blob: 82c797dd6520db516216e052ed7abc142bacf1de [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:
Dave Chinnerfd074842011-04-08 12:45:07 +1000764 if (!xfs_ail_min_lsn(log->l_ailp) &&
Dave Chinnerb6f8dd42010-04-13 15:06:44 +1000765 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
Dave Chinnerfd074842011-04-08 12:45:07 +1000804 tail_lsn = xfs_ail_min_lsn(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 *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 l = iclog->ic_log;
881
882 /*
883 * Race to shutdown the filesystem if we see an error.
884 */
885 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
886 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
887 xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
888 XFS_BUF_STALE(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +1000889 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 /*
891 * This flag will be propagated to the trans-committed
892 * callback routines to let them know that the log-commit
893 * didn't succeed.
894 */
895 aborted = XFS_LI_ABORTED;
896 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
897 aborted = XFS_LI_ABORTED;
898 }
David Chinner3db296f2007-05-14 18:24:16 +1000899
900 /* log I/O is always issued ASYNC */
901 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 xlog_state_done_syncing(iclog, aborted);
David Chinner3db296f2007-05-14 18:24:16 +1000903 /*
904 * do not reference the buffer (bp) here as we could race
905 * with it being freed after writing the unmount record to the
906 * log.
907 */
908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909} /* xlog_iodone */
910
911/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 * Return size of each in-core log record buffer.
913 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200914 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 *
916 * If the filesystem blocksize is too large, we may need to choose a
917 * larger size since the directory code currently logs entire blocks.
918 */
919
920STATIC void
921xlog_get_iclog_buffer_size(xfs_mount_t *mp,
922 xlog_t *log)
923{
924 int size;
925 int xhdrs;
926
Eric Sandeen1cb51252007-08-16 16:24:43 +1000927 if (mp->m_logbufs <= 0)
928 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
929 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100930 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 /*
933 * Buffer size passed in from mount system call.
934 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100935 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 size = log->l_iclog_size = mp->m_logbsize;
937 log->l_iclog_size_log = 0;
938 while (size != 1) {
939 log->l_iclog_size_log++;
940 size >>= 1;
941 }
942
Eric Sandeen62118702008-03-06 13:44:28 +1100943 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200944 /* # headers = size / 32k
945 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 */
947
948 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
949 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
950 xhdrs++;
951 log->l_iclog_hsize = xhdrs << BBSHIFT;
952 log->l_iclog_heads = xhdrs;
953 } else {
954 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
955 log->l_iclog_hsize = BBSIZE;
956 log->l_iclog_heads = 1;
957 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100958 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960
Malcolm Parsons9da096f2009-03-29 09:55:42 +0200961 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +1000962 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
963 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
965 /* the default log size is 16k or 32k which is one header sector */
966 log->l_iclog_hsize = BBSIZE;
967 log->l_iclog_heads = 1;
968
Christoph Hellwig7153f8b2009-02-09 08:36:46 +0100969done:
970 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100971 if (mp->m_logbufs == 0)
972 mp->m_logbufs = log->l_iclog_bufs;
973 if (mp->m_logbsize == 0)
974 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975} /* xlog_get_iclog_buffer_size */
976
977
978/*
979 * This routine initializes some of the log structure for a given mount point.
980 * Its primary purpose is to fill in enough, so recovery can occur. However,
981 * some other stuff may be filled in too.
982 */
983STATIC xlog_t *
984xlog_alloc_log(xfs_mount_t *mp,
985 xfs_buftarg_t *log_target,
986 xfs_daddr_t blk_offset,
987 int num_bblks)
988{
989 xlog_t *log;
990 xlog_rec_header_t *head;
991 xlog_in_core_t **iclogp;
992 xlog_in_core_t *iclog, *prev_iclog=NULL;
993 xfs_buf_t *bp;
994 int i;
Dave Chinnera6cb7672009-04-06 18:39:27 +0200995 int error = ENOMEM;
Alex Elder69ce58f2010-04-20 17:09:59 +1000996 uint log2_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Dave Chinner644c3562008-11-10 16:50:24 +1100998 log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200999 if (!log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001000 xfs_warn(mp, "Log allocation failed: No memory!");
Dave Chinnera6cb7672009-04-06 18:39:27 +02001001 goto out;
1002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 log->l_mp = mp;
1005 log->l_targ = log_target;
1006 log->l_logsize = BBTOB(num_bblks);
1007 log->l_logBBstart = blk_offset;
1008 log->l_logBBsize = num_bblks;
1009 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1010 log->l_flags |= XLOG_ACTIVE_RECOVERY;
1011
1012 log->l_prev_block = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001014 xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
1015 xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
Dave Chinnera69ed032010-12-21 12:08:20 +11001017 xlog_assign_grant_head(&log->l_grant_reserve_head, 1, 0);
1018 xlog_assign_grant_head(&log->l_grant_write_head, 1, 0);
Dave Chinner10547942010-12-21 12:02:25 +11001019 INIT_LIST_HEAD(&log->l_reserveq);
1020 INIT_LIST_HEAD(&log->l_writeq);
Dave Chinner3f16b982010-12-21 12:29:01 +11001021 spin_lock_init(&log->l_grant_reserve_lock);
1022 spin_lock_init(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
Dave Chinnera6cb7672009-04-06 18:39:27 +02001024 error = EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001025 if (xfs_sb_version_hassector(&mp->m_sb)) {
Alex Elder69ce58f2010-04-20 17:09:59 +10001026 log2_size = mp->m_sb.sb_logsectlog;
1027 if (log2_size < BBSHIFT) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001028 xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
1029 log2_size, BBSHIFT);
Alex Elder69ce58f2010-04-20 17:09:59 +10001030 goto out_free_log;
1031 }
1032
1033 log2_size -= BBSHIFT;
1034 if (log2_size > mp->m_sectbb_log) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001035 xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
1036 log2_size, mp->m_sectbb_log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001037 goto out_free_log;
1038 }
1039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 /* for larger sector sizes, must have v2 or external log */
Alex Elder69ce58f2010-04-20 17:09:59 +10001041 if (log2_size && log->l_logBBstart > 0 &&
1042 !xfs_sb_version_haslogv2(&mp->m_sb)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001043 xfs_warn(mp,
1044 "log sector size (0x%x) invalid for configuration.",
1045 log2_size);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001046 goto out_free_log;
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
Alex Elder69ce58f2010-04-20 17:09:59 +10001049 log->l_sectBBsize = 1 << log2_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
1051 xlog_get_iclog_buffer_size(mp, log);
1052
Dave Chinnera6cb7672009-04-06 18:39:27 +02001053 error = ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001055 if (!bp)
1056 goto out_free_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001059 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 log->l_xbuf = bp;
1061
Eric Sandeen007c61c2007-10-11 17:43:56 +10001062 spin_lock_init(&log->l_icloglock);
Dave Chinnereb40a872010-12-21 12:09:01 +11001063 init_waitqueue_head(&log->l_flush_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1066 ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
1067
1068 iclogp = &log->l_iclog;
1069 /*
1070 * The amount of memory to allocate for the iclog structure is
1071 * rather funky due to the way the structure is defined. It is
1072 * done this way so that we can use different sizes for machines
1073 * with different amounts of memory. See the definition of
1074 * xlog_in_core_t in xfs_log_priv.h for details.
1075 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 ASSERT(log->l_iclog_size >= 4096);
1077 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001078 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1079 if (!*iclogp)
1080 goto out_free_iclog;
1081
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 iclog->ic_prev = prev_iclog;
1084 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001085
Dave Chinner686865f2010-09-24 20:07:47 +10001086 bp = xfs_buf_get_uncached(mp->m_logdev_targp,
1087 log->l_iclog_size, 0);
Dave Chinner644c3562008-11-10 16:50:24 +11001088 if (!bp)
1089 goto out_free_iclog;
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +02001090
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001091 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001092 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001093 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001094#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
David Chinner4679b2d2008-04-10 12:18:54 +10001096#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 head = &iclog->ic_header;
1098 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001099 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1100 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001101 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001102 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001104 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
1108 iclog->ic_state = XLOG_STATE_ACTIVE;
1109 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001110 atomic_set(&iclog->ic_refcnt, 0);
1111 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001113 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
1115 ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
Christoph Hellwig0c842ad2011-07-08 14:36:19 +02001116 ASSERT(xfs_buf_islocked(iclog->ic_bp));
Dave Chinnereb40a872010-12-21 12:09:01 +11001117 init_waitqueue_head(&iclog->ic_force_wait);
1118 init_waitqueue_head(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120 iclogp = &iclog->ic_next;
1121 }
1122 *iclogp = log->l_iclog; /* complete ring */
1123 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1124
Dave Chinner71e330b2010-05-21 14:37:18 +10001125 error = xlog_cil_init(log);
1126 if (error)
1127 goto out_free_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001129
1130out_free_iclog:
1131 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1132 prev_iclog = iclog->ic_next;
Dave Chinnereb40a872010-12-21 12:09:01 +11001133 if (iclog->ic_bp)
Dave Chinner644c3562008-11-10 16:50:24 +11001134 xfs_buf_free(iclog->ic_bp);
Dave Chinner644c3562008-11-10 16:50:24 +11001135 kmem_free(iclog);
1136 }
1137 spinlock_destroy(&log->l_icloglock);
Dave Chinner644c3562008-11-10 16:50:24 +11001138 xfs_buf_free(log->l_xbuf);
1139out_free_log:
1140 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001141out:
1142 return ERR_PTR(-error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143} /* xlog_alloc_log */
1144
1145
1146/*
1147 * Write out the commit record of a transaction associated with the given
1148 * ticket. Return the lsn of the commit record.
1149 */
1150STATIC int
Dave Chinner55b66332010-03-23 11:43:17 +11001151xlog_commit_record(
1152 struct log *log,
1153 struct xlog_ticket *ticket,
1154 struct xlog_in_core **iclog,
1155 xfs_lsn_t *commitlsnp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
Dave Chinner55b66332010-03-23 11:43:17 +11001157 struct xfs_mount *mp = log->l_mp;
1158 int error;
1159 struct xfs_log_iovec reg = {
1160 .i_addr = NULL,
1161 .i_len = 0,
1162 .i_type = XLOG_REG_TYPE_COMMIT,
1163 };
1164 struct xfs_log_vec vec = {
1165 .lv_niovecs = 1,
1166 .lv_iovecp = &reg,
1167 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 ASSERT_ALWAYS(iclog);
Dave Chinner55b66332010-03-23 11:43:17 +11001170 error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
1171 XLOG_COMMIT_TRANS);
1172 if (error)
Nathan Scott7d04a332006-06-09 14:58:38 +10001173 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +01001174 return error;
Dave Chinner55b66332010-03-23 11:43:17 +11001175}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
1177/*
1178 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1179 * log space. This code pushes on the lsn which would supposedly free up
1180 * the 25% which we want to leave free. We may need to adopt a policy which
1181 * pushes on an lsn which is further along in the log once we reach the high
1182 * water mark. In this manner, we would be creating a low water mark.
1183 */
David Chinnera8272ce2007-11-23 16:28:09 +11001184STATIC void
Dave Chinner2ced19c2010-12-21 12:09:20 +11001185xlog_grant_push_ail(
1186 struct log *log,
1187 int need_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188{
Dave Chinner2ced19c2010-12-21 12:09:20 +11001189 xfs_lsn_t threshold_lsn = 0;
Dave Chinner84f3c682010-12-03 22:11:29 +11001190 xfs_lsn_t last_sync_lsn;
Dave Chinner2ced19c2010-12-21 12:09:20 +11001191 int free_blocks;
1192 int free_bytes;
1193 int threshold_block;
1194 int threshold_cycle;
1195 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Dave Chinner2ced19c2010-12-21 12:09:20 +11001197 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Dave Chinner2ced19c2010-12-21 12:09:20 +11001199 free_bytes = xlog_space_left(log, &log->l_grant_reserve_head);
1200 free_blocks = BTOBBT(free_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Dave Chinner2ced19c2010-12-21 12:09:20 +11001202 /*
1203 * Set the threshold for the minimum number of free blocks in the
1204 * log to the maximum of what the caller needs, one quarter of the
1205 * log, and 256 blocks.
1206 */
1207 free_threshold = BTOBB(need_bytes);
1208 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1209 free_threshold = MAX(free_threshold, 256);
1210 if (free_blocks >= free_threshold)
1211 return;
1212
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001213 xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
1214 &threshold_block);
1215 threshold_block += free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (threshold_block >= log->l_logBBsize) {
Dave Chinner2ced19c2010-12-21 12:09:20 +11001217 threshold_block -= log->l_logBBsize;
1218 threshold_cycle += 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
Dave Chinner2ced19c2010-12-21 12:09:20 +11001220 threshold_lsn = xlog_assign_lsn(threshold_cycle,
1221 threshold_block);
1222 /*
1223 * Don't pass in an lsn greater than the lsn of the last
Dave Chinner84f3c682010-12-03 22:11:29 +11001224 * log record known to be on disk. Use a snapshot of the last sync lsn
1225 * so that it doesn't change between the compare and the set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 */
Dave Chinner84f3c682010-12-03 22:11:29 +11001227 last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
1228 if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
1229 threshold_lsn = last_sync_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Dave Chinner2ced19c2010-12-21 12:09:20 +11001231 /*
1232 * Get the transaction layer to kick the dirty buffers out to
1233 * disk asynchronously. No point in trying to do this if
1234 * the filesystem is shutting down.
1235 */
1236 if (!XLOG_FORCED_SHUTDOWN(log))
Dave Chinnerfd074842011-04-08 12:45:07 +10001237 xfs_ail_push(log->l_ailp, threshold_lsn);
Dave Chinner2ced19c2010-12-21 12:09:20 +11001238}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Christoph Hellwig873ff552010-01-13 22:17:57 +00001240/*
1241 * The bdstrat callback function for log bufs. This gives us a central
1242 * place to trap bufs in case we get hit by a log I/O error and need to
1243 * shutdown. Actually, in practice, even when we didn't get a log error,
1244 * we transition the iclogs to IOERROR state *after* flushing all existing
1245 * iclogs to disk. This is because we don't want anymore new transactions to be
1246 * started or completed afterwards.
1247 */
1248STATIC int
1249xlog_bdstrat(
1250 struct xfs_buf *bp)
1251{
1252 struct xlog_in_core *iclog;
1253
1254 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1255 if (iclog->ic_state & XLOG_STATE_IOERROR) {
1256 XFS_BUF_ERROR(bp, EIO);
1257 XFS_BUF_STALE(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001258 xfs_buf_ioend(bp, 0);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001259 /*
1260 * It would seem logical to return EIO here, but we rely on
1261 * the log state machine to propagate I/O errors instead of
1262 * doing it here.
1263 */
1264 return 0;
1265 }
1266
Christoph Hellwig873ff552010-01-13 22:17:57 +00001267 xfs_buf_iorequest(bp);
1268 return 0;
1269}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271/*
1272 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1273 * fashion. Previously, we should have moved the current iclog
1274 * ptr in the log to point to the next available iclog. This allows further
1275 * write to continue while this code syncs out an iclog ready to go.
1276 * Before an in-core log can be written out, the data section must be scanned
1277 * to save away the 1st word of each BBSIZE block into the header. We replace
1278 * it with the current cycle count. Each BBSIZE block is tagged with the
1279 * cycle count because there in an implicit assumption that drives will
1280 * guarantee that entire 512 byte blocks get written at once. In other words,
1281 * we can't have part of a 512 byte block written and part not written. By
1282 * tagging each block, we will know which blocks are valid when recovering
1283 * after an unclean shutdown.
1284 *
1285 * This routine is single threaded on the iclog. No other thread can be in
1286 * this routine with the same iclog. Changing contents of iclog can there-
1287 * fore be done without grabbing the state machine lock. Updating the global
1288 * log will require grabbing the lock though.
1289 *
1290 * The entire log manager uses a logical block numbering scheme. Only
1291 * log_sync (and then only bwrite()) know about the fact that the log may
1292 * not start with block zero on a given device. The log block start offset
1293 * is added immediately before calling bwrite().
1294 */
1295
David Chinnera8272ce2007-11-23 16:28:09 +11001296STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297xlog_sync(xlog_t *log,
1298 xlog_in_core_t *iclog)
1299{
1300 xfs_caddr_t dptr; /* pointer to byte sized element */
1301 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001302 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 uint count; /* byte count of bwrite */
1304 uint count_init; /* initial count before roundup */
1305 int roundoff; /* roundoff to BB or stripe */
1306 int split = 0; /* split write into two regions */
1307 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001308 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001311 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 /* Add for LR header */
1314 count_init = log->l_iclog_hsize + iclog->ic_offset;
1315
1316 /* Round out the log write size */
1317 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1318 /* we have a v2 stripe unit to use */
1319 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1320 } else {
1321 count = BBTOB(BTOBB(count_init));
1322 }
1323 roundoff = count - count_init;
1324 ASSERT(roundoff >= 0);
1325 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1326 roundoff < log->l_mp->m_sb.sb_logsunit)
1327 ||
1328 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1329 roundoff < BBTOB(1)));
1330
1331 /* move grant heads by roundoff in sync */
Dave Chinnera69ed032010-12-21 12:08:20 +11001332 xlog_grant_add_space(log, &log->l_grant_reserve_head, roundoff);
1333 xlog_grant_add_space(log, &log->l_grant_write_head, roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 /* put cycle number in every block */
1336 xlog_pack_data(log, iclog, roundoff);
1337
1338 /* real byte length */
1339 if (v2) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001340 iclog->ic_header.h_len =
1341 cpu_to_be32(iclog->ic_offset + roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001343 iclog->ic_header.h_len =
1344 cpu_to_be32(iclog->ic_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 }
1346
Nathan Scottf5faad72006-07-28 17:04:44 +10001347 bp = iclog->ic_bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001348 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1351
1352 /* Do we need to split this write into 2 parts? */
1353 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1354 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1355 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1356 iclog->ic_bwritecnt = 2; /* split into 2 writes */
1357 } else {
1358 iclog->ic_bwritecnt = 1;
1359 }
David Chinner511105b2007-05-24 15:21:57 +10001360 XFS_BUF_SET_COUNT(bp, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */
Nathan Scottf5faad72006-07-28 17:04:44 +10001362 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 XFS_BUF_BUSY(bp);
1364 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001365 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwig651701d2010-06-28 10:34:34 -04001366
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001367 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001368 bp->b_flags |= XBF_FUA;
1369
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001370 /*
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001371 * Flush the data device before flushing the log to make
1372 * sure all meta data written back from the AIL actually made
1373 * it to disk before stamping the new log tail LSN into the
1374 * log buffer. For an external log we need to issue the
1375 * flush explicitly, and unfortunately synchronously here;
1376 * for an internal log we can simply use the block layer
1377 * state machine for preflushes.
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001378 */
1379 if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
1380 xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001381 else
1382 bp->b_flags |= XBF_FLUSH;
Christoph Hellwiga27a2632011-06-16 12:02:23 +00001383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1386 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1387
1388 xlog_verify_iclog(log, iclog, count, B_TRUE);
1389
1390 /* account for log which doesn't start at block #0 */
1391 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1392 /*
1393 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1394 * is shutting down.
1395 */
1396 XFS_BUF_WRITE(bp);
1397
Christoph Hellwig873ff552010-01-13 22:17:57 +00001398 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
1400 XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001401 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001404 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
1406 XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
1407 (__psint_t)count), split);
1408 XFS_BUF_SET_FSPRIVATE(bp, iclog);
Nathan Scottf5faad72006-07-28 17:04:44 +10001409 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 XFS_BUF_BUSY(bp);
1411 XFS_BUF_ASYNC(bp);
Christoph Hellwig1d5ae5d2011-07-08 14:36:32 +02001412 bp->b_flags |= XBF_SYNCIO;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001413 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
Christoph Hellwige163cbd2011-07-08 14:36:36 +02001414 bp->b_flags |= XBF_FUA;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 dptr = XFS_BUF_PTR(bp);
1416 /*
1417 * Bump the cycle numbers at the start of each block
1418 * since this part of the buffer is at the start of
1419 * a new cycle. Watch out for the header magic number
1420 * case, though.
1421 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001422 for (i = 0; i < split; i += BBSIZE) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001423 be32_add_cpu((__be32 *)dptr, 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001424 if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001425 be32_add_cpu((__be32 *)dptr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 dptr += BBSIZE;
1427 }
1428
1429 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1430 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1431
Nathan Scottc41564b2006-03-29 08:55:14 +10001432 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1434 XFS_BUF_WRITE(bp);
Christoph Hellwig873ff552010-01-13 22:17:57 +00001435 if ((error = xlog_bdstrat(bp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
1437 bp, XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001438 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
1440 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001441 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442} /* xlog_sync */
1443
1444
1445/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001446 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 */
David Chinnera8272ce2007-11-23 16:28:09 +11001448STATIC void
Nathan Scottc41564b2006-03-29 08:55:14 +10001449xlog_dealloc_log(xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 int i;
1453
Dave Chinner71e330b2010-05-21 14:37:18 +10001454 xlog_cil_destroy(log);
1455
Dave Chinner44396472011-04-21 09:34:27 +00001456 /*
1457 * always need to ensure that the extra buffer does not point to memory
1458 * owned by another log buffer before we free it.
1459 */
1460 xfs_buf_set_empty(log->l_xbuf, log->l_iclog_size);
1461 xfs_buf_free(log->l_xbuf);
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 iclog = log->l_iclog;
1464 for (i=0; i<log->l_iclog_bufs; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 xfs_buf_free(iclog->ic_bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001467 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 iclog = next_iclog;
1469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 spinlock_destroy(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001473 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001474} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
1476/*
1477 * Update counters atomically now that memcpy is done.
1478 */
1479/* ARGSUSED */
1480static inline void
1481xlog_state_finish_copy(xlog_t *log,
1482 xlog_in_core_t *iclog,
1483 int record_cnt,
1484 int copy_bytes)
1485{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001486 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001488 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 iclog->ic_offset += copy_bytes;
1490
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001491 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492} /* xlog_state_finish_copy */
1493
1494
1495
1496
1497/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001498 * print out info relating to regions written which consume
1499 * the reservation
1500 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001501void
1502xlog_print_tic_res(
1503 struct xfs_mount *mp,
1504 struct xlog_ticket *ticket)
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001505{
1506 uint i;
1507 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1508
1509 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1510 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1511 "bformat",
1512 "bchunk",
1513 "efi_format",
1514 "efd_format",
1515 "iformat",
1516 "icore",
1517 "iext",
1518 "ibroot",
1519 "ilocal",
1520 "iattr_ext",
1521 "iattr_broot",
1522 "iattr_local",
1523 "qformat",
1524 "dquot",
1525 "quotaoff",
1526 "LR header",
1527 "unmount",
1528 "commit",
1529 "trans header"
1530 };
1531 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1532 "SETATTR_NOT_SIZE",
1533 "SETATTR_SIZE",
1534 "INACTIVE",
1535 "CREATE",
1536 "CREATE_TRUNC",
1537 "TRUNCATE_FILE",
1538 "REMOVE",
1539 "LINK",
1540 "RENAME",
1541 "MKDIR",
1542 "RMDIR",
1543 "SYMLINK",
1544 "SET_DMATTRS",
1545 "GROWFS",
1546 "STRAT_WRITE",
1547 "DIOSTRAT",
1548 "WRITE_SYNC",
1549 "WRITEID",
1550 "ADDAFORK",
1551 "ATTRINVAL",
1552 "ATRUNCATE",
1553 "ATTR_SET",
1554 "ATTR_RM",
1555 "ATTR_FLAG",
1556 "CLEAR_AGI_BUCKET",
1557 "QM_SBCHANGE",
1558 "DUMMY1",
1559 "DUMMY2",
1560 "QM_QUOTAOFF",
1561 "QM_DQALLOC",
1562 "QM_SETQLIM",
1563 "QM_DQCLUSTER",
1564 "QM_QINOCREATE",
1565 "QM_QUOTAOFF_END",
1566 "SB_UNIT",
1567 "FSYNC_TS",
1568 "GROWFSRT_ALLOC",
1569 "GROWFSRT_ZERO",
1570 "GROWFSRT_FREE",
1571 "SWAPEXT"
1572 };
1573
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001574 xfs_warn(mp,
1575 "xfs_log_write: reservation summary:\n"
1576 " trans type = %s (%u)\n"
1577 " unit res = %d bytes\n"
1578 " current res = %d bytes\n"
1579 " total reg = %u bytes (o/flow = %u bytes)\n"
1580 " ophdrs = %u (ophdr space = %u bytes)\n"
1581 " ophdr + reg = %u bytes\n"
1582 " num regions = %u\n",
1583 ((ticket->t_trans_type <= 0 ||
1584 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1585 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1586 ticket->t_trans_type,
1587 ticket->t_unit_res,
1588 ticket->t_curr_res,
1589 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1590 ticket->t_res_num_ophdrs, ophdr_spc,
1591 ticket->t_res_arr_sum +
1592 ticket->t_res_o_flow + ophdr_spc,
1593 ticket->t_res_num);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001594
1595 for (i = 0; i < ticket->t_res_num; i++) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001596 uint r_type = ticket->t_res_arr[i].r_type;
1597 xfs_warn(mp, "region[%u]: %s - %u bytes\n", i,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001598 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1599 "bad-rtype" : res_type_str[r_type-1]),
1600 ticket->t_res_arr[i].r_len);
1601 }
Dave Chinner169a7b02010-05-07 11:05:31 +10001602
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001603 xfs_alert_tag(mp, XFS_PTAG_LOGRES,
Dave Chinner169a7b02010-05-07 11:05:31 +10001604 "xfs_log_write: reservation ran out. Need to up reservation");
1605 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001606}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001607
1608/*
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001609 * Calculate the potential space needed by the log vector. Each region gets
1610 * its own xlog_op_header_t and may need to be double word aligned.
1611 */
1612static int
1613xlog_write_calc_vec_length(
1614 struct xlog_ticket *ticket,
Dave Chinner55b66332010-03-23 11:43:17 +11001615 struct xfs_log_vec *log_vector)
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001616{
Dave Chinner55b66332010-03-23 11:43:17 +11001617 struct xfs_log_vec *lv;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001618 int headers = 0;
1619 int len = 0;
1620 int i;
1621
1622 /* acct for start rec of xact */
1623 if (ticket->t_flags & XLOG_TIC_INITED)
1624 headers++;
1625
Dave Chinner55b66332010-03-23 11:43:17 +11001626 for (lv = log_vector; lv; lv = lv->lv_next) {
1627 headers += lv->lv_niovecs;
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001628
Dave Chinner55b66332010-03-23 11:43:17 +11001629 for (i = 0; i < lv->lv_niovecs; i++) {
1630 struct xfs_log_iovec *vecp = &lv->lv_iovecp[i];
1631
1632 len += vecp->i_len;
1633 xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
1634 }
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001635 }
1636
1637 ticket->t_res_num_ophdrs += headers;
1638 len += headers * sizeof(struct xlog_op_header);
1639
1640 return len;
1641}
1642
1643/*
1644 * If first write for transaction, insert start record We can't be trying to
1645 * commit if we are inited. We can't have any "partial_copy" if we are inited.
1646 */
1647static int
1648xlog_write_start_rec(
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001649 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001650 struct xlog_ticket *ticket)
1651{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001652 if (!(ticket->t_flags & XLOG_TIC_INITED))
1653 return 0;
1654
1655 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1656 ophdr->oh_clientid = ticket->t_clientid;
1657 ophdr->oh_len = 0;
1658 ophdr->oh_flags = XLOG_START_TRANS;
1659 ophdr->oh_res2 = 0;
1660
1661 ticket->t_flags &= ~XLOG_TIC_INITED;
1662
1663 return sizeof(struct xlog_op_header);
1664}
1665
1666static xlog_op_header_t *
1667xlog_write_setup_ophdr(
1668 struct log *log,
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001669 struct xlog_op_header *ophdr,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001670 struct xlog_ticket *ticket,
1671 uint flags)
1672{
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001673 ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
1674 ophdr->oh_clientid = ticket->t_clientid;
1675 ophdr->oh_res2 = 0;
1676
1677 /* are we copying a commit or unmount record? */
1678 ophdr->oh_flags = flags;
1679
1680 /*
1681 * We've seen logs corrupted with bad transaction client ids. This
1682 * makes sure that XFS doesn't generate them on. Turn this into an EIO
1683 * and shut down the filesystem.
1684 */
1685 switch (ophdr->oh_clientid) {
1686 case XFS_TRANSACTION:
1687 case XFS_VOLUME:
1688 case XFS_LOG:
1689 break;
1690 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001691 xfs_warn(log->l_mp,
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001692 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1693 ophdr->oh_clientid, ticket);
1694 return NULL;
1695 }
1696
1697 return ophdr;
1698}
1699
1700/*
1701 * Set up the parameters of the region copy into the log. This has
1702 * to handle region write split across multiple log buffers - this
1703 * state is kept external to this function so that this code can
1704 * can be written in an obvious, self documenting manner.
1705 */
1706static int
1707xlog_write_setup_copy(
1708 struct xlog_ticket *ticket,
1709 struct xlog_op_header *ophdr,
1710 int space_available,
1711 int space_required,
1712 int *copy_off,
1713 int *copy_len,
1714 int *last_was_partial_copy,
1715 int *bytes_consumed)
1716{
1717 int still_to_copy;
1718
1719 still_to_copy = space_required - *bytes_consumed;
1720 *copy_off = *bytes_consumed;
1721
1722 if (still_to_copy <= space_available) {
1723 /* write of region completes here */
1724 *copy_len = still_to_copy;
1725 ophdr->oh_len = cpu_to_be32(*copy_len);
1726 if (*last_was_partial_copy)
1727 ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
1728 *last_was_partial_copy = 0;
1729 *bytes_consumed = 0;
1730 return 0;
1731 }
1732
1733 /* partial write of region, needs extra log op header reservation */
1734 *copy_len = space_available;
1735 ophdr->oh_len = cpu_to_be32(*copy_len);
1736 ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
1737 if (*last_was_partial_copy)
1738 ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
1739 *bytes_consumed += *copy_len;
1740 (*last_was_partial_copy)++;
1741
1742 /* account for new log op header */
1743 ticket->t_curr_res -= sizeof(struct xlog_op_header);
1744 ticket->t_res_num_ophdrs++;
1745
1746 return sizeof(struct xlog_op_header);
1747}
1748
1749static int
1750xlog_write_copy_finish(
1751 struct log *log,
1752 struct xlog_in_core *iclog,
1753 uint flags,
1754 int *record_cnt,
1755 int *data_cnt,
1756 int *partial_copy,
1757 int *partial_copy_len,
1758 int log_offset,
1759 struct xlog_in_core **commit_iclog)
1760{
1761 if (*partial_copy) {
1762 /*
1763 * This iclog has already been marked WANT_SYNC by
1764 * xlog_state_get_iclog_space.
1765 */
1766 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1767 *record_cnt = 0;
1768 *data_cnt = 0;
1769 return xlog_state_release_iclog(log, iclog);
1770 }
1771
1772 *partial_copy = 0;
1773 *partial_copy_len = 0;
1774
1775 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
1776 /* no more space in this iclog - push it. */
1777 xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
1778 *record_cnt = 0;
1779 *data_cnt = 0;
1780
1781 spin_lock(&log->l_icloglock);
1782 xlog_state_want_sync(log, iclog);
1783 spin_unlock(&log->l_icloglock);
1784
1785 if (!commit_iclog)
1786 return xlog_state_release_iclog(log, iclog);
1787 ASSERT(flags & XLOG_COMMIT_TRANS);
1788 *commit_iclog = iclog;
1789 }
1790
1791 return 0;
1792}
1793
1794/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 * Write some region out to in-core log
1796 *
1797 * This will be called when writing externally provided regions or when
1798 * writing out a commit record for a given transaction.
1799 *
1800 * General algorithm:
1801 * 1. Find total length of this write. This may include adding to the
1802 * lengths passed in.
1803 * 2. Check whether we violate the tickets reservation.
1804 * 3. While writing to this iclog
1805 * A. Reserve as much space in this iclog as can get
1806 * B. If this is first write, save away start lsn
1807 * C. While writing this region:
1808 * 1. If first write of transaction, write start record
1809 * 2. Write log operation header (header per region)
1810 * 3. Find out if we can fit entire region into this iclog
1811 * 4. Potentially, verify destination memcpy ptr
1812 * 5. Memcpy (partial) region
1813 * 6. If partial copy, release iclog; otherwise, continue
1814 * copying more regions into current iclog
1815 * 4. Mark want sync bit (in simulation mode)
1816 * 5. Release iclog for potential flush to on-disk log.
1817 *
1818 * ERRORS:
1819 * 1. Panic if reservation is overrun. This should never happen since
1820 * reservation amounts are generated internal to the filesystem.
1821 * NOTES:
1822 * 1. Tickets are single threaded data structures.
1823 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1824 * syncing routine. When a single log_write region needs to span
1825 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1826 * on all log operation writes which don't contain the end of the
1827 * region. The XLOG_END_TRANS bit is used for the in-core log
1828 * operation which contains the end of the continued log_write region.
1829 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1830 * we don't really know exactly how much space will be used. As a result,
1831 * we don't update ic_offset until the end when we know exactly how many
1832 * bytes have been written out.
1833 */
Dave Chinner71e330b2010-05-21 14:37:18 +10001834int
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001835xlog_write(
Dave Chinner55b66332010-03-23 11:43:17 +11001836 struct log *log,
1837 struct xfs_log_vec *log_vector,
Christoph Hellwig35a8a722010-02-15 23:34:54 +00001838 struct xlog_ticket *ticket,
1839 xfs_lsn_t *start_lsn,
1840 struct xlog_in_core **commit_iclog,
1841 uint flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001843 struct xlog_in_core *iclog = NULL;
Dave Chinner55b66332010-03-23 11:43:17 +11001844 struct xfs_log_iovec *vecp;
1845 struct xfs_log_vec *lv;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001846 int len;
1847 int index;
1848 int partial_copy = 0;
1849 int partial_copy_len = 0;
1850 int contwr = 0;
1851 int record_cnt = 0;
1852 int data_cnt = 0;
1853 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001855 *start_lsn = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Dave Chinner55b66332010-03-23 11:43:17 +11001857 len = xlog_write_calc_vec_length(ticket, log_vector);
Dave Chinner71e330b2010-05-21 14:37:18 +10001858 if (log->l_cilp) {
1859 /*
1860 * Region headers and bytes are already accounted for.
1861 * We only need to take into account start records and
1862 * split regions in this function.
1863 */
1864 if (ticket->t_flags & XLOG_TIC_INITED)
1865 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1866
1867 /*
1868 * Commit record headers need to be accounted for. These
1869 * come in as separate writes so are easy to detect.
1870 */
1871 if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
1872 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1873 } else
1874 ticket->t_curr_res -= len;
1875
1876 if (ticket->t_curr_res < 0)
Dave Chinner55b66332010-03-23 11:43:17 +11001877 xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
Dave Chinner55b66332010-03-23 11:43:17 +11001879 index = 0;
1880 lv = log_vector;
1881 vecp = lv->lv_iovecp;
1882 while (lv && index < lv->lv_niovecs) {
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001883 void *ptr;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001884 int log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001886 error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
1887 &contwr, &log_offset);
1888 if (error)
1889 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001891 ASSERT(log_offset <= iclog->ic_size - 1);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001892 ptr = iclog->ic_datap + log_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001894 /* start_lsn is the first lsn written to. That's all we need. */
1895 if (!*start_lsn)
1896 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001898 /*
1899 * This loop writes out as many regions as can fit in the amount
1900 * of space which was allocated by xlog_state_get_iclog_space().
1901 */
Dave Chinner55b66332010-03-23 11:43:17 +11001902 while (lv && index < lv->lv_niovecs) {
1903 struct xfs_log_iovec *reg = &vecp[index];
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001904 struct xlog_op_header *ophdr;
1905 int start_rec_copy;
1906 int copy_len;
1907 int copy_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Dave Chinner55b66332010-03-23 11:43:17 +11001909 ASSERT(reg->i_len % sizeof(__int32_t) == 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001910 ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001912 start_rec_copy = xlog_write_start_rec(ptr, ticket);
1913 if (start_rec_copy) {
1914 record_cnt++;
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001915 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001916 start_rec_copy);
1917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001919 ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
1920 if (!ophdr)
1921 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001923 xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001924 sizeof(struct xlog_op_header));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001926 len += xlog_write_setup_copy(ticket, ophdr,
1927 iclog->ic_size-log_offset,
Dave Chinner55b66332010-03-23 11:43:17 +11001928 reg->i_len,
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001929 &copy_off, &copy_len,
1930 &partial_copy,
1931 &partial_copy_len);
1932 xlog_verify_dest_ptr(log, ptr);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001933
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001934 /* copy region */
1935 ASSERT(copy_len >= 0);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11001936 memcpy(ptr, reg->i_addr + copy_off, copy_len);
1937 xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len);
Dave Chinnerb5203cd2010-03-23 11:29:44 +11001938
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001939 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
1940 record_cnt++;
1941 data_cnt += contwr ? copy_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001943 error = xlog_write_copy_finish(log, iclog, flags,
1944 &record_cnt, &data_cnt,
1945 &partial_copy,
1946 &partial_copy_len,
1947 log_offset,
1948 commit_iclog);
1949 if (error)
1950 return error;
1951
1952 /*
1953 * if we had a partial copy, we need to get more iclog
1954 * space but we don't want to increment the region
1955 * index because there is still more is this region to
1956 * write.
1957 *
1958 * If we completed writing this region, and we flushed
1959 * the iclog (indicated by resetting of the record
1960 * count), then we also need to get more log space. If
1961 * this was the last record, though, we are done and
1962 * can just return.
1963 */
1964 if (partial_copy)
1965 break;
1966
Dave Chinner55b66332010-03-23 11:43:17 +11001967 if (++index == lv->lv_niovecs) {
1968 lv = lv->lv_next;
1969 index = 0;
1970 if (lv)
1971 vecp = lv->lv_iovecp;
1972 }
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001973 if (record_cnt == 0) {
Dave Chinner55b66332010-03-23 11:43:17 +11001974 if (!lv)
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001975 return 0;
1976 break;
1977 }
1978 }
1979 }
1980
1981 ASSERT(len == 0);
1982
1983 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1984 if (!commit_iclog)
1985 return xlog_state_release_iclog(log, iclog);
1986
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 ASSERT(flags & XLOG_COMMIT_TRANS);
1988 *commit_iclog = iclog;
1989 return 0;
Christoph Hellwig99428ad2010-03-23 11:35:45 +11001990}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
1992
1993/*****************************************************************************
1994 *
1995 * State Machine functions
1996 *
1997 *****************************************************************************
1998 */
1999
2000/* Clean iclogs starting from the head. This ordering must be
2001 * maintained, so an iclog doesn't become ACTIVE beyond one that
2002 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10002003 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 * iclog is trying to sync to disk.
2005 *
2006 * State Change: DIRTY -> ACTIVE
2007 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002008STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009xlog_state_clean_log(xlog_t *log)
2010{
2011 xlog_in_core_t *iclog;
2012 int changed = 0;
2013
2014 iclog = log->l_iclog;
2015 do {
2016 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2017 iclog->ic_state = XLOG_STATE_ACTIVE;
2018 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002019 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 /*
2021 * If the number of ops in this iclog indicate it just
2022 * contains the dummy transaction, we can
2023 * change state into IDLE (the second time around).
2024 * Otherwise we should change the state into
2025 * NEED a dummy.
2026 * We don't need to cover the dummy.
2027 */
2028 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002029 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2030 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 changed = 1;
2032 } else {
2033 /*
2034 * We have two dirty iclogs so start over
2035 * This could also be num of ops indicates
2036 * this is not the dummy going out.
2037 */
2038 changed = 2;
2039 }
2040 iclog->ic_header.h_num_logops = 0;
2041 memset(iclog->ic_header.h_cycle_data, 0,
2042 sizeof(iclog->ic_header.h_cycle_data));
2043 iclog->ic_header.h_lsn = 0;
2044 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2045 /* do nothing */;
2046 else
2047 break; /* stop cleaning */
2048 iclog = iclog->ic_next;
2049 } while (iclog != log->l_iclog);
2050
2051 /* log is locked when we are called */
2052 /*
2053 * Change state for the dummy log recording.
2054 * We usually go to NEED. But we go to NEED2 if the changed indicates
2055 * we are done writing the dummy record.
2056 * If we are done with the second dummy recored (DONE2), then
2057 * we go to IDLE.
2058 */
2059 if (changed) {
2060 switch (log->l_covered_state) {
2061 case XLOG_STATE_COVER_IDLE:
2062 case XLOG_STATE_COVER_NEED:
2063 case XLOG_STATE_COVER_NEED2:
2064 log->l_covered_state = XLOG_STATE_COVER_NEED;
2065 break;
2066
2067 case XLOG_STATE_COVER_DONE:
2068 if (changed == 1)
2069 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2070 else
2071 log->l_covered_state = XLOG_STATE_COVER_NEED;
2072 break;
2073
2074 case XLOG_STATE_COVER_DONE2:
2075 if (changed == 1)
2076 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2077 else
2078 log->l_covered_state = XLOG_STATE_COVER_NEED;
2079 break;
2080
2081 default:
2082 ASSERT(0);
2083 }
2084 }
2085} /* xlog_state_clean_log */
2086
2087STATIC xfs_lsn_t
2088xlog_get_lowest_lsn(
2089 xlog_t *log)
2090{
2091 xlog_in_core_t *lsn_log;
2092 xfs_lsn_t lowest_lsn, lsn;
2093
2094 lsn_log = log->l_iclog;
2095 lowest_lsn = 0;
2096 do {
2097 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002098 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 if ((lsn && !lowest_lsn) ||
2100 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2101 lowest_lsn = lsn;
2102 }
2103 }
2104 lsn_log = lsn_log->ic_next;
2105 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002106 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
2108
2109
2110STATIC void
2111xlog_state_do_callback(
2112 xlog_t *log,
2113 int aborted,
2114 xlog_in_core_t *ciclog)
2115{
2116 xlog_in_core_t *iclog;
2117 xlog_in_core_t *first_iclog; /* used to know when we've
2118 * processed all iclogs once */
2119 xfs_log_callback_t *cb, *cb_next;
2120 int flushcnt = 0;
2121 xfs_lsn_t lowest_lsn;
2122 int ioerrors; /* counter: iclogs with errors */
2123 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2124 int funcdidcallbacks; /* flag: function did callbacks */
2125 int repeats; /* for issuing console warnings if
2126 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002127 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002129 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 first_iclog = iclog = log->l_iclog;
2131 ioerrors = 0;
2132 funcdidcallbacks = 0;
2133 repeats = 0;
2134
2135 do {
2136 /*
2137 * Scan all iclogs starting with the one pointed to by the
2138 * log. Reset this starting point each time the log is
2139 * unlocked (during callbacks).
2140 *
2141 * Keep looping through iclogs until one full pass is made
2142 * without running any callbacks.
2143 */
2144 first_iclog = log->l_iclog;
2145 iclog = log->l_iclog;
2146 loopdidcallbacks = 0;
2147 repeats++;
2148
2149 do {
2150
2151 /* skip all iclogs in the ACTIVE & DIRTY states */
2152 if (iclog->ic_state &
2153 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2154 iclog = iclog->ic_next;
2155 continue;
2156 }
2157
2158 /*
2159 * Between marking a filesystem SHUTDOWN and stopping
2160 * the log, we do flush all iclogs to disk (if there
2161 * wasn't a log I/O error). So, we do want things to
2162 * go smoothly in case of just a SHUTDOWN w/o a
2163 * LOG_IO_ERROR.
2164 */
2165 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2166 /*
2167 * Can only perform callbacks in order. Since
2168 * this iclog is not in the DONE_SYNC/
2169 * DO_CALLBACK state, we skip the rest and
2170 * just try to clean up. If we set our iclog
2171 * to DO_CALLBACK, we will not process it when
2172 * we retry since a previous iclog is in the
2173 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002174 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 */
2176 if (!(iclog->ic_state &
2177 (XLOG_STATE_DONE_SYNC |
2178 XLOG_STATE_DO_CALLBACK))) {
2179 if (ciclog && (ciclog->ic_state ==
2180 XLOG_STATE_DONE_SYNC)) {
2181 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2182 }
2183 break;
2184 }
2185 /*
2186 * We now have an iclog that is in either the
2187 * DO_CALLBACK or DONE_SYNC states. The other
2188 * states (WANT_SYNC, SYNCING, or CALLBACK were
2189 * caught by the above if and are going to
2190 * clean (i.e. we aren't doing their callbacks)
2191 * see the above if.
2192 */
2193
2194 /*
2195 * We will do one more check here to see if we
2196 * have chased our tail around.
2197 */
2198
2199 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002200 if (lowest_lsn &&
2201 XFS_LSN_CMP(lowest_lsn,
Dave Chinner84f3c682010-12-03 22:11:29 +11002202 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 iclog = iclog->ic_next;
2204 continue; /* Leave this iclog for
2205 * another thread */
2206 }
2207
2208 iclog->ic_state = XLOG_STATE_CALLBACK;
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Dave Chinner84f3c682010-12-03 22:11:29 +11002211 /*
2212 * update the last_sync_lsn before we drop the
2213 * icloglock to ensure we are the only one that
2214 * can update it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 */
Dave Chinner84f3c682010-12-03 22:11:29 +11002216 ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
2217 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2218 atomic64_set(&log->l_last_sync_lsn,
2219 be64_to_cpu(iclog->ic_header.h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
Dave Chinner84f3c682010-12-03 22:11:29 +11002221 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 ioerrors++;
Dave Chinner84f3c682010-12-03 22:11:29 +11002223
2224 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
David Chinner114d23a2008-04-10 12:18:39 +10002226 /*
2227 * Keep processing entries in the callback list until
2228 * we come around and it is empty. We need to
2229 * atomically see that the list is empty and change the
2230 * state to DIRTY so that we don't miss any more
2231 * callbacks being added.
2232 */
2233 spin_lock(&iclog->ic_callback_lock);
2234 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002235 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 iclog->ic_callback_tail = &(iclog->ic_callback);
2237 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002238 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002241 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 cb_next = cb->cb_next;
2243 cb->cb_func(cb->cb_arg, aborted);
2244 }
David Chinner114d23a2008-04-10 12:18:39 +10002245 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 cb = iclog->ic_callback;
2247 }
2248
2249 loopdidcallbacks++;
2250 funcdidcallbacks++;
2251
David Chinner114d23a2008-04-10 12:18:39 +10002252 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002253 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002254 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2256 iclog->ic_state = XLOG_STATE_DIRTY;
2257
2258 /*
2259 * Transition from DIRTY to ACTIVE if applicable.
2260 * NOP if STATE_IOERROR.
2261 */
2262 xlog_state_clean_log(log);
2263
2264 /* wake up threads waiting in xfs_log_force() */
Dave Chinnereb40a872010-12-21 12:09:01 +11002265 wake_up_all(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
2267 iclog = iclog->ic_next;
2268 } while (first_iclog != iclog);
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002269
2270 if (repeats > 5000) {
2271 flushcnt += repeats;
2272 repeats = 0;
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002273 xfs_warn(log->l_mp,
Nathan Scotta3c6685e2006-09-28 11:02:14 +10002274 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002275 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
2277 } while (!ioerrors && loopdidcallbacks);
2278
2279 /*
2280 * make one last gasp attempt to see if iclogs are being left in
2281 * limbo..
2282 */
2283#ifdef DEBUG
2284 if (funcdidcallbacks) {
2285 first_iclog = iclog = log->l_iclog;
2286 do {
2287 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2288 /*
2289 * Terminate the loop if iclogs are found in states
2290 * which will cause other threads to clean up iclogs.
2291 *
2292 * SYNCING - i/o completion will go through logs
2293 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002294 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 * IOERROR - give up hope all ye who enter here
2296 */
2297 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2298 iclog->ic_state == XLOG_STATE_SYNCING ||
2299 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2300 iclog->ic_state == XLOG_STATE_IOERROR )
2301 break;
2302 iclog = iclog->ic_next;
2303 } while (first_iclog != iclog);
2304 }
2305#endif
2306
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002307 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2308 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002309 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002310
2311 if (wake)
Dave Chinnereb40a872010-12-21 12:09:01 +11002312 wake_up_all(&log->l_flush_wait);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002313}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
2315
2316/*
2317 * Finish transitioning this iclog to the dirty state.
2318 *
2319 * Make sure that we completely execute this routine only when this is
2320 * the last call to the iclog. There is a good chance that iclog flushes,
2321 * when we reach the end of the physical log, get turned into 2 separate
2322 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2323 * routine. By using the reference count bwritecnt, we guarantee that only
2324 * the second completion goes through.
2325 *
2326 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002327 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 */
David Chinnera8272ce2007-11-23 16:28:09 +11002329STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330xlog_state_done_syncing(
2331 xlog_in_core_t *iclog,
2332 int aborted)
2333{
2334 xlog_t *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002336 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337
2338 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2339 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002340 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2342
2343
2344 /*
2345 * If we got an error, either on the first buffer, or in the case of
2346 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2347 * and none should ever be attempted to be written to disk
2348 * again.
2349 */
2350 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2351 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002352 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 return;
2354 }
2355 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2356 }
2357
2358 /*
2359 * Someone could be sleeping prior to writing out the next
2360 * iclog buffer, we wake them all, one will get to do the
2361 * I/O, the others get to wait for the result.
2362 */
Dave Chinnereb40a872010-12-21 12:09:01 +11002363 wake_up_all(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002364 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2366} /* xlog_state_done_syncing */
2367
2368
2369/*
2370 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002371 * sleep. We wait on the flush queue on the head iclog as that should be
2372 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2373 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 *
2375 * The in-core logs are used in a circular fashion. They are not used
2376 * out-of-order even when an iclog past the head is free.
2377 *
2378 * return:
2379 * * log_offset where xlog_write() can start writing into the in-core
2380 * log's data space.
2381 * * in-core log pointer to which xlog_write() should write.
2382 * * boolean indicating this is a continued write to an in-core log.
2383 * If this is the last write, then the in-core log's offset field
2384 * needs to be incremented, depending on the amount of data which
2385 * is copied.
2386 */
David Chinnera8272ce2007-11-23 16:28:09 +11002387STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388xlog_state_get_iclog_space(xlog_t *log,
2389 int len,
2390 xlog_in_core_t **iclogp,
2391 xlog_ticket_t *ticket,
2392 int *continued_write,
2393 int *logoffsetp)
2394{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 int log_offset;
2396 xlog_rec_header_t *head;
2397 xlog_in_core_t *iclog;
2398 int error;
2399
2400restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002401 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002403 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 return XFS_ERROR(EIO);
2405 }
2406
2407 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002408 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002410
2411 /* Wait for log writes to have flushed */
Dave Chinnereb40a872010-12-21 12:09:01 +11002412 xlog_wait(&log->l_flush_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 goto restart;
2414 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 head = &iclog->ic_header;
2417
David Chinner155cc6b2008-03-06 13:44:14 +11002418 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 log_offset = iclog->ic_offset;
2420
2421 /* On the 1st write to an iclog, figure out lsn. This works
2422 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2423 * committing to. If the offset is set, that's how many blocks
2424 * must be written.
2425 */
2426 if (log_offset == 0) {
2427 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002428 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002429 log->l_iclog_hsize,
2430 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002431 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2432 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002433 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 ASSERT(log->l_curr_block >= 0);
2435 }
2436
2437 /* If there is enough room to write everything, then do it. Otherwise,
2438 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2439 * bit is on, so this will get flushed out. Don't update ic_offset
2440 * until you know exactly how many bytes get copied. Therefore, wait
2441 * until later to update ic_offset.
2442 *
2443 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2444 * can fit into remaining data section.
2445 */
2446 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2447 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2448
Dave Chinner49641f12008-07-11 17:43:55 +10002449 /*
2450 * If I'm the only one writing to this iclog, sync it to disk.
2451 * We need to do an atomic compare and decrement here to avoid
2452 * racing with concurrent atomic_dec_and_lock() calls in
2453 * xlog_state_release_iclog() when there is more than one
2454 * reference to the iclog.
2455 */
2456 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2457 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002458 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002459 error = xlog_state_release_iclog(log, iclog);
2460 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002461 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002463 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 }
2465 goto restart;
2466 }
2467
2468 /* Do we have enough room to write the full amount in the remainder
2469 * of this iclog? Or must we continue a write on the next iclog and
2470 * mark this iclog as completely taken? In the case where we switch
2471 * iclogs (to mark it taken), this particular iclog will release/sync
2472 * to disk in xlog_write().
2473 */
2474 if (len <= iclog->ic_size - iclog->ic_offset) {
2475 *continued_write = 0;
2476 iclog->ic_offset += len;
2477 } else {
2478 *continued_write = 1;
2479 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2480 }
2481 *iclogp = iclog;
2482
2483 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002484 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
2486 *logoffsetp = log_offset;
2487 return 0;
2488} /* xlog_state_get_iclog_space */
2489
2490/*
2491 * Atomically get the log space required for a log ticket.
2492 *
2493 * Once a ticket gets put onto the reserveq, it will only return after
2494 * the needed reservation is satisfied.
Dave Chinner3f16b982010-12-21 12:29:01 +11002495 *
2496 * This function is structured so that it has a lock free fast path. This is
2497 * necessary because every new transaction reservation will come through this
2498 * path. Hence any lock will be globally hot if we take it unconditionally on
2499 * every pass.
2500 *
2501 * As tickets are only ever moved on and off the reserveq under the
2502 * l_grant_reserve_lock, we only need to take that lock if we are going
2503 * to add the ticket to the queue and sleep. We can avoid taking the lock if the
2504 * ticket was never added to the reserveq because the t_queue list head will be
2505 * empty and we hold the only reference to it so it can safely be checked
2506 * unlocked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 */
2508STATIC int
2509xlog_grant_log_space(xlog_t *log,
2510 xlog_ticket_t *tic)
2511{
2512 int free_bytes;
2513 int need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
2515#ifdef DEBUG
2516 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2517 panic("grant Recovery problem");
2518#endif
2519
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002520 trace_xfs_log_grant_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Dave Chinner3f16b982010-12-21 12:29:01 +11002522 need_bytes = tic->t_unit_res;
2523 if (tic->t_flags & XFS_LOG_PERM_RESERV)
2524 need_bytes *= tic->t_ocnt;
2525
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 /* something is already sleeping; insert new transaction at end */
Dave Chinner3f16b982010-12-21 12:29:01 +11002527 if (!list_empty_careful(&log->l_reserveq)) {
2528 spin_lock(&log->l_grant_reserve_lock);
2529 /* recheck the queue now we are locked */
2530 if (list_empty(&log->l_reserveq)) {
2531 spin_unlock(&log->l_grant_reserve_lock);
2532 goto redo;
2533 }
Dave Chinner10547942010-12-21 12:02:25 +11002534 list_add_tail(&tic->t_queue, &log->l_reserveq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002535
2536 trace_xfs_log_grant_sleep1(log, tic);
2537
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 /*
2539 * Gotta check this before going to sleep, while we're
2540 * holding the grant lock.
2541 */
2542 if (XLOG_FORCED_SHUTDOWN(log))
2543 goto error_return;
2544
2545 XFS_STATS_INC(xs_sleep_logspace);
Dave Chinner3f16b982010-12-21 12:29:01 +11002546 xlog_wait(&tic->t_wait, &log->l_grant_reserve_lock);
Dave Chinnereb40a872010-12-21 12:09:01 +11002547
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 /*
2549 * If we got an error, and the filesystem is shutting down,
2550 * we'll catch it down below. So just continue...
2551 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002552 trace_xfs_log_grant_wake1(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554
2555redo:
2556 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002557 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
Dave Chinnera69ed032010-12-21 12:08:20 +11002559 free_bytes = xlog_space_left(log, &log->l_grant_reserve_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 if (free_bytes < need_bytes) {
Dave Chinner3f16b982010-12-21 12:29:01 +11002561 spin_lock(&log->l_grant_reserve_lock);
Dave Chinner10547942010-12-21 12:02:25 +11002562 if (list_empty(&tic->t_queue))
2563 list_add_tail(&tic->t_queue, &log->l_reserveq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002564
2565 trace_xfs_log_grant_sleep2(log, tic);
2566
Dave Chinner9d7fef72009-04-06 18:42:59 +02002567 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Dave Chinner3f16b982010-12-21 12:29:01 +11002570 xlog_grant_push_ail(log, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002571
Dave Chinner3f16b982010-12-21 12:29:01 +11002572 XFS_STATS_INC(xs_sleep_logspace);
2573 xlog_wait(&tic->t_wait, &log->l_grant_reserve_lock);
2574
2575 trace_xfs_log_grant_wake2(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 goto redo;
Dave Chinner10547942010-12-21 12:02:25 +11002577 }
2578
Dave Chinner3f16b982010-12-21 12:29:01 +11002579 if (!list_empty(&tic->t_queue)) {
2580 spin_lock(&log->l_grant_reserve_lock);
2581 list_del_init(&tic->t_queue);
2582 spin_unlock(&log->l_grant_reserve_lock);
2583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
2585 /* we've got enough space */
Dave Chinnera69ed032010-12-21 12:08:20 +11002586 xlog_grant_add_space(log, &log->l_grant_reserve_head, need_bytes);
2587 xlog_grant_add_space(log, &log->l_grant_write_head, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002588 trace_xfs_log_grant_exit(log, tic);
Dave Chinner3f336c62010-12-21 12:02:52 +11002589 xlog_verify_grant_tail(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 return 0;
2591
Dave Chinner3f16b982010-12-21 12:29:01 +11002592error_return_unlocked:
2593 spin_lock(&log->l_grant_reserve_lock);
2594error_return:
Dave Chinner10547942010-12-21 12:02:25 +11002595 list_del_init(&tic->t_queue);
Dave Chinner3f16b982010-12-21 12:29:01 +11002596 spin_unlock(&log->l_grant_reserve_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002597 trace_xfs_log_grant_error(log, tic);
2598
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 /*
2600 * If we are failing, make sure the ticket doesn't have any
2601 * current reservations. We don't want to add this back when
2602 * the ticket/transaction gets cancelled.
2603 */
2604 tic->t_curr_res = 0;
2605 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 return XFS_ERROR(EIO);
2607} /* xlog_grant_log_space */
2608
2609
2610/*
2611 * Replenish the byte reservation required by moving the grant write head.
2612 *
Dave Chinner3f16b982010-12-21 12:29:01 +11002613 * Similar to xlog_grant_log_space, the function is structured to have a lock
2614 * free fast path.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 */
2616STATIC int
2617xlog_regrant_write_log_space(xlog_t *log,
2618 xlog_ticket_t *tic)
2619{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 int free_bytes, need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621
2622 tic->t_curr_res = tic->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002623 xlog_tic_reset_res(tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
2625 if (tic->t_cnt > 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01002626 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
2628#ifdef DEBUG
2629 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2630 panic("regrant Recovery problem");
2631#endif
2632
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002633 trace_xfs_log_regrant_write_enter(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002635 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
2637 /* If there are other waiters on the queue then give them a
2638 * chance at logspace before us. Wake up the first waiters,
2639 * if we do not wake up all the waiters then go to sleep waiting
2640 * for more free space, otherwise try to get some space for
2641 * this transaction.
2642 */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002643 need_bytes = tic->t_unit_res;
Dave Chinner3f16b982010-12-21 12:29:01 +11002644 if (!list_empty_careful(&log->l_writeq)) {
Dave Chinner10547942010-12-21 12:02:25 +11002645 struct xlog_ticket *ntic;
Dave Chinner3f16b982010-12-21 12:29:01 +11002646
2647 spin_lock(&log->l_grant_write_lock);
Dave Chinnera69ed032010-12-21 12:08:20 +11002648 free_bytes = xlog_space_left(log, &log->l_grant_write_head);
Dave Chinner10547942010-12-21 12:02:25 +11002649 list_for_each_entry(ntic, &log->l_writeq, t_queue) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);
2651
2652 if (free_bytes < ntic->t_unit_res)
2653 break;
2654 free_bytes -= ntic->t_unit_res;
Dave Chinnereb40a872010-12-21 12:09:01 +11002655 wake_up(&ntic->t_wait);
Dave Chinner10547942010-12-21 12:02:25 +11002656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
Dave Chinner10547942010-12-21 12:02:25 +11002658 if (ntic != list_first_entry(&log->l_writeq,
2659 struct xlog_ticket, t_queue)) {
2660 if (list_empty(&tic->t_queue))
2661 list_add_tail(&tic->t_queue, &log->l_writeq);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002662 trace_xfs_log_regrant_write_sleep1(log, tic);
2663
Dave Chinner2ced19c2010-12-21 12:09:20 +11002664 xlog_grant_push_ail(log, need_bytes);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002665
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 XFS_STATS_INC(xs_sleep_logspace);
Dave Chinner3f16b982010-12-21 12:29:01 +11002667 xlog_wait(&tic->t_wait, &log->l_grant_write_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002668 trace_xfs_log_regrant_write_wake1(log, tic);
Dave Chinner3f16b982010-12-21 12:29:01 +11002669 } else
2670 spin_unlock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 }
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673redo:
2674 if (XLOG_FORCED_SHUTDOWN(log))
Dave Chinner3f16b982010-12-21 12:29:01 +11002675 goto error_return_unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676
Dave Chinnera69ed032010-12-21 12:08:20 +11002677 free_bytes = xlog_space_left(log, &log->l_grant_write_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 if (free_bytes < need_bytes) {
Dave Chinner3f16b982010-12-21 12:29:01 +11002679 spin_lock(&log->l_grant_write_lock);
Dave Chinner10547942010-12-21 12:02:25 +11002680 if (list_empty(&tic->t_queue))
2681 list_add_tail(&tic->t_queue, &log->l_writeq);
Dave Chinner3f16b982010-12-21 12:29:01 +11002682
2683 if (XLOG_FORCED_SHUTDOWN(log))
2684 goto error_return;
2685
Dave Chinner2ced19c2010-12-21 12:09:20 +11002686 xlog_grant_push_ail(log, need_bytes);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 XFS_STATS_INC(xs_sleep_logspace);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002689 trace_xfs_log_regrant_write_sleep2(log, tic);
Dave Chinner3f16b982010-12-21 12:29:01 +11002690 xlog_wait(&tic->t_wait, &log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002692 trace_xfs_log_regrant_write_wake2(log, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 goto redo;
Dave Chinner10547942010-12-21 12:02:25 +11002694 }
2695
Dave Chinner3f16b982010-12-21 12:29:01 +11002696 if (!list_empty(&tic->t_queue)) {
2697 spin_lock(&log->l_grant_write_lock);
2698 list_del_init(&tic->t_queue);
2699 spin_unlock(&log->l_grant_write_lock);
2700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002702 /* we've got enough space */
Dave Chinnera69ed032010-12-21 12:08:20 +11002703 xlog_grant_add_space(log, &log->l_grant_write_head, need_bytes);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002704 trace_xfs_log_regrant_write_exit(log, tic);
Dave Chinner3f336c62010-12-21 12:02:52 +11002705 xlog_verify_grant_tail(log);
Jesper Juhl014c2542006-01-15 02:37:08 +01002706 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
2708
Dave Chinner3f16b982010-12-21 12:29:01 +11002709 error_return_unlocked:
2710 spin_lock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 error_return:
Dave Chinner10547942010-12-21 12:02:25 +11002712 list_del_init(&tic->t_queue);
Dave Chinner3f16b982010-12-21 12:29:01 +11002713 spin_unlock(&log->l_grant_write_lock);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002714 trace_xfs_log_regrant_write_error(log, tic);
2715
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 /*
2717 * If we are failing, make sure the ticket doesn't have any
2718 * current reservations. We don't want to add this back when
2719 * the ticket/transaction gets cancelled.
2720 */
2721 tic->t_curr_res = 0;
2722 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 return XFS_ERROR(EIO);
2724} /* xlog_regrant_write_log_space */
2725
2726
2727/* The first cnt-1 times through here we don't need to
2728 * move the grant write head because the permanent
2729 * reservation has reserved cnt times the unit amount.
2730 * Release part of current permanent unit reservation and
2731 * reset current reservation to be one units worth. Also
2732 * move grant reservation head forward.
2733 */
2734STATIC void
2735xlog_regrant_reserve_log_space(xlog_t *log,
2736 xlog_ticket_t *ticket)
2737{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002738 trace_xfs_log_regrant_reserve_enter(log, ticket);
2739
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 if (ticket->t_cnt > 0)
2741 ticket->t_cnt--;
2742
Dave Chinnera69ed032010-12-21 12:08:20 +11002743 xlog_grant_sub_space(log, &log->l_grant_reserve_head,
2744 ticket->t_curr_res);
2745 xlog_grant_sub_space(log, &log->l_grant_write_head,
2746 ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002748 xlog_tic_reset_res(ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002749
2750 trace_xfs_log_regrant_reserve_sub(log, ticket);
2751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 /* just return if we still have some of the pre-reserved space */
Dave Chinnerd0eb2f32010-12-21 12:29:14 +11002753 if (ticket->t_cnt > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
Dave Chinnera69ed032010-12-21 12:08:20 +11002756 xlog_grant_add_space(log, &log->l_grant_reserve_head,
2757 ticket->t_unit_res);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002758
2759 trace_xfs_log_regrant_reserve_exit(log, ticket);
2760
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002762 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763} /* xlog_regrant_reserve_log_space */
2764
2765
2766/*
2767 * Give back the space left from a reservation.
2768 *
2769 * All the information we need to make a correct determination of space left
2770 * is present. For non-permanent reservations, things are quite easy. The
2771 * count should have been decremented to zero. We only need to deal with the
2772 * space remaining in the current reservation part of the ticket. If the
2773 * ticket contains a permanent reservation, there may be left over space which
2774 * needs to be released. A count of N means that N-1 refills of the current
2775 * reservation can be done before we need to ask for more space. The first
2776 * one goes to fill up the first current reservation. Once we run out of
2777 * space, the count will stay at zero and the only space remaining will be
2778 * in the current reservation field.
2779 */
2780STATIC void
2781xlog_ungrant_log_space(xlog_t *log,
2782 xlog_ticket_t *ticket)
2783{
Dave Chinner663e4962010-12-21 12:06:05 +11002784 int bytes;
2785
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (ticket->t_cnt > 0)
2787 ticket->t_cnt--;
2788
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002789 trace_xfs_log_ungrant_enter(log, ticket);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002790 trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Dave Chinner663e4962010-12-21 12:06:05 +11002792 /*
2793 * If this is a permanent reservation ticket, we may be able to free
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794 * up more space based on the remaining count.
2795 */
Dave Chinner663e4962010-12-21 12:06:05 +11002796 bytes = ticket->t_curr_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (ticket->t_cnt > 0) {
2798 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Dave Chinner663e4962010-12-21 12:06:05 +11002799 bytes += ticket->t_unit_res*ticket->t_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 }
2801
Dave Chinnera69ed032010-12-21 12:08:20 +11002802 xlog_grant_sub_space(log, &log->l_grant_reserve_head, bytes);
2803 xlog_grant_sub_space(log, &log->l_grant_write_head, bytes);
Dave Chinner663e4962010-12-21 12:06:05 +11002804
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00002805 trace_xfs_log_ungrant_exit(log, ticket);
2806
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 xfs_log_move_tail(log->l_mp, 1);
2808} /* xlog_ungrant_log_space */
2809
2810
2811/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 * Flush iclog to disk if this is the last reference to the given iclog and
2813 * the WANT_SYNC bit is set.
2814 *
2815 * When this function is entered, the iclog is not necessarily in the
2816 * WANT_SYNC state. It may be sitting around waiting to get filled.
2817 *
2818 *
2819 */
David Chinnera8272ce2007-11-23 16:28:09 +11002820STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11002821xlog_state_release_iclog(
2822 xlog_t *log,
2823 xlog_in_core_t *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 int sync = 0; /* do we sync? */
2826
David Chinner155cc6b2008-03-06 13:44:14 +11002827 if (iclog->ic_state & XLOG_STATE_IOERROR)
2828 return XFS_ERROR(EIO);
2829
2830 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
2831 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
2832 return 0;
2833
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002835 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 return XFS_ERROR(EIO);
2837 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
2839 iclog->ic_state == XLOG_STATE_WANT_SYNC);
2840
David Chinner155cc6b2008-03-06 13:44:14 +11002841 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11002842 /* update tail before writing to iclog */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11002843 xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 sync++;
2845 iclog->ic_state = XLOG_STATE_SYNCING;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11002846 iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
2847 xlog_verify_tail_lsn(log, iclog, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 /* cycle incremented when incrementing curr_block */
2849 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002850 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
2852 /*
2853 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10002854 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2856 * this iclog has consistent data, so we ignore IOERROR
2857 * flags after this point.
2858 */
David Chinnerb5893342008-03-06 13:44:06 +11002859 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002861 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862} /* xlog_state_release_iclog */
2863
2864
2865/*
2866 * This routine will mark the current iclog in the ring as WANT_SYNC
2867 * and move the current iclog pointer to the next iclog in the ring.
2868 * When this routine is called from xlog_state_get_iclog_space(), the
2869 * exact size of the iclog has not yet been determined. All we know is
2870 * that every data block. We have run out of space in this log record.
2871 */
2872STATIC void
2873xlog_state_switch_iclogs(xlog_t *log,
2874 xlog_in_core_t *iclog,
2875 int eventual_size)
2876{
2877 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2878 if (!eventual_size)
2879 eventual_size = iclog->ic_offset;
2880 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002881 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 log->l_prev_block = log->l_curr_block;
2883 log->l_prev_cycle = log->l_curr_cycle;
2884
2885 /* roll log?: ic_offset changed later */
2886 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
2887
2888 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11002889 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 log->l_mp->m_sb.sb_logsunit > 1) {
2891 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
2892 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
2893 }
2894
2895 if (log->l_curr_block >= log->l_logBBsize) {
2896 log->l_curr_cycle++;
2897 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
2898 log->l_curr_cycle++;
2899 log->l_curr_block -= log->l_logBBsize;
2900 ASSERT(log->l_curr_block >= 0);
2901 }
2902 ASSERT(iclog == log->l_iclog);
2903 log->l_iclog = iclog->ic_next;
2904} /* xlog_state_switch_iclogs */
2905
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906/*
2907 * Write out all data in the in-core log as of this exact moment in time.
2908 *
2909 * Data may be written to the in-core log during this call. However,
2910 * we don't guarantee this data will be written out. A change from past
2911 * implementation means this routine will *not* write out zero length LRs.
2912 *
2913 * Basically, we try and perform an intelligent scan of the in-core logs.
2914 * If we determine there is no flushable data, we just return. There is no
2915 * flushable data if:
2916 *
2917 * 1. the current iclog is active and has no data; the previous iclog
2918 * is in the active or dirty state.
2919 * 2. the current iclog is drity, and the previous iclog is in the
2920 * active or dirty state.
2921 *
David Chinner12017fa2008-08-13 16:34:31 +10002922 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 *
2924 * 1. the current iclog is not in the active nor dirty state.
2925 * 2. the current iclog dirty, and the previous iclog is not in the
2926 * active nor dirty state.
2927 * 3. the current iclog is active, and there is another thread writing
2928 * to this particular iclog.
2929 * 4. a) the current iclog is active and has no other writers
2930 * b) when we return from flushing out this iclog, it is still
2931 * not in the active nor dirty state.
2932 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002933int
2934_xfs_log_force(
2935 struct xfs_mount *mp,
2936 uint flags,
2937 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002939 struct log *log = mp->m_log;
2940 struct xlog_in_core *iclog;
2941 xfs_lsn_t lsn;
2942
2943 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
Dave Chinnera44f13ed2010-08-24 11:40:03 +10002945 if (log->l_cilp)
2946 xlog_cil_force(log);
Dave Chinner71e330b2010-05-21 14:37:18 +10002947
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002948 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949
2950 iclog = log->l_iclog;
2951 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002952 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 return XFS_ERROR(EIO);
2954 }
2955
2956 /* If the head iclog is not active nor dirty, we just attach
2957 * ourselves to the head and go to sleep.
2958 */
2959 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2960 iclog->ic_state == XLOG_STATE_DIRTY) {
2961 /*
2962 * If the head is dirty or (active and empty), then
2963 * we need to look at the previous iclog. If the previous
2964 * iclog is active or dirty we are done. There is nothing
2965 * to sync out. Otherwise, we attach ourselves to the
2966 * previous iclog and go to sleep.
2967 */
2968 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11002969 (atomic_read(&iclog->ic_refcnt) == 0
2970 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 iclog = iclog->ic_prev;
2972 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2973 iclog->ic_state == XLOG_STATE_DIRTY)
2974 goto no_sleep;
2975 else
2976 goto maybe_sleep;
2977 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11002978 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 /* We are the only one with access to this
2980 * iclog. Flush it out now. There should
2981 * be a roundoff of zero to show that someone
2982 * has already taken care of the roundoff from
2983 * the previous sync.
2984 */
David Chinner155cc6b2008-03-06 13:44:14 +11002985 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002986 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002988 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
2990 if (xlog_state_release_iclog(log, iclog))
2991 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00002992
2993 if (log_flushed)
2994 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002995 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002996 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 iclog->ic_state != XLOG_STATE_DIRTY)
2998 goto maybe_sleep;
2999 else
3000 goto no_sleep;
3001 } else {
3002 /* Someone else is writing to this iclog.
3003 * Use its call to flush out the data. However,
3004 * the other thread may not force out this LR,
3005 * so we mark it WANT_SYNC.
3006 */
3007 xlog_state_switch_iclogs(log, iclog, 0);
3008 goto maybe_sleep;
3009 }
3010 }
3011 }
3012
3013 /* By the time we come around again, the iclog could've been filled
3014 * which would give it another lsn. If we have a new lsn, just
3015 * return because the relevant data has been flushed.
3016 */
3017maybe_sleep:
3018 if (flags & XFS_LOG_SYNC) {
3019 /*
3020 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003021 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 * Then we check again after waking up, in case our
3023 * sleep was disturbed by a bad news.
3024 */
3025 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003026 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 return XFS_ERROR(EIO);
3028 }
3029 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003030 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 /*
3032 * No need to grab the log lock here since we're
3033 * only deciding whether or not to return EIO
3034 * and the memory read should be atomic.
3035 */
3036 if (iclog->ic_state & XLOG_STATE_IOERROR)
3037 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003038 if (log_flushed)
3039 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 } else {
3041
3042no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003043 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 }
3045 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003046}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
3048/*
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003049 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3050 * about errors or whether the log was flushed or not. This is the normal
3051 * interface to use when trying to unpin items or move the log forward.
3052 */
3053void
3054xfs_log_force(
3055 xfs_mount_t *mp,
3056 uint flags)
3057{
3058 int error;
3059
3060 error = _xfs_log_force(mp, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003061 if (error)
3062 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003063}
3064
3065/*
3066 * Force the in-core log to disk for a specific LSN.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 *
3068 * Find in-core log with lsn.
3069 * If it is in the DIRTY state, just return.
3070 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3071 * state and go to sleep or return.
3072 * If it is in any other state, go to sleep or return.
3073 *
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003074 * Synchronous forces are implemented with a signal variable. All callers
3075 * to force a given lsn to disk will wait on a the sv attached to the
3076 * specific in-core log. When given in-core log finally completes its
3077 * write to disk, that thread will wake up all threads waiting on the
3078 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003080int
3081_xfs_log_force_lsn(
3082 struct xfs_mount *mp,
3083 xfs_lsn_t lsn,
3084 uint flags,
3085 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086{
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003087 struct log *log = mp->m_log;
3088 struct xlog_in_core *iclog;
3089 int already_slept = 0;
3090
3091 ASSERT(lsn != 0);
3092
3093 XFS_STATS_INC(xs_log_force);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094
Dave Chinner71e330b2010-05-21 14:37:18 +10003095 if (log->l_cilp) {
Dave Chinnera44f13ed2010-08-24 11:40:03 +10003096 lsn = xlog_cil_force_lsn(log, lsn);
Dave Chinner71e330b2010-05-21 14:37:18 +10003097 if (lsn == NULLCOMMITLSN)
3098 return 0;
3099 }
3100
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101try_again:
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003102 spin_lock(&log->l_icloglock);
3103 iclog = log->l_iclog;
3104 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003105 spin_unlock(&log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003106 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 }
3108
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003109 do {
3110 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3111 iclog = iclog->ic_next;
3112 continue;
3113 }
3114
3115 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3116 spin_unlock(&log->l_icloglock);
3117 return 0;
3118 }
3119
3120 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3121 /*
3122 * We sleep here if we haven't already slept (e.g.
3123 * this is the first time we've looked at the correct
3124 * iclog buf) and the buffer before us is going to
3125 * be sync'ed. The reason for this is that if we
3126 * are doing sync transactions here, by waiting for
3127 * the previous I/O to complete, we can allow a few
3128 * more transactions into this iclog before we close
3129 * it down.
3130 *
3131 * Otherwise, we mark the buffer WANT_SYNC, and bump
3132 * up the refcnt so we can release the log (which
3133 * drops the ref count). The state switch keeps new
3134 * transaction commits from using this buffer. When
3135 * the current commits finish writing into the buffer,
3136 * the refcount will drop to zero and the buffer will
3137 * go out then.
3138 */
3139 if (!already_slept &&
3140 (iclog->ic_prev->ic_state &
3141 (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
3142 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3143
3144 XFS_STATS_INC(xs_log_force_sleep);
3145
Dave Chinnereb40a872010-12-21 12:09:01 +11003146 xlog_wait(&iclog->ic_prev->ic_write_wait,
3147 &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003148 if (log_flushed)
3149 *log_flushed = 1;
3150 already_slept = 1;
3151 goto try_again;
3152 }
David Chinner155cc6b2008-03-06 13:44:14 +11003153 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003155 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 if (xlog_state_release_iclog(log, iclog))
3157 return XFS_ERROR(EIO);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003158 if (log_flushed)
3159 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003160 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003163 if ((flags & XFS_LOG_SYNC) && /* sleep */
3164 !(iclog->ic_state &
3165 (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3166 /*
3167 * Don't wait on completion if we know that we've
3168 * gotten a log write error.
3169 */
3170 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3171 spin_unlock(&log->l_icloglock);
3172 return XFS_ERROR(EIO);
3173 }
3174 XFS_STATS_INC(xs_log_force_sleep);
Dave Chinnereb40a872010-12-21 12:09:01 +11003175 xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003176 /*
3177 * No need to grab the log lock here since we're
3178 * only deciding whether or not to return EIO
3179 * and the memory read should be atomic.
3180 */
3181 if (iclog->ic_state & XLOG_STATE_IOERROR)
3182 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003183
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003184 if (log_flushed)
3185 *log_flushed = 1;
3186 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003187 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 }
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003189
3190 return 0;
3191 } while (iclog != log->l_iclog);
3192
3193 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 return 0;
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003195}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003197/*
3198 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3199 * about errors or whether the log was flushed or not. This is the normal
3200 * interface to use when trying to unpin items or move the log forward.
3201 */
3202void
3203xfs_log_force_lsn(
3204 xfs_mount_t *mp,
3205 xfs_lsn_t lsn,
3206 uint flags)
3207{
3208 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003209
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003210 error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003211 if (error)
3212 xfs_warn(mp, "%s: error %d returned.", __func__, error);
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003213}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
3215/*
3216 * Called when we want to mark the current iclog as being ready to sync to
3217 * disk.
3218 */
David Chinnera8272ce2007-11-23 16:28:09 +11003219STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3221{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003222 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223
3224 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3225 xlog_state_switch_iclogs(log, iclog, 0);
3226 } else {
3227 ASSERT(iclog->ic_state &
3228 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3229 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003230}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231
3232
3233/*****************************************************************************
3234 *
3235 * TICKET functions
3236 *
3237 *****************************************************************************
3238 */
3239
3240/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003241 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003243void
3244xfs_log_ticket_put(
3245 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003247 ASSERT(atomic_read(&ticket->t_ref) > 0);
Dave Chinnereb40a872010-12-21 12:09:01 +11003248 if (atomic_dec_and_test(&ticket->t_ref))
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003249 kmem_zone_free(xfs_log_ticket_zone, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003250}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003252xlog_ticket_t *
3253xfs_log_ticket_get(
3254 xlog_ticket_t *ticket)
3255{
3256 ASSERT(atomic_read(&ticket->t_ref) > 0);
3257 atomic_inc(&ticket->t_ref);
3258 return ticket;
3259}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260
3261/*
David Chinnereb01c9c2008-04-10 12:18:46 +10003262 * Allocate and initialise a new log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 */
Dave Chinner71e330b2010-05-21 14:37:18 +10003264xlog_ticket_t *
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003265xlog_ticket_alloc(
3266 struct log *log,
3267 int unit_bytes,
3268 int cnt,
3269 char client,
Dave Chinner3383ca52010-05-07 11:04:17 +10003270 uint xflags,
3271 int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272{
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003273 struct xlog_ticket *tic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 uint num_headers;
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003275 int iclog_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276
Dave Chinner3383ca52010-05-07 11:04:17 +10003277 tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
David Chinnereb01c9c2008-04-10 12:18:46 +10003278 if (!tic)
3279 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
3281 /*
3282 * Permanent reservations have up to 'cnt'-1 active log operations
3283 * in the log. A unit in this case is the amount of space for one
3284 * of these log operations. Normal reservations have a cnt of 1
3285 * and their unit amount is the total amount of space required.
3286 *
3287 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003288 * which occupy space in the on-disk log.
3289 *
3290 * Normal form of a transaction is:
3291 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3292 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3293 *
3294 * We need to account for all the leadup data and trailer data
3295 * around the transaction data.
3296 * And then we need to account for the worst case in terms of using
3297 * more space.
3298 * The worst case will happen if:
3299 * - the placement of the transaction happens to be such that the
3300 * roundoff is at its maximum
3301 * - the transaction data is synced before the commit record is synced
3302 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3303 * Therefore the commit record is in its own Log Record.
3304 * This can happen as the commit record is called with its
3305 * own region to xlog_write().
3306 * This then means that in the worst case, roundoff can happen for
3307 * the commit-rec as well.
3308 * The commit-rec is smaller than padding in this scenario and so it is
3309 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 */
3311
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003312 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003314 unit_bytes += sizeof(xfs_trans_header_t);
3315
3316 /* for start-rec */
3317 unit_bytes += sizeof(xlog_op_header_t);
3318
Dave Chinner9b9fc2b72010-03-23 11:21:11 +11003319 /*
3320 * for LR headers - the space for data in an iclog is the size minus
3321 * the space used for the headers. If we use the iclog size, then we
3322 * undercalculate the number of headers required.
3323 *
3324 * Furthermore - the addition of op headers for split-recs might
3325 * increase the space required enough to require more log and op
3326 * headers, so take that into account too.
3327 *
3328 * IMPORTANT: This reservation makes the assumption that if this
3329 * transaction is the first in an iclog and hence has the LR headers
3330 * accounted to it, then the remaining space in the iclog is
3331 * exclusively for this transaction. i.e. if the transaction is larger
3332 * than the iclog, it will be the only thing in that iclog.
3333 * Fundamentally, this means we must pass the entire log vector to
3334 * xlog_write to guarantee this.
3335 */
3336 iclog_space = log->l_iclog_size - log->l_iclog_hsize;
3337 num_headers = howmany(unit_bytes, iclog_space);
3338
3339 /* for split-recs - ophdrs added when data split over LRs */
3340 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3341
3342 /* add extra header reservations if we overrun */
3343 while (!num_headers ||
3344 howmany(unit_bytes, iclog_space) > num_headers) {
3345 unit_bytes += sizeof(xlog_op_header_t);
3346 num_headers++;
3347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 unit_bytes += log->l_iclog_hsize * num_headers;
3349
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003350 /* for commit-rec LR header - note: padding will subsume the ophdr */
3351 unit_bytes += log->l_iclog_hsize;
3352
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003353 /* for roundoff padding for transaction data and one for commit record */
Eric Sandeen62118702008-03-06 13:44:28 +11003354 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003355 log->l_mp->m_sb.sb_logsunit > 1) {
3356 /* log su roundoff */
3357 unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
3358 } else {
3359 /* BB roundoff */
3360 unit_bytes += 2*BBSIZE;
3361 }
3362
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003363 atomic_set(&tic->t_ref, 1);
Dave Chinner10547942010-12-21 12:02:25 +11003364 INIT_LIST_HEAD(&tic->t_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365 tic->t_unit_res = unit_bytes;
3366 tic->t_curr_res = unit_bytes;
3367 tic->t_cnt = cnt;
3368 tic->t_ocnt = cnt;
Dave Chinnerf9837102010-04-14 15:47:55 +10003369 tic->t_tid = random32();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 tic->t_clientid = client;
3371 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003372 tic->t_trans_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373 if (xflags & XFS_LOG_PERM_RESERV)
3374 tic->t_flags |= XLOG_TIC_PERM_RESERV;
Dave Chinnereb40a872010-12-21 12:09:01 +11003375 init_waitqueue_head(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376
Christoph Hellwig0adba532007-08-30 17:21:46 +10003377 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003378
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003380}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382
3383/******************************************************************************
3384 *
3385 * Log debug routines
3386 *
3387 ******************************************************************************
3388 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003389#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390/*
3391 * Make sure that the destination ptr is within the valid data region of
3392 * one of the iclogs. This uses backup pointers stored in a different
3393 * part of the log in case we trash the log structure.
3394 */
3395void
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003396xlog_verify_dest_ptr(
3397 struct log *log,
3398 char *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399{
3400 int i;
3401 int good_ptr = 0;
3402
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003403 for (i = 0; i < log->l_iclog_bufs; i++) {
3404 if (ptr >= log->l_iclog_bak[i] &&
3405 ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 good_ptr++;
3407 }
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003408
3409 if (!good_ptr)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003410 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Christoph Hellwige6b1f272010-03-23 11:47:38 +11003411}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003413/*
3414 * Check to make sure the grant write head didn't just over lap the tail. If
3415 * the cycles are the same, we can't be overlapping. Otherwise, make sure that
3416 * the cycles differ by exactly one and check the byte count.
3417 *
3418 * This check is run unlocked, so can give false positives. Rather than assert
3419 * on failures, use a warn-once flag and a panic tag to allow the admin to
3420 * determine if they want to panic the machine when such an error occurs. For
3421 * debug kernels this will have the same effect as using an assert but, unlinke
3422 * an assert, it can be turned off at runtime.
3423 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424STATIC void
Dave Chinner3f336c62010-12-21 12:02:52 +11003425xlog_verify_grant_tail(
3426 struct log *log)
3427{
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003428 int tail_cycle, tail_blocks;
Dave Chinnera69ed032010-12-21 12:08:20 +11003429 int cycle, space;
Dave Chinner3f336c62010-12-21 12:02:52 +11003430
Dave Chinnera69ed032010-12-21 12:08:20 +11003431 xlog_crack_grant_head(&log->l_grant_write_head, &cycle, &space);
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11003432 xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
3433 if (tail_cycle != cycle) {
Dave Chinnerda8a1a42011-04-08 12:45:07 +10003434 if (cycle - 1 != tail_cycle &&
3435 !(log->l_flags & XLOG_TAIL_WARN)) {
3436 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3437 "%s: cycle - 1 != tail_cycle", __func__);
3438 log->l_flags |= XLOG_TAIL_WARN;
3439 }
3440
3441 if (space > BBTOB(tail_blocks) &&
3442 !(log->l_flags & XLOG_TAIL_WARN)) {
3443 xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
3444 "%s: space > BBTOB(tail_blocks)", __func__);
3445 log->l_flags |= XLOG_TAIL_WARN;
3446 }
Dave Chinner3f336c62010-12-21 12:02:52 +11003447 }
3448}
3449
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450/* check if it will fit */
3451STATIC void
3452xlog_verify_tail_lsn(xlog_t *log,
3453 xlog_in_core_t *iclog,
3454 xfs_lsn_t tail_lsn)
3455{
3456 int blocks;
3457
3458 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3459 blocks =
3460 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3461 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003462 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 } else {
3464 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3465
3466 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003467 xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468
3469 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3470 if (blocks < BTOBB(iclog->ic_offset) + 1)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003471 xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 }
3473} /* xlog_verify_tail_lsn */
3474
3475/*
3476 * Perform a number of checks on the iclog before writing to disk.
3477 *
3478 * 1. Make sure the iclogs are still circular
3479 * 2. Make sure we have a good magic number
3480 * 3. Make sure we don't have magic numbers in the data
3481 * 4. Check fields of each log operation header for:
3482 * A. Valid client identifier
3483 * B. tid ptr value falls in valid ptr space (user space code)
3484 * C. Length in log record header is correct according to the
3485 * individual operation headers within record.
3486 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3487 * log, check the preceding blocks of the physical log to make sure all
3488 * the cycle numbers agree with the current cycle number.
3489 */
3490STATIC void
3491xlog_verify_iclog(xlog_t *log,
3492 xlog_in_core_t *iclog,
3493 int count,
3494 boolean_t syncing)
3495{
3496 xlog_op_header_t *ophead;
3497 xlog_in_core_t *icptr;
3498 xlog_in_core_2_t *xhdr;
3499 xfs_caddr_t ptr;
3500 xfs_caddr_t base_ptr;
3501 __psint_t field_offset;
3502 __uint8_t clientid;
3503 int len, i, j, k, op_len;
3504 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505
3506 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003507 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 icptr = log->l_iclog;
3509 for (i=0; i < log->l_iclog_bufs; i++) {
Christoph Hellwig4b809162007-08-16 15:37:36 +10003510 if (icptr == NULL)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003511 xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 icptr = icptr->ic_next;
3513 }
3514 if (icptr != log->l_iclog)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003515 xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003516 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517
3518 /* check log magic numbers */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003519 if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003520 xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003522 ptr = (xfs_caddr_t) &iclog->ic_header;
3523 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 ptr += BBSIZE) {
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003525 if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003526 xfs_emerg(log->l_mp, "%s: unexpected magic num",
3527 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 }
3529
3530 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003531 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003532 ptr = iclog->ic_datap;
3533 base_ptr = ptr;
3534 ophead = (xlog_op_header_t *)ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003535 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 for (i = 0; i < len; i++) {
3537 ophead = (xlog_op_header_t *)ptr;
3538
3539 /* clientid is only 1 byte */
3540 field_offset = (__psint_t)
3541 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3542 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3543 clientid = ophead->oh_clientid;
3544 } else {
3545 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3546 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3547 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3548 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003549 clientid = xlog_get_client_id(
3550 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003552 clientid = xlog_get_client_id(
3553 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554 }
3555 }
3556 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003557 xfs_warn(log->l_mp,
3558 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3559 __func__, clientid, ophead,
3560 (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561
3562 /* check length */
3563 field_offset = (__psint_t)
3564 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3565 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003566 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 } else {
3568 idx = BTOBBT((__psint_t)&ophead->oh_len -
3569 (__psint_t)iclog->ic_datap);
3570 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3571 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3572 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003573 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003575 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 }
3577 }
3578 ptr += sizeof(xlog_op_header_t) + op_len;
3579 }
3580} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003581#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582
3583/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003584 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585 */
3586STATIC int
3587xlog_state_ioerror(
3588 xlog_t *log)
3589{
3590 xlog_in_core_t *iclog, *ic;
3591
3592 iclog = log->l_iclog;
3593 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3594 /*
3595 * Mark all the incore logs IOERROR.
3596 * From now on, no log flushes will result.
3597 */
3598 ic = iclog;
3599 do {
3600 ic->ic_state = XLOG_STATE_IOERROR;
3601 ic = ic->ic_next;
3602 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003603 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604 }
3605 /*
3606 * Return non-zero, if state transition has already happened.
3607 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003608 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003609}
3610
3611/*
3612 * This is called from xfs_force_shutdown, when we're forcibly
3613 * shutting down the filesystem, typically because of an IO error.
3614 * Our main objectives here are to make sure that:
3615 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3616 * parties to find out, 'atomically'.
3617 * b. those who're sleeping on log reservations, pinned objects and
3618 * other resources get woken up, and be told the bad news.
3619 * c. nothing new gets queued up after (a) and (b) are done.
3620 * d. if !logerror, flush the iclogs to disk, then seal them off
3621 * for business.
Dave Chinner9da1ab12010-05-17 15:51:59 +10003622 *
3623 * Note: for delayed logging the !logerror case needs to flush the regions
3624 * held in memory out to the iclogs before flushing them to disk. This needs
3625 * to be done before the log is marked as shutdown, otherwise the flush to the
3626 * iclogs will fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 */
3628int
3629xfs_log_force_umount(
3630 struct xfs_mount *mp,
3631 int logerror)
3632{
3633 xlog_ticket_t *tic;
3634 xlog_t *log;
3635 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636
3637 log = mp->m_log;
3638
3639 /*
3640 * If this happens during log recovery, don't worry about
3641 * locking; the log isn't open for business yet.
3642 */
3643 if (!log ||
3644 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3645 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003646 if (mp->m_sb_bp)
3647 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003648 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 }
3650
3651 /*
3652 * Somebody could've already done the hard work for us.
3653 * No need to get locks for this.
3654 */
3655 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3656 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003657 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658 }
3659 retval = 0;
Dave Chinner9da1ab12010-05-17 15:51:59 +10003660
3661 /*
3662 * Flush the in memory commit item list before marking the log as
3663 * being shut down. We need to do it in this order to ensure all the
3664 * completed transactions are flushed to disk with the xfs_log_force()
3665 * call below.
3666 */
3667 if (!logerror && (mp->m_flags & XFS_MOUNT_DELAYLOG))
Dave Chinnera44f13ed2010-08-24 11:40:03 +10003668 xlog_cil_force(log);
Dave Chinner9da1ab12010-05-17 15:51:59 +10003669
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670 /*
Dave Chinner3f16b982010-12-21 12:29:01 +11003671 * mark the filesystem and the as in a shutdown state and wake
3672 * everybody up to tell them the bad news.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003674 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003676 if (mp->m_sb_bp)
3677 XFS_BUF_DONE(mp->m_sb_bp);
3678
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 /*
3680 * This flag is sort of redundant because of the mount flag, but
3681 * it's good to maintain the separation between the log and the rest
3682 * of XFS.
3683 */
3684 log->l_flags |= XLOG_IO_ERROR;
3685
3686 /*
3687 * If we hit a log error, we want to mark all the iclogs IOERROR
3688 * while we're still holding the loglock.
3689 */
3690 if (logerror)
3691 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003692 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693
3694 /*
Dave Chinner10547942010-12-21 12:02:25 +11003695 * We don't want anybody waiting for log reservations after this. That
3696 * means we have to wake up everybody queued up on reserveq as well as
3697 * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3698 * we don't enqueue anything once the SHUTDOWN flag is set, and this
Dave Chinner3f16b982010-12-21 12:29:01 +11003699 * action is protected by the grant locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700 */
Dave Chinner3f16b982010-12-21 12:29:01 +11003701 spin_lock(&log->l_grant_reserve_lock);
Dave Chinner10547942010-12-21 12:02:25 +11003702 list_for_each_entry(tic, &log->l_reserveq, t_queue)
Dave Chinnereb40a872010-12-21 12:09:01 +11003703 wake_up(&tic->t_wait);
Dave Chinner3f16b982010-12-21 12:29:01 +11003704 spin_unlock(&log->l_grant_reserve_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003705
Dave Chinner3f16b982010-12-21 12:29:01 +11003706 spin_lock(&log->l_grant_write_lock);
Dave Chinner10547942010-12-21 12:02:25 +11003707 list_for_each_entry(tic, &log->l_writeq, t_queue)
Dave Chinnereb40a872010-12-21 12:09:01 +11003708 wake_up(&tic->t_wait);
Dave Chinner3f16b982010-12-21 12:29:01 +11003709 spin_unlock(&log->l_grant_write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003711 if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 ASSERT(!logerror);
3713 /*
3714 * Force the incore logs to disk before shutting the
3715 * log down completely.
3716 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003717 _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
3718
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003719 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003720 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003721 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003722 }
3723 /*
3724 * Wake up everybody waiting on xfs_log_force.
3725 * Callback all log item committed functions as if the
3726 * log writes were completed.
3727 */
3728 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3729
3730#ifdef XFSERRORDEBUG
3731 {
3732 xlog_in_core_t *iclog;
3733
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003734 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003735 iclog = log->l_iclog;
3736 do {
3737 ASSERT(iclog->ic_callback == 0);
3738 iclog = iclog->ic_next;
3739 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003740 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741 }
3742#endif
3743 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01003744 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745}
3746
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003747STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748xlog_iclogs_empty(xlog_t *log)
3749{
3750 xlog_in_core_t *iclog;
3751
3752 iclog = log->l_iclog;
3753 do {
3754 /* endianness does not matter here, zero is zero in
3755 * any language.
3756 */
3757 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01003758 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 iclog = iclog->ic_next;
3760 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003761 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762}