * marked as with WANT_SYNC.
182 */
183
184/*
185 * This routine is called when a user of a log manager ticket is done with
186 * the reservation. If the ticket was ever used, then a commit record for
187 * the associated transaction is written out as a log operation header with
188 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
189 * a given ticket. If the ticket was one with a permanent reservation, then
190 * a few operations are done differently. Permanent reservation tickets by
191 * default don't release the reservation. They just commit the current
192 * transaction with the belief that the reservation is still needed. A flag
193 * must be passed in before permanent reservations are actually released.
194 * When these type of tickets are not released, they need to be set into
195 * the inited state again. By doing this, a start record will be written
196 * out when the next write occurs.
197 */
198xfs_lsn_t
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000199xfs_log_done(
200 struct xfs_mount *mp,
201 struct xlog_ticket *ticket,
202 struct xlog_in_core **iclog,
203 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000205 struct log *log = mp->m_log;
206 xfs_lsn_t lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 if (XLOG_FORCED_SHUTDOWN(log) ||
209 /*
210 * If nothing was ever written, don't write out commit record.
211 * If we get an error, just continue and give back the log ticket.
212 */
213 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
Dave Chinner55b66332010-03-23 11:43:17 +1100214 (xlog_commit_record(log, ticket, iclog, &lsn)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 lsn = (xfs_lsn_t) -1;
216 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
217 flags |= XFS_LOG_REL_PERM_RESERV;
218 }
219 }
220
221
222 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
223 (flags & XFS_LOG_REL_PERM_RESERV)) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000224 trace_xfs_log_done_nonperm(log, ticket);
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000227 * Release ticket if not permanent reservation or a specific
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 * request has been made to release a permanent reservation.
229 */
230 xlog_ungrant_log_space(log, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100231 xfs_log_ticket_put(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 } else {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000233 trace_xfs_log_done_perm(log, ticket);
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 xlog_regrant_reserve_log_space(log, ticket);
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000236 /* If this ticket was a permanent reservation and we aren't
237 * trying to release it, reset the inited flags; so next time
238 * we write, a start record will be written out.
239 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 ticket->t_flags |= XLOG_TIC_INITED;
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000241 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 return lsn;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000244}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246/*
247 * Attaches a new iclog I/O completion callback routine during
248 * transaction commit. If the log is in error state, a non-zero
249 * return code is handed back and the caller is responsible for
250 * executing the callback at an appropriate time.
251 */
252int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000253xfs_log_notify(
254 struct xfs_mount *mp,
255 struct xlog_in_core *iclog,
256 xfs_log_callback_t *cb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000258 int abortflg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
David Chinner114d23a2008-04-10 12:18:39 +1000260 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
262 if (!abortflg) {
263 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
264 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
265 cb->cb_next = NULL;
266 *(iclog->ic_callback_tail) = cb;
267 iclog->ic_callback_tail = &(cb->cb_next);
268 }
David Chinner114d23a2008-04-10 12:18:39 +1000269 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 return abortflg;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000271}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
273int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000274xfs_log_release_iclog(
275 struct xfs_mount *mp,
276 struct xlog_in_core *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000278 if (xlog_state_release_iclog(mp->m_log, iclog)) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000279 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +0100280 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 }
282
283 return 0;
284}
285
286/*
287 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
288 * to the reservation.
289 * 2. Potentially, push buffers at tail of log to disk.
290 *
291 * Each reservation is going to reserve extra space for a log record header.
292 * When writes happen to the on-disk log, we don't subtract the length of the
293 * log record header from any reservation. By wasting space in each
294 * reservation, we prevent over allocation problems.
295 */
296int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000297xfs_log_reserve(
298 struct xfs_mount *mp,
299 int unit_bytes,
300 int cnt,
301 struct xlog_ticket **ticket,
302 __uint8_t client,
303 uint flags,
304 uint t_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000306 struct log *log = mp->m_log;
307 struct xlog_ticket *internal_ticket;
308 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 if (XLOG_FORCED_SHUTDOWN(log))
313 return XFS_ERROR(EIO);
314
315 XFS_STATS_INC(xs_try_logspace);
316
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 if (*ticket != NULL) {
319 ASSERT(flags & XFS_LOG_PERM_RESERV);
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000320 internal_ticket = *ticket;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000321
Dave Chinner524ee362010-05-07 11:05:05 +1000322 /*
323 * this is a new transaction on the ticket, so we need to
324 * change the transaction ID so that the next transaction has a
325 * different TID in the log. Just add one to the existing tid
326 * so that we can see chains of rolling transactions in the log
327 * easily.
328 */
329 internal_ticket->t_tid++;
330
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000331 trace_xfs_log_reserve(log, internal_ticket);
332
Dave Chinner2ced19c2010-12-21 12:09:20 +1100333 xlog_grant_push_ail(log, internal_ticket->t_unit_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 retval = xlog_regrant_write_log_space(log, internal_ticket);
335 } else {
336 /* may sleep if need to allocate more tickets */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100337 internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
Dave Chinner3383ca52010-05-07 11:04:17 +1000338 client, flags,
339 KM_SLEEP|KM_MAYFAIL);
David Chinnereb01c9c2008-04-10 12:18:46 +1000340 if (!internal_ticket)
341 return XFS_ERROR(ENOMEM);
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000342 internal_ticket->t_trans_type = t_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 *ticket = internal_ticket;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000344
345 trace_xfs_log_reserve(log, internal_ticket);
346
Dave Chinner2ced19c2010-12-21 12:09:20 +1100347 xlog_grant_push_ail(log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 (internal_ticket->t_unit_res *
349 internal_ticket->t_cnt));
350 retval = xlog_grant_log_space(log, internal_ticket);
351 }
352
353 return retval;
354} /* xfs_log_reserve */
355
356
357/*
358 * Mount a log filesystem
359 *
360 * mp - ubiquitous xfs mount point structure
361 * log_target - buftarg of on-disk log device
362 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
363 * num_bblocks - Number of BBSIZE blocks in on-disk log
364 *
365 * Return error or zero.
366 */
367int
David Chinner249a8c12008-02-05 12:13:32 +1100368xfs_log_mount(
369 xfs_mount_t *mp,
370 xfs_buftarg_t *log_target,
371 xfs_daddr_t blk_offset,
372 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
David Chinner249a8c12008-02-05 12:13:32 +1100374 int error;
375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100377 xfs_notice(mp, "Mounting Filesystem");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100379 xfs_notice(mp,
380"Mounting filesystem in no-recovery mode. Filesystem will be inconsistent.");
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000381 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
383
384 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200385 if (IS_ERR(mp->m_log)) {
386 error = -PTR_ERR(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100387 goto out;
388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 /*
David Chinner249a8c12008-02-05 12:13:32 +1100391 * Initialize the AIL now we have a log.
392 */
David Chinner249a8c12008-02-05 12:13:32 +1100393 error = xfs_trans_ail_init(mp);
394 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100395 xfs_warn(mp, "AIL initialisation failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100396 goto out_free_log;
David Chinner249a8c12008-02-05 12:13:32 +1100397 }
David Chinnera9c21c12008-10-30 17:39:35 +1100398 mp->m_log->l_ailp = mp->m_ail;
David Chinner249a8c12008-02-05 12:13:32 +1100399
400 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * skip log recovery on a norecovery mount. pretend it all
402 * just worked.
403 */
404 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
David Chinner249a8c12008-02-05 12:13:32 +1100405 int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000408 mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
Eric Sandeen65be6052006-01-11 15:34:19 +1100410 error = xlog_recover(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000413 mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100415 xfs_warn(mp, "log mount/recovery failed: error %d",
416 error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100417 goto out_destroy_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
419 }
420
421 /* Normal transactions can now occur */
422 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
423
Dave Chinner71e330b2010-05-21 14:37:18 +1000424 /*
425 * Now the log has been fully initialised and we know were our
426 * space grant counters are, we can initialise the permanent ticket
427 * needed for delayed logging to work.
428 */
429 xlog_cil_init_post_recovery(mp->m_log);
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 return 0;
Christoph Hellwig26430752009-02-12 19:55:48 +0100432
433out_destroy_ail:
434 xfs_trans_ail_destroy(mp);
435out_free_log:
436 xlog_dealloc_log(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100437out:
David Chinner249a8c12008-02-05 12:13:32 +1100438 return error;
Christoph Hellwig26430752009-02-12 19:55:48 +0100439}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
441/*
442 * Finish the recovery of the file system. This is separate from
443 * the xfs_log_mount() call, because it depends on the code in
444 * xfs_mountfs() to read in the root and real-time bitmap inodes
445 * between calling xfs_log_mount() and here.
446 *
447 * mp - ubiquitous xfs mount point structure
448 */
449int
Christoph Hellwig42490232008-08-13 16:49:32 +1000450xfs_log_mount_finish(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
452 int error;
453
454 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
Christoph Hellwig42490232008-08-13 16:49:32 +1000455 error = xlog_recover_finish(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 else {
457 error = 0;
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000458 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
460
461 return error;
462}
463
464/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 * Final log writes as part of unmount.
466 *
467 * Mark the filesystem clean as unmount happens. Note that during relocation
468 * this routine needs to be executed as part of source-bag while the
469 * deallocation must not be done until source-end.
470 */
471
472/*
473 * Unmount record used to have a string "Unmount filesystem--" in the
474 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
475 * We just write the magic number now since that particular field isn't
476 * currently architecture converted and "nUmount" is a bit foo.
477 * As far as I know, there weren't any dependencies on the old behaviour.
478 */
479
480int
481xfs_log_unmount_write(xfs_mount_t *mp)
482{
483 xlog_t *log = mp->m_log;
484 xlog_in_core_t *iclog;
485#ifdef DEBUG
486 xlog_in_core_t *first_iclog;
487#endif
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000488 xlog_ticket_t *tic = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 xfs_lsn_t lsn;
490 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 /*
493 * Don't write out unmount record on read-only mounts.
494 * Or, if we are doing a forced umount (typically because of IO errors).
495 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000496 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 return 0;
498
Christoph Hellwiga14a3482010-01-19 09:56:46 +0000499 error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
David Chinnerb911ca02008-04-10 12:24:30 +1000500 ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502#ifdef DEBUG
503 first_iclog = iclog = log->l_iclog;
504 do {
505 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
506 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
507 ASSERT(iclog->ic_offset == 0);
508 }
509 iclog = iclog->ic_next;
510 } while (iclog != first_iclog);
511#endif
512 if (! (XLOG_FORCED_SHUTDOWN(log))) {
Tim Shimmin955e47a2006-09-28 11:04:16 +1000513 error = xfs_log_reserve(mp, 600, 1, &tic,
514 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (!error) {
Dave Chinner55b66332010-03-23 11:43:17 +1100516 /* the data section must be 32 bit size aligned */
517 struct {
518 __uint16_t magic;
519 __uint16_t pad1;
520 __uint32_t pad2; /* may as well make it 64 bits */
521 } magic = {
522 .magic = XLOG_UNMOUNT_TYPE,
523 };
524 struct xfs_log_iovec reg = {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000525 .i_addr = &magic,
Dave Chinner55b66332010-03-23 11:43:17 +1100526 .i_len = sizeof(magic),
527 .i_type = XLOG_REG_TYPE_UNMOUNT,
528 };
529 struct xfs_log_vec vec = {
530 .lv_niovecs = 1,
531 .lv_iovecp = &reg,
532 };
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 /* remove inited flag */
Dave Chinner55b66332010-03-23 11:43:17 +1100535 tic->t_flags = 0;
536 error = xlog_write(log, &vec, tic, &lsn,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 NULL, XLOG_UNMOUNT_TRANS);
538 /*
539 * At this point, we're umounting anyway,
540 * so there's no point in transitioning log state
541 * to IOERROR. Just continue...
542 */
543 }
544
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100545 if (error)
546 xfs_alert(mp, "%s: unmount record failed", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000549 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100551 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100553 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000554 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000556 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
558 iclog->ic_state == XLOG_STATE_DIRTY)) {
559 if (!XLOG_FORCED_SHUTDOWN(log)) {
Dave Chinnereb40a872010-12-21 12:09:01 +1100560 xlog_wait(&iclog->ic_force_wait,
561 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000563 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
565 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000566 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
Tim Shimmin955e47a2006-09-28 11:04:16 +1000568 if (tic) {
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000569 trace_xfs_log_umount_write(log, tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000570 xlog_ungrant_log_space(log, tic);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100571 xfs_log_ticket_put(tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 } else {
574 /*
575 * We're already in forced_shutdown mode, couldn't
576 * even attempt to write out the unmount transaction.
577 *
578 * Go through the motions of sync'ing and releasing
579 * the iclog, even though no I/O will actually happen,
Nathan Scottc41564b2006-03-29 08:55:14 +1000580 * we need to wait for other log I/Os that may already
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * be in progress. Do this as a separate section of
582 * code so we'll know if we ever get stuck here that
583 * we're in this odd situation of trying to unmount
584 * a file system that went into forced_shutdown as
585 * the result of an unmount..
586 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000587 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100589 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100592 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000593 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000595 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
598 || iclog->ic_state == XLOG_STATE_DIRTY
599 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
600
Dave Chinnereb40a872010-12-21 12:09:01 +1100601 xlog_wait(&iclog->ic_force_wait,
602 &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000604 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
606 }
607
David Chinner1bb7d6b2008-04-10 12:24:38 +1000608 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609} /* xfs_log_unmount_write */
610
611/*
612 * Deallocate log structures for unmount/relocation.
David Chinner249a8c12008-02-05 12:13:32 +1100613 *
614 * We need to stop the aild from running before we destroy
615 * and deallocate the log as the aild references the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 */
617void
Christoph Hellwig21b699c2009-03-16 08:19:29 +0100618xfs_log_unmount(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
David Chinner249a8c12008-02-05 12:13:32 +1100620 xfs_trans_ail_destroy(mp);
Nathan Scottc41564b2006-03-29 08:55:14 +1000621 xlog_dealloc_log(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622}
623
Dave Chinner43f5efc2010-03-23 10:10:00 +1100624void
625xfs_log_item_init(
626 struct xfs_mount *mp,
627 struct xfs_log_item *item,
628 int type,
629 struct xfs_item_ops *ops)
630{
631 item->li_mountp = mp;
632 item->li_ailp = mp->m_ail;
633 item->li_type = type;
634 item->li_ops = ops;
Dave Chinner71e330b2010-05-21 14:37:18 +1000635 item->li_lv = NULL;
636
637 INIT_LIST_HEAD(&item->li_ail);
638 INIT_LIST_HEAD(&item->li_cil);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100639}
640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641/*
642 * Write region vectors to log. The write happens using the space reservation
643 * of the ticket (tic). It is not a requirement that all writes for a given
Dave Chinner9b9fc2b72010-03-23 11:21:11 +1100644 * transaction occur with one call to xfs_log_write(). However, it is important
645 * to note that the transaction reservation code makes an assumption about the
646 * number of log headers a transaction requires that may be violated if you
647 * don't pass all the transaction vectors in one call....
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 */
649int
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000650xfs_log_write(
651 struct xfs_mount *mp,
652 struct xfs_log_iovec reg[],
653 int nentries,
654 struct xlog_ticket *tic,
655 xfs_lsn_t *start_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000657 struct log *log = mp->m_log;
658 int error;
Dave Chinner55b66332010-03-23 11:43:17 +1100659 struct xfs_log_vec vec = {
660 .lv_niovecs = nentries,
661 .lv_iovecp = reg,
662 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (XLOG_FORCED_SHUTDOWN(log))
665 return XFS_ERROR(EIO);
666
Dave Chinner55b66332010-03-23 11:43:17 +1100667 error = xlog_write(log, &vec, tic, start_lsn, NULL, 0);
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000668 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +1000669 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +0100670 return error;
Christoph Hellwig35a8a722010-02-15 23:34:54 +0000671}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
673void
674xfs_log_move_tail(xfs_mount_t *mp,
675 xfs_lsn_t tail_lsn)
676{
677 xlog_ticket_t *tic;
678 xlog_t *log = mp->m_log;
Dave Chinnera69ed032010-12-21 12:08:20 +1100679 int need_bytes, free_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (XLOG_FORCED_SHUTDOWN(log))
682 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Dave Chinner84f3c682010-12-03 22:11:29 +1100684 if (tail_lsn == 0)
685 tail_lsn = atomic64_read(&log->l_last_sync_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100687 /* tail_lsn == 1 implies that we weren't passed a valid value. */
688 if (tail_lsn != 1)
689 atomic64_set(&log->l_tail_lsn, tail_lsn);
690
Dave Chinner3f16b982010-12-21 12:29:01 +1100691 if (!list_empty_careful(&log->l_writeq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692#ifdef DEBUG
693 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
694 panic("Recovery problem");
695#endif
Dave Chinner3f16b982010-12-21 12:29:01 +1100696 spin_lock(&log->l_grant_write_lock);
Dave Chinnera69ed032010-12-21 12:08:20 +1100697 free_bytes = xlog_space_left(log, &log->l_grant_write_head);
Dave Chinner10547942010-12-21 12:02:25 +1100698 list_for_each_entry(tic, &log->l_writeq, t_queue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
700
701 if (free_bytes < tic->t_unit_res && tail_lsn != 1)
702 break;
703 tail_lsn = 0;
704 free_bytes -= tic->t_unit_res;
Dave Chinner3f16b982010-12-21 12:29:01 +1100705 trace_xfs_log_regrant_write_wake_up(log, tic);
Dave Chinnereb40a872010-12-21 12:09:01 +1100706 wake_up(&tic->t_wait);
Dave Chinner10547942010-12-21 12:02:25 +1100707 }
Dave Chinner3f16b982010-12-21 12:29:01 +1100708 spin_unlock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Dave Chinner10547942010-12-21 12:02:25 +1100710
Dave Chinner3f16b982010-12-21 12:29:01 +1100711 if (!list_empty_careful(&log->l_reserveq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712#ifdef DEBUG
713 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
714 panic("Recovery problem");
715#endif
Dave Chinner3f16b982010-12-21 12:29:01 +1100716 spin_lock(&log->l_grant_reserve_lock);
Dave Chinnera69ed032010-12-21 12:08:20 +1100717 free_bytes = xlog_space_left(log, &log->l_grant_reserve_head);
Dave Chinner10547942010-12-21 12:02:25 +1100718 list_for_each_entry(tic, &log->l_reserveq, t_queue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
720 need_bytes = tic->t_unit_res*tic->t_cnt;
721 else
722 need_bytes = tic->t_unit_res;
723 if (free_bytes < need_bytes && tail_lsn != 1)
724 break;
725 tail_lsn = 0;
726 free_bytes -= need_bytes;
Dave Chinner3f16b982010-12-21 12:29:01 +1100727 trace_xfs_log_grant_wake_up(log, tic);
Dave Chinnereb40a872010-12-21 12:09:01 +1100728 wake_up(&tic->t_wait);
Dave Chinner10547942010-12-21 12:02:25 +1100729 }
Dave Chinner3f16b982010-12-21 12:29:01 +1100730 spin_unlock(&log->l_grant_reserve_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
Dave Chinner3f16b982010-12-21 12:29:01 +1100732}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734/*
735 * Determine if we have a transaction that has gone to disk
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000736 * that needs to be covered. To begin the transition to the idle state
737 * firstly the log needs to be idle (no AIL and nothing in the iclogs).
738 * If we are then in a state where covering is needed, the caller is informed
739 * that dummy transactions are required to move the log into the idle state.
740 *
741 * Because this is called as part of the sync process, we should also indicate
742 * that dummy transactions should be issued in anything but the covered or
743 * idle states. This ensures that the log tail is accurately reflected in
744 * the log at the end of the sync, hence if a crash occurrs avoids replay
745 * of transactions where the metadata is already on disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
747int
748xfs_log_need_covered(xfs_mount_t *mp)
749{
David Chinner27d8d5f2008-10-30 17:38:39 +1100750 int needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 xlog_t *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
David Chinner92821e22007-05-24 15:26:31 +1000753 if (!xfs_fs_writable(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return 0;
755
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000756 spin_lock(&log->l_icloglock);
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000757 switch (log->l_covered_state) {
758 case XLOG_STATE_COVER_DONE:
759 case XLOG_STATE_COVER_DONE2:
760 case XLOG_STATE_COVER_IDLE:
761 break;
762 case XLOG_STATE_COVER_NEED:
763 case XLOG_STATE_COVER_NEED2:
764 if (!xfs_trans_ail_tail(log->l_ailp) &&
765 xlog_iclogs_empty(log)) {
766 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
767 log->l_covered_state = XLOG_STATE_COVER_DONE;
768 else
769 log->l_covered_state = XLOG_STATE_COVER_DONE2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 }
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000771 /* FALLTHRU */
772 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 needed = 1;
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000774 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000776 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +0100777 return needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
780/******************************************************************************
781 *
782 * local routines
783 *
784 ******************************************************************************
785 */
786
787/* xfs_trans_tail_ail returns 0 when there is nothing in the list.
788 * The log manager must keep track of the last LR which was committed
789 * to disk. The lsn of this LR will become the new tail_lsn whenever
790 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
791 * the situation where stuff could be written into the log but nothing
792 * was ever in the AIL when asked. Eventually, we panic since the
793 * tail hits the head.
794 *
795 * We may be holding the log iclog lock upon entering this routine.
796 */
797xfs_lsn_t
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100798xlog_assign_tail_lsn(
799 struct xfs_mount *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100801 xfs_lsn_t tail_lsn;
802 struct log *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
David Chinner5b00f142008-10-30 17:39:00 +1100804 tail_lsn = xfs_trans_ail_tail(mp->m_ail);
Dave Chinner84f3c682010-12-03 22:11:29 +1100805 if (!tail_lsn)
806 tail_lsn = atomic64_read(&log->l_last_sync_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100808 atomic64_set(&log->l_tail_lsn, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return tail_lsn;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100810}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812/*
813 * Return the space in the log between the tail and the head. The head
814 * is passed in the cycle/bytes formal parms. In the special case where
815 * the reserve head has wrapped passed the tail, this calculation is no
816 * longer valid. In this case, just return 0 which means there is no space
817 * in the log. This works for all places where this function is called
818 * with the reserve head. Of course, if the write head were to ever
819 * wrap the tail, we should blow up. Rather than catch this case here,
820 * we depend on other ASSERTions in other parts of the code. XXXmiken
821 *
822 * This code also handles the case where the reservation head is behind
823 * the tail. The details of this case are described below, but the end
824 * result is that we return the size of the log as the amount of space left.
825 */
David Chinnera8272ce2007-11-23 16:28:09 +1100826STATIC int
Dave Chinnera69ed032010-12-21 12:08:20 +1100827xlog_space_left(
828 struct log *log,
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100829 atomic64_t *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
Dave Chinnera69ed032010-12-21 12:08:20 +1100831 int free_bytes;
832 int tail_bytes;
833 int tail_cycle;
834 int head_cycle;
835 int head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Dave Chinnera69ed032010-12-21 12:08:20 +1100837 xlog_crack_grant_head(head, &head_cycle, &head_bytes);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100838 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
839 tail_bytes = BBTOB(tail_bytes);
Dave Chinnera69ed032010-12-21 12:08:20 +1100840 if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
841 free_bytes = log->l_logsize - (head_bytes - tail_bytes);
842 else if (tail_cycle + 1 < head_cycle)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 return 0;
Dave Chinnera69ed032010-12-21 12:08:20 +1100844 else if (tail_cycle < head_cycle) {
845 ASSERT(tail_cycle == (head_cycle - 1));
846 free_bytes = tail_bytes - head_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 } else {
848 /*
849 * The reservation head is behind the tail.
850 * In this case we just want to return the size of the
851 * log as the amount of space left.
852 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100853 xfs_alert(log->l_mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 "xlog_space_left: head behind tail\n"
855 " tail_cycle = %d, tail_bytes = %d\n"
856 " GH cycle = %d, GH bytes = %d",
Dave Chinnera69ed032010-12-21 12:08:20 +1100857 tail_cycle, tail_bytes, head_cycle, head_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 ASSERT(0);
859 free_bytes = log->l_logsize;
860 }
861 return free_bytes;
Dave Chinnera69ed032010-12-21 12:08:20 +1100862}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864
865/*
866 * Log function which is called when an io completes.
867 *
868 * The log manager needs its own routine, in order to control what
869 * happens with the buffer after the write completes.
870 */
871void
872xlog_iodone(xfs_buf_t *bp)
873{
874 xlog_in_core_t *iclog;
875 xlog_t *l;
876 int aborted;
877
878 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
879 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
880 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
881 aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 l = iclog->ic_log;
883
884 /*
885 * Race to shutdown the filesystem if we see an error.
886 */
887 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
888 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
889 xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
890 XFS_BUF_STALE(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +1000891 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 /*
893 * This flag will be propagated to the trans-committed
894 * callback routines to let them know that the log-commit
895 * didn't succeed.
896 */
897 aborted = XFS_LI_ABORTED;
898 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
899 aborted = XFS_LI_ABORTED;
900 }
David Chinner3db296f2007-05-14 18:24:16 +1000901
902 /* log I/O is always issued ASYNC */
903 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 xlog_state_done_syncing(iclog, aborted);
David Chinner3db296f2007-05-14 18:24:16 +1000905 /*
906 * do not reference the buffer (bp) here as we could race
907 * with it being freed after writing the unmount record to the
908 * log.
909 */
910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911} /* xlog_iodone */
912
913/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 * Return size of each in-core log record buffer.
915 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200916 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 *
918 * If the filesystem blocksize is too large, we may need to choose a
919 * larger size since the directory code currently logs entire blocks.
920 */
921
922STATIC void
923xlog_get_iclog_buffer_size(xfs_mount_t *mp,
924 xlog_t *log)
925{
926 int size;
927 int xhdrs;
928
Eric Sandeen1cb51252007-08-16 16:24:43 +1000929 if (mp->m_logbufs <= 0)
930 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
931 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100932 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 /*
935 * Buffer size passed in from mount system call.
936 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100937 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 size = log->l_iclog_size = mp->m_logbsize;
939 log->l_iclog_size_log = 0;
940 while (size != 1) {
941 log->l_iclog_size_log++;
942 size >>= 1;
943 }
944
Eric Sandeen62118702008-03-06 13:44:28 +1100945 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200946 /* # headers = size / 32k
947 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 */
949
950 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
951 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
952 xhdrs++;
953 log->l_iclog_hsize = xhdrs << BBSHIFT;
954 log->l_iclog_heads = xhdrs;
955 } else {
956 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
957 log->l_iclog_hsize = BBSIZE;
958 log->l_iclog_heads = 1;
959 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100960 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 }
962
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200963 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +1000964 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
965 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /* the default log size is 16k or 32k which is one header sector */
968 log->l_iclog_hsize = BBSIZE;
969 log->l_iclog_heads = 1;
970
Christoph Hellwig7153f8b2009-02-09 08:36:46 +0100971done:
972 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100973 if (mp->m_logbufs == 0)
974 mp->m_logbufs = log->l_iclog_bufs;
975 if (mp->m_logbsize == 0)
976 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977} /* xlog_get_iclog_buffer_size */
978
979
980/*
981 * This routine initializes some of the log structure for a given mount point.
982 * Its primary purpose is to fill in enough, so recovery can occur. However,
983 * some other stuff may be filled in too.
984 */
985STATIC xlog_t *
986xlog_alloc_log(xfs_mount_t *mp,
987 xfs_buftarg_t *log_target,
988 xfs_daddr_t blk_offset,
989 int num_bblks)
990{
991 xlog_t *log;
992 xlog_rec_header_t *head;
993 xlog_in_core_t **iclogp;
994 xlog_in_core_t *iclog, *prev_iclog=NULL;
995 xfs_buf_t *bp;
996 int i;
Dave Chinnera6cb7672009-04-06 18:39:27 +0200997 int error = ENOMEM;
Alex Elder69ce58f2010-04-20 17:09:59 +1000998 uint log2_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Dave Chinner644c3562008-11-10 16:50:24 +11001000 log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001001 if (!log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001002 xfs_warn(mp, "Log allocation failed: No memory!");
Dave Chinnera6cb7672009-04-06 18:39:27 +02001003 goto out;
1004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 log->l_mp = mp;
1007 log->l_targ = log_target;
1008 log->l_logsize = BBTOB(num_bblks);
1009 log->l_logBBstart = blk_offset;
1010 log->l_logBBsize = num_bblks;
1011 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1012 log->l_flags |= XLOG_ACTIVE_RECOVERY;
1013
1014 log->l_prev_block = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001016 xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1017 xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
Dave Chinnera69ed032010-12-21 12:08:20 +11001019 xlog_assign_grant_head(&log->l_grant_reserve_head, 1, 0);
1020 xlog_assign_grant_head(&log->l_grant_write_head, 1, 0);
Dave Chinner10547942010-12-21 12:02:25 +11001021 INIT_LIST_HEAD(&log->l_reserveq);
1022 INIT_LIST_HEAD(&log->l_writeq);
Dave Chinner3f16b982010-12-21 12:29:01 +11001023 spin_lock_init(&log->l_grant_reserve_lock);
1024 spin_lock_init(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Dave Chinnera6cb7672009-04-06 18:39:27 +02001026 error = EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001027 if (xfs_sb_version_hassector(&mp->m_sb)) {
Alex Elder69ce58f2010-04-20 17:09:59 +10001028 log2_size = mp->m_sb.sb_logsectlog;
1029 if (log2_size < BBSHIFT) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001030 xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1031 log2_size, BBSHIFT);
Alex Elder69ce58f2010-04-20 17:09:59 +10001032 goto out_free_log;
1033 }
1034
1035 log2_size -= BBSHIFT;
1036 if (log2_size > mp->m_sectbb_log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001037 xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1038 log2_size, mp->m_sectbb_log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001039 goto out_free_log;
1040 }
1041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /* for larger sector sizes, must have v2 or external log */
Alex Elder69ce58f2010-04-20 17:09:59 +10001043 if (log2_size && log->l_logBBstart > 0 &&
1044 !xfs_sb_version_haslogv2(&mp->m_sb)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001045 xfs_warn(mp,
1046 "log sector size (0x%x) invalid for configuration.",
1047 log2_size);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001048 goto out_free_log;
1049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Alex Elder69ce58f2010-04-20 17:09:59 +10001051 log->l_sectBBsize = 1 << log2_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
1053 xlog_get_iclog_buffer_size(mp, log);
1054
Dave Chinnera6cb7672009-04-06 18:39:27 +02001055 error = ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001057 if (!bp)
1058 goto out_free_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1061 ASSERT(XFS_BUF_ISBUSY(bp));
1062 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
1063 log->l_xbuf = bp;
1064
Eric Sandeen007c61c2007-10-11 17:43:56 +10001065 spin_lock_init(&log->l_icloglock);
Dave Chinnereb40a872010-12-21 12:09:01 +11001066 init_waitqueue_head(&log->l_flush_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1069 ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
1070
1071 iclogp = &log->l_iclog;
1072 /*
1073 * The amount of memory to allocate for the iclog structure is
1074 * rather funky due to the way the structure is defined. It is
1075 * done this way so that we can use different sizes for machines
1076 * with different amounts of memory. See the definition of
1077 * xlog_in_core_t in xfs_log_priv.h for details.
1078 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 ASSERT(log->l_iclog_size >= 4096);
1080 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001081 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1082 if (!*iclogp)
1083 goto out_free_iclog;
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 iclog->ic_prev = prev_iclog;
1087 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001088
Dave Chinner686865f2010-09-24 20:07:47 +10001089 bp = xfs_buf_get_uncached(mp->m_logdev_targp,
1090 log->l_iclog_size, 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001091 if (!bp)
1092 goto out_free_iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001093 if (!XFS_BUF_CPSEMA(bp))
1094 ASSERT(0);
1095 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001096 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1097 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001098 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001099#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
David Chinner4679b2d2008-04-10 12:18:54 +10001101#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 head = &iclog->ic_header;
1103 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001104 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1105 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001106 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001107 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001109 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
1113 iclog->ic_state = XLOG_STATE_ACTIVE;
1114 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001115 atomic_set(&iclog->ic_refcnt, 0);
1116 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001118 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
1121 ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
Dave Chinnereb40a872010-12-21 12:09:01 +11001122 init_waitqueue_head(&iclog->ic_force_wait);
1123 init_waitqueue_head(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 iclogp = &iclog->ic_next;
1126 }
1127 *iclogp = log->l_iclog; /* complete ring */
1128 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1129
Dave Chinner71e330b2010-05-21 14:37:18 +10001130 error = xlog_cil_init(log);
1131 if (error)
1132 goto out_free_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001134
1135out_free_iclog:
1136 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1137 prev_iclog = iclog->ic_next;
Dave Chinnereb40a872010-12-21 12:09:01 +11001138 if (iclog->ic_bp)
Dave Chinner644c3562008-11-10 16:50:24 +11001139 xfs_buf_free(iclog->ic_bp);
Dave Chinner644c3562008-11-10 16:50:24 +11001140 kmem_free(iclog);
1141 }
1142 spinlock_destroy(&log->l_icloglock);
Dave Chinner644c3562008-11-10 16:50:24 +11001143 xfs_buf_free(log->l_xbuf);
1144out_free_log:
1145 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001146out:
1147 return ERR_PTR(-error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148} /* xlog_alloc_log */
1149
1150
1151/*
1152 * Write out the commit record of a transaction associated with the given
1153 * ticket. Return the lsn of the commit record.
1154 */
1155STATIC int
Dave Chinner55b66332010-03-23 11:43:17 +11001156xlog_commit_record(
1157 struct log *log,
1158 struct xlog_ticket *ticket,
1159 struct xlog_in_core **iclog,
1160 xfs_lsn_t *commitlsnp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
Dave Chinner55b66332010-03-23 11:43:17 +11001162 struct xfs_mount *mp = log->l_mp;
1163 int error;
1164 struct xfs_log_iovec reg = {
1165 .i_addr = NULL,
1166 .i_len = 0,
1167 .i_type = XLOG_REG_TYPE_COMMIT,
1168 };
1169 struct xfs_log_vec vec = {
1170 .lv_niovecs = 1,
1171 .lv_iovecp = &reg,
1172 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 ASSERT_ALWAYS(iclog);
Dave Chinner55b66332010-03-23 11:43:17 +11001175 error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1176 XLOG_COMMIT_TRANS);
1177 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +10001178 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +01001179 return error;
Dave Chinner55b66332010-03-23 11:43:17 +11001180}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182/*
1183 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1184 * log space. This code pushes on the lsn which would supposedly free up
1185 * the 25% which we want to leave free. We may need to adopt a policy which
1186 * pushes on an lsn which is further along in the log once we reach the high
1187 * water mark. In this manner, we would be creating a low water mark.
1188 */
David Chinnera8272ce2007-11-23 16:28:09 +11001189STATIC void
Dave Chinner2ced19c2010-12-21 12:09:20 +11001190xlog_grant_push_ail(
1191 struct log *log,
1192 int need_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193{
Dave Chinner2ced19c2010-12-21 12:09:20 +11001194 xfs_lsn_t threshold_lsn = 0;
Dave Chinner84f3c682010-12-03 22:11:29 +11001195 xfs_lsn_t last_sync_lsn;
Dave Chinner2ced19c2010-12-21 12:09:20 +11001196 int free_blocks;
1197 int free_bytes;
1198 int threshold_block;
1199 int threshold_cycle;
1200 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Dave Chinner2ced19c2010-12-21 12:09:20 +11001202 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Dave Chinner2ced19c2010-12-21 12:09:20 +11001204 free_bytes = xlog_space_left(log, &log->l_grant_reserve_head);
1205 free_blocks = BTOBBT(free_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Dave Chinner2ced19c2010-12-21 12:09:20 +11001207 /*
1208 * Set the threshold for the minimum number of free blocks in the
1209 * log to the maximum of what the caller needs, one quarter of the
1210 * log, and 256 blocks.
1211 */
1212 free_threshold = BTOBB(need_bytes);
1213 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1214 free_threshold = MAX(free_threshold, 256);
1215 if (free_blocks >= free_threshold)
1216 return;
1217
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001218 xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1219 &threshold_block);
1220 threshold_block += free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 if (threshold_block >= log->l_logBBsize) {
Dave Chinner2ced19c2010-12-21 12:09:20 +11001222 threshold_block -= log->l_logBBsize;
1223 threshold_cycle += 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
Dave Chinner2ced19c2010-12-21 12:09:20 +11001225 threshold_lsn = xlog_assign_lsn(threshold_cycle,
1226 threshold_block);
1227 /*
1228 * Don't pass in an lsn greater than the lsn of the last
Dave Chinner84f3c682010-12-03 22:11:29 +11001229 * log record known to be on disk. Use a snapshot of the last sync lsn
1230 * so that it doesn't change between the compare and the set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 */
Dave Chinner84f3c682010-12-03 22:11:29 +11001232 last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1233 if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1234 threshold_lsn = last_sync_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Dave Chinner2ced19c2010-12-21 12:09:20 +11001236 /*
1237 * Get the transaction layer to kick the dirty buffers out to
1238 * disk asynchronously. No point in trying to do this if
1239 * the filesystem is shutting down.
1240 */
1241 if (!XLOG_FORCED_SHUTDOWN(log))
1242 xfs_trans_ail_push(log->l_ailp, threshold_lsn);
1243}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Christoph Hellwig873ff552010-01-13 22:17:57 +00001245/*
1246 * The bdstrat callback function for log bufs. This gives us a central
1247 * place to trap bufs in case we get hit by a log I/O error and need to
1248 * shutdown. Actually, in practice, even when we didn't get a log error,
1249 * we transition the iclogs to IOERROR state *after* flushing all existing
1250 * iclogs to disk. This is because we don't want anymore new transactions to be
1251 * started or completed afterwards.
1252 */
1253STATIC int
1254xlog_bdstrat(
1255 struct xfs_buf *bp)
1256{
1257 struct xlog_in_core *iclog;
1258
1259 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1260 if (iclog->ic_state & XLOG_STATE_IOERROR) {
1261 XFS_BUF_ERROR(bp, EIO);
1262 XFS_BUF_STALE(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001263 xfs_buf_ioend(bp, 0);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001264 /*
1265 * It would seem logical to return EIO here, but we rely on
1266 * the log state machine to propagate I/O errors instead of
1267 * doing it here.
1268 */
1269 return 0;
1270 }
1271
1272 bp->b_flags |= _XBF_RUN_QUEUES;
1273 xfs_buf_iorequest(bp);
1274 return 0;
1275}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
1277/*
1278 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1279 * fashion. Previously, we should have moved the current iclog
1280 * ptr in the log to point to the next available iclog. This allows further
1281 * write to continue while this code syncs out an iclog ready to go.
1282 * Before an in-core log can be written out, the data section must be scanned
1283 * to save away the 1st word of each BBSIZE block into the header. We replace
1284 * it with the current cycle count. Each BBSIZE block is tagged with the
1285 * cycle count because there in an implicit assumption that drives will
1286 * guarantee that entire 512 byte blocks get written at once. In other words,
1287 * we can't have part of a 512 byte block written and part not written. By
1288 * tagging each block, we will know which blocks are valid when recovering
1289 * after an unclean shutdown.
1290 *
1291 * This routine is single threaded on the iclog. No other thread can be in
1292 * this routine with the same iclog. Changing contents of iclog can there-
1293 * fore be done without grabbing the state machine lock. Updating the global
1294 * log will require grabbing the lock though.
1295 *
1296 * The entire log manager uses a logical block numbering scheme. Only
1297 * log_sync (and then only bwrite()) know about the fact that the log may
1298 * not start with block zero on a given device. The log block start offset
1299 * is added immediately before calling bwrite().
1300 */
1301
David Chinnera8272ce2007-11-23 16:28:09 +11001302STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303xlog_sync(xlog_t *log,
1304 xlog_in_core_t *iclog)
1305{
1306 xfs_caddr_t dptr; /* pointer to byte sized element */
1307 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001308 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 uint count; /* byte count of bwrite */
1310 uint count_init; /* initial count before roundup */
1311 int roundoff; /* roundoff to BB or stripe */
1312 int split = 0; /* split write into two regions */
1313 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001314 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001317 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
1319 /* Add for LR header */
1320 count_init = log->l_iclog_hsize + iclog->ic_offset;
1321
1322 /* Round out the log write size */
1323 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1324 /* we have a v2 stripe unit to use */
1325 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1326 } else {
1327 count = BBTOB(BTOBB(count_init));
1328 }
1329 roundoff = count - count_init;
1330 ASSERT(roundoff >= 0);
1331 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1332 roundoff < log->l_mp->m_sb.sb_logsunit)
1333 ||
1334 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1335 roundoff < BBTOB(1)));
1336
1337 /* move grant heads by roundoff in sync */
Dave Chinnera69ed032010-12-21 12:08:20 +11001338 xlog_grant_add_space(log, &log->l_grant_reserve_head, roundoff);
1339 xlog_grant_add_space(log, &log->l_grant_write_head, roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
1341 /* put cycle number in every block */
1342 xlog_pack_data(log, iclog, roundoff);
1343
1344 /* real byte length */
1345 if (v2) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001346 iclog->ic_header.h_len =
1347 cpu_to_be32(iclog->ic_offset + roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001349 iclog->ic_header.h_len =
1350 cpu_to_be32(iclog->ic_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352
Nathan Scottf5faad72006-07-28 17:04:44 +10001353 bp = iclog->ic_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
1355 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001356 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1359
1360 /* Do we need to split this write into 2 parts? */
1361 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1362 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1363 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1364 iclog->ic_bwritecnt = 2; /* split into 2 writes */
1365 } else {
1366 iclog->ic_bwritecnt = 1;
1367 }
David Chinner511105b2007-05-24 15:21:57 +10001368 XFS_BUF_SET_COUNT(bp, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */
Nathan Scottf5faad72006-07-28 17:04:44 +10001370 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 XFS_BUF_BUSY(bp);
1372 XFS_BUF_ASYNC(bp);
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001373 bp->b_flags |= XBF_LOG_BUFFER;
Christoph Hellwig651701d2010-06-28 10:34:34 -04001374
1375 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001376 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
1378 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1379 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1380
1381 xlog_verify_iclog(log, iclog, count, B_TRUE);
1382
1383 /* account for log which doesn't start at block #0 */
1384 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1385 /*
1386 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1387 * is shutting down.
1388 */
1389 XFS_BUF_WRITE(bp);
1390
Christoph Hellwig873ff552010-01-13 22:17:57 +00001391 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
1393 XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001394 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
1396 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001397 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
1399 (unsigned long)1);
1400 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1401 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
1402 XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
1403 (__psint_t)count), split);
1404 XFS_BUF_SET_FSPRIVATE(bp, iclog);
Nathan Scottf5faad72006-07-28 17:04:44 +10001405 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 XFS_BUF_BUSY(bp);
1407 XFS_BUF_ASYNC(bp);
Dave Chinner2ee1aba2009-11-24 18:03:15 +00001408 bp->b_flags |= XBF_LOG_BUFFER;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001409 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1410 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 dptr = XFS_BUF_PTR(bp);
1412 /*
1413 * Bump the cycle numbers at the start of each block
1414 * since this part of the buffer is at the start of
1415 * a new cycle. Watch out for the header magic number
1416 * case, though.
1417 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001418 for (i = 0; i < split; i += BBSIZE) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001419 be32_add_cpu((__be32 *)dptr, 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001420 if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001421 be32_add_cpu((__be32 *)dptr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 dptr += BBSIZE;
1423 }
1424
1425 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1426 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1427
Nathan Scottc41564b2006-03-29 08:55:14 +10001428 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1430 XFS_BUF_WRITE(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001431 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
1433 bp, XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001434 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
1436 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001437 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438} /* xlog_sync */
1439
1440
1441/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001442 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 */
David Chinnera8272ce2007-11-23 16:28:09 +11001444STATIC void
Nathan Scottc41564b2006-03-29 08:55:14 +10001445xlog_dealloc_log(xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
1447 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 int i;
1449
Dave Chinner71e330b2010-05-21 14:37:18 +10001450 xlog_cil_destroy(log);
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 iclog = log->l_iclog;
1453 for (i=0; i<log->l_iclog_bufs; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 xfs_buf_free(iclog->ic_bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001456 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 iclog = next_iclog;
1458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 spinlock_destroy(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 xfs_buf_free(log->l_xbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001463 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001464} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466/*
1467 * Update counters atomically now that memcpy is done.
1468 */
1469/* ARGSUSED */
1470static inline void
1471xlog_state_finish_copy(xlog_t *log,
1472 xlog_in_core_t *iclog,
1473 int record_cnt,
1474 int copy_bytes)
1475{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001476 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001478 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 iclog->ic_offset += copy_bytes;
1480
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001481 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482} /* xlog_state_finish_copy */
1483
1484
1485
1486
1487/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001488 * print out info relating to regions written which consume
1489 * the reservation
1490 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001491void
1492xlog_print_tic_res(
1493 struct xfs_mount *mp,
1494 struct xlog_ticket *ticket)
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001495{
1496 uint i;
1497 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1498
1499 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1500 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1501 "bformat",
1502 "bchunk",
1503 "efi_format",
1504 "efd_format",
1505 "iformat",
1506 "icore",
1507 "iext",
1508 "ibroot",
1509 "ilocal",
1510 "iattr_ext",
1511 "iattr_broot",
1512 "iattr_local",
1513 "qformat",
1514 "dquot",
1515 "quotaoff",
1516 "LR header",
1517 "unmount",
1518 "commit",
1519 "trans header"
1520 };
1521 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1522 "SETATTR_NOT_SIZE",
1523 "SETATTR_SIZE",
1524 "INACTIVE",
1525 "CREATE",
1526 "CREATE_TRUNC",
1527 "TRUNCATE_FILE",
1528 "REMOVE",
1529 "LINK",
1530 "RENAME",
1531 "MKDIR",
1532 "RMDIR",
1533 "SYMLINK",
1534 "SET_DMATTRS",
1535 "GROWFS",
1536 "STRAT_WRITE",
1537 "DIOSTRAT",
1538 "WRITE_SYNC",
1539 "WRITEID",
1540 "ADDAFORK",
1541 "ATTRINVAL",
1542 "ATRUNCATE",
1543 "ATTR_SET",
1544 "ATTR_RM",
1545 "ATTR_FLAG",
1546 "CLEAR_AGI_BUCKET",
1547 "QM_SBCHANGE",
1548 "DUMMY1",
1549 "DUMMY2",
1550 "QM_QUOTAOFF",
1551 "QM_DQALLOC",
1552 "QM_SETQLIM",
1553 "QM_DQCLUSTER",
1554 "QM_QINOCREATE",
1555 "QM_QUOTAOFF_END",
1556 "SB_UNIT",
1557 "FSYNC_TS",
1558 "GROWFSRT_ALLOC",
1559 "GROWFSRT_ZERO",
1560 "GROWFSRT_FREE",
1561 "SWAPEXT"
1562 };
1563
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001564 xfs_warn(mp,
1565 "xfs_log_write: reservation summary:\n"
1566 " trans type = %s (%u)\n"
1567 " unit res = %d bytes\n"
1568 " current res = %d bytes\n"
1569 " total reg = %u bytes (o/flow = %u bytes)\n"
1570 " ophdrs = %u (ophdr space = %u bytes)\n"
1571 " ophdr + reg = %u bytes\n"
1572 " num regions = %u\n",
1573 ((ticket->t_trans_type <= 0 ||
1574 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1575 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1576 ticket->t_trans_type,
1577 ticket->t_unit_res,
1578 ticket->t_curr_res,
1579 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1580 ticket->t_res_num_ophdrs, ophdr_spc,
1581 ticket->t_res_arr_sum +
1582 ticket->t_res_o_flow + ophdr_spc,
1583 ticket->t_res_num);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001584
1585 for (i = 0; i < ticket->t_res_num; i++) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001586 uint r_type = ticket->t_res_arr[i].r_type;
1587 xfs_warn(mp, "region[%u]: %s - %u bytes\n", i,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001588 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1589 "bad-rtype" : res_type_str[r_type-1]),
1590 ticket->t_res_arr[i].r_len);
1591 }
Dave Chinner169a7b02010-05-07 11:05:31 +10001592
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001593 xfs_alert_tag(mp, XFS_PTAG_LOGRES,
Dave Chinner169a7b02010-05-07 11:05:31 +10001594 "xfs_log_write: reservation ran out. Need to up reservation");
1595 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001596}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001597
1598/*
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001599 * Calculate the potential space needed by the log vector. Each region gets
1600 * its own xlog_op_header_t and may need to be double word aligned.
1601 */
1602static int
1603xlog_write_calc_vec_length(
1604 struct xlog_ticket *ticket,
Dave Chinner55b66332010-03-23 11:43:17 +11001605 struct xfs_log_vec *log_vector)
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001606{
Dave Chinner55b66332010-03-23 11:43:17 +11001607 struct xfs_log_vec *lv;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001608 int headers = 0;
1609 int len = 0;
1610 int i;
1611
1612 /* acct for start rec of xact */
1613 if (ticket->t_flags & XLOG_TIC_INITED)
1614 headers++;
1615
Dave Chinner55b66332010-03-23 11:43:17 +11001616 for (lv = log_vector; lv; lv = lv->lv_next) {
1617 headers += lv->lv_niovecs;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001618
Dave Chinner55b66332010-03-23 11:43:17 +11001619 for (i = 0; i < lv->lv_niovecs; i++) {
1620 struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
1621
1622 len += vecp->i_len;
1623 xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
1624 }
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001625 }
1626
1627 ticket->t_res_num_ophdrs += headers;
1628 len += headers * sizeof(struct xlog_op_header);
1629
1630 return len;
1631}
1632
1633/*
1634 * If first write for transaction, insert start record We can't be trying to
1635 * commit if we are inited. We can't have any "partial_copy" if we are inited.
1636 */
1637static int
1638xlog_write_start_rec(
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001639 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001640 struct xlog_ticket *ticket)
1641{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001642 if (!(ticket->t_flags & XLOG_TIC_INITED))
1643 return 0;
1644
1645 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1646 ophdr->oh_clientid = ticket->t_clientid;
1647 ophdr->oh_len = 0;
1648 ophdr->oh_flags = XLOG_START_TRANS;
1649 ophdr->oh_res2 = 0;
1650
1651 ticket->t_flags &= ~XLOG_TIC_INITED;
1652
1653 return sizeof(struct xlog_op_header);
1654}
1655
1656static xlog_op_header_t *
1657xlog_write_setup_ophdr(
1658 struct log *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001659 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001660 struct xlog_ticket *ticket,
1661 uint flags)
1662{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001663 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1664 ophdr->oh_clientid = ticket->t_clientid;
1665 ophdr->oh_res2 = 0;
1666
1667 /* are we copying a commit or unmount record? */
1668 ophdr->oh_flags = flags;
1669
1670 /*
1671 * We've seen logs corrupted with bad transaction client ids. This
1672 * makes sure that XFS doesn't generate them on. Turn this into an EIO
1673 * and shut down the filesystem.
1674 */
1675 switch (ophdr->oh_clientid) {
1676 case XFS_TRANSACTION:
1677 case XFS_VOLUME:
1678 case XFS_LOG:
1679 break;
1680 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001681 xfs_warn(log->l_mp,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001682 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1683 ophdr->oh_clientid, ticket);
1684 return NULL;
1685 }
1686
1687 return ophdr;
1688}
1689
1690/*
1691 * Set up the parameters of the region copy into the log. This has
1692 * to handle region write split across multiple log buffers - this
1693 * state is kept external to this function so that this code can
1694 * can be written in an obvious, self documenting manner.
1695 */
1696static int
1697xlog_write_setup_copy(
1698 struct xlog_ticket *ticket,
1699 struct xlog_op_header *ophdr,
1700 int space_available,
1701 int space_required,
1702 int *copy_off,
1703 int *copy_len,
1704 int *last_was_partial_copy,
1705 int *bytes_consumed)
1706{
1707 int still_to_copy;
1708
1709 still_to_copy = space_required - *bytes_consumed;
1710 *copy_off = *bytes_consumed;
1711
1712 if (still_to_copy <= space_available) {
1713 /* write of region completes here */
1714 *copy_len = still_to_copy;
1715 ophdr->oh_len = cpu_to_be32(*copy_len);
1716 if (*last_was_partial_copy)
1717 ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
1718 *last_was_partial_copy = 0;
1719 *bytes_consumed = 0;
1720 return 0;
1721 }
1722
1723 /* partial write of region, needs extra log op header reservation */
1724 *copy_len = space_available;
1725 ophdr->oh_len = cpu_to_be32(*copy_len);
1726 ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
1727 if (*last_was_partial_copy)
1728 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
1729 *bytes_consumed += *copy_len;
1730 (*last_was_partial_copy)++;
1731
1732 /* account for new log op header */
1733 ticket->t_curr_res -= sizeof(struct xlog_op_header);
1734 ticket->t_res_num_ophdrs++;
1735
1736 return sizeof(struct xlog_op_header);
1737}
1738
1739static int
1740xlog_write_copy_finish(
1741 struct log *log,
1742 struct xlog_in_core *iclog,
1743 uint flags,
1744 int *record_cnt,
1745 int *data_cnt,
1746 int *partial_copy,
1747 int *partial_copy_len,
1748 int log_offset,
1749 struct xlog_in_core **commit_iclog)
1750{
1751 if (*partial_copy) {
1752 /*
1753 * This iclog has already been marked WANT_SYNC by
1754 * xlog_state_get_iclog_space.
1755 */
1756 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1757 *record_cnt = 0;
1758 *data_cnt = 0;
1759 return xlog_state_release_iclog(log, iclog);
1760 }
1761
1762 *partial_copy = 0;
1763 *partial_copy_len = 0;
1764
1765 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
1766 /* no more space in this iclog - push it. */
1767 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1768 *record_cnt = 0;
1769 *data_cnt = 0;
1770
1771 spin_lock(&log->l_icloglock);
1772 xlog_state_want_sync(log, iclog);
1773 spin_unlock(&log->l_icloglock);
1774
1775 if (!commit_iclog)
1776 return xlog_state_release_iclog(log, iclog);
1777 ASSERT(flags & XLOG_COMMIT_TRANS);
1778 *commit_iclog = iclog;
1779 }
1780
1781 return 0;
1782}
1783
1784/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 * Write some region out to in-core log
1786 *
1787 * This will be called when writing externally provided regions or when
1788 * writing out a commit record for a given transaction.
1789 *
1790 * General algorithm:
1791 * 1. Find total length of this write. This may include adding to the
1792 * lengths passed in.
1793 * 2. Check whether we violate the tickets reservation.
1794 * 3. While writing to this iclog
1795 * A. Reserve as much space in this iclog as can get
1796 * B. If this is first write, save away start lsn
1797 * C. While writing this region:
1798 * 1. If first write of transaction, write start record
1799 * 2. Write log operation header (header per region)
1800 * 3. Find out if we can fit entire region into this iclog
1801 * 4. Potentially, verify destination memcpy ptr
1802 * 5. Memcpy (partial) region
1803 * 6. If partial copy, release iclog; otherwise, continue
1804 * copying more regions into current iclog
1805 * 4. Mark want sync bit (in simulation mode)
1806 * 5. Release iclog for potential flush to on-disk log.
1807 *
1808 * ERRORS:
1809 * 1. Panic if reservation is overrun. This should never happen since
1810 * reservation amounts are generated internal to the filesystem.
1811 * NOTES:
1812 * 1. Tickets are single threaded data structures.
1813 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1814 * syncing routine. When a single log_write region needs to span
1815 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1816 * on all log operation writes which don't contain the end of the
1817 * region. The XLOG_END_TRANS bit is used for the in-core log
1818 * operation which contains the end of the continued log_write region.
1819 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1820 * we don't really know exactly how much space will be used. As a result,
1821 * we don't update ic_offset until the end when we know exactly how many
1822 * bytes have been written out.
1823 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001824int
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001825xlog_write(
Dave Chinner55b66332010-03-23 11:43:17 +11001826 struct log *log,
1827 struct xfs_log_vec *log_vector,
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001828 struct xlog_ticket *ticket,
1829 xfs_lsn_t *start_lsn,
1830 struct xlog_in_core **commit_iclog,
1831 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832{
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001833 struct xlog_in_core *iclog = NULL;
Dave Chinner55b66332010-03-23 11:43:17 +11001834 struct xfs_log_iovec *vecp;
1835 struct xfs_log_vec *lv;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001836 int len;
1837 int index;
1838 int partial_copy = 0;
1839 int partial_copy_len = 0;
1840 int contwr = 0;
1841 int record_cnt = 0;
1842 int data_cnt = 0;
1843 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001845 *start_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Dave Chinner55b66332010-03-23 11:43:17 +11001847 len = xlog_write_calc_vec_length(ticket, log_vector);
Dave Chinner71e330b2010-05-21 14:37:18 +10001848 if (log->l_cilp) {
1849 /*
1850 * Region headers and bytes are already accounted for.
1851 * We only need to take into account start records and
1852 * split regions in this function.
1853 */
1854 if (ticket->t_flags & XLOG_TIC_INITED)
1855 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1856
1857 /*
1858 * Commit record headers need to be accounted for. These
1859 * come in as separate writes so are easy to detect.
1860 */
1861 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
1862 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1863 } else
1864 ticket->t_curr_res -= len;
1865
1866 if (ticket->t_curr_res < 0)
Dave Chinner55b66332010-03-23 11:43:17 +11001867 xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Dave Chinner55b66332010-03-23 11:43:17 +11001869 index = 0;
1870 lv = log_vector;
1871 vecp = lv->lv_iovecp;
1872 while (lv && index < lv->lv_niovecs) {
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001873 void *ptr;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001874 int log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001876 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
1877 &contwr, &log_offset);
1878 if (error)
1879 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001881 ASSERT(log_offset <= iclog->ic_size - 1);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001882 ptr = iclog->ic_datap + log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001884 /* start_lsn is the first lsn written to. That's all we need. */
1885 if (!*start_lsn)
1886 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001888 /*
1889 * This loop writes out as many regions as can fit in the amount
1890 * of space which was allocated by xlog_state_get_iclog_space().
1891 */
Dave Chinner55b66332010-03-23 11:43:17 +11001892 while (lv && index < lv->lv_niovecs) {
1893 struct xfs_log_iovec *reg = &vecp[index];
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001894 struct xlog_op_header *ophdr;
1895 int start_rec_copy;
1896 int copy_len;
1897 int copy_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
Dave Chinner55b66332010-03-23 11:43:17 +11001899 ASSERT(reg->i_len % sizeof(__int32_t) == 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001900 ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001902 start_rec_copy = xlog_write_start_rec(ptr, ticket);
1903 if (start_rec_copy) {
1904 record_cnt++;
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001905 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001906 start_rec_copy);
1907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001909 ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
1910 if (!ophdr)
1911 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001913 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001914 sizeof(struct xlog_op_header));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001916 len += xlog_write_setup_copy(ticket, ophdr,
1917 iclog->ic_size-log_offset,
Dave Chinner55b66332010-03-23 11:43:17 +11001918 reg->i_len,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001919 &copy_off, &copy_len,
1920 &partial_copy,
1921 &partial_copy_len);
1922 xlog_verify_dest_ptr(log, ptr);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001923
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001924 /* copy region */
1925 ASSERT(copy_len >= 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001926 memcpy(ptr, reg->i_addr + copy_off, copy_len);
1927 xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001928
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001929 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
1930 record_cnt++;
1931 data_cnt += contwr ? copy_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001933 error = xlog_write_copy_finish(log, iclog, flags,
1934 &record_cnt, &data_cnt,
1935 &partial_copy,
1936 &partial_copy_len,
1937 log_offset,
1938 commit_iclog);
1939 if (error)
1940 return error;
1941
1942 /*
1943 * if we had a partial copy, we need to get more iclog
1944 * space but we don't want to increment the region
1945 * index because there is still more is this region to
1946 * write.
1947 *
1948 * If we completed writing this region, and we flushed
1949 * the iclog (indicated by resetting of the record
1950 * count), then we also need to get more log space. If
1951 * this was the last record, though, we are done and
1952 * can just return.
1953 */
1954 if (partial_copy)
1955 break;
1956
Dave Chinner55b66332010-03-23 11:43:17 +11001957 if (++index == lv->lv_niovecs) {
1958 lv = lv->lv_next;
1959 index = 0;
1960 if (lv)
1961 vecp = lv->lv_iovecp;
1962 }
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001963 if (record_cnt == 0) {
Dave Chinner55b66332010-03-23 11:43:17 +11001964 if (!lv)
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001965 return 0;
1966 break;
1967 }
1968 }
1969 }
1970
1971 ASSERT(len == 0);
1972
1973 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1974 if (!commit_iclog)
1975 return xlog_state_release_iclog(log, iclog);
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 ASSERT(flags & XLOG_COMMIT_TRANS);
1978 *commit_iclog = iclog;
1979 return 0;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001980}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
1982
1983/*****************************************************************************
1984 *
1985 * State Machine functions
1986 *
1987 *****************************************************************************
1988 */
1989
1990/* Clean iclogs starting from the head. This ordering must be
1991 * maintained, so an iclog doesn't become ACTIVE beyond one that
1992 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10001993 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 * iclog is trying to sync to disk.
1995 *
1996 * State Change: DIRTY -> ACTIVE
1997 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001998STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999xlog_state_clean_log(xlog_t *log)
2000{
2001 xlog_in_core_t *iclog;
2002 int changed = 0;
2003
2004 iclog = log->l_iclog;
2005 do {
2006 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2007 iclog->ic_state = XLOG_STATE_ACTIVE;
2008 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002009 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 /*
2011 * If the number of ops in this iclog indicate it just
2012 * contains the dummy transaction, we can
2013 * change state into IDLE (the second time around).
2014 * Otherwise we should change the state into
2015 * NEED a dummy.
2016 * We don't need to cover the dummy.
2017 */
2018 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002019 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2020 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 changed = 1;
2022 } else {
2023 /*
2024 * We have two dirty iclogs so start over
2025 * This could also be num of ops indicates
2026 * this is not the dummy going out.
2027 */
2028 changed = 2;
2029 }
2030 iclog->ic_header.h_num_logops = 0;
2031 memset(iclog->ic_header.h_cycle_data, 0,
2032 sizeof(iclog->ic_header.h_cycle_data));
2033 iclog->ic_header.h_lsn = 0;
2034 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2035 /* do nothing */;
2036 else
2037 break; /* stop cleaning */
2038 iclog = iclog->ic_next;
2039 } while (iclog != log->l_iclog);
2040
2041 /* log is locked when we are called */
2042 /*
2043 * Change state for the dummy log recording.
2044 * We usually go to NEED. But we go to NEED2 if the changed indicates
2045 * we are done writing the dummy record.
2046 * If we are done with the second dummy recored (DONE2), then
2047 * we go to IDLE.
2048 */
2049 if (changed) {
2050 switch (log->l_covered_state) {
2051 case XLOG_STATE_COVER_IDLE:
2052 case XLOG_STATE_COVER_NEED:
2053 case XLOG_STATE_COVER_NEED2:
2054 log->l_covered_state = XLOG_STATE_COVER_NEED;
2055 break;
2056
2057 case XLOG_STATE_COVER_DONE:
2058 if (changed == 1)
2059 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2060 else
2061 log->l_covered_state = XLOG_STATE_COVER_NEED;
2062 break;
2063
2064 case XLOG_STATE_COVER_DONE2:
2065 if (changed == 1)
2066 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2067 else
2068 log->l_covered_state = XLOG_STATE_COVER_NEED;
2069 break;
2070
2071 default:
2072 ASSERT(0);
2073 }
2074 }
2075} /* xlog_state_clean_log */
2076
2077STATIC xfs_lsn_t
2078xlog_get_lowest_lsn(
2079 xlog_t *log)
2080{
2081 xlog_in_core_t *lsn_log;
2082 xfs_lsn_t lowest_lsn, lsn;
2083
2084 lsn_log = log->l_iclog;
2085 lowest_lsn = 0;
2086 do {
2087 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002088 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 if ((lsn && !lowest_lsn) ||
2090 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2091 lowest_lsn = lsn;
2092 }
2093 }
2094 lsn_log = lsn_log->ic_next;
2095 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002096 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097}
2098
2099
2100STATIC void
2101xlog_state_do_callback(
2102 xlog_t *log,
2103 int aborted,
2104 xlog_in_core_t *ciclog)
2105{
2106 xlog_in_core_t *iclog;
2107 xlog_in_core_t *first_iclog; /* used to know when we've
2108 * processed all iclogs once */
2109 xfs_log_callback_t *cb, *cb_next;
2110 int flushcnt = 0;
2111 xfs_lsn_t lowest_lsn;
2112 int ioerrors; /* counter: iclogs with errors */
2113 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2114 int funcdidcallbacks; /* flag: function did callbacks */
2115 int repeats; /* for issuing console warnings if
2116 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002117 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002119 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 first_iclog = iclog = log->l_iclog;
2121 ioerrors = 0;
2122 funcdidcallbacks = 0;
2123 repeats = 0;
2124
2125 do {
2126 /*
2127 * Scan all iclogs starting with the one pointed to by the
2128 * log. Reset this starting point each time the log is
2129 * unlocked (during callbacks).
2130 *
2131 * Keep looping through iclogs until one full pass is made
2132 * without running any callbacks.
2133 */
2134 first_iclog = log->l_iclog;
2135 iclog = log->l_iclog;
2136 loopdidcallbacks = 0;
2137 repeats++;
2138
2139 do {
2140
2141 /* skip all iclogs in the ACTIVE & DIRTY states */
2142 if (iclog->ic_state &
2143 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2144 iclog = iclog->ic_next;
2145 continue;
2146 }
2147
2148 /*
2149 * Between marking a filesystem SHUTDOWN and stopping
2150 * the log, we do flush all iclogs to disk (if there
2151 * wasn't a log I/O error). So, we do want things to
2152 * go smoothly in case of just a SHUTDOWN w/o a
2153 * LOG_IO_ERROR.
2154 */
2155 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2156 /*
2157 * Can only perform callbacks in order. Since
2158 * this iclog is not in the DONE_SYNC/
2159 * DO_CALLBACK state, we skip the rest and
2160 * just try to clean up. If we set our iclog
2161 * to DO_CALLBACK, we will not process it when
2162 * we retry since a previous iclog is in the
2163 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002164 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 */
2166 if (!(iclog->ic_state &
2167 (XLOG_STATE_DONE_SYNC |
2168 XLOG_STATE_DO_CALLBACK))) {
2169 if (ciclog && (ciclog->ic_state ==
2170 XLOG_STATE_DONE_SYNC)) {
2171 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2172 }
2173 break;
2174 }
2175 /*
2176 * We now have an iclog that is in either the
2177 * DO_CALLBACK or DONE_SYNC states. The other
2178 * states (WANT_SYNC, SYNCING, or CALLBACK were
2179 * caught by the above if and are going to
2180 * clean (i.e. we aren't doing their callbacks)
2181 * see the above if.
2182 */
2183
2184 /*
2185 * We will do one more check here to see if we
2186 * have chased our tail around.
2187 */
2188
2189 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002190 if (lowest_lsn &&
2191 XFS_LSN_CMP(lowest_lsn,
Dave Chinner84f3c682010-12-03 22:11:29 +11002192 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 iclog = iclog->ic_next;
2194 continue; /* Leave this iclog for
2195 * another thread */
2196 }
2197
2198 iclog->ic_state = XLOG_STATE_CALLBACK;
2199
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
Dave Chinner84f3c682010-12-03 22:11:29 +11002201 /*
2202 * update the last_sync_lsn before we drop the
2203 * icloglock to ensure we are the only one that
2204 * can update it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 */
Dave Chinner84f3c682010-12-03 22:11:29 +11002206 ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2207 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2208 atomic64_set(&log->l_last_sync_lsn,
2209 be64_to_cpu(iclog->ic_header.h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Dave Chinner84f3c682010-12-03 22:11:29 +11002211 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 ioerrors++;
Dave Chinner84f3c682010-12-03 22:11:29 +11002213
2214 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
David Chinner114d23a2008-04-10 12:18:39 +10002216 /*
2217 * Keep processing entries in the callback list until
2218 * we come around and it is empty. We need to
2219 * atomically see that the list is empty and change the
2220 * state to DIRTY so that we don't miss any more
2221 * callbacks being added.
2222 */
2223 spin_lock(&iclog->ic_callback_lock);
2224 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002225 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 iclog->ic_callback_tail = &(iclog->ic_callback);
2227 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002228 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
2230 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002231 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 cb_next = cb->cb_next;
2233 cb->cb_func(cb->cb_arg, aborted);
2234 }
David Chinner114d23a2008-04-10 12:18:39 +10002235 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 cb = iclog->ic_callback;
2237 }
2238
2239 loopdidcallbacks++;
2240 funcdidcallbacks++;
2241
David Chinner114d23a2008-04-10 12:18:39 +10002242 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002243 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002244 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2246 iclog->ic_state = XLOG_STATE_DIRTY;
2247
2248 /*
2249 * Transition from DIRTY to ACTIVE if applicable.
2250 * NOP if STATE_IOERROR.
2251 */
2252 xlog_state_clean_log(log);
2253
2254 /* wake up threads waiting in xfs_log_force() */
Dave Chinnereb40a872010-12-21 12:09:01 +11002255 wake_up_all(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
2257 iclog = iclog->ic_next;
2258 } while (first_iclog != iclog);
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002259
2260 if (repeats > 5000) {
2261 flushcnt += repeats;
2262 repeats = 0;
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002263 xfs_warn(log->l_mp,
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002264 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002265 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 }
2267 } while (!ioerrors && loopdidcallbacks);
2268
2269 /*
2270 * make one last gasp attempt to see if iclogs are being left in
2271 * limbo..
2272 */
2273#ifdef DEBUG
2274 if (funcdidcallbacks) {
2275 first_iclog = iclog = log->l_iclog;
2276 do {
2277 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2278 /*
2279 * Terminate the loop if iclogs are found in states
2280 * which will cause other threads to clean up iclogs.
2281 *
2282 * SYNCING - i/o completion will go through logs
2283 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002284 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 * IOERROR - give up hope all ye who enter here
2286 */
2287 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2288 iclog->ic_state == XLOG_STATE_SYNCING ||
2289 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2290 iclog->ic_state == XLOG_STATE_IOERROR )
2291 break;
2292 iclog = iclog->ic_next;
2293 } while (first_iclog != iclog);
2294 }
2295#endif
2296
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002297 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2298 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002299 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002300
2301 if (wake)
Dave Chinnereb40a872010-12-21 12:09:01 +11002302 wake_up_all(&log->l_flush_wait);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002303}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305
2306/*
2307 * Finish transitioning this iclog to the dirty state.
2308 *
2309 * Make sure that we completely execute this routine only when this is
2310 * the last call to the iclog. There is a good chance that iclog flushes,
2311 * when we reach the end of the physical log, get turned into 2 separate
2312 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2313 * routine. By using the reference count bwritecnt, we guarantee that only
2314 * the second completion goes through.
2315 *
2316 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002317 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 */
David Chinnera8272ce2007-11-23 16:28:09 +11002319STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320xlog_state_done_syncing(
2321 xlog_in_core_t *iclog,
2322 int aborted)
2323{
2324 xlog_t *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002326 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327
2328 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2329 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002330 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2332
2333
2334 /*
2335 * If we got an error, either on the first buffer, or in the case of
2336 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2337 * and none should ever be attempted to be written to disk
2338 * again.
2339 */
2340 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2341 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002342 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return;
2344 }
2345 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2346 }
2347
2348 /*
2349 * Someone could be sleeping prior to writing out the next
2350 * iclog buffer, we wake them all, one will get to do the
2351 * I/O, the others get to wait for the result.
2352 */
Dave Chinnereb40a872010-12-21 12:09:01 +11002353 wake_up_all(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002354 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2356} /* xlog_state_done_syncing */
2357
2358
2359/*
2360 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002361 * sleep. We wait on the flush queue on the head iclog as that should be
2362 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2363 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 *
2365 * The in-core logs are used in a circular fashion. They are not used
2366 * out-of-order even when an iclog past the head is free.
2367 *
2368 * return:
2369 * * log_offset where xlog_write() can start writing into the in-core
2370 * log's data space.
2371 * * in-core log pointer to which xlog_write() should write.
2372 * * boolean indicating this is a continued write to an in-core log.
2373 * If this is the last write, then the in-core log's offset field
2374 * needs to be incremented, depending on the amount of data which
2375 * is copied.
2376 */
David Chinnera8272ce2007-11-23 16:28:09 +11002377STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378xlog_state_get_iclog_space(xlog_t *log,
2379 int len,
2380 xlog_in_core_t **iclogp,
2381 xlog_ticket_t *ticket,
2382 int *continued_write,
2383 int *logoffsetp)
2384{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 int log_offset;
2386 xlog_rec_header_t *head;
2387 xlog_in_core_t *iclog;
2388 int error;
2389
2390restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002391 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002393 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 return XFS_ERROR(EIO);
2395 }
2396
2397 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002398 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002400
2401 /* Wait for log writes to have flushed */
Dave Chinnereb40a872010-12-21 12:09:01 +11002402 xlog_wait(&log->l_flush_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 goto restart;
2404 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002405
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 head = &iclog->ic_header;
2407
David Chinner155cc6b2008-03-06 13:44:14 +11002408 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 log_offset = iclog->ic_offset;
2410
2411 /* On the 1st write to an iclog, figure out lsn. This works
2412 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2413 * committing to. If the offset is set, that's how many blocks
2414 * must be written.
2415 */
2416 if (log_offset == 0) {
2417 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002418 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002419 log->l_iclog_hsize,
2420 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002421 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2422 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002423 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 ASSERT(log->l_curr_block >= 0);
2425 }
2426
2427 /* If there is enough room to write everything, then do it. Otherwise,
2428 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2429 * bit is on, so this will get flushed out. Don't update ic_offset
2430 * until you know exactly how many bytes get copied. Therefore, wait
2431 * until later to update ic_offset.
2432 *
2433 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2434 * can fit into remaining data section.
2435 */
2436 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2437 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2438
Dave Chinner49641f12008-07-11 17:43:55 +10002439 /*
2440 * If I'm the only one writing to this iclog, sync it to disk.
2441 * We need to do an atomic compare and decrement here to avoid
2442 * racing with concurrent atomic_dec_and_lock() calls in
2443 * xlog_state_release_iclog() when there is more than one
2444 * reference to the iclog.
2445 */
2446 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2447 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002448 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002449 error = xlog_state_release_iclog(log, iclog);
2450 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002451 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002453 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 }
2455 goto restart;
2456 }
2457
2458 /* Do we have enough room to write the full amount in the remainder
2459 * of this iclog? Or must we continue a write on the next iclog and
2460 * mark this iclog as completely taken? In the case where we switch
2461 * iclogs (to mark it taken), this particular iclog will release/sync
2462 * to disk in xlog_write().
2463 */
2464 if (len <= iclog->ic_size - iclog->ic_offset) {
2465 *continued_write = 0;
2466 iclog->ic_offset += len;
2467 } else {
2468 *continued_write = 1;
2469 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2470 }
2471 *iclogp = iclog;
2472
2473 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002474 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475
2476 *logoffsetp = log_offset;
2477 return 0;
2478} /* xlog_state_get_iclog_space */
2479
2480/*
2481 * Atomically get the log space required for a log ticket.
2482 *
2483 * Once a ticket gets put onto the reserveq, it will only return after
2484 * the needed reservation is satisfied.
Dave Chinner3f16b982010-12-21 12:29:01 +11002485 *
2486 * This function is structured so that it has a lock free fast path. This is
2487 * necessary because every new transaction reservation will come through this
2488 * path. Hence any lock will be globally hot if we take it unconditionally on
2489 * every pass.
2490 *
2491 * As tickets are only ever moved on and off the reserveq under the
2492 * l_grant_reserve_lock, we only need to take that lock if we are going
2493 * to add the ticket to the queue and sleep. We can avoid taking the lock if the
2494 * ticket was never added to the reserveq because the t_queue list head will be
2495 * empty and we hold the only reference to it so it can safely be checked
2496 * unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 */
2498STATIC int
2499xlog_grant_log_space(xlog_t *log,
2500 xlog_ticket_t *tic)
2501{
2502 int free_bytes;
2503 int need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504
2505#ifdef DEBUG
2506 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2507 panic("grant Recovery problem");
2508#endif
2509
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002510 trace_xfs_log_grant_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Dave Chinner3f16b982010-12-21 12:29:01 +11002512 need_bytes = tic->t_unit_res;
2513 if (tic->t_flags & XFS_LOG_PERM_RESERV)
2514 need_bytes *= tic->t_ocnt;
2515
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 /* something is already sleeping; insert new transaction at end */
Dave Chinner3f16b982010-12-21 12:29:01 +11002517 if (!list_empty_careful(&log->l_reserveq)) {
2518 spin_lock(&log->l_grant_reserve_lock);
2519 /* recheck the queue now we are locked */
2520 if (list_empty(&log->l_reserveq)) {
2521 spin_unlock(&log->l_grant_reserve_lock);
2522 goto redo;
2523 }
Dave Chinner10547942010-12-21 12:02:25 +11002524 list_add_tail(&tic->t_queue, &log->l_reserveq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002525
2526 trace_xfs_log_grant_sleep1(log, tic);
2527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 /*
2529 * Gotta check this before going to sleep, while we're
2530 * holding the grant lock.
2531 */
2532 if (XLOG_FORCED_SHUTDOWN(log))
2533 goto error_return;
2534
2535 XFS_STATS_INC(xs_sleep_logspace);
Dave Chinner3f16b982010-12-21 12:29:01 +11002536 xlog_wait(&tic->t_wait, &log->l_grant_reserve_lock);
Dave Chinnereb40a872010-12-21 12:09:01 +11002537
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 /*
2539 * If we got an error, and the filesystem is shutting down,
2540 * we'll catch it down below. So just continue...
2541 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002542 trace_xfs_log_grant_wake1(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
2545redo:
2546 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002547 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Dave Chinnera69ed032010-12-21 12:08:20 +11002549 free_bytes = xlog_space_left(log, &log->l_grant_reserve_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 if (free_bytes < need_bytes) {
Dave Chinner3f16b982010-12-21 12:29:01 +11002551 spin_lock(&log->l_grant_reserve_lock);
Dave Chinner10547942010-12-21 12:02:25 +11002552 if (list_empty(&tic->t_queue))
2553 list_add_tail(&tic->t_queue, &log->l_reserveq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002554
2555 trace_xfs_log_grant_sleep2(log, tic);
2556
Dave Chinner9d7fef72009-04-06 18:42:59 +02002557 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
Dave Chinner3f16b982010-12-21 12:29:01 +11002560 xlog_grant_push_ail(log, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002561
Dave Chinner3f16b982010-12-21 12:29:01 +11002562 XFS_STATS_INC(xs_sleep_logspace);
2563 xlog_wait(&tic->t_wait, &log->l_grant_reserve_lock);
2564
2565 trace_xfs_log_grant_wake2(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 goto redo;
Dave Chinner10547942010-12-21 12:02:25 +11002567 }
2568
Dave Chinner3f16b982010-12-21 12:29:01 +11002569 if (!list_empty(&tic->t_queue)) {
2570 spin_lock(&log->l_grant_reserve_lock);
2571 list_del_init(&tic->t_queue);
2572 spin_unlock(&log->l_grant_reserve_lock);
2573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574
2575 /* we've got enough space */
Dave Chinnera69ed032010-12-21 12:08:20 +11002576 xlog_grant_add_space(log, &log->l_grant_reserve_head, need_bytes);
2577 xlog_grant_add_space(log, &log->l_grant_write_head, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002578 trace_xfs_log_grant_exit(log, tic);
Dave Chinner3f336c62010-12-21 12:02:52 +11002579 xlog_verify_grant_tail(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 return 0;
2581
Dave Chinner3f16b982010-12-21 12:29:01 +11002582error_return_unlocked:
2583 spin_lock(&log->l_grant_reserve_lock);
2584error_return:
Dave Chinner10547942010-12-21 12:02:25 +11002585 list_del_init(&tic->t_queue);
Dave Chinner3f16b982010-12-21 12:29:01 +11002586 spin_unlock(&log->l_grant_reserve_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002587 trace_xfs_log_grant_error(log, tic);
2588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 /*
2590 * If we are failing, make sure the ticket doesn't have any
2591 * current reservations. We don't want to add this back when
2592 * the ticket/transaction gets cancelled.
2593 */
2594 tic->t_curr_res = 0;
2595 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 return XFS_ERROR(EIO);
2597} /* xlog_grant_log_space */
2598
2599
2600/*
2601 * Replenish the byte reservation required by moving the grant write head.
2602 *
Dave Chinner3f16b982010-12-21 12:29:01 +11002603 * Similar to xlog_grant_log_space, the function is structured to have a lock
2604 * free fast path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 */
2606STATIC int
2607xlog_regrant_write_log_space(xlog_t *log,
2608 xlog_ticket_t *tic)
2609{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 int free_bytes, need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
2612 tic->t_curr_res = tic->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002613 xlog_tic_reset_res(tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
2615 if (tic->t_cnt > 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01002616 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617
2618#ifdef DEBUG
2619 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2620 panic("regrant Recovery problem");
2621#endif
2622
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002623 trace_xfs_log_regrant_write_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002625 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
2627 /* If there are other waiters on the queue then give them a
2628 * chance at logspace before us. Wake up the first waiters,
2629 * if we do not wake up all the waiters then go to sleep waiting
2630 * for more free space, otherwise try to get some space for
2631 * this transaction.
2632 */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002633 need_bytes = tic->t_unit_res;
Dave Chinner3f16b982010-12-21 12:29:01 +11002634 if (!list_empty_careful(&log->l_writeq)) {
Dave Chinner10547942010-12-21 12:02:25 +11002635 struct xlog_ticket *ntic;
Dave Chinner3f16b982010-12-21 12:29:01 +11002636
2637 spin_lock(&log->l_grant_write_lock);
Dave Chinnera69ed032010-12-21 12:08:20 +11002638 free_bytes = xlog_space_left(log, &log->l_grant_write_head);
Dave Chinner10547942010-12-21 12:02:25 +11002639 list_for_each_entry(ntic, &log->l_writeq, t_queue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);
2641
2642 if (free_bytes < ntic->t_unit_res)
2643 break;
2644 free_bytes -= ntic->t_unit_res;
Dave Chinnereb40a872010-12-21 12:09:01 +11002645 wake_up(&ntic->t_wait);
Dave Chinner10547942010-12-21 12:02:25 +11002646 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
Dave Chinner10547942010-12-21 12:02:25 +11002648 if (ntic != list_first_entry(&log->l_writeq,
2649 struct xlog_ticket, t_queue)) {
2650 if (list_empty(&tic->t_queue))
2651 list_add_tail(&tic->t_queue, &log->l_writeq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002652 trace_xfs_log_regrant_write_sleep1(log, tic);
2653
Dave Chinner2ced19c2010-12-21 12:09:20 +11002654 xlog_grant_push_ail(log, need_bytes);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002655
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 XFS_STATS_INC(xs_sleep_logspace);
Dave Chinner3f16b982010-12-21 12:29:01 +11002657 xlog_wait(&tic->t_wait, &log->l_grant_write_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002658 trace_xfs_log_regrant_write_wake1(log, tic);
Dave Chinner3f16b982010-12-21 12:29:01 +11002659 } else
2660 spin_unlock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661 }
2662
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663redo:
2664 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002665 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
Dave Chinnera69ed032010-12-21 12:08:20 +11002667 free_bytes = xlog_space_left(log, &log->l_grant_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 if (free_bytes < need_bytes) {
Dave Chinner3f16b982010-12-21 12:29:01 +11002669 spin_lock(&log->l_grant_write_lock);
Dave Chinner10547942010-12-21 12:02:25 +11002670 if (list_empty(&tic->t_queue))
2671 list_add_tail(&tic->t_queue, &log->l_writeq);
Dave Chinner3f16b982010-12-21 12:29:01 +11002672
2673 if (XLOG_FORCED_SHUTDOWN(log))
2674 goto error_return;
2675
Dave Chinner2ced19c2010-12-21 12:09:20 +11002676 xlog_grant_push_ail(log, need_bytes);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002677
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 XFS_STATS_INC(xs_sleep_logspace);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002679 trace_xfs_log_regrant_write_sleep2(log, tic);
Dave Chinner3f16b982010-12-21 12:29:01 +11002680 xlog_wait(&tic->t_wait, &log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002682 trace_xfs_log_regrant_write_wake2(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 goto redo;
Dave Chinner10547942010-12-21 12:02:25 +11002684 }
2685
Dave Chinner3f16b982010-12-21 12:29:01 +11002686 if (!list_empty(&tic->t_queue)) {
2687 spin_lock(&log->l_grant_write_lock);
2688 list_del_init(&tic->t_queue);
2689 spin_unlock(&log->l_grant_write_lock);
2690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002692 /* we've got enough space */
Dave Chinnera69ed032010-12-21 12:08:20 +11002693 xlog_grant_add_space(log, &log->l_grant_write_head, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002694 trace_xfs_log_regrant_write_exit(log, tic);
Dave Chinner3f336c62010-12-21 12:02:52 +11002695 xlog_verify_grant_tail(log);
Jesper Juhl014c2542006-01-15 02:37:08 +01002696 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697
2698
Dave Chinner3f16b982010-12-21 12:29:01 +11002699 error_return_unlocked:
2700 spin_lock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 error_return:
Dave Chinner10547942010-12-21 12:02:25 +11002702 list_del_init(&tic->t_queue);
Dave Chinner3f16b982010-12-21 12:29:01 +11002703 spin_unlock(&log->l_grant_write_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002704 trace_xfs_log_regrant_write_error(log, tic);
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 /*
2707 * If we are failing, make sure the ticket doesn't have any
2708 * current reservations. We don't want to add this back when
2709 * the ticket/transaction gets cancelled.
2710 */
2711 tic->t_curr_res = 0;
2712 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 return XFS_ERROR(EIO);
2714} /* xlog_regrant_write_log_space */
2715
2716
2717/* The first cnt-1 times through here we don't need to
2718 * move the grant write head because the permanent
2719 * reservation has reserved cnt times the unit amount.
2720 * Release part of current permanent unit reservation and
2721 * reset current reservation to be one units worth. Also
2722 * move grant reservation head forward.
2723 */
2724STATIC void
2725xlog_regrant_reserve_log_space(xlog_t *log,
2726 xlog_ticket_t *ticket)
2727{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002728 trace_xfs_log_regrant_reserve_enter(log, ticket);
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 if (ticket->t_cnt > 0)
2731 ticket->t_cnt--;
2732
Dave Chinnera69ed032010-12-21 12:08:20 +11002733 xlog_grant_sub_space(log, &log->l_grant_reserve_head,
2734 ticket->t_curr_res);
2735 xlog_grant_sub_space(log, &log->l_grant_write_head,
2736 ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002738 xlog_tic_reset_res(ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002739
2740 trace_xfs_log_regrant_reserve_sub(log, ticket);
2741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 /* just return if we still have some of the pre-reserved space */
Dave Chinnerd0eb2f32010-12-21 12:29:14 +11002743 if (ticket->t_cnt > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745
Dave Chinnera69ed032010-12-21 12:08:20 +11002746 xlog_grant_add_space(log, &log->l_grant_reserve_head,
2747 ticket->t_unit_res);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002748
2749 trace_xfs_log_regrant_reserve_exit(log, ticket);
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002752 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753} /* xlog_regrant_reserve_log_space */
2754
2755
2756/*
2757 * Give back the space left from a reservation.
2758 *
2759 * All the information we need to make a correct determination of space left
2760 * is present. For non-permanent reservations, things are quite easy. The
2761 * count should have been decremented to zero. We only need to deal with the
2762 * space remaining in the current reservation part of the ticket. If the
2763 * ticket contains a permanent reservation, there may be left over space which
2764 * needs to be released. A count of N means that N-1 refills of the current
2765 * reservation can be done before we need to ask for more space. The first
2766 * one goes to fill up the first current reservation. Once we run out of
2767 * space, the count will stay at zero and the only space remaining will be
2768 * in the current reservation field.
2769 */
2770STATIC void
2771xlog_ungrant_log_space(xlog_t *log,
2772 xlog_ticket_t *ticket)
2773{
Dave Chinner663e4962010-12-21 12:06:05 +11002774 int bytes;
2775
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (ticket->t_cnt > 0)
2777 ticket->t_cnt--;
2778
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002779 trace_xfs_log_ungrant_enter(log, ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002780 trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
Dave Chinner663e4962010-12-21 12:06:05 +11002782 /*
2783 * If this is a permanent reservation ticket, we may be able to free
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 * up more space based on the remaining count.
2785 */
Dave Chinner663e4962010-12-21 12:06:05 +11002786 bytes = ticket->t_curr_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 if (ticket->t_cnt > 0) {
2788 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Dave Chinner663e4962010-12-21 12:06:05 +11002789 bytes += ticket->t_unit_res*ticket->t_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 }
2791
Dave Chinnera69ed032010-12-21 12:08:20 +11002792 xlog_grant_sub_space(log, &log->l_grant_reserve_head, bytes);
2793 xlog_grant_sub_space(log, &log->l_grant_write_head, bytes);
Dave Chinner663e4962010-12-21 12:06:05 +11002794
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002795 trace_xfs_log_ungrant_exit(log, ticket);
2796
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 xfs_log_move_tail(log->l_mp, 1);
2798} /* xlog_ungrant_log_space */
2799
2800
2801/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 * Flush iclog to disk if this is the last reference to the given iclog and
2803 * the WANT_SYNC bit is set.
2804 *
2805 * When this function is entered, the iclog is not necessarily in the
2806 * WANT_SYNC state. It may be sitting around waiting to get filled.
2807 *
2808 *
2809 */
David Chinnera8272ce2007-11-23 16:28:09 +11002810STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11002811xlog_state_release_iclog(
2812 xlog_t *log,
2813 xlog_in_core_t *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 int sync = 0; /* do we sync? */
2816
David Chinner155cc6b2008-03-06 13:44:14 +11002817 if (iclog->ic_state & XLOG_STATE_IOERROR)
2818 return XFS_ERROR(EIO);
2819
2820 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
2821 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
2822 return 0;
2823
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002825 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 return XFS_ERROR(EIO);
2827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
2829 iclog->ic_state == XLOG_STATE_WANT_SYNC);
2830
David Chinner155cc6b2008-03-06 13:44:14 +11002831 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11002832 /* update tail before writing to iclog */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11002833 xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 sync++;
2835 iclog->ic_state = XLOG_STATE_SYNCING;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11002836 iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
2837 xlog_verify_tail_lsn(log, iclog, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 /* cycle incremented when incrementing curr_block */
2839 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002840 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841
2842 /*
2843 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10002844 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2846 * this iclog has consistent data, so we ignore IOERROR
2847 * flags after this point.
2848 */
David Chinnerb5893342008-03-06 13:44:06 +11002849 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002851 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852} /* xlog_state_release_iclog */
2853
2854
2855/*
2856 * This routine will mark the current iclog in the ring as WANT_SYNC
2857 * and move the current iclog pointer to the next iclog in the ring.
2858 * When this routine is called from xlog_state_get_iclog_space(), the
2859 * exact size of the iclog has not yet been determined. All we know is
2860 * that every data block. We have run out of space in this log record.
2861 */
2862STATIC void
2863xlog_state_switch_iclogs(xlog_t *log,
2864 xlog_in_core_t *iclog,
2865 int eventual_size)
2866{
2867 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2868 if (!eventual_size)
2869 eventual_size = iclog->ic_offset;
2870 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002871 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 log->l_prev_block = log->l_curr_block;
2873 log->l_prev_cycle = log->l_curr_cycle;
2874
2875 /* roll log?: ic_offset changed later */
2876 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
2877
2878 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11002879 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 log->l_mp->m_sb.sb_logsunit > 1) {
2881 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
2882 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
2883 }
2884
2885 if (log->l_curr_block >= log->l_logBBsize) {
2886 log->l_curr_cycle++;
2887 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
2888 log->l_curr_cycle++;
2889 log->l_curr_block -= log->l_logBBsize;
2890 ASSERT(log->l_curr_block >= 0);
2891 }
2892 ASSERT(iclog == log->l_iclog);
2893 log->l_iclog = iclog->ic_next;
2894} /* xlog_state_switch_iclogs */
2895
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896/*
2897 * Write out all data in the in-core log as of this exact moment in time.
2898 *
2899 * Data may be written to the in-core log during this call. However,
2900 * we don't guarantee this data will be written out. A change from past
2901 * implementation means this routine will *not* write out zero length LRs.
2902 *
2903 * Basically, we try and perform an intelligent scan of the in-core logs.
2904 * If we determine there is no flushable data, we just return. There is no
2905 * flushable data if:
2906 *
2907 * 1. the current iclog is active and has no data; the previous iclog
2908 * is in the active or dirty state.
2909 * 2. the current iclog is drity, and the previous iclog is in the
2910 * active or dirty state.
2911 *
David Chinner12017fa2008-08-13 16:34:31 +10002912 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 *
2914 * 1. the current iclog is not in the active nor dirty state.
2915 * 2. the current iclog dirty, and the previous iclog is not in the
2916 * active nor dirty state.
2917 * 3. the current iclog is active, and there is another thread writing
2918 * to this particular iclog.
2919 * 4. a) the current iclog is active and has no other writers
2920 * b) when we return from flushing out this iclog, it is still
2921 * not in the active nor dirty state.
2922 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002923int
2924_xfs_log_force(
2925 struct xfs_mount *mp,
2926 uint flags,
2927 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002929 struct log *log = mp->m_log;
2930 struct xlog_in_core *iclog;
2931 xfs_lsn_t lsn;
2932
2933 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934
Dave Chinnera44f13e2010-08-24 11:40:03 +10002935 if (log->l_cilp)
2936 xlog_cil_force(log);
Dave Chinner71e330b2010-05-21 14:37:18 +10002937
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002938 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
2940 iclog = log->l_iclog;
2941 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002942 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 return XFS_ERROR(EIO);
2944 }
2945
2946 /* If the head iclog is not active nor dirty, we just attach
2947 * ourselves to the head and go to sleep.
2948 */
2949 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2950 iclog->ic_state == XLOG_STATE_DIRTY) {
2951 /*
2952 * If the head is dirty or (active and empty), then
2953 * we need to look at the previous iclog. If the previous
2954 * iclog is active or dirty we are done. There is nothing
2955 * to sync out. Otherwise, we attach ourselves to the
2956 * previous iclog and go to sleep.
2957 */
2958 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11002959 (atomic_read(&iclog->ic_refcnt) == 0
2960 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 iclog = iclog->ic_prev;
2962 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2963 iclog->ic_state == XLOG_STATE_DIRTY)
2964 goto no_sleep;
2965 else
2966 goto maybe_sleep;
2967 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11002968 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 /* We are the only one with access to this
2970 * iclog. Flush it out now. There should
2971 * be a roundoff of zero to show that someone
2972 * has already taken care of the roundoff from
2973 * the previous sync.
2974 */
David Chinner155cc6b2008-03-06 13:44:14 +11002975 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002976 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002978 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979
2980 if (xlog_state_release_iclog(log, iclog))
2981 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002982
2983 if (log_flushed)
2984 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002985 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002986 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 iclog->ic_state != XLOG_STATE_DIRTY)
2988 goto maybe_sleep;
2989 else
2990 goto no_sleep;
2991 } else {
2992 /* Someone else is writing to this iclog.
2993 * Use its call to flush out the data. However,
2994 * the other thread may not force out this LR,
2995 * so we mark it WANT_SYNC.
2996 */
2997 xlog_state_switch_iclogs(log, iclog, 0);
2998 goto maybe_sleep;
2999 }
3000 }
3001 }
3002
3003 /* By the time we come around again, the iclog could've been filled
3004 * which would give it another lsn. If we have a new lsn, just
3005 * return because the relevant data has been flushed.
3006 */
3007maybe_sleep:
3008 if (flags & XFS_LOG_SYNC) {
3009 /*
3010 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003011 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003012 * Then we check again after waking up, in case our
3013 * sleep was disturbed by a bad news.
3014 */
3015 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003016 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 return XFS_ERROR(EIO);
3018 }
3019 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003020 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 /*
3022 * No need to grab the log lock here since we're
3023 * only deciding whether or not to return EIO
3024 * and the memory read should be atomic.
3025 */
3026 if (iclog->ic_state & XLOG_STATE_IOERROR)
3027 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003028 if (log_flushed)
3029 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 } else {
3031
3032no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003033 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 }
3035 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003036}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
3038/*
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003039 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3040 * about errors or whether the log was flushed or not. This is the normal
3041 * interface to use when trying to unpin items or move the log forward.
3042 */
3043void
3044xfs_log_force(
3045 xfs_mount_t *mp,
3046 uint flags)
3047{
3048 int error;
3049
3050 error = _xfs_log_force(mp, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003051 if (error)
3052 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003053}
3054
3055/*
3056 * Force the in-core log to disk for a specific LSN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 *
3058 * Find in-core log with lsn.
3059 * If it is in the DIRTY state, just return.
3060 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3061 * state and go to sleep or return.
3062 * If it is in any other state, go to sleep or return.
3063 *
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003064 * Synchronous forces are implemented with a signal variable. All callers
3065 * to force a given lsn to disk will wait on a the sv attached to the
3066 * specific in-core log. When given in-core log finally completes its
3067 * write to disk, that thread will wake up all threads waiting on the
3068 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003070int
3071_xfs_log_force_lsn(
3072 struct xfs_mount *mp,
3073 xfs_lsn_t lsn,
3074 uint flags,
3075 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003077 struct log *log = mp->m_log;
3078 struct xlog_in_core *iclog;
3079 int already_slept = 0;
3080
3081 ASSERT(lsn != 0);
3082
3083 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084
Dave Chinner71e330b2010-05-21 14:37:18 +10003085 if (log->l_cilp) {
Dave Chinnera44f13e2010-08-24 11:40:03 +10003086 lsn = xlog_cil_force_lsn(log, lsn);
Dave Chinner71e330b2010-05-21 14:37:18 +10003087 if (lsn == NULLCOMMITLSN)
3088 return 0;
3089 }
3090
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091try_again:
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003092 spin_lock(&log->l_icloglock);
3093 iclog = log->l_iclog;
3094 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003095 spin_unlock(&log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003096 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 }
3098
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003099 do {
3100 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3101 iclog = iclog->ic_next;
3102 continue;
3103 }
3104
3105 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3106 spin_unlock(&log->l_icloglock);
3107 return 0;
3108 }
3109
3110 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3111 /*
3112 * We sleep here if we haven't already slept (e.g.
3113 * this is the first time we've looked at the correct
3114 * iclog buf) and the buffer before us is going to
3115 * be sync'ed. The reason for this is that if we
3116 * are doing sync transactions here, by waiting for
3117 * the previous I/O to complete, we can allow a few
3118 * more transactions into this iclog before we close
3119 * it down.
3120 *
3121 * Otherwise, we mark the buffer WANT_SYNC, and bump
3122 * up the refcnt so we can release the log (which
3123 * drops the ref count). The state switch keeps new
3124 * transaction commits from using this buffer. When
3125 * the current commits finish writing into the buffer,
3126 * the refcount will drop to zero and the buffer will
3127 * go out then.
3128 */
3129 if (!already_slept &&
3130 (iclog->ic_prev->ic_state &
3131 (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3132 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3133
3134 XFS_STATS_INC(xs_log_force_sleep);
3135
Dave Chinnereb40a872010-12-21 12:09:01 +11003136 xlog_wait(&iclog->ic_prev->ic_write_wait,
3137 &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003138 if (log_flushed)
3139 *log_flushed = 1;
3140 already_slept = 1;
3141 goto try_again;
3142 }
David Chinner155cc6b2008-03-06 13:44:14 +11003143 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003145 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 if (xlog_state_release_iclog(log, iclog))
3147 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003148 if (log_flushed)
3149 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003150 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003153 if ((flags & XFS_LOG_SYNC) && /* sleep */
3154 !(iclog->ic_state &
3155 (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3156 /*
3157 * Don't wait on completion if we know that we've
3158 * gotten a log write error.
3159 */
3160 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3161 spin_unlock(&log->l_icloglock);
3162 return XFS_ERROR(EIO);
3163 }
3164 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003165 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003166 /*
3167 * No need to grab the log lock here since we're
3168 * only deciding whether or not to return EIO
3169 * and the memory read should be atomic.
3170 */
3171 if (iclog->ic_state & XLOG_STATE_IOERROR)
3172 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003174 if (log_flushed)
3175 *log_flushed = 1;
3176 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003177 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 }
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003179
3180 return 0;
3181 } while (iclog != log->l_iclog);
3182
3183 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003185}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003187/*
3188 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3189 * about errors or whether the log was flushed or not. This is the normal
3190 * interface to use when trying to unpin items or move the log forward.
3191 */
3192void
3193xfs_log_force_lsn(
3194 xfs_mount_t *mp,
3195 xfs_lsn_t lsn,
3196 uint flags)
3197{
3198 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003200 error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003201 if (error)
3202 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003203}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204
3205/*
3206 * Called when we want to mark the current iclog as being ready to sync to
3207 * disk.
3208 */
David Chinnera8272ce2007-11-23 16:28:09 +11003209STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3211{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003212 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213
3214 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3215 xlog_state_switch_iclogs(log, iclog, 0);
3216 } else {
3217 ASSERT(iclog->ic_state &
3218 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3219 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003220}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221
3222
3223/*****************************************************************************
3224 *
3225 * TICKET functions
3226 *
3227 *****************************************************************************
3228 */
3229
3230/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003231 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003233void
3234xfs_log_ticket_put(
3235 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003237 ASSERT(atomic_read(&ticket->t_ref) > 0);
Dave Chinnereb40a872010-12-21 12:09:01 +11003238 if (atomic_dec_and_test(&ticket->t_ref))
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003239 kmem_zone_free(xfs_log_ticket_zone, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003240}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003242xlog_ticket_t *
3243xfs_log_ticket_get(
3244 xlog_ticket_t *ticket)
3245{
3246 ASSERT(atomic_read(&ticket->t_ref) > 0);
3247 atomic_inc(&ticket->t_ref);
3248 return ticket;
3249}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250
Dave Chinner955833c2010-05-14 21:41:46 +10003251xlog_tid_t
3252xfs_log_get_trans_ident(
3253 struct xfs_trans *tp)
3254{
3255 return tp->t_ticket->t_tid;
3256}
3257
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258/*
David Chinnereb01c9c2008-04-10 12:18:46 +10003259 * Allocate and initialise a new log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 */
Dave Chinner71e330b2010-05-21 14:37:18 +10003261xlog_ticket_t *
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003262xlog_ticket_alloc(
3263 struct log *log,
3264 int unit_bytes,
3265 int cnt,
3266 char client,
Dave Chinner3383ca52010-05-07 11:04:17 +10003267 uint xflags,
3268 int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269{
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003270 struct xlog_ticket *tic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 uint num_headers;
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003272 int iclog_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273
Dave Chinner3383ca52010-05-07 11:04:17 +10003274 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
David Chinnereb01c9c2008-04-10 12:18:46 +10003275 if (!tic)
3276 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
3278 /*
3279 * Permanent reservations have up to 'cnt'-1 active log operations
3280 * in the log. A unit in this case is the amount of space for one
3281 * of these log operations. Normal reservations have a cnt of 1
3282 * and their unit amount is the total amount of space required.
3283 *
3284 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003285 * which occupy space in the on-disk log.
3286 *
3287 * Normal form of a transaction is:
3288 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3289 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3290 *
3291 * We need to account for all the leadup data and trailer data
3292 * around the transaction data.
3293 * And then we need to account for the worst case in terms of using
3294 * more space.
3295 * The worst case will happen if:
3296 * - the placement of the transaction happens to be such that the
3297 * roundoff is at its maximum
3298 * - the transaction data is synced before the commit record is synced
3299 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3300 * Therefore the commit record is in its own Log Record.
3301 * This can happen as the commit record is called with its
3302 * own region to xlog_write().
3303 * This then means that in the worst case, roundoff can happen for
3304 * the commit-rec as well.
3305 * The commit-rec is smaller than padding in this scenario and so it is
3306 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 */
3308
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003309 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003311 unit_bytes += sizeof(xfs_trans_header_t);
3312
3313 /* for start-rec */
3314 unit_bytes += sizeof(xlog_op_header_t);
3315
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003316 /*
3317 * for LR headers - the space for data in an iclog is the size minus
3318 * the space used for the headers. If we use the iclog size, then we
3319 * undercalculate the number of headers required.
3320 *
3321 * Furthermore - the addition of op headers for split-recs might
3322 * increase the space required enough to require more log and op
3323 * headers, so take that into account too.
3324 *
3325 * IMPORTANT: This reservation makes the assumption that if this
3326 * transaction is the first in an iclog and hence has the LR headers
3327 * accounted to it, then the remaining space in the iclog is
3328 * exclusively for this transaction. i.e. if the transaction is larger
3329 * than the iclog, it will be the only thing in that iclog.
3330 * Fundamentally, this means we must pass the entire log vector to
3331 * xlog_write to guarantee this.
3332 */
3333 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3334 num_headers = howmany(unit_bytes, iclog_space);
3335
3336 /* for split-recs - ophdrs added when data split over LRs */
3337 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3338
3339 /* add extra header reservations if we overrun */
3340 while (!num_headers ||
3341 howmany(unit_bytes, iclog_space) > num_headers) {
3342 unit_bytes += sizeof(xlog_op_header_t);
3343 num_headers++;
3344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 unit_bytes += log->l_iclog_hsize * num_headers;
3346
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003347 /* for commit-rec LR header - note: padding will subsume the ophdr */
3348 unit_bytes += log->l_iclog_hsize;
3349
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003350 /* for roundoff padding for transaction data and one for commit record */
Eric Sandeen62118702008-03-06 13:44:28 +11003351 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003352 log->l_mp->m_sb.sb_logsunit > 1) {
3353 /* log su roundoff */
3354 unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
3355 } else {
3356 /* BB roundoff */
3357 unit_bytes += 2*BBSIZE;
3358 }
3359
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003360 atomic_set(&tic->t_ref, 1);
Dave Chinner10547942010-12-21 12:02:25 +11003361 INIT_LIST_HEAD(&tic->t_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 tic->t_unit_res = unit_bytes;
3363 tic->t_curr_res = unit_bytes;
3364 tic->t_cnt = cnt;
3365 tic->t_ocnt = cnt;
Dave Chinnerf9837102010-04-14 15:47:55 +10003366 tic->t_tid = random32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 tic->t_clientid = client;
3368 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003369 tic->t_trans_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 if (xflags & XFS_LOG_PERM_RESERV)
3371 tic->t_flags |= XLOG_TIC_PERM_RESERV;
Dave Chinnereb40a872010-12-21 12:09:01 +11003372 init_waitqueue_head(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
Christoph Hellwig0adba532007-08-30 17:21:46 +10003374 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003375
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003377}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378
3379
3380/******************************************************************************
3381 *
3382 * Log debug routines
3383 *
3384 ******************************************************************************
3385 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003386#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387/*
3388 * Make sure that the destination ptr is within the valid data region of
3389 * one of the iclogs. This uses backup pointers stored in a different
3390 * part of the log in case we trash the log structure.
3391 */
3392void
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003393xlog_verify_dest_ptr(
3394 struct log *log,
3395 char *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396{
3397 int i;
3398 int good_ptr = 0;
3399
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003400 for (i = 0; i < log->l_iclog_bufs; i++) {
3401 if (ptr >= log->l_iclog_bak[i] &&
3402 ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 good_ptr++;
3404 }
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003405
3406 if (!good_ptr)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003407 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003408}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409
3410STATIC void
Dave Chinner3f336c62010-12-21 12:02:52 +11003411xlog_verify_grant_tail(
3412 struct log *log)
3413{
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003414 int tail_cycle, tail_blocks;
Dave Chinnera69ed032010-12-21 12:08:20 +11003415 int cycle, space;
Dave Chinner3f336c62010-12-21 12:02:52 +11003416
3417 /*
3418 * Check to make sure the grant write head didn't just over lap the
3419 * tail. If the cycles are the same, we can't be overlapping.
3420 * Otherwise, make sure that the cycles differ by exactly one and
3421 * check the byte count.
3422 */
Dave Chinnera69ed032010-12-21 12:08:20 +11003423 xlog_crack_grant_head(&log->l_grant_write_head, &cycle, &space);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003424 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3425 if (tail_cycle != cycle) {
3426 ASSERT(cycle - 1 == tail_cycle);
3427 ASSERT(space <= BBTOB(tail_blocks));
Dave Chinner3f336c62010-12-21 12:02:52 +11003428 }
3429}
3430
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431/* check if it will fit */
3432STATIC void
3433xlog_verify_tail_lsn(xlog_t *log,
3434 xlog_in_core_t *iclog,
3435 xfs_lsn_t tail_lsn)
3436{
3437 int blocks;
3438
3439 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3440 blocks =
3441 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3442 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003443 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 } else {
3445 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3446
3447 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003448 xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449
3450 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3451 if (blocks < BTOBB(iclog->ic_offset) + 1)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003452 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 }
3454} /* xlog_verify_tail_lsn */
3455
3456/*
3457 * Perform a number of checks on the iclog before writing to disk.
3458 *
3459 * 1. Make sure the iclogs are still circular
3460 * 2. Make sure we have a good magic number
3461 * 3. Make sure we don't have magic numbers in the data
3462 * 4. Check fields of each log operation header for:
3463 * A. Valid client identifier
3464 * B. tid ptr value falls in valid ptr space (user space code)
3465 * C. Length in log record header is correct according to the
3466 * individual operation headers within record.
3467 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3468 * log, check the preceding blocks of the physical log to make sure all
3469 * the cycle numbers agree with the current cycle number.
3470 */
3471STATIC void
3472xlog_verify_iclog(xlog_t *log,
3473 xlog_in_core_t *iclog,
3474 int count,
3475 boolean_t syncing)
3476{
3477 xlog_op_header_t *ophead;
3478 xlog_in_core_t *icptr;
3479 xlog_in_core_2_t *xhdr;
3480 xfs_caddr_t ptr;
3481 xfs_caddr_t base_ptr;
3482 __psint_t field_offset;
3483 __uint8_t clientid;
3484 int len, i, j, k, op_len;
3485 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486
3487 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003488 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 icptr = log->l_iclog;
3490 for (i=0; i < log->l_iclog_bufs; i++) {
Christoph Hellwig4b809162007-08-16 15:37:36 +10003491 if (icptr == NULL)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003492 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 icptr = icptr->ic_next;
3494 }
3495 if (icptr != log->l_iclog)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003496 xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003497 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498
3499 /* check log magic numbers */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003500 if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003501 xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003503 ptr = (xfs_caddr_t) &iclog->ic_header;
3504 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 ptr += BBSIZE) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003506 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003507 xfs_emerg(log->l_mp, "%s: unexpected magic num",
3508 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509 }
3510
3511 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003512 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513 ptr = iclog->ic_datap;
3514 base_ptr = ptr;
3515 ophead = (xlog_op_header_t *)ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003516 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 for (i = 0; i < len; i++) {
3518 ophead = (xlog_op_header_t *)ptr;
3519
3520 /* clientid is only 1 byte */
3521 field_offset = (__psint_t)
3522 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3523 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3524 clientid = ophead->oh_clientid;
3525 } else {
3526 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3527 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3528 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3529 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003530 clientid = xlog_get_client_id(
3531 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003533 clientid = xlog_get_client_id(
3534 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535 }
3536 }
3537 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003538 xfs_warn(log->l_mp,
3539 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3540 __func__, clientid, ophead,
3541 (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
3543 /* check length */
3544 field_offset = (__psint_t)
3545 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3546 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003547 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 } else {
3549 idx = BTOBBT((__psint_t)&ophead->oh_len -
3550 (__psint_t)iclog->ic_datap);
3551 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3552 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3553 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003554 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003556 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 }
3558 }
3559 ptr += sizeof(xlog_op_header_t) + op_len;
3560 }
3561} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003562#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563
3564/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003565 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566 */
3567STATIC int
3568xlog_state_ioerror(
3569 xlog_t *log)
3570{
3571 xlog_in_core_t *iclog, *ic;
3572
3573 iclog = log->l_iclog;
3574 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3575 /*
3576 * Mark all the incore logs IOERROR.
3577 * From now on, no log flushes will result.
3578 */
3579 ic = iclog;
3580 do {
3581 ic->ic_state = XLOG_STATE_IOERROR;
3582 ic = ic->ic_next;
3583 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003584 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585 }
3586 /*
3587 * Return non-zero, if state transition has already happened.
3588 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003589 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590}
3591
3592/*
3593 * This is called from xfs_force_shutdown, when we're forcibly
3594 * shutting down the filesystem, typically because of an IO error.
3595 * Our main objectives here are to make sure that:
3596 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3597 * parties to find out, 'atomically'.
3598 * b. those who're sleeping on log reservations, pinned objects and
3599 * other resources get woken up, and be told the bad news.
3600 * c. nothing new gets queued up after (a) and (b) are done.
3601 * d. if !logerror, flush the iclogs to disk, then seal them off
3602 * for business.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003603 *
3604 * Note: for delayed logging the !logerror case needs to flush the regions
3605 * held in memory out to the iclogs before flushing them to disk. This needs
3606 * to be done before the log is marked as shutdown, otherwise the flush to the
3607 * iclogs will fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 */
3609int
3610xfs_log_force_umount(
3611 struct xfs_mount *mp,
3612 int logerror)
3613{
3614 xlog_ticket_t *tic;
3615 xlog_t *log;
3616 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617
3618 log = mp->m_log;
3619
3620 /*
3621 * If this happens during log recovery, don't worry about
3622 * locking; the log isn't open for business yet.
3623 */
3624 if (!log ||
3625 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3626 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003627 if (mp->m_sb_bp)
3628 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003629 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630 }
3631
3632 /*
3633 * Somebody could've already done the hard work for us.
3634 * No need to get locks for this.
3635 */
3636 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3637 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003638 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639 }
3640 retval = 0;
Dave Chinner9da1ab12010-05-17 15:51:59 +10003641
3642 /*
3643 * Flush the in memory commit item list before marking the log as
3644 * being shut down. We need to do it in this order to ensure all the
3645 * completed transactions are flushed to disk with the xfs_log_force()
3646 * call below.
3647 */
3648 if (!logerror && (mp->m_flags & XFS_MOUNT_DELAYLOG))
Dave Chinnera44f13e2010-08-24 11:40:03 +10003649 xlog_cil_force(log);
Dave Chinner9da1ab12010-05-17 15:51:59 +10003650
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 /*
Dave Chinner3f16b982010-12-21 12:29:01 +11003652 * mark the filesystem and the as in a shutdown state and wake
3653 * everybody up to tell them the bad news.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003655 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003657 if (mp->m_sb_bp)
3658 XFS_BUF_DONE(mp->m_sb_bp);
3659
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 /*
3661 * This flag is sort of redundant because of the mount flag, but
3662 * it's good to maintain the separation between the log and the rest
3663 * of XFS.
3664 */
3665 log->l_flags |= XLOG_IO_ERROR;
3666
3667 /*
3668 * If we hit a log error, we want to mark all the iclogs IOERROR
3669 * while we're still holding the loglock.
3670 */
3671 if (logerror)
3672 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003673 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674
3675 /*
Dave Chinner10547942010-12-21 12:02:25 +11003676 * We don't want anybody waiting for log reservations after this. That
3677 * means we have to wake up everybody queued up on reserveq as well as
3678 * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3679 * we don't enqueue anything once the SHUTDOWN flag is set, and this
Dave Chinner3f16b982010-12-21 12:29:01 +11003680 * action is protected by the grant locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003681 */
Dave Chinner3f16b982010-12-21 12:29:01 +11003682 spin_lock(&log->l_grant_reserve_lock);
Dave Chinner10547942010-12-21 12:02:25 +11003683 list_for_each_entry(tic, &log->l_reserveq, t_queue)
Dave Chinnereb40a872010-12-21 12:09:01 +11003684 wake_up(&tic->t_wait);
Dave Chinner3f16b982010-12-21 12:29:01 +11003685 spin_unlock(&log->l_grant_reserve_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003686
Dave Chinner3f16b982010-12-21 12:29:01 +11003687 spin_lock(&log->l_grant_write_lock);
Dave Chinner10547942010-12-21 12:02:25 +11003688 list_for_each_entry(tic, &log->l_writeq, t_queue)
Dave Chinnereb40a872010-12-21 12:09:01 +11003689 wake_up(&tic->t_wait);
Dave Chinner3f16b982010-12-21 12:29:01 +11003690 spin_unlock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003692 if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693 ASSERT(!logerror);
3694 /*
3695 * Force the incore logs to disk before shutting the
3696 * log down completely.
3697 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003698 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
3699
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003700 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003702 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 }
3704 /*
3705 * Wake up everybody waiting on xfs_log_force.
3706 * Callback all log item committed functions as if the
3707 * log writes were completed.
3708 */
3709 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3710
3711#ifdef XFSERRORDEBUG
3712 {
3713 xlog_in_core_t *iclog;
3714
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003715 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 iclog = log->l_iclog;
3717 do {
3718 ASSERT(iclog->ic_callback == 0);
3719 iclog = iclog->ic_next;
3720 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003721 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 }
3723#endif
3724 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01003725 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726}
3727
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003728STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003729xlog_iclogs_empty(xlog_t *log)
3730{
3731 xlog_in_core_t *iclog;
3732
3733 iclog = log->l_iclog;
3734 do {
3735 /* endianness does not matter here, zero is zero in
3736 * any language.
3737 */
3738 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01003739 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 iclog = iclog->ic_next;
3741 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003742 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743}