blob: 9dbdff3ea484bf008d1b634a53a2f42aedb3effc [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"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dmapi.h"
29#include "xfs_mount.h"
30#include "xfs_error.h"
31#include "xfs_log_priv.h"
32#include "xfs_buf_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_alloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_log_recover.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_trans_priv.h"
Nathan Scotta844f452005-11-02 14:38:42 +110038#include "xfs_dir2_sf.h"
39#include "xfs_attr_sf.h"
40#include "xfs_dinode.h"
41#include "xfs_inode.h"
42#include "xfs_rw.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
David Chinnereb01c9c2008-04-10 12:18:46 +100044kmem_zone_t *xfs_log_ticket_zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46#define xlog_write_adv_cnt(ptr, len, off, bytes) \
47 { (ptr) += (bytes); \
48 (len) -= (bytes); \
49 (off) += (bytes);}
50
51/* Local miscellaneous function prototypes */
52STATIC int xlog_bdstrat_cb(struct xfs_buf *);
53STATIC int xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket,
54 xlog_in_core_t **, xfs_lsn_t *);
55STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp,
56 xfs_buftarg_t *log_target,
57 xfs_daddr_t blk_offset,
58 int num_bblks);
59STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes);
60STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
Nathan Scottc41564b2006-03-29 08:55:14 +100061STATIC void xlog_dealloc_log(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062STATIC int xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[],
63 int nentries, xfs_log_ticket_t tic,
64 xfs_lsn_t *start_lsn,
65 xlog_in_core_t **commit_iclog,
66 uint flags);
67
68/* local state machine functions */
69STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
70STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog);
71STATIC int xlog_state_get_iclog_space(xlog_t *log,
72 int len,
73 xlog_in_core_t **iclog,
74 xlog_ticket_t *ticket,
75 int *continued_write,
76 int *logoffsetp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077STATIC int xlog_state_release_iclog(xlog_t *log,
78 xlog_in_core_t *iclog);
79STATIC void xlog_state_switch_iclogs(xlog_t *log,
80 xlog_in_core_t *iclog,
81 int eventual_size);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +110082STATIC int xlog_state_sync(xlog_t *log,
83 xfs_lsn_t lsn,
84 uint flags,
85 int *log_flushed);
86STATIC int xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087STATIC void xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog);
88
89/* local functions to manipulate grant head */
90STATIC int xlog_grant_log_space(xlog_t *log,
91 xlog_ticket_t *xtic);
92STATIC void xlog_grant_push_ail(xfs_mount_t *mp,
93 int need_bytes);
94STATIC void xlog_regrant_reserve_log_space(xlog_t *log,
95 xlog_ticket_t *ticket);
96STATIC int xlog_regrant_write_log_space(xlog_t *log,
97 xlog_ticket_t *ticket);
98STATIC void xlog_ungrant_log_space(xlog_t *log,
99 xlog_ticket_t *ticket);
100
101
102/* local ticket functions */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100103STATIC xlog_ticket_t *xlog_ticket_alloc(xlog_t *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 int unit_bytes,
105 int count,
106 char clientid,
107 uint flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100109#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110STATIC void xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr);
111STATIC void xlog_verify_grant_head(xlog_t *log, int equals);
112STATIC void xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog,
113 int count, boolean_t syncing);
114STATIC void xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog,
115 xfs_lsn_t tail_lsn);
116#else
117#define xlog_verify_dest_ptr(a,b)
118#define xlog_verify_grant_head(a,b)
119#define xlog_verify_iclog(a,b,c,d)
120#define xlog_verify_tail_lsn(a,b,c)
121#endif
122
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000123STATIC int xlog_iclogs_empty(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#if defined(XFS_LOG_TRACE)
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +1000126
127#define XLOG_TRACE_LOGGRANT_SIZE 2048
128#define XLOG_TRACE_ICLOG_SIZE 256
129
130void
131xlog_trace_loggrant_alloc(xlog_t *log)
132{
133 log->l_grant_trace = ktrace_alloc(XLOG_TRACE_LOGGRANT_SIZE, KM_NOFS);
134}
135
136void
137xlog_trace_loggrant_dealloc(xlog_t *log)
138{
139 ktrace_free(log->l_grant_trace);
140}
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142void
143xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
144{
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000145 unsigned long cnts;
146
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000147 /* ticket counts are 1 byte each */
148 cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150 ktrace_enter(log->l_grant_trace,
151 (void *)tic,
152 (void *)log->l_reserve_headq,
153 (void *)log->l_write_headq,
154 (void *)((unsigned long)log->l_grant_reserve_cycle),
155 (void *)((unsigned long)log->l_grant_reserve_bytes),
156 (void *)((unsigned long)log->l_grant_write_cycle),
157 (void *)((unsigned long)log->l_grant_write_bytes),
158 (void *)((unsigned long)log->l_curr_cycle),
159 (void *)((unsigned long)log->l_curr_block),
160 (void *)((unsigned long)CYCLE_LSN(log->l_tail_lsn)),
161 (void *)((unsigned long)BLOCK_LSN(log->l_tail_lsn)),
162 (void *)string,
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000163 (void *)((unsigned long)tic->t_trans_type),
164 (void *)cnts,
165 (void *)((unsigned long)tic->t_curr_res),
166 (void *)((unsigned long)tic->t_unit_res));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167}
168
169void
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +1000170xlog_trace_iclog_alloc(xlog_in_core_t *iclog)
171{
172 iclog->ic_trace = ktrace_alloc(XLOG_TRACE_ICLOG_SIZE, KM_NOFS);
173}
174
175void
176xlog_trace_iclog_dealloc(xlog_in_core_t *iclog)
177{
178 ktrace_free(iclog->ic_trace);
179}
180
181void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
183{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 ktrace_enter(iclog->ic_trace,
185 (void *)((unsigned long)state),
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100186 (void *)((unsigned long)current_pid()),
187 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
188 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
189 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
190 (void *)NULL, (void *)NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192#else
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +1000193
194#define xlog_trace_loggrant_alloc(log)
195#define xlog_trace_loggrant_dealloc(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196#define xlog_trace_loggrant(log,tic,string)
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +1000197
198#define xlog_trace_iclog_alloc(iclog)
199#define xlog_trace_iclog_dealloc(iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200#define xlog_trace_iclog(iclog,state)
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +1000201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202#endif /* XFS_LOG_TRACE */
203
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100204
205static void
206xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
207{
208 if (*qp) {
209 tic->t_next = (*qp);
210 tic->t_prev = (*qp)->t_prev;
211 (*qp)->t_prev->t_next = tic;
212 (*qp)->t_prev = tic;
213 } else {
214 tic->t_prev = tic->t_next = tic;
215 *qp = tic;
216 }
217
218 tic->t_flags |= XLOG_TIC_IN_Q;
219}
220
221static void
222xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
223{
224 if (tic == tic->t_next) {
225 *qp = NULL;
226 } else {
227 *qp = tic->t_next;
228 tic->t_next->t_prev = tic->t_prev;
229 tic->t_prev->t_next = tic->t_next;
230 }
231
232 tic->t_next = tic->t_prev = NULL;
233 tic->t_flags &= ~XLOG_TIC_IN_Q;
234}
235
236static void
237xlog_grant_sub_space(struct log *log, int bytes)
238{
239 log->l_grant_write_bytes -= bytes;
240 if (log->l_grant_write_bytes < 0) {
241 log->l_grant_write_bytes += log->l_logsize;
242 log->l_grant_write_cycle--;
243 }
244
245 log->l_grant_reserve_bytes -= bytes;
246 if ((log)->l_grant_reserve_bytes < 0) {
247 log->l_grant_reserve_bytes += log->l_logsize;
248 log->l_grant_reserve_cycle--;
249 }
250
251}
252
253static void
254xlog_grant_add_space_write(struct log *log, int bytes)
255{
Michael Nishimotod729eae2008-05-19 16:34:20 +1000256 int tmp = log->l_logsize - log->l_grant_write_bytes;
257 if (tmp > bytes)
258 log->l_grant_write_bytes += bytes;
259 else {
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100260 log->l_grant_write_cycle++;
Michael Nishimotod729eae2008-05-19 16:34:20 +1000261 log->l_grant_write_bytes = bytes - tmp;
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100262 }
263}
264
265static void
266xlog_grant_add_space_reserve(struct log *log, int bytes)
267{
Michael Nishimotod729eae2008-05-19 16:34:20 +1000268 int tmp = log->l_logsize - log->l_grant_reserve_bytes;
269 if (tmp > bytes)
270 log->l_grant_reserve_bytes += bytes;
271 else {
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100272 log->l_grant_reserve_cycle++;
Michael Nishimotod729eae2008-05-19 16:34:20 +1000273 log->l_grant_reserve_bytes = bytes - tmp;
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100274 }
275}
276
277static inline void
278xlog_grant_add_space(struct log *log, int bytes)
279{
280 xlog_grant_add_space_write(log, bytes);
281 xlog_grant_add_space_reserve(log, bytes);
282}
283
Christoph Hellwig0adba532007-08-30 17:21:46 +1000284static void
285xlog_tic_reset_res(xlog_ticket_t *tic)
286{
287 tic->t_res_num = 0;
288 tic->t_res_arr_sum = 0;
289 tic->t_res_num_ophdrs = 0;
290}
291
292static void
293xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
294{
295 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
296 /* add to overflow and start again */
297 tic->t_res_o_flow += tic->t_res_arr_sum;
298 tic->t_res_num = 0;
299 tic->t_res_arr_sum = 0;
300 }
301
302 tic->t_res_arr[tic->t_res_num].r_len = len;
303 tic->t_res_arr[tic->t_res_num].r_type = type;
304 tic->t_res_arr_sum += len;
305 tic->t_res_num++;
306}
Christoph Hellwigdd954c62006-01-11 15:34:50 +1100307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * NOTES:
310 *
311 * 1. currblock field gets updated at startup and after in-core logs
312 * marked as with WANT_SYNC.
313 */
314
315/*
316 * This routine is called when a user of a log manager ticket is done with
317 * the reservation. If the ticket was ever used, then a commit record for
318 * the associated transaction is written out as a log operation header with
319 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
320 * a given ticket. If the ticket was one with a permanent reservation, then
321 * a few operations are done differently. Permanent reservation tickets by
322 * default don't release the reservation. They just commit the current
323 * transaction with the belief that the reservation is still needed. A flag
324 * must be passed in before permanent reservations are actually released.
325 * When these type of tickets are not released, they need to be set into
326 * the inited state again. By doing this, a start record will be written
327 * out when the next write occurs.
328 */
329xfs_lsn_t
330xfs_log_done(xfs_mount_t *mp,
331 xfs_log_ticket_t xtic,
332 void **iclog,
333 uint flags)
334{
335 xlog_t *log = mp->m_log;
336 xlog_ticket_t *ticket = (xfs_log_ticket_t) xtic;
337 xfs_lsn_t lsn = 0;
338
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 if (XLOG_FORCED_SHUTDOWN(log) ||
340 /*
341 * If nothing was ever written, don't write out commit record.
342 * If we get an error, just continue and give back the log ticket.
343 */
344 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
345 (xlog_commit_record(mp, ticket,
346 (xlog_in_core_t **)iclog, &lsn)))) {
347 lsn = (xfs_lsn_t) -1;
348 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
349 flags |= XFS_LOG_REL_PERM_RESERV;
350 }
351 }
352
353
354 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
355 (flags & XFS_LOG_REL_PERM_RESERV)) {
356 /*
Nathan Scottc41564b2006-03-29 08:55:14 +1000357 * Release ticket if not permanent reservation or a specific
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 * request has been made to release a permanent reservation.
359 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000360 xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 xlog_ungrant_log_space(log, ticket);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100362 xfs_log_ticket_put(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 } else {
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000364 xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 xlog_regrant_reserve_log_space(log, ticket);
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000366 /* If this ticket was a permanent reservation and we aren't
367 * trying to release it, reset the inited flags; so next time
368 * we write, a start record will be written out.
369 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 ticket->t_flags |= XLOG_TIC_INITED;
Lachlan McIlroyc6a7b0f2008-08-13 16:52:50 +1000371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373 return lsn;
374} /* xfs_log_done */
375
376
377/*
378 * Force the in-core log to disk. If flags == XFS_LOG_SYNC,
379 * the force is done synchronously.
380 *
381 * Asynchronous forces are implemented by setting the WANT_SYNC
382 * bit in the appropriate in-core log and then returning.
383 *
David Chinner12017fa2008-08-13 16:34:31 +1000384 * Synchronous forces are implemented with a signal variable. All callers
385 * to force a given lsn to disk will wait on a the sv attached to the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 * specific in-core log. When given in-core log finally completes its
387 * write to disk, that thread will wake up all threads waiting on the
David Chinner12017fa2008-08-13 16:34:31 +1000388 * sv.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390int
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100391_xfs_log_force(
392 xfs_mount_t *mp,
393 xfs_lsn_t lsn,
394 uint flags,
395 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396{
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100397 xlog_t *log = mp->m_log;
398 int dummy;
399
400 if (!log_flushed)
401 log_flushed = &dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 ASSERT(flags & XFS_LOG_FORCE);
404
405 XFS_STATS_INC(xs_log_force);
406
Christoph Hellwigf538d4d2005-11-02 10:26:59 +1100407 if (log->l_flags & XLOG_IO_ERROR)
408 return XFS_ERROR(EIO);
409 if (lsn == 0)
410 return xlog_state_sync_all(log, flags, log_flushed);
411 else
412 return xlog_state_sync(log, lsn, flags, log_flushed);
David Chinnerb911ca02008-04-10 12:24:30 +1000413} /* _xfs_log_force */
414
415/*
416 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
417 * about errors or whether the log was flushed or not. This is the normal
418 * interface to use when trying to unpin items or move the log forward.
419 */
420void
421xfs_log_force(
422 xfs_mount_t *mp,
423 xfs_lsn_t lsn,
424 uint flags)
425{
426 int error;
427 error = _xfs_log_force(mp, lsn, flags, NULL);
428 if (error) {
429 xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
430 "error %d returned.", error);
431 }
432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
435/*
436 * Attaches a new iclog I/O completion callback routine during
437 * transaction commit. If the log is in error state, a non-zero
438 * return code is handed back and the caller is responsible for
439 * executing the callback at an appropriate time.
440 */
441int
442xfs_log_notify(xfs_mount_t *mp, /* mount of partition */
443 void *iclog_hndl, /* iclog to hang callback off */
444 xfs_log_callback_t *cb)
445{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000447 int abortflg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
David Chinner114d23a2008-04-10 12:18:39 +1000449 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
451 if (!abortflg) {
452 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
453 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
454 cb->cb_next = NULL;
455 *(iclog->ic_callback_tail) = cb;
456 iclog->ic_callback_tail = &(cb->cb_next);
457 }
David Chinner114d23a2008-04-10 12:18:39 +1000458 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 return abortflg;
460} /* xfs_log_notify */
461
462int
463xfs_log_release_iclog(xfs_mount_t *mp,
464 void *iclog_hndl)
465{
466 xlog_t *log = mp->m_log;
467 xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl;
468
469 if (xlog_state_release_iclog(log, iclog)) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000470 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Jesper Juhl014c2542006-01-15 02:37:08 +0100471 return EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 }
473
474 return 0;
475}
476
477/*
478 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
479 * to the reservation.
480 * 2. Potentially, push buffers at tail of log to disk.
481 *
482 * Each reservation is going to reserve extra space for a log record header.
483 * When writes happen to the on-disk log, we don't subtract the length of the
484 * log record header from any reservation. By wasting space in each
485 * reservation, we prevent over allocation problems.
486 */
487int
488xfs_log_reserve(xfs_mount_t *mp,
489 int unit_bytes,
490 int cnt,
491 xfs_log_ticket_t *ticket,
492 __uint8_t client,
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000493 uint flags,
494 uint t_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
496 xlog_t *log = mp->m_log;
497 xlog_ticket_t *internal_ticket;
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100498 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
501 ASSERT((flags & XFS_LOG_NOSLEEP) == 0);
502
503 if (XLOG_FORCED_SHUTDOWN(log))
504 return XFS_ERROR(EIO);
505
506 XFS_STATS_INC(xs_try_logspace);
507
508 if (*ticket != NULL) {
509 ASSERT(flags & XFS_LOG_PERM_RESERV);
510 internal_ticket = (xlog_ticket_t *)*ticket;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000511 xlog_trace_loggrant(log, internal_ticket, "xfs_log_reserve: existing ticket (permanent trans)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
513 retval = xlog_regrant_write_log_space(log, internal_ticket);
514 } else {
515 /* may sleep if need to allocate more tickets */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100516 internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 client, flags);
David Chinnereb01c9c2008-04-10 12:18:46 +1000518 if (!internal_ticket)
519 return XFS_ERROR(ENOMEM);
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000520 internal_ticket->t_trans_type = t_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 *ticket = internal_ticket;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000522 xlog_trace_loggrant(log, internal_ticket,
523 (internal_ticket->t_flags & XLOG_TIC_PERM_RESERV) ?
524 "xfs_log_reserve: create new ticket (permanent trans)" :
525 "xfs_log_reserve: create new ticket");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 xlog_grant_push_ail(mp,
527 (internal_ticket->t_unit_res *
528 internal_ticket->t_cnt));
529 retval = xlog_grant_log_space(log, internal_ticket);
530 }
531
532 return retval;
533} /* xfs_log_reserve */
534
535
536/*
537 * Mount a log filesystem
538 *
539 * mp - ubiquitous xfs mount point structure
540 * log_target - buftarg of on-disk log device
541 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
542 * num_bblocks - Number of BBSIZE blocks in on-disk log
543 *
544 * Return error or zero.
545 */
546int
David Chinner249a8c12008-02-05 12:13:32 +1100547xfs_log_mount(
548 xfs_mount_t *mp,
549 xfs_buftarg_t *log_target,
550 xfs_daddr_t blk_offset,
551 int num_bblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
David Chinner249a8c12008-02-05 12:13:32 +1100553 int error;
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
556 cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
557 else {
558 cmn_err(CE_NOTE,
559 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
560 mp->m_fsname);
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000561 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
563
564 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
Dave Chinnera6cb7672009-04-06 18:39:27 +0200565 if (IS_ERR(mp->m_log)) {
566 error = -PTR_ERR(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100567 goto out;
568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 /*
David Chinner249a8c12008-02-05 12:13:32 +1100571 * Initialize the AIL now we have a log.
572 */
David Chinner249a8c12008-02-05 12:13:32 +1100573 error = xfs_trans_ail_init(mp);
574 if (error) {
575 cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100576 goto out_free_log;
David Chinner249a8c12008-02-05 12:13:32 +1100577 }
David Chinnera9c21c12008-10-30 17:39:35 +1100578 mp->m_log->l_ailp = mp->m_ail;
David Chinner249a8c12008-02-05 12:13:32 +1100579
580 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * skip log recovery on a norecovery mount. pretend it all
582 * just worked.
583 */
584 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
David Chinner249a8c12008-02-05 12:13:32 +1100585 int readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000588 mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Eric Sandeen65be6052006-01-11 15:34:19 +1100590 error = xlog_recover(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 if (readonly)
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000593 mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 if (error) {
595 cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error);
Christoph Hellwig26430752009-02-12 19:55:48 +0100596 goto out_destroy_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598 }
599
600 /* Normal transactions can now occur */
601 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 return 0;
Christoph Hellwig26430752009-02-12 19:55:48 +0100604
605out_destroy_ail:
606 xfs_trans_ail_destroy(mp);
607out_free_log:
608 xlog_dealloc_log(mp->m_log);
Dave Chinner644c3562008-11-10 16:50:24 +1100609out:
David Chinner249a8c12008-02-05 12:13:32 +1100610 return error;
Christoph Hellwig26430752009-02-12 19:55:48 +0100611}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
613/*
614 * Finish the recovery of the file system. This is separate from
615 * the xfs_log_mount() call, because it depends on the code in
616 * xfs_mountfs() to read in the root and real-time bitmap inodes
617 * between calling xfs_log_mount() and here.
618 *
619 * mp - ubiquitous xfs mount point structure
620 */
621int
Christoph Hellwig42490232008-08-13 16:49:32 +1000622xfs_log_mount_finish(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
624 int error;
625
626 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
Christoph Hellwig42490232008-08-13 16:49:32 +1000627 error = xlog_recover_finish(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 else {
629 error = 0;
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000630 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
632
633 return error;
634}
635
636/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 * Final log writes as part of unmount.
638 *
639 * Mark the filesystem clean as unmount happens. Note that during relocation
640 * this routine needs to be executed as part of source-bag while the
641 * deallocation must not be done until source-end.
642 */
643
644/*
645 * Unmount record used to have a string "Unmount filesystem--" in the
646 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
647 * We just write the magic number now since that particular field isn't
648 * currently architecture converted and "nUmount" is a bit foo.
649 * As far as I know, there weren't any dependencies on the old behaviour.
650 */
651
652int
653xfs_log_unmount_write(xfs_mount_t *mp)
654{
655 xlog_t *log = mp->m_log;
656 xlog_in_core_t *iclog;
657#ifdef DEBUG
658 xlog_in_core_t *first_iclog;
659#endif
660 xfs_log_iovec_t reg[1];
661 xfs_log_ticket_t tic = NULL;
662 xfs_lsn_t lsn;
663 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665 /* the data section must be 32 bit size aligned */
666 struct {
667 __uint16_t magic;
668 __uint16_t pad1;
669 __uint32_t pad2; /* may as well make it 64 bits */
670 } magic = { XLOG_UNMOUNT_TYPE, 0, 0 };
671
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 /*
673 * Don't write out unmount record on read-only mounts.
674 * Or, if we are doing a forced umount (typically because of IO errors).
675 */
Christoph Hellwigbd186aa2007-08-30 17:21:12 +1000676 if (mp->m_flags & XFS_MOUNT_RDONLY)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 return 0;
678
David Chinnerb911ca02008-04-10 12:24:30 +1000679 error = _xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC, NULL);
680 ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682#ifdef DEBUG
683 first_iclog = iclog = log->l_iclog;
684 do {
685 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
686 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
687 ASSERT(iclog->ic_offset == 0);
688 }
689 iclog = iclog->ic_next;
690 } while (iclog != first_iclog);
691#endif
692 if (! (XLOG_FORCED_SHUTDOWN(log))) {
693 reg[0].i_addr = (void*)&magic;
694 reg[0].i_len = sizeof(magic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000695 XLOG_VEC_SET_TYPE(&reg[0], XLOG_REG_TYPE_UNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Tim Shimmin955e47a2006-09-28 11:04:16 +1000697 error = xfs_log_reserve(mp, 600, 1, &tic,
698 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (!error) {
700 /* remove inited flag */
701 ((xlog_ticket_t *)tic)->t_flags = 0;
702 error = xlog_write(mp, reg, 1, tic, &lsn,
703 NULL, XLOG_UNMOUNT_TRANS);
704 /*
705 * At this point, we're umounting anyway,
706 * so there's no point in transitioning log state
707 * to IOERROR. Just continue...
708 */
709 }
710
711 if (error) {
712 xfs_fs_cmn_err(CE_ALERT, mp,
713 "xfs_log_unmount: unmount record failed");
714 }
715
716
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000717 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100719 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100721 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000722 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000724 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
726 iclog->ic_state == XLOG_STATE_DIRTY)) {
727 if (!XLOG_FORCED_SHUTDOWN(log)) {
David Chinner12017fa2008-08-13 16:34:31 +1000728 sv_wait(&iclog->ic_force_wait, PMEM,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 &log->l_icloglock, s);
730 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000731 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
733 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000734 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
Tim Shimmin955e47a2006-09-28 11:04:16 +1000736 if (tic) {
737 xlog_trace_loggrant(log, tic, "unmount rec");
738 xlog_ungrant_log_space(log, tic);
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100739 xfs_log_ticket_put(tic);
Tim Shimmin955e47a2006-09-28 11:04:16 +1000740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 } else {
742 /*
743 * We're already in forced_shutdown mode, couldn't
744 * even attempt to write out the unmount transaction.
745 *
746 * Go through the motions of sync'ing and releasing
747 * the iclog, even though no I/O will actually happen,
Nathan Scottc41564b2006-03-29 08:55:14 +1000748 * we need to wait for other log I/Os that may already
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 * be in progress. Do this as a separate section of
750 * code so we'll know if we ever get stuck here that
751 * we're in this odd situation of trying to unmount
752 * a file system that went into forced_shutdown as
753 * the result of an unmount..
754 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000755 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 iclog = log->l_iclog;
David Chinner155cc6b2008-03-06 13:44:14 +1100757 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
759 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +0100760 spin_unlock(&log->l_icloglock);
David Chinner1bb7d6b2008-04-10 12:24:38 +1000761 error = xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000763 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
766 || iclog->ic_state == XLOG_STATE_DIRTY
767 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
768
David Chinner12017fa2008-08-13 16:34:31 +1000769 sv_wait(&iclog->ic_force_wait, PMEM,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 &log->l_icloglock, s);
771 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000772 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 }
774 }
775
David Chinner1bb7d6b2008-04-10 12:24:38 +1000776 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777} /* xfs_log_unmount_write */
778
779/*
780 * Deallocate log structures for unmount/relocation.
David Chinner249a8c12008-02-05 12:13:32 +1100781 *
782 * We need to stop the aild from running before we destroy
783 * and deallocate the log as the aild references the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 */
785void
Christoph Hellwig21b699c2009-03-16 08:19:29 +0100786xfs_log_unmount(xfs_mount_t *mp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
David Chinner249a8c12008-02-05 12:13:32 +1100788 xfs_trans_ail_destroy(mp);
Nathan Scottc41564b2006-03-29 08:55:14 +1000789 xlog_dealloc_log(mp->m_log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
792/*
793 * Write region vectors to log. The write happens using the space reservation
794 * of the ticket (tic). It is not a requirement that all writes for a given
795 * transaction occur with one call to xfs_log_write().
796 */
797int
798xfs_log_write(xfs_mount_t * mp,
799 xfs_log_iovec_t reg[],
800 int nentries,
801 xfs_log_ticket_t tic,
802 xfs_lsn_t *start_lsn)
803{
804 int error;
805 xlog_t *log = mp->m_log;
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 if (XLOG_FORCED_SHUTDOWN(log))
808 return XFS_ERROR(EIO);
809
810 if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
Nathan Scott7d04a332006-06-09 14:58:38 +1000811 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 }
Jesper Juhl014c2542006-01-15 02:37:08 +0100813 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814} /* xfs_log_write */
815
816
817void
818xfs_log_move_tail(xfs_mount_t *mp,
819 xfs_lsn_t tail_lsn)
820{
821 xlog_ticket_t *tic;
822 xlog_t *log = mp->m_log;
823 int need_bytes, free_bytes, cycle, bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (XLOG_FORCED_SHUTDOWN(log))
826 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 if (tail_lsn == 0) {
829 /* needed since sync_lsn is 64 bits */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000830 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 tail_lsn = log->l_last_sync_lsn;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000832 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000835 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
837 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
838 * tail_lsn.
839 */
840 if (tail_lsn != 1) {
841 log->l_tail_lsn = tail_lsn;
842 }
843
844 if ((tic = log->l_write_headq)) {
845#ifdef DEBUG
846 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
847 panic("Recovery problem");
848#endif
849 cycle = log->l_grant_write_cycle;
850 bytes = log->l_grant_write_bytes;
851 free_bytes = xlog_space_left(log, cycle, bytes);
852 do {
853 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
854
855 if (free_bytes < tic->t_unit_res && tail_lsn != 1)
856 break;
857 tail_lsn = 0;
858 free_bytes -= tic->t_unit_res;
David Chinner12017fa2008-08-13 16:34:31 +1000859 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 tic = tic->t_next;
861 } while (tic != log->l_write_headq);
862 }
863 if ((tic = log->l_reserve_headq)) {
864#ifdef DEBUG
865 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
866 panic("Recovery problem");
867#endif
868 cycle = log->l_grant_reserve_cycle;
869 bytes = log->l_grant_reserve_bytes;
870 free_bytes = xlog_space_left(log, cycle, bytes);
871 do {
872 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
873 need_bytes = tic->t_unit_res*tic->t_cnt;
874 else
875 need_bytes = tic->t_unit_res;
876 if (free_bytes < need_bytes && tail_lsn != 1)
877 break;
878 tail_lsn = 0;
879 free_bytes -= need_bytes;
David Chinner12017fa2008-08-13 16:34:31 +1000880 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 tic = tic->t_next;
882 } while (tic != log->l_reserve_headq);
883 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000884 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885} /* xfs_log_move_tail */
886
887/*
888 * Determine if we have a transaction that has gone to disk
889 * that needs to be covered. Log activity needs to be idle (no AIL and
890 * nothing in the iclogs). And, we need to be in the right state indicating
891 * something has gone out.
892 */
893int
894xfs_log_need_covered(xfs_mount_t *mp)
895{
David Chinner27d8d5f2008-10-30 17:38:39 +1100896 int needed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 xlog_t *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
David Chinner92821e22007-05-24 15:26:31 +1000899 if (!xfs_fs_writable(mp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return 0;
901
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000902 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
904 (log->l_covered_state == XLOG_STATE_COVER_NEED2))
David Chinnera9c21c12008-10-30 17:39:35 +1100905 && !xfs_trans_ail_tail(log->l_ailp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 && xlog_iclogs_empty(log)) {
907 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
908 log->l_covered_state = XLOG_STATE_COVER_DONE;
909 else {
910 ASSERT(log->l_covered_state == XLOG_STATE_COVER_NEED2);
911 log->l_covered_state = XLOG_STATE_COVER_DONE2;
912 }
913 needed = 1;
914 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000915 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +0100916 return needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917}
918
919/******************************************************************************
920 *
921 * local routines
922 *
923 ******************************************************************************
924 */
925
926/* xfs_trans_tail_ail returns 0 when there is nothing in the list.
927 * The log manager must keep track of the last LR which was committed
928 * to disk. The lsn of this LR will become the new tail_lsn whenever
929 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
930 * the situation where stuff could be written into the log but nothing
931 * was ever in the AIL when asked. Eventually, we panic since the
932 * tail hits the head.
933 *
934 * We may be holding the log iclog lock upon entering this routine.
935 */
936xfs_lsn_t
937xlog_assign_tail_lsn(xfs_mount_t *mp)
938{
939 xfs_lsn_t tail_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 xlog_t *log = mp->m_log;
941
David Chinner5b00f142008-10-30 17:39:00 +1100942 tail_lsn = xfs_trans_ail_tail(mp->m_ail);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000943 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (tail_lsn != 0) {
945 log->l_tail_lsn = tail_lsn;
946 } else {
947 tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn;
948 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +1000949 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951 return tail_lsn;
952} /* xlog_assign_tail_lsn */
953
954
955/*
956 * Return the space in the log between the tail and the head. The head
957 * is passed in the cycle/bytes formal parms. In the special case where
958 * the reserve head has wrapped passed the tail, this calculation is no
959 * longer valid. In this case, just return 0 which means there is no space
960 * in the log. This works for all places where this function is called
961 * with the reserve head. Of course, if the write head were to ever
962 * wrap the tail, we should blow up. Rather than catch this case here,
963 * we depend on other ASSERTions in other parts of the code. XXXmiken
964 *
965 * This code also handles the case where the reservation head is behind
966 * the tail. The details of this case are described below, but the end
967 * result is that we return the size of the log as the amount of space left.
968 */
David Chinnera8272ce2007-11-23 16:28:09 +1100969STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970xlog_space_left(xlog_t *log, int cycle, int bytes)
971{
972 int free_bytes;
973 int tail_bytes;
974 int tail_cycle;
975
976 tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn));
977 tail_cycle = CYCLE_LSN(log->l_tail_lsn);
978 if ((tail_cycle == cycle) && (bytes >= tail_bytes)) {
979 free_bytes = log->l_logsize - (bytes - tail_bytes);
980 } else if ((tail_cycle + 1) < cycle) {
981 return 0;
982 } else if (tail_cycle < cycle) {
983 ASSERT(tail_cycle == (cycle - 1));
984 free_bytes = tail_bytes - bytes;
985 } else {
986 /*
987 * The reservation head is behind the tail.
988 * In this case we just want to return the size of the
989 * log as the amount of space left.
990 */
991 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
992 "xlog_space_left: head behind tail\n"
993 " tail_cycle = %d, tail_bytes = %d\n"
994 " GH cycle = %d, GH bytes = %d",
995 tail_cycle, tail_bytes, cycle, bytes);
996 ASSERT(0);
997 free_bytes = log->l_logsize;
998 }
999 return free_bytes;
1000} /* xlog_space_left */
1001
1002
1003/*
1004 * Log function which is called when an io completes.
1005 *
1006 * The log manager needs its own routine, in order to control what
1007 * happens with the buffer after the write completes.
1008 */
1009void
1010xlog_iodone(xfs_buf_t *bp)
1011{
1012 xlog_in_core_t *iclog;
1013 xlog_t *l;
1014 int aborted;
1015
1016 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1017 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
1018 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1019 aborted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 l = iclog->ic_log;
1021
1022 /*
Christoph Hellwig73f6aa42008-10-10 17:28:29 +11001023 * If the _XFS_BARRIER_FAILED flag was set by a lower
1024 * layer, it means the underlying device no longer supports
David Chinner0bfefc42007-05-14 18:24:23 +10001025 * barrier I/O. Warn loudly and turn off barriers.
1026 */
Christoph Hellwig73f6aa42008-10-10 17:28:29 +11001027 if (bp->b_flags & _XFS_BARRIER_FAILED) {
1028 bp->b_flags &= ~_XFS_BARRIER_FAILED;
David Chinner0bfefc42007-05-14 18:24:23 +10001029 l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
1030 xfs_fs_cmn_err(CE_WARN, l->l_mp,
1031 "xlog_iodone: Barriers are no longer supported"
1032 " by device. Disabling barriers\n");
1033 xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp);
1034 }
1035
1036 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 * Race to shutdown the filesystem if we see an error.
1038 */
1039 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
1040 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
1041 xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
1042 XFS_BUF_STALE(bp);
Nathan Scott7d04a332006-06-09 14:58:38 +10001043 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /*
1045 * This flag will be propagated to the trans-committed
1046 * callback routines to let them know that the log-commit
1047 * didn't succeed.
1048 */
1049 aborted = XFS_LI_ABORTED;
1050 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
1051 aborted = XFS_LI_ABORTED;
1052 }
David Chinner3db296f2007-05-14 18:24:16 +10001053
1054 /* log I/O is always issued ASYNC */
1055 ASSERT(XFS_BUF_ISASYNC(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 xlog_state_done_syncing(iclog, aborted);
David Chinner3db296f2007-05-14 18:24:16 +10001057 /*
1058 * do not reference the buffer (bp) here as we could race
1059 * with it being freed after writing the unmount record to the
1060 * log.
1061 */
1062
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063} /* xlog_iodone */
1064
1065/*
1066 * The bdstrat callback function for log bufs. This gives us a central
1067 * place to trap bufs in case we get hit by a log I/O error and need to
1068 * shutdown. Actually, in practice, even when we didn't get a log error,
1069 * we transition the iclogs to IOERROR state *after* flushing all existing
1070 * iclogs to disk. This is because we don't want anymore new transactions to be
1071 * started or completed afterwards.
1072 */
1073STATIC int
1074xlog_bdstrat_cb(struct xfs_buf *bp)
1075{
1076 xlog_in_core_t *iclog;
1077
1078 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1079
1080 if ((iclog->ic_state & XLOG_STATE_IOERROR) == 0) {
1081 /* note for irix bstrat will need struct bdevsw passed
1082 * Fix the following macro if the code ever is merged
1083 */
1084 XFS_bdstrat(bp);
1085 return 0;
1086 }
1087
1088 xfs_buftrace("XLOG__BDSTRAT IOERROR", bp);
1089 XFS_BUF_ERROR(bp, EIO);
1090 XFS_BUF_STALE(bp);
1091 xfs_biodone(bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01001092 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
1094
1095}
1096
1097/*
1098 * Return size of each in-core log record buffer.
1099 *
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001100 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 *
1102 * If the filesystem blocksize is too large, we may need to choose a
1103 * larger size since the directory code currently logs entire blocks.
1104 */
1105
1106STATIC void
1107xlog_get_iclog_buffer_size(xfs_mount_t *mp,
1108 xlog_t *log)
1109{
1110 int size;
1111 int xhdrs;
1112
Eric Sandeen1cb51252007-08-16 16:24:43 +10001113 if (mp->m_logbufs <= 0)
1114 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
1115 else
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001116 log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117
1118 /*
1119 * Buffer size passed in from mount system call.
1120 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001121 if (mp->m_logbsize > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 size = log->l_iclog_size = mp->m_logbsize;
1123 log->l_iclog_size_log = 0;
1124 while (size != 1) {
1125 log->l_iclog_size_log++;
1126 size >>= 1;
1127 }
1128
Eric Sandeen62118702008-03-06 13:44:28 +11001129 if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001130 /* # headers = size / 32k
1131 * one header holds cycles from 32k of data
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 */
1133
1134 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
1135 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1136 xhdrs++;
1137 log->l_iclog_hsize = xhdrs << BBSHIFT;
1138 log->l_iclog_heads = xhdrs;
1139 } else {
1140 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1141 log->l_iclog_hsize = BBSIZE;
1142 log->l_iclog_heads = 1;
1143 }
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001144 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 }
1146
Malcolm Parsons9da096f2009-03-29 09:55:42 +02001147 /* All machines use 32kB buffers by default. */
Eric Sandeen1cb51252007-08-16 16:24:43 +10001148 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1149 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151 /* the default log size is 16k or 32k which is one header sector */
1152 log->l_iclog_hsize = BBSIZE;
1153 log->l_iclog_heads = 1;
1154
Christoph Hellwig7153f8b2009-02-09 08:36:46 +01001155done:
1156 /* are we being asked to make the sizes selected above visible? */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11001157 if (mp->m_logbufs == 0)
1158 mp->m_logbufs = log->l_iclog_bufs;
1159 if (mp->m_logbsize == 0)
1160 mp->m_logbsize = log->l_iclog_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161} /* xlog_get_iclog_buffer_size */
1162
1163
1164/*
1165 * This routine initializes some of the log structure for a given mount point.
1166 * Its primary purpose is to fill in enough, so recovery can occur. However,
1167 * some other stuff may be filled in too.
1168 */
1169STATIC xlog_t *
1170xlog_alloc_log(xfs_mount_t *mp,
1171 xfs_buftarg_t *log_target,
1172 xfs_daddr_t blk_offset,
1173 int num_bblks)
1174{
1175 xlog_t *log;
1176 xlog_rec_header_t *head;
1177 xlog_in_core_t **iclogp;
1178 xlog_in_core_t *iclog, *prev_iclog=NULL;
1179 xfs_buf_t *bp;
1180 int i;
1181 int iclogsize;
Dave Chinnera6cb7672009-04-06 18:39:27 +02001182 int error = ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Dave Chinner644c3562008-11-10 16:50:24 +11001184 log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001185 if (!log) {
1186 xlog_warn("XFS: Log allocation failed: No memory!");
1187 goto out;
1188 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 log->l_mp = mp;
1191 log->l_targ = log_target;
1192 log->l_logsize = BBTOB(num_bblks);
1193 log->l_logBBstart = blk_offset;
1194 log->l_logBBsize = num_bblks;
1195 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1196 log->l_flags |= XLOG_ACTIVE_RECOVERY;
1197
1198 log->l_prev_block = -1;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001199 log->l_tail_lsn = xlog_assign_lsn(1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1201 log->l_last_sync_lsn = log->l_tail_lsn;
1202 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
1203 log->l_grant_reserve_cycle = 1;
1204 log->l_grant_write_cycle = 1;
1205
Dave Chinnera6cb7672009-04-06 18:39:27 +02001206 error = EFSCORRUPTED;
Eric Sandeen62118702008-03-06 13:44:28 +11001207 if (xfs_sb_version_hassector(&mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT;
Dave Chinnera6cb7672009-04-06 18:39:27 +02001209 if (log->l_sectbb_log < 0 ||
1210 log->l_sectbb_log > mp->m_sectbb_log) {
1211 xlog_warn("XFS: Log sector size (0x%x) out of range.",
1212 log->l_sectbb_log);
1213 goto out_free_log;
1214 }
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 /* for larger sector sizes, must have v2 or external log */
Dave Chinnera6cb7672009-04-06 18:39:27 +02001217 if (log->l_sectbb_log != 0 &&
1218 (log->l_logBBstart != 0 &&
1219 !xfs_sb_version_haslogv2(&mp->m_sb))) {
1220 xlog_warn("XFS: log sector size (0x%x) invalid "
1221 "for configuration.", log->l_sectbb_log);
1222 goto out_free_log;
1223 }
1224 if (mp->m_sb.sb_logsectlog < BBSHIFT) {
1225 xlog_warn("XFS: Log sector log (0x%x) too small.",
1226 mp->m_sb.sb_logsectlog);
1227 goto out_free_log;
1228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
1230 log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1;
1231
1232 xlog_get_iclog_buffer_size(mp, log);
1233
Dave Chinnera6cb7672009-04-06 18:39:27 +02001234 error = ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001236 if (!bp)
1237 goto out_free_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
1239 XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
1240 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1241 ASSERT(XFS_BUF_ISBUSY(bp));
1242 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
1243 log->l_xbuf = bp;
1244
Eric Sandeen007c61c2007-10-11 17:43:56 +10001245 spin_lock_init(&log->l_icloglock);
1246 spin_lock_init(&log->l_grant_lock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10001247 sv_init(&log->l_flush_wait, 0, "flush_wait");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +10001249 xlog_trace_loggrant_alloc(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1251 ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
1252
1253 iclogp = &log->l_iclog;
1254 /*
1255 * The amount of memory to allocate for the iclog structure is
1256 * rather funky due to the way the structure is defined. It is
1257 * done this way so that we can use different sizes for machines
1258 * with different amounts of memory. See the definition of
1259 * xlog_in_core_t in xfs_log_priv.h for details.
1260 */
1261 iclogsize = log->l_iclog_size;
1262 ASSERT(log->l_iclog_size >= 4096);
1263 for (i=0; i < log->l_iclog_bufs; i++) {
Dave Chinner644c3562008-11-10 16:50:24 +11001264 *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
1265 if (!*iclogp)
1266 goto out_free_iclog;
1267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 iclog = *iclogp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 iclog->ic_prev = prev_iclog;
1270 prev_iclog = iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001271
1272 bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
Dave Chinner644c3562008-11-10 16:50:24 +11001273 if (!bp)
1274 goto out_free_iclog;
Christoph Hellwig1fa40b02007-05-14 18:23:50 +10001275 if (!XFS_BUF_CPSEMA(bp))
1276 ASSERT(0);
1277 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
1278 XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
1279 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1280 iclog->ic_bp = bp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001281 iclog->ic_data = bp->b_addr;
David Chinner4679b2d2008-04-10 12:18:54 +10001282#ifdef DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
David Chinner4679b2d2008-04-10 12:18:54 +10001284#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 head = &iclog->ic_header;
1286 memset(head, 0, sizeof(xlog_rec_header_t));
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001287 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1288 head->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001289 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001290 head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001292 head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
1296 iclog->ic_state = XLOG_STATE_ACTIVE;
1297 iclog->ic_log = log;
David Chinner114d23a2008-04-10 12:18:39 +10001298 atomic_set(&iclog->ic_refcnt, 0);
1299 spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 iclog->ic_callback_tail = &(iclog->ic_callback);
Christoph Hellwigb28708d2008-11-28 14:23:38 +11001301 iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
1304 ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
David Chinner12017fa2008-08-13 16:34:31 +10001305 sv_init(&iclog->ic_force_wait, SV_DEFAULT, "iclog-force");
1306 sv_init(&iclog->ic_write_wait, SV_DEFAULT, "iclog-write");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +10001308 xlog_trace_iclog_alloc(iclog);
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 iclogp = &iclog->ic_next;
1311 }
1312 *iclogp = log->l_iclog; /* complete ring */
1313 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1314
1315 return log;
Dave Chinner644c3562008-11-10 16:50:24 +11001316
1317out_free_iclog:
1318 for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
1319 prev_iclog = iclog->ic_next;
1320 if (iclog->ic_bp) {
1321 sv_destroy(&iclog->ic_force_wait);
1322 sv_destroy(&iclog->ic_write_wait);
1323 xfs_buf_free(iclog->ic_bp);
1324 xlog_trace_iclog_dealloc(iclog);
1325 }
1326 kmem_free(iclog);
1327 }
1328 spinlock_destroy(&log->l_icloglock);
1329 spinlock_destroy(&log->l_grant_lock);
1330 xlog_trace_loggrant_dealloc(log);
1331 xfs_buf_free(log->l_xbuf);
1332out_free_log:
1333 kmem_free(log);
Dave Chinnera6cb7672009-04-06 18:39:27 +02001334out:
1335 return ERR_PTR(-error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336} /* xlog_alloc_log */
1337
1338
1339/*
1340 * Write out the commit record of a transaction associated with the given
1341 * ticket. Return the lsn of the commit record.
1342 */
1343STATIC int
1344xlog_commit_record(xfs_mount_t *mp,
1345 xlog_ticket_t *ticket,
1346 xlog_in_core_t **iclog,
1347 xfs_lsn_t *commitlsnp)
1348{
1349 int error;
1350 xfs_log_iovec_t reg[1];
1351
1352 reg[0].i_addr = NULL;
1353 reg[0].i_len = 0;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001354 XLOG_VEC_SET_TYPE(&reg[0], XLOG_REG_TYPE_COMMIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356 ASSERT_ALWAYS(iclog);
1357 if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
1358 iclog, XLOG_COMMIT_TRANS))) {
Nathan Scott7d04a332006-06-09 14:58:38 +10001359 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001361 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362} /* xlog_commit_record */
1363
1364
1365/*
1366 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1367 * log space. This code pushes on the lsn which would supposedly free up
1368 * the 25% which we want to leave free. We may need to adopt a policy which
1369 * pushes on an lsn which is further along in the log once we reach the high
1370 * water mark. In this manner, we would be creating a low water mark.
1371 */
David Chinnera8272ce2007-11-23 16:28:09 +11001372STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373xlog_grant_push_ail(xfs_mount_t *mp,
1374 int need_bytes)
1375{
1376 xlog_t *log = mp->m_log; /* pointer to the log */
1377 xfs_lsn_t tail_lsn; /* lsn of the log tail */
1378 xfs_lsn_t threshold_lsn = 0; /* lsn we'd like to be at */
1379 int free_blocks; /* free blocks left to write to */
1380 int free_bytes; /* free bytes left to write to */
1381 int threshold_block; /* block in lsn we'd like to be at */
1382 int threshold_cycle; /* lsn cycle we'd like to be at */
1383 int free_threshold;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
1386
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001387 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 free_bytes = xlog_space_left(log,
1389 log->l_grant_reserve_cycle,
1390 log->l_grant_reserve_bytes);
1391 tail_lsn = log->l_tail_lsn;
1392 free_blocks = BTOBBT(free_bytes);
1393
1394 /*
1395 * Set the threshold for the minimum number of free blocks in the
1396 * log to the maximum of what the caller needs, one quarter of the
1397 * log, and 256 blocks.
1398 */
1399 free_threshold = BTOBB(need_bytes);
1400 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1401 free_threshold = MAX(free_threshold, 256);
1402 if (free_blocks < free_threshold) {
1403 threshold_block = BLOCK_LSN(tail_lsn) + free_threshold;
1404 threshold_cycle = CYCLE_LSN(tail_lsn);
1405 if (threshold_block >= log->l_logBBsize) {
1406 threshold_block -= log->l_logBBsize;
1407 threshold_cycle += 1;
1408 }
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001409 threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
1411 /* Don't pass in an lsn greater than the lsn of the last
1412 * log record known to be on disk.
1413 */
1414 if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0)
1415 threshold_lsn = log->l_last_sync_lsn;
1416 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001417 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
1419 /*
1420 * Get the transaction layer to kick the dirty buffers out to
1421 * disk asynchronously. No point in trying to do this if
1422 * the filesystem is shutting down.
1423 */
1424 if (threshold_lsn &&
1425 !XLOG_FORCED_SHUTDOWN(log))
David Chinner783a2f62008-10-30 17:39:58 +11001426 xfs_trans_ail_push(log->l_ailp, threshold_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427} /* xlog_grant_push_ail */
1428
1429
1430/*
1431 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1432 * fashion. Previously, we should have moved the current iclog
1433 * ptr in the log to point to the next available iclog. This allows further
1434 * write to continue while this code syncs out an iclog ready to go.
1435 * Before an in-core log can be written out, the data section must be scanned
1436 * to save away the 1st word of each BBSIZE block into the header. We replace
1437 * it with the current cycle count. Each BBSIZE block is tagged with the
1438 * cycle count because there in an implicit assumption that drives will
1439 * guarantee that entire 512 byte blocks get written at once. In other words,
1440 * we can't have part of a 512 byte block written and part not written. By
1441 * tagging each block, we will know which blocks are valid when recovering
1442 * after an unclean shutdown.
1443 *
1444 * This routine is single threaded on the iclog. No other thread can be in
1445 * this routine with the same iclog. Changing contents of iclog can there-
1446 * fore be done without grabbing the state machine lock. Updating the global
1447 * log will require grabbing the lock though.
1448 *
1449 * The entire log manager uses a logical block numbering scheme. Only
1450 * log_sync (and then only bwrite()) know about the fact that the log may
1451 * not start with block zero on a given device. The log block start offset
1452 * is added immediately before calling bwrite().
1453 */
1454
David Chinnera8272ce2007-11-23 16:28:09 +11001455STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456xlog_sync(xlog_t *log,
1457 xlog_in_core_t *iclog)
1458{
1459 xfs_caddr_t dptr; /* pointer to byte sized element */
1460 xfs_buf_t *bp;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001461 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 uint count; /* byte count of bwrite */
1463 uint count_init; /* initial count before roundup */
1464 int roundoff; /* roundoff to BB or stripe */
1465 int split = 0; /* split write into two regions */
1466 int error;
Eric Sandeen62118702008-03-06 13:44:28 +11001467 int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 XFS_STATS_INC(xs_log_writes);
David Chinner155cc6b2008-03-06 13:44:14 +11001470 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472 /* Add for LR header */
1473 count_init = log->l_iclog_hsize + iclog->ic_offset;
1474
1475 /* Round out the log write size */
1476 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1477 /* we have a v2 stripe unit to use */
1478 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1479 } else {
1480 count = BBTOB(BTOBB(count_init));
1481 }
1482 roundoff = count - count_init;
1483 ASSERT(roundoff >= 0);
1484 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1485 roundoff < log->l_mp->m_sb.sb_logsunit)
1486 ||
1487 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1488 roundoff < BBTOB(1)));
1489
1490 /* move grant heads by roundoff in sync */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001491 spin_lock(&log->l_grant_lock);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11001492 xlog_grant_add_space(log, roundoff);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10001493 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 /* put cycle number in every block */
1496 xlog_pack_data(log, iclog, roundoff);
1497
1498 /* real byte length */
1499 if (v2) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001500 iclog->ic_header.h_len =
1501 cpu_to_be32(iclog->ic_offset + roundoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001503 iclog->ic_header.h_len =
1504 cpu_to_be32(iclog->ic_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 }
1506
Nathan Scottf5faad72006-07-28 17:04:44 +10001507 bp = iclog->ic_bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
1509 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001510 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
1512 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1513
1514 /* Do we need to split this write into 2 parts? */
1515 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1516 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1517 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1518 iclog->ic_bwritecnt = 2; /* split into 2 writes */
1519 } else {
1520 iclog->ic_bwritecnt = 1;
1521 }
David Chinner511105b2007-05-24 15:21:57 +10001522 XFS_BUF_SET_COUNT(bp, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */
Nathan Scottf5faad72006-07-28 17:04:44 +10001524 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 XFS_BUF_BUSY(bp);
1526 XFS_BUF_ASYNC(bp);
1527 /*
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001528 * Do an ordered write for the log block.
Nathan Scottf5faad72006-07-28 17:04:44 +10001529 * Its unnecessary to flush the first split block in the log wrap case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 */
Nathan Scottf5faad72006-07-28 17:04:44 +10001531 if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER))
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001532 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
1534 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1535 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1536
1537 xlog_verify_iclog(log, iclog, count, B_TRUE);
1538
1539 /* account for log which doesn't start at block #0 */
1540 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1541 /*
1542 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1543 * is shutting down.
1544 */
1545 XFS_BUF_WRITE(bp);
1546
1547 if ((error = XFS_bwrite(bp))) {
1548 xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
1549 XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001550 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
1552 if (split) {
Nathan Scottf5faad72006-07-28 17:04:44 +10001553 bp = iclog->ic_log->l_xbuf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
1555 (unsigned long)1);
1556 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1557 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
1558 XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
1559 (__psint_t)count), split);
1560 XFS_BUF_SET_FSPRIVATE(bp, iclog);
Nathan Scottf5faad72006-07-28 17:04:44 +10001561 XFS_BUF_ZEROFLAGS(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 XFS_BUF_BUSY(bp);
1563 XFS_BUF_ASYNC(bp);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11001564 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1565 XFS_BUF_ORDERED(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 dptr = XFS_BUF_PTR(bp);
1567 /*
1568 * Bump the cycle numbers at the start of each block
1569 * since this part of the buffer is at the start of
1570 * a new cycle. Watch out for the header magic number
1571 * case, though.
1572 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001573 for (i = 0; i < split; i += BBSIZE) {
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001574 be32_add_cpu((__be32 *)dptr, 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001575 if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001576 be32_add_cpu((__be32 *)dptr, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 dptr += BBSIZE;
1578 }
1579
1580 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1581 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1582
Nathan Scottc41564b2006-03-29 08:55:14 +10001583 /* account for internal log which doesn't start at block #0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1585 XFS_BUF_WRITE(bp);
1586 if ((error = XFS_bwrite(bp))) {
1587 xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
1588 bp, XFS_BUF_ADDR(bp));
Jesper Juhl014c2542006-01-15 02:37:08 +01001589 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
1591 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001592 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593} /* xlog_sync */
1594
1595
1596/*
Nathan Scottc41564b2006-03-29 08:55:14 +10001597 * Deallocate a log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 */
David Chinnera8272ce2007-11-23 16:28:09 +11001599STATIC void
Nathan Scottc41564b2006-03-29 08:55:14 +10001600xlog_dealloc_log(xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
1602 xlog_in_core_t *iclog, *next_iclog;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 int i;
1604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 iclog = log->l_iclog;
1606 for (i=0; i<log->l_iclog_bufs; i++) {
David Chinner12017fa2008-08-13 16:34:31 +10001607 sv_destroy(&iclog->ic_force_wait);
1608 sv_destroy(&iclog->ic_write_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 xfs_buf_free(iclog->ic_bp);
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +10001610 xlog_trace_iclog_dealloc(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 next_iclog = iclog->ic_next;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001612 kmem_free(iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 iclog = next_iclog;
1614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 spinlock_destroy(&log->l_icloglock);
1616 spinlock_destroy(&log->l_grant_lock);
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 xfs_buf_free(log->l_xbuf);
Lachlan McIlroy31bd61f2008-09-17 16:45:37 +10001619 xlog_trace_loggrant_dealloc(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 log->l_mp->m_log = NULL;
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10001621 kmem_free(log);
Nathan Scottc41564b2006-03-29 08:55:14 +10001622} /* xlog_dealloc_log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624/*
1625 * Update counters atomically now that memcpy is done.
1626 */
1627/* ARGSUSED */
1628static inline void
1629xlog_state_finish_copy(xlog_t *log,
1630 xlog_in_core_t *iclog,
1631 int record_cnt,
1632 int copy_bytes)
1633{
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001634 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Marcin Slusarz413d57c2008-02-13 15:03:29 -08001636 be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 iclog->ic_offset += copy_bytes;
1638
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10001639 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640} /* xlog_state_finish_copy */
1641
1642
1643
1644
1645/*
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001646 * print out info relating to regions written which consume
1647 * the reservation
1648 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001649STATIC void
1650xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
1651{
1652 uint i;
1653 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1654
1655 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1656 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1657 "bformat",
1658 "bchunk",
1659 "efi_format",
1660 "efd_format",
1661 "iformat",
1662 "icore",
1663 "iext",
1664 "ibroot",
1665 "ilocal",
1666 "iattr_ext",
1667 "iattr_broot",
1668 "iattr_local",
1669 "qformat",
1670 "dquot",
1671 "quotaoff",
1672 "LR header",
1673 "unmount",
1674 "commit",
1675 "trans header"
1676 };
1677 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1678 "SETATTR_NOT_SIZE",
1679 "SETATTR_SIZE",
1680 "INACTIVE",
1681 "CREATE",
1682 "CREATE_TRUNC",
1683 "TRUNCATE_FILE",
1684 "REMOVE",
1685 "LINK",
1686 "RENAME",
1687 "MKDIR",
1688 "RMDIR",
1689 "SYMLINK",
1690 "SET_DMATTRS",
1691 "GROWFS",
1692 "STRAT_WRITE",
1693 "DIOSTRAT",
1694 "WRITE_SYNC",
1695 "WRITEID",
1696 "ADDAFORK",
1697 "ATTRINVAL",
1698 "ATRUNCATE",
1699 "ATTR_SET",
1700 "ATTR_RM",
1701 "ATTR_FLAG",
1702 "CLEAR_AGI_BUCKET",
1703 "QM_SBCHANGE",
1704 "DUMMY1",
1705 "DUMMY2",
1706 "QM_QUOTAOFF",
1707 "QM_DQALLOC",
1708 "QM_SETQLIM",
1709 "QM_DQCLUSTER",
1710 "QM_QINOCREATE",
1711 "QM_QUOTAOFF_END",
1712 "SB_UNIT",
1713 "FSYNC_TS",
1714 "GROWFSRT_ALLOC",
1715 "GROWFSRT_ZERO",
1716 "GROWFSRT_FREE",
1717 "SWAPEXT"
1718 };
1719
1720 xfs_fs_cmn_err(CE_WARN, mp,
1721 "xfs_log_write: reservation summary:\n"
1722 " trans type = %s (%u)\n"
1723 " unit res = %d bytes\n"
1724 " current res = %d bytes\n"
1725 " total reg = %u bytes (o/flow = %u bytes)\n"
1726 " ophdrs = %u (ophdr space = %u bytes)\n"
1727 " ophdr + reg = %u bytes\n"
1728 " num regions = %u\n",
1729 ((ticket->t_trans_type <= 0 ||
1730 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1731 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1732 ticket->t_trans_type,
1733 ticket->t_unit_res,
1734 ticket->t_curr_res,
1735 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1736 ticket->t_res_num_ophdrs, ophdr_spc,
1737 ticket->t_res_arr_sum +
Tim Shimmin12598452006-01-11 21:02:47 +11001738 ticket->t_res_o_flow + ophdr_spc,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001739 ticket->t_res_num);
1740
1741 for (i = 0; i < ticket->t_res_num; i++) {
Tim Shimmin12598452006-01-11 21:02:47 +11001742 uint r_type = ticket->t_res_arr[i].r_type;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001743 cmn_err(CE_WARN,
1744 "region[%u]: %s - %u bytes\n",
1745 i,
1746 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1747 "bad-rtype" : res_type_str[r_type-1]),
1748 ticket->t_res_arr[i].r_len);
1749 }
1750}
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001751
1752/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 * Write some region out to in-core log
1754 *
1755 * This will be called when writing externally provided regions or when
1756 * writing out a commit record for a given transaction.
1757 *
1758 * General algorithm:
1759 * 1. Find total length of this write. This may include adding to the
1760 * lengths passed in.
1761 * 2. Check whether we violate the tickets reservation.
1762 * 3. While writing to this iclog
1763 * A. Reserve as much space in this iclog as can get
1764 * B. If this is first write, save away start lsn
1765 * C. While writing this region:
1766 * 1. If first write of transaction, write start record
1767 * 2. Write log operation header (header per region)
1768 * 3. Find out if we can fit entire region into this iclog
1769 * 4. Potentially, verify destination memcpy ptr
1770 * 5. Memcpy (partial) region
1771 * 6. If partial copy, release iclog; otherwise, continue
1772 * copying more regions into current iclog
1773 * 4. Mark want sync bit (in simulation mode)
1774 * 5. Release iclog for potential flush to on-disk log.
1775 *
1776 * ERRORS:
1777 * 1. Panic if reservation is overrun. This should never happen since
1778 * reservation amounts are generated internal to the filesystem.
1779 * NOTES:
1780 * 1. Tickets are single threaded data structures.
1781 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1782 * syncing routine. When a single log_write region needs to span
1783 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1784 * on all log operation writes which don't contain the end of the
1785 * region. The XLOG_END_TRANS bit is used for the in-core log
1786 * operation which contains the end of the continued log_write region.
1787 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1788 * we don't really know exactly how much space will be used. As a result,
1789 * we don't update ic_offset until the end when we know exactly how many
1790 * bytes have been written out.
1791 */
David Chinnera8272ce2007-11-23 16:28:09 +11001792STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793xlog_write(xfs_mount_t * mp,
1794 xfs_log_iovec_t reg[],
1795 int nentries,
1796 xfs_log_ticket_t tic,
1797 xfs_lsn_t *start_lsn,
1798 xlog_in_core_t **commit_iclog,
1799 uint flags)
1800{
Nathan Scott5493a0f2006-06-28 11:17:28 +10001801 xlog_t *log = mp->m_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 xlog_ticket_t *ticket = (xlog_ticket_t *)tic;
Nathan Scott5493a0f2006-06-28 11:17:28 +10001803 xlog_in_core_t *iclog = NULL; /* ptr to current in-core log */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 xlog_op_header_t *logop_head; /* ptr to log operation header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 __psint_t ptr; /* copy address into data region */
1806 int len; /* # xlog_write() bytes 2 still copy */
1807 int index; /* region index currently copying */
1808 int log_offset; /* offset (from 0) into data region */
1809 int start_rec_copy; /* # bytes to copy for start record */
1810 int partial_copy; /* did we split a region? */
1811 int partial_copy_len;/* # bytes copied if split region */
1812 int need_copy; /* # bytes need to memcpy this region */
1813 int copy_len; /* # bytes actually memcpy'ing */
1814 int copy_off; /* # bytes from entry start */
1815 int contwr; /* continued write of in-core log? */
1816 int error;
1817 int record_cnt = 0, data_cnt = 0;
1818
1819 partial_copy_len = partial_copy = 0;
1820
1821 /* Calculate potential maximum space. Each region gets its own
1822 * xlog_op_header_t and may need to be double word aligned.
1823 */
1824 len = 0;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001825 if (ticket->t_flags & XLOG_TIC_INITED) { /* acct for start rec of xact */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 len += sizeof(xlog_op_header_t);
Christoph Hellwig0adba532007-08-30 17:21:46 +10001827 ticket->t_res_num_ophdrs++;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
1830 for (index = 0; index < nentries; index++) {
1831 len += sizeof(xlog_op_header_t); /* each region gets >= 1 */
Christoph Hellwig0adba532007-08-30 17:21:46 +10001832 ticket->t_res_num_ophdrs++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 len += reg[index].i_len;
Christoph Hellwig0adba532007-08-30 17:21:46 +10001834 xlog_tic_add_region(ticket, reg[index].i_len, reg[index].i_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 }
1836 contwr = *start_lsn = 0;
1837
1838 if (ticket->t_curr_res < len) {
Tim Shimmin7e9c6392005-09-02 16:42:05 +10001839 xlog_print_tic_res(mp, ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840#ifdef DEBUG
1841 xlog_panic(
1842 "xfs_log_write: reservation ran out. Need to up reservation");
1843#else
1844 /* Customer configurable panic */
1845 xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
1846 "xfs_log_write: reservation ran out. Need to up reservation");
1847 /* If we did not panic, shutdown the filesystem */
Nathan Scott7d04a332006-06-09 14:58:38 +10001848 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849#endif
1850 } else
1851 ticket->t_curr_res -= len;
1852
1853 for (index = 0; index < nentries; ) {
1854 if ((error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
1855 &contwr, &log_offset)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001856 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
1858 ASSERT(log_offset <= iclog->ic_size - 1);
1859 ptr = (__psint_t) ((char *)iclog->ic_datap+log_offset);
1860
1861 /* start_lsn is the first lsn written to. That's all we need. */
1862 if (! *start_lsn)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001863 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 /* This loop writes out as many regions as can fit in the amount
1866 * of space which was allocated by xlog_state_get_iclog_space().
1867 */
1868 while (index < nentries) {
1869 ASSERT(reg[index].i_len % sizeof(__int32_t) == 0);
1870 ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0);
1871 start_rec_copy = 0;
1872
1873 /* If first write for transaction, insert start record.
1874 * We can't be trying to commit if we are inited. We can't
1875 * have any "partial_copy" if we are inited.
1876 */
1877 if (ticket->t_flags & XLOG_TIC_INITED) {
1878 logop_head = (xlog_op_header_t *)ptr;
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10001879 logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 logop_head->oh_clientid = ticket->t_clientid;
1881 logop_head->oh_len = 0;
1882 logop_head->oh_flags = XLOG_START_TRANS;
1883 logop_head->oh_res2 = 0;
1884 ticket->t_flags &= ~XLOG_TIC_INITED; /* clear bit */
1885 record_cnt++;
1886
1887 start_rec_copy = sizeof(xlog_op_header_t);
1888 xlog_write_adv_cnt(ptr, len, log_offset, start_rec_copy);
1889 }
1890
1891 /* Copy log operation header directly into data section */
1892 logop_head = (xlog_op_header_t *)ptr;
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10001893 logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 logop_head->oh_clientid = ticket->t_clientid;
1895 logop_head->oh_res2 = 0;
1896
1897 /* header copied directly */
1898 xlog_write_adv_cnt(ptr, len, log_offset, sizeof(xlog_op_header_t));
1899
1900 /* are we copying a commit or unmount record? */
1901 logop_head->oh_flags = flags;
1902
1903 /*
1904 * We've seen logs corrupted with bad transaction client
1905 * ids. This makes sure that XFS doesn't generate them on.
1906 * Turn this into an EIO and shut down the filesystem.
1907 */
1908 switch (logop_head->oh_clientid) {
1909 case XFS_TRANSACTION:
1910 case XFS_VOLUME:
1911 case XFS_LOG:
1912 break;
1913 default:
1914 xfs_fs_cmn_err(CE_WARN, mp,
1915 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1916 logop_head->oh_clientid, tic);
1917 return XFS_ERROR(EIO);
1918 }
1919
1920 /* Partial write last time? => (partial_copy != 0)
1921 * need_copy is the amount we'd like to copy if everything could
1922 * fit in the current memcpy.
1923 */
1924 need_copy = reg[index].i_len - partial_copy_len;
1925
1926 copy_off = partial_copy_len;
1927 if (need_copy <= iclog->ic_size - log_offset) { /*complete write */
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10001928 copy_len = need_copy;
1929 logop_head->oh_len = cpu_to_be32(copy_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 if (partial_copy)
1931 logop_head->oh_flags|= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
1932 partial_copy_len = partial_copy = 0;
1933 } else { /* partial write */
1934 copy_len = iclog->ic_size - log_offset;
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10001935 logop_head->oh_len = cpu_to_be32(copy_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 logop_head->oh_flags |= XLOG_CONTINUE_TRANS;
1937 if (partial_copy)
1938 logop_head->oh_flags |= XLOG_WAS_CONT_TRANS;
1939 partial_copy_len += copy_len;
1940 partial_copy++;
1941 len += sizeof(xlog_op_header_t); /* from splitting of region */
1942 /* account for new log op header */
1943 ticket->t_curr_res -= sizeof(xlog_op_header_t);
Christoph Hellwig0adba532007-08-30 17:21:46 +10001944 ticket->t_res_num_ophdrs++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 }
1946 xlog_verify_dest_ptr(log, ptr);
1947
1948 /* copy region */
1949 ASSERT(copy_len >= 0);
1950 memcpy((xfs_caddr_t)ptr, reg[index].i_addr + copy_off, copy_len);
1951 xlog_write_adv_cnt(ptr, len, log_offset, copy_len);
1952
1953 /* make copy_len total bytes copied, including headers */
1954 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
1955 record_cnt++;
1956 data_cnt += contwr ? copy_len : 0;
1957 if (partial_copy) { /* copied partial region */
1958 /* already marked WANT_SYNC by xlog_state_get_iclog_space */
1959 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1960 record_cnt = data_cnt = 0;
1961 if ((error = xlog_state_release_iclog(log, iclog)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001962 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 break; /* don't increment index */
1964 } else { /* copied entire region */
1965 index++;
1966 partial_copy_len = partial_copy = 0;
1967
1968 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
1969 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1970 record_cnt = data_cnt = 0;
Christoph Hellwig39e2def2008-12-03 12:20:28 +01001971 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 xlog_state_want_sync(log, iclog);
Christoph Hellwig39e2def2008-12-03 12:20:28 +01001973 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 if (commit_iclog) {
1975 ASSERT(flags & XLOG_COMMIT_TRANS);
1976 *commit_iclog = iclog;
1977 } else if ((error = xlog_state_release_iclog(log, iclog)))
Jesper Juhl014c2542006-01-15 02:37:08 +01001978 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (index == nentries)
1980 return 0; /* we are done */
1981 else
1982 break;
1983 }
1984 } /* if (partial_copy) */
1985 } /* while (index < nentries) */
1986 } /* for (index = 0; index < nentries; ) */
1987 ASSERT(len == 0);
1988
1989 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1990 if (commit_iclog) {
1991 ASSERT(flags & XLOG_COMMIT_TRANS);
1992 *commit_iclog = iclog;
1993 return 0;
1994 }
Jesper Juhl014c2542006-01-15 02:37:08 +01001995 return xlog_state_release_iclog(log, iclog);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996} /* xlog_write */
1997
1998
1999/*****************************************************************************
2000 *
2001 * State Machine functions
2002 *
2003 *****************************************************************************
2004 */
2005
2006/* Clean iclogs starting from the head. This ordering must be
2007 * maintained, so an iclog doesn't become ACTIVE beyond one that
2008 * is SYNCING. This is also required to maintain the notion that we use
David Chinner12017fa2008-08-13 16:34:31 +10002009 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 * iclog is trying to sync to disk.
2011 *
2012 * State Change: DIRTY -> ACTIVE
2013 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10002014STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015xlog_state_clean_log(xlog_t *log)
2016{
2017 xlog_in_core_t *iclog;
2018 int changed = 0;
2019
2020 iclog = log->l_iclog;
2021 do {
2022 if (iclog->ic_state == XLOG_STATE_DIRTY) {
2023 iclog->ic_state = XLOG_STATE_ACTIVE;
2024 iclog->ic_offset = 0;
David Chinner114d23a2008-04-10 12:18:39 +10002025 ASSERT(iclog->ic_callback == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 /*
2027 * If the number of ops in this iclog indicate it just
2028 * contains the dummy transaction, we can
2029 * change state into IDLE (the second time around).
2030 * Otherwise we should change the state into
2031 * NEED a dummy.
2032 * We don't need to cover the dummy.
2033 */
2034 if (!changed &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002035 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
2036 XLOG_COVER_OPS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 changed = 1;
2038 } else {
2039 /*
2040 * We have two dirty iclogs so start over
2041 * This could also be num of ops indicates
2042 * this is not the dummy going out.
2043 */
2044 changed = 2;
2045 }
2046 iclog->ic_header.h_num_logops = 0;
2047 memset(iclog->ic_header.h_cycle_data, 0,
2048 sizeof(iclog->ic_header.h_cycle_data));
2049 iclog->ic_header.h_lsn = 0;
2050 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
2051 /* do nothing */;
2052 else
2053 break; /* stop cleaning */
2054 iclog = iclog->ic_next;
2055 } while (iclog != log->l_iclog);
2056
2057 /* log is locked when we are called */
2058 /*
2059 * Change state for the dummy log recording.
2060 * We usually go to NEED. But we go to NEED2 if the changed indicates
2061 * we are done writing the dummy record.
2062 * If we are done with the second dummy recored (DONE2), then
2063 * we go to IDLE.
2064 */
2065 if (changed) {
2066 switch (log->l_covered_state) {
2067 case XLOG_STATE_COVER_IDLE:
2068 case XLOG_STATE_COVER_NEED:
2069 case XLOG_STATE_COVER_NEED2:
2070 log->l_covered_state = XLOG_STATE_COVER_NEED;
2071 break;
2072
2073 case XLOG_STATE_COVER_DONE:
2074 if (changed == 1)
2075 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2076 else
2077 log->l_covered_state = XLOG_STATE_COVER_NEED;
2078 break;
2079
2080 case XLOG_STATE_COVER_DONE2:
2081 if (changed == 1)
2082 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2083 else
2084 log->l_covered_state = XLOG_STATE_COVER_NEED;
2085 break;
2086
2087 default:
2088 ASSERT(0);
2089 }
2090 }
2091} /* xlog_state_clean_log */
2092
2093STATIC xfs_lsn_t
2094xlog_get_lowest_lsn(
2095 xlog_t *log)
2096{
2097 xlog_in_core_t *lsn_log;
2098 xfs_lsn_t lowest_lsn, lsn;
2099
2100 lsn_log = log->l_iclog;
2101 lowest_lsn = 0;
2102 do {
2103 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002104 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if ((lsn && !lowest_lsn) ||
2106 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2107 lowest_lsn = lsn;
2108 }
2109 }
2110 lsn_log = lsn_log->ic_next;
2111 } while (lsn_log != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002112 return lowest_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113}
2114
2115
2116STATIC void
2117xlog_state_do_callback(
2118 xlog_t *log,
2119 int aborted,
2120 xlog_in_core_t *ciclog)
2121{
2122 xlog_in_core_t *iclog;
2123 xlog_in_core_t *first_iclog; /* used to know when we've
2124 * processed all iclogs once */
2125 xfs_log_callback_t *cb, *cb_next;
2126 int flushcnt = 0;
2127 xfs_lsn_t lowest_lsn;
2128 int ioerrors; /* counter: iclogs with errors */
2129 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2130 int funcdidcallbacks; /* flag: function did callbacks */
2131 int repeats; /* for issuing console warnings if
2132 * looping too many times */
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002133 int wake = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002135 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 first_iclog = iclog = log->l_iclog;
2137 ioerrors = 0;
2138 funcdidcallbacks = 0;
2139 repeats = 0;
2140
2141 do {
2142 /*
2143 * Scan all iclogs starting with the one pointed to by the
2144 * log. Reset this starting point each time the log is
2145 * unlocked (during callbacks).
2146 *
2147 * Keep looping through iclogs until one full pass is made
2148 * without running any callbacks.
2149 */
2150 first_iclog = log->l_iclog;
2151 iclog = log->l_iclog;
2152 loopdidcallbacks = 0;
2153 repeats++;
2154
2155 do {
2156
2157 /* skip all iclogs in the ACTIVE & DIRTY states */
2158 if (iclog->ic_state &
2159 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2160 iclog = iclog->ic_next;
2161 continue;
2162 }
2163
2164 /*
2165 * Between marking a filesystem SHUTDOWN and stopping
2166 * the log, we do flush all iclogs to disk (if there
2167 * wasn't a log I/O error). So, we do want things to
2168 * go smoothly in case of just a SHUTDOWN w/o a
2169 * LOG_IO_ERROR.
2170 */
2171 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2172 /*
2173 * Can only perform callbacks in order. Since
2174 * this iclog is not in the DONE_SYNC/
2175 * DO_CALLBACK state, we skip the rest and
2176 * just try to clean up. If we set our iclog
2177 * to DO_CALLBACK, we will not process it when
2178 * we retry since a previous iclog is in the
2179 * CALLBACK and the state cannot change since
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002180 * we are holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 */
2182 if (!(iclog->ic_state &
2183 (XLOG_STATE_DONE_SYNC |
2184 XLOG_STATE_DO_CALLBACK))) {
2185 if (ciclog && (ciclog->ic_state ==
2186 XLOG_STATE_DONE_SYNC)) {
2187 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2188 }
2189 break;
2190 }
2191 /*
2192 * We now have an iclog that is in either the
2193 * DO_CALLBACK or DONE_SYNC states. The other
2194 * states (WANT_SYNC, SYNCING, or CALLBACK were
2195 * caught by the above if and are going to
2196 * clean (i.e. we aren't doing their callbacks)
2197 * see the above if.
2198 */
2199
2200 /*
2201 * We will do one more check here to see if we
2202 * have chased our tail around.
2203 */
2204
2205 lowest_lsn = xlog_get_lowest_lsn(log);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002206 if (lowest_lsn &&
2207 XFS_LSN_CMP(lowest_lsn,
2208 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 iclog = iclog->ic_next;
2210 continue; /* Leave this iclog for
2211 * another thread */
2212 }
2213
2214 iclog->ic_state = XLOG_STATE_CALLBACK;
2215
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002216 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 /* l_last_sync_lsn field protected by
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002219 * l_grant_lock. Don't worry about iclog's lsn.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 * No one else can be here except us.
2221 */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002222 spin_lock(&log->l_grant_lock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002223 ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn,
2224 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2225 log->l_last_sync_lsn =
2226 be64_to_cpu(iclog->ic_header.h_lsn);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002227 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 } else {
David Chinner114d23a2008-04-10 12:18:39 +10002230 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 ioerrors++;
2232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
David Chinner114d23a2008-04-10 12:18:39 +10002234 /*
2235 * Keep processing entries in the callback list until
2236 * we come around and it is empty. We need to
2237 * atomically see that the list is empty and change the
2238 * state to DIRTY so that we don't miss any more
2239 * callbacks being added.
2240 */
2241 spin_lock(&iclog->ic_callback_lock);
2242 cb = iclog->ic_callback;
Christoph Hellwig4b809162007-08-16 15:37:36 +10002243 while (cb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 iclog->ic_callback_tail = &(iclog->ic_callback);
2245 iclog->ic_callback = NULL;
David Chinner114d23a2008-04-10 12:18:39 +10002246 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 /* perform callbacks in the order given */
Christoph Hellwig4b809162007-08-16 15:37:36 +10002249 for (; cb; cb = cb_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 cb_next = cb->cb_next;
2251 cb->cb_func(cb->cb_arg, aborted);
2252 }
David Chinner114d23a2008-04-10 12:18:39 +10002253 spin_lock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 cb = iclog->ic_callback;
2255 }
2256
2257 loopdidcallbacks++;
2258 funcdidcallbacks++;
2259
David Chinner114d23a2008-04-10 12:18:39 +10002260 spin_lock(&log->l_icloglock);
Christoph Hellwig4b809162007-08-16 15:37:36 +10002261 ASSERT(iclog->ic_callback == NULL);
David Chinner114d23a2008-04-10 12:18:39 +10002262 spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2264 iclog->ic_state = XLOG_STATE_DIRTY;
2265
2266 /*
2267 * Transition from DIRTY to ACTIVE if applicable.
2268 * NOP if STATE_IOERROR.
2269 */
2270 xlog_state_clean_log(log);
2271
2272 /* wake up threads waiting in xfs_log_force() */
David Chinner12017fa2008-08-13 16:34:31 +10002273 sv_broadcast(&iclog->ic_force_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
2275 iclog = iclog->ic_next;
2276 } while (first_iclog != iclog);
Nathan Scotta3c66852006-09-28 11:02:14 +10002277
2278 if (repeats > 5000) {
2279 flushcnt += repeats;
2280 repeats = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 xfs_fs_cmn_err(CE_WARN, log->l_mp,
Nathan Scotta3c66852006-09-28 11:02:14 +10002282 "%s: possible infinite loop (%d iterations)",
Harvey Harrison34a622b2008-04-10 12:19:21 +10002283 __func__, flushcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 }
2285 } while (!ioerrors && loopdidcallbacks);
2286
2287 /*
2288 * make one last gasp attempt to see if iclogs are being left in
2289 * limbo..
2290 */
2291#ifdef DEBUG
2292 if (funcdidcallbacks) {
2293 first_iclog = iclog = log->l_iclog;
2294 do {
2295 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2296 /*
2297 * Terminate the loop if iclogs are found in states
2298 * which will cause other threads to clean up iclogs.
2299 *
2300 * SYNCING - i/o completion will go through logs
2301 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002302 * l_icloglock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 * IOERROR - give up hope all ye who enter here
2304 */
2305 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2306 iclog->ic_state == XLOG_STATE_SYNCING ||
2307 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2308 iclog->ic_state == XLOG_STATE_IOERROR )
2309 break;
2310 iclog = iclog->ic_next;
2311 } while (first_iclog != iclog);
2312 }
2313#endif
2314
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002315 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2316 wake = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002317 spin_unlock(&log->l_icloglock);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002318
2319 if (wake)
2320 sv_broadcast(&log->l_flush_wait);
2321}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
2323
2324/*
2325 * Finish transitioning this iclog to the dirty state.
2326 *
2327 * Make sure that we completely execute this routine only when this is
2328 * the last call to the iclog. There is a good chance that iclog flushes,
2329 * when we reach the end of the physical log, get turned into 2 separate
2330 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2331 * routine. By using the reference count bwritecnt, we guarantee that only
2332 * the second completion goes through.
2333 *
2334 * Callbacks could take time, so they are done outside the scope of the
David Chinner12017fa2008-08-13 16:34:31 +10002335 * global state machine log lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 */
David Chinnera8272ce2007-11-23 16:28:09 +11002337STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338xlog_state_done_syncing(
2339 xlog_in_core_t *iclog,
2340 int aborted)
2341{
2342 xlog_t *log = iclog->ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002344 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2347 iclog->ic_state == XLOG_STATE_IOERROR);
David Chinner155cc6b2008-03-06 13:44:14 +11002348 ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2350
2351
2352 /*
2353 * If we got an error, either on the first buffer, or in the case of
2354 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2355 * and none should ever be attempted to be written to disk
2356 * again.
2357 */
2358 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2359 if (--iclog->ic_bwritecnt == 1) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002360 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return;
2362 }
2363 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2364 }
2365
2366 /*
2367 * Someone could be sleeping prior to writing out the next
2368 * iclog buffer, we wake them all, one will get to do the
2369 * I/O, the others get to wait for the result.
2370 */
David Chinner12017fa2008-08-13 16:34:31 +10002371 sv_broadcast(&iclog->ic_write_wait);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002372 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2374} /* xlog_state_done_syncing */
2375
2376
2377/*
2378 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
David Chinner12017fa2008-08-13 16:34:31 +10002379 * sleep. We wait on the flush queue on the head iclog as that should be
2380 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2381 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 *
2383 * The in-core logs are used in a circular fashion. They are not used
2384 * out-of-order even when an iclog past the head is free.
2385 *
2386 * return:
2387 * * log_offset where xlog_write() can start writing into the in-core
2388 * log's data space.
2389 * * in-core log pointer to which xlog_write() should write.
2390 * * boolean indicating this is a continued write to an in-core log.
2391 * If this is the last write, then the in-core log's offset field
2392 * needs to be incremented, depending on the amount of data which
2393 * is copied.
2394 */
David Chinnera8272ce2007-11-23 16:28:09 +11002395STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396xlog_state_get_iclog_space(xlog_t *log,
2397 int len,
2398 xlog_in_core_t **iclogp,
2399 xlog_ticket_t *ticket,
2400 int *continued_write,
2401 int *logoffsetp)
2402{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 int log_offset;
2404 xlog_rec_header_t *head;
2405 xlog_in_core_t *iclog;
2406 int error;
2407
2408restart:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002409 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002411 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 return XFS_ERROR(EIO);
2413 }
2414
2415 iclog = log->l_iclog;
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002416 if (iclog->ic_state != XLOG_STATE_ACTIVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH);
2418 XFS_STATS_INC(xs_log_noiclogs);
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002419
2420 /* Wait for log writes to have flushed */
2421 sv_wait(&log->l_flush_wait, 0, &log->l_icloglock, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 goto restart;
2423 }
Matthew Wilcoxd748c622008-05-19 16:34:27 +10002424
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 head = &iclog->ic_header;
2426
David Chinner155cc6b2008-03-06 13:44:14 +11002427 atomic_inc(&iclog->ic_refcnt); /* prevents sync */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 log_offset = iclog->ic_offset;
2429
2430 /* On the 1st write to an iclog, figure out lsn. This works
2431 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2432 * committing to. If the offset is set, that's how many blocks
2433 * must be written.
2434 */
2435 if (log_offset == 0) {
2436 ticket->t_curr_res -= log->l_iclog_hsize;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002437 xlog_tic_add_region(ticket,
Tim Shimmin7e9c6392005-09-02 16:42:05 +10002438 log->l_iclog_hsize,
2439 XLOG_REG_TYPE_LRHEADER);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002440 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2441 head->h_lsn = cpu_to_be64(
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10002442 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 ASSERT(log->l_curr_block >= 0);
2444 }
2445
2446 /* If there is enough room to write everything, then do it. Otherwise,
2447 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2448 * bit is on, so this will get flushed out. Don't update ic_offset
2449 * until you know exactly how many bytes get copied. Therefore, wait
2450 * until later to update ic_offset.
2451 *
2452 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2453 * can fit into remaining data section.
2454 */
2455 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2456 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2457
Dave Chinner49641f12008-07-11 17:43:55 +10002458 /*
2459 * If I'm the only one writing to this iclog, sync it to disk.
2460 * We need to do an atomic compare and decrement here to avoid
2461 * racing with concurrent atomic_dec_and_lock() calls in
2462 * xlog_state_release_iclog() when there is more than one
2463 * reference to the iclog.
2464 */
2465 if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
2466 /* we are the only one */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002467 spin_unlock(&log->l_icloglock);
Dave Chinner49641f12008-07-11 17:43:55 +10002468 error = xlog_state_release_iclog(log, iclog);
2469 if (error)
Jesper Juhl014c2542006-01-15 02:37:08 +01002470 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 } else {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002472 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 }
2474 goto restart;
2475 }
2476
2477 /* Do we have enough room to write the full amount in the remainder
2478 * of this iclog? Or must we continue a write on the next iclog and
2479 * mark this iclog as completely taken? In the case where we switch
2480 * iclogs (to mark it taken), this particular iclog will release/sync
2481 * to disk in xlog_write().
2482 */
2483 if (len <= iclog->ic_size - iclog->ic_offset) {
2484 *continued_write = 0;
2485 iclog->ic_offset += len;
2486 } else {
2487 *continued_write = 1;
2488 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2489 }
2490 *iclogp = iclog;
2491
2492 ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002493 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494
2495 *logoffsetp = log_offset;
2496 return 0;
2497} /* xlog_state_get_iclog_space */
2498
2499/*
2500 * Atomically get the log space required for a log ticket.
2501 *
2502 * Once a ticket gets put onto the reserveq, it will only return after
2503 * the needed reservation is satisfied.
2504 */
2505STATIC int
2506xlog_grant_log_space(xlog_t *log,
2507 xlog_ticket_t *tic)
2508{
2509 int free_bytes;
2510 int need_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511#ifdef DEBUG
2512 xfs_lsn_t tail_lsn;
2513#endif
2514
2515
2516#ifdef DEBUG
2517 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2518 panic("grant Recovery problem");
2519#endif
2520
2521 /* Is there space or do we need to sleep? */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002522 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: enter");
2524
2525 /* something is already sleeping; insert new transaction at end */
2526 if (log->l_reserve_headq) {
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002527 xlog_ins_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 xlog_trace_loggrant(log, tic,
2529 "xlog_grant_log_space: sleep 1");
2530 /*
2531 * Gotta check this before going to sleep, while we're
2532 * holding the grant lock.
2533 */
2534 if (XLOG_FORCED_SHUTDOWN(log))
2535 goto error_return;
2536
2537 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002538 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 /*
2540 * If we got an error, and the filesystem is shutting down,
2541 * we'll catch it down below. So just continue...
2542 */
2543 xlog_trace_loggrant(log, tic,
2544 "xlog_grant_log_space: wake 1");
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002545 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547 if (tic->t_flags & XFS_LOG_PERM_RESERV)
2548 need_bytes = tic->t_unit_res*tic->t_ocnt;
2549 else
2550 need_bytes = tic->t_unit_res;
2551
2552redo:
2553 if (XLOG_FORCED_SHUTDOWN(log))
2554 goto error_return;
2555
2556 free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle,
2557 log->l_grant_reserve_bytes);
2558 if (free_bytes < need_bytes) {
2559 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002560 xlog_ins_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 xlog_trace_loggrant(log, tic,
2562 "xlog_grant_log_space: sleep 2");
Dave Chinner9d7fef72009-04-06 18:42:59 +02002563 spin_unlock(&log->l_grant_lock);
2564 xlog_grant_push_ail(log->l_mp, need_bytes);
2565 spin_lock(&log->l_grant_lock);
2566
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002568 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Dave Chinner9d7fef72009-04-06 18:42:59 +02002570 spin_lock(&log->l_grant_lock);
2571 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
2574 xlog_trace_loggrant(log, tic,
2575 "xlog_grant_log_space: wake 2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 goto redo;
2577 } else if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002578 xlog_del_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579
2580 /* we've got enough space */
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002581 xlog_grant_add_space(log, need_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582#ifdef DEBUG
2583 tail_lsn = log->l_tail_lsn;
2584 /*
2585 * Check to make sure the grant write head didn't just over lap the
2586 * tail. If the cycles are the same, we can't be overlapping.
2587 * Otherwise, make sure that the cycles differ by exactly one and
2588 * check the byte count.
2589 */
2590 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2591 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2592 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2593 }
2594#endif
2595 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: exit");
2596 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002597 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 return 0;
2599
2600 error_return:
2601 if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002602 xlog_del_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: err_ret");
2604 /*
2605 * If we are failing, make sure the ticket doesn't have any
2606 * current reservations. We don't want to add this back when
2607 * the ticket/transaction gets cancelled.
2608 */
2609 tic->t_curr_res = 0;
2610 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002611 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 return XFS_ERROR(EIO);
2613} /* xlog_grant_log_space */
2614
2615
2616/*
2617 * Replenish the byte reservation required by moving the grant write head.
2618 *
2619 *
2620 */
2621STATIC int
2622xlog_regrant_write_log_space(xlog_t *log,
2623 xlog_ticket_t *tic)
2624{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 int free_bytes, need_bytes;
2626 xlog_ticket_t *ntic;
2627#ifdef DEBUG
2628 xfs_lsn_t tail_lsn;
2629#endif
2630
2631 tic->t_curr_res = tic->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002632 xlog_tic_reset_res(tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
2634 if (tic->t_cnt > 0)
Jesper Juhl014c2542006-01-15 02:37:08 +01002635 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
2637#ifdef DEBUG
2638 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2639 panic("regrant Recovery problem");
2640#endif
2641
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002642 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: enter");
2644
2645 if (XLOG_FORCED_SHUTDOWN(log))
2646 goto error_return;
2647
2648 /* If there are other waiters on the queue then give them a
2649 * chance at logspace before us. Wake up the first waiters,
2650 * if we do not wake up all the waiters then go to sleep waiting
2651 * for more free space, otherwise try to get some space for
2652 * this transaction.
2653 */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002654 need_bytes = tic->t_unit_res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if ((ntic = log->l_write_headq)) {
2656 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2657 log->l_grant_write_bytes);
2658 do {
2659 ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);
2660
2661 if (free_bytes < ntic->t_unit_res)
2662 break;
2663 free_bytes -= ntic->t_unit_res;
David Chinner12017fa2008-08-13 16:34:31 +10002664 sv_signal(&ntic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 ntic = ntic->t_next;
2666 } while (ntic != log->l_write_headq);
2667
2668 if (ntic != log->l_write_headq) {
2669 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002670 xlog_ins_ticketq(&log->l_write_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671
2672 xlog_trace_loggrant(log, tic,
2673 "xlog_regrant_write_log_space: sleep 1");
Dave Chinner9d7fef72009-04-06 18:42:59 +02002674 spin_unlock(&log->l_grant_lock);
2675 xlog_grant_push_ail(log->l_mp, need_bytes);
2676 spin_lock(&log->l_grant_lock);
2677
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002679 sv_wait(&tic->t_wait, PINOD|PLTWAIT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 &log->l_grant_lock, s);
2681
2682 /* If we're shutting down, this tic is already
2683 * off the queue */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002684 spin_lock(&log->l_grant_lock);
2685 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687
2688 xlog_trace_loggrant(log, tic,
2689 "xlog_regrant_write_log_space: wake 1");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 }
2691 }
2692
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693redo:
2694 if (XLOG_FORCED_SHUTDOWN(log))
2695 goto error_return;
2696
2697 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2698 log->l_grant_write_bytes);
2699 if (free_bytes < need_bytes) {
2700 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002701 xlog_ins_ticketq(&log->l_write_headq, tic);
Dave Chinner9d7fef72009-04-06 18:42:59 +02002702 spin_unlock(&log->l_grant_lock);
2703 xlog_grant_push_ail(log->l_mp, need_bytes);
2704 spin_lock(&log->l_grant_lock);
2705
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 XFS_STATS_INC(xs_sleep_logspace);
David Chinner12017fa2008-08-13 16:34:31 +10002707 sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
2709 /* If we're shutting down, this tic is already off the queue */
Dave Chinner9d7fef72009-04-06 18:42:59 +02002710 spin_lock(&log->l_grant_lock);
2711 if (XLOG_FORCED_SHUTDOWN(log))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 goto error_return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713
2714 xlog_trace_loggrant(log, tic,
2715 "xlog_regrant_write_log_space: wake 2");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 goto redo;
2717 } else if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002718 xlog_del_ticketq(&log->l_write_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002720 /* we've got enough space */
2721 xlog_grant_add_space_write(log, need_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722#ifdef DEBUG
2723 tail_lsn = log->l_tail_lsn;
2724 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2725 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2726 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2727 }
2728#endif
2729
2730 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: exit");
2731 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002732 spin_unlock(&log->l_grant_lock);
Jesper Juhl014c2542006-01-15 02:37:08 +01002733 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
2735
2736 error_return:
2737 if (tic->t_flags & XLOG_TIC_IN_Q)
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002738 xlog_del_ticketq(&log->l_reserve_headq, tic);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: err_ret");
2740 /*
2741 * If we are failing, make sure the ticket doesn't have any
2742 * current reservations. We don't want to add this back when
2743 * the ticket/transaction gets cancelled.
2744 */
2745 tic->t_curr_res = 0;
2746 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002747 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 return XFS_ERROR(EIO);
2749} /* xlog_regrant_write_log_space */
2750
2751
2752/* The first cnt-1 times through here we don't need to
2753 * move the grant write head because the permanent
2754 * reservation has reserved cnt times the unit amount.
2755 * Release part of current permanent unit reservation and
2756 * reset current reservation to be one units worth. Also
2757 * move grant reservation head forward.
2758 */
2759STATIC void
2760xlog_regrant_reserve_log_space(xlog_t *log,
2761 xlog_ticket_t *ticket)
2762{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 xlog_trace_loggrant(log, ticket,
2764 "xlog_regrant_reserve_log_space: enter");
2765 if (ticket->t_cnt > 0)
2766 ticket->t_cnt--;
2767
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002768 spin_lock(&log->l_grant_lock);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002769 xlog_grant_sub_space(log, ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002771 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 xlog_trace_loggrant(log, ticket,
2773 "xlog_regrant_reserve_log_space: sub current res");
2774 xlog_verify_grant_head(log, 1);
2775
2776 /* just return if we still have some of the pre-reserved space */
2777 if (ticket->t_cnt > 0) {
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002778 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 return;
2780 }
2781
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002782 xlog_grant_add_space_reserve(log, ticket->t_unit_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 xlog_trace_loggrant(log, ticket,
2784 "xlog_regrant_reserve_log_space: exit");
2785 xlog_verify_grant_head(log, 0);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002786 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 ticket->t_curr_res = ticket->t_unit_res;
Christoph Hellwig0adba532007-08-30 17:21:46 +10002788 xlog_tic_reset_res(ticket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789} /* xlog_regrant_reserve_log_space */
2790
2791
2792/*
2793 * Give back the space left from a reservation.
2794 *
2795 * All the information we need to make a correct determination of space left
2796 * is present. For non-permanent reservations, things are quite easy. The
2797 * count should have been decremented to zero. We only need to deal with the
2798 * space remaining in the current reservation part of the ticket. If the
2799 * ticket contains a permanent reservation, there may be left over space which
2800 * needs to be released. A count of N means that N-1 refills of the current
2801 * reservation can be done before we need to ask for more space. The first
2802 * one goes to fill up the first current reservation. Once we run out of
2803 * space, the count will stay at zero and the only space remaining will be
2804 * in the current reservation field.
2805 */
2806STATIC void
2807xlog_ungrant_log_space(xlog_t *log,
2808 xlog_ticket_t *ticket)
2809{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 if (ticket->t_cnt > 0)
2811 ticket->t_cnt--;
2812
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002813 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: enter");
2815
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002816 xlog_grant_sub_space(log, ticket->t_curr_res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
2818 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: sub current");
2819
2820 /* If this is a permanent reservation ticket, we may be able to free
2821 * up more space based on the remaining count.
2822 */
2823 if (ticket->t_cnt > 0) {
2824 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
Christoph Hellwigdd954c62006-01-11 15:34:50 +11002825 xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 }
2827
2828 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: exit");
2829 xlog_verify_grant_head(log, 1);
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10002830 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 xfs_log_move_tail(log->l_mp, 1);
2832} /* xlog_ungrant_log_space */
2833
2834
2835/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 * Flush iclog to disk if this is the last reference to the given iclog and
2837 * the WANT_SYNC bit is set.
2838 *
2839 * When this function is entered, the iclog is not necessarily in the
2840 * WANT_SYNC state. It may be sitting around waiting to get filled.
2841 *
2842 *
2843 */
David Chinnera8272ce2007-11-23 16:28:09 +11002844STATIC int
David Chinnerb5893342008-03-06 13:44:06 +11002845xlog_state_release_iclog(
2846 xlog_t *log,
2847 xlog_in_core_t *iclog)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 int sync = 0; /* do we sync? */
2850
David Chinner155cc6b2008-03-06 13:44:14 +11002851 if (iclog->ic_state & XLOG_STATE_IOERROR)
2852 return XFS_ERROR(EIO);
2853
2854 ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
2855 if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
2856 return 0;
2857
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002859 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 return XFS_ERROR(EIO);
2861 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
2863 iclog->ic_state == XLOG_STATE_WANT_SYNC);
2864
David Chinner155cc6b2008-03-06 13:44:14 +11002865 if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
David Chinnerb5893342008-03-06 13:44:06 +11002866 /* update tail before writing to iclog */
2867 xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 sync++;
2869 iclog->ic_state = XLOG_STATE_SYNCING;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002870 iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn);
2872 /* cycle incremented when incrementing curr_block */
2873 }
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002874 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
2876 /*
2877 * We let the log lock go, so it's possible that we hit a log I/O
Nathan Scottc41564b2006-03-29 08:55:14 +10002878 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2880 * this iclog has consistent data, so we ignore IOERROR
2881 * flags after this point.
2882 */
David Chinnerb5893342008-03-06 13:44:06 +11002883 if (sync)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 return xlog_sync(log, iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01002885 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886} /* xlog_state_release_iclog */
2887
2888
2889/*
2890 * This routine will mark the current iclog in the ring as WANT_SYNC
2891 * and move the current iclog pointer to the next iclog in the ring.
2892 * When this routine is called from xlog_state_get_iclog_space(), the
2893 * exact size of the iclog has not yet been determined. All we know is
2894 * that every data block. We have run out of space in this log record.
2895 */
2896STATIC void
2897xlog_state_switch_iclogs(xlog_t *log,
2898 xlog_in_core_t *iclog,
2899 int eventual_size)
2900{
2901 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2902 if (!eventual_size)
2903 eventual_size = iclog->ic_offset;
2904 iclog->ic_state = XLOG_STATE_WANT_SYNC;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002905 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 log->l_prev_block = log->l_curr_block;
2907 log->l_prev_cycle = log->l_curr_cycle;
2908
2909 /* roll log?: ic_offset changed later */
2910 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
2911
2912 /* Round up to next log-sunit */
Eric Sandeen62118702008-03-06 13:44:28 +11002913 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 log->l_mp->m_sb.sb_logsunit > 1) {
2915 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
2916 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
2917 }
2918
2919 if (log->l_curr_block >= log->l_logBBsize) {
2920 log->l_curr_cycle++;
2921 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
2922 log->l_curr_cycle++;
2923 log->l_curr_block -= log->l_logBBsize;
2924 ASSERT(log->l_curr_block >= 0);
2925 }
2926 ASSERT(iclog == log->l_iclog);
2927 log->l_iclog = iclog->ic_next;
2928} /* xlog_state_switch_iclogs */
2929
2930
2931/*
2932 * Write out all data in the in-core log as of this exact moment in time.
2933 *
2934 * Data may be written to the in-core log during this call. However,
2935 * we don't guarantee this data will be written out. A change from past
2936 * implementation means this routine will *not* write out zero length LRs.
2937 *
2938 * Basically, we try and perform an intelligent scan of the in-core logs.
2939 * If we determine there is no flushable data, we just return. There is no
2940 * flushable data if:
2941 *
2942 * 1. the current iclog is active and has no data; the previous iclog
2943 * is in the active or dirty state.
2944 * 2. the current iclog is drity, and the previous iclog is in the
2945 * active or dirty state.
2946 *
David Chinner12017fa2008-08-13 16:34:31 +10002947 * We may sleep if:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 *
2949 * 1. the current iclog is not in the active nor dirty state.
2950 * 2. the current iclog dirty, and the previous iclog is not in the
2951 * active nor dirty state.
2952 * 3. the current iclog is active, and there is another thread writing
2953 * to this particular iclog.
2954 * 4. a) the current iclog is active and has no other writers
2955 * b) when we return from flushing out this iclog, it is still
2956 * not in the active nor dirty state.
2957 */
2958STATIC int
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11002959xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960{
2961 xlog_in_core_t *iclog;
2962 xfs_lsn_t lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002964 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
2966 iclog = log->l_iclog;
2967 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10002968 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 return XFS_ERROR(EIO);
2970 }
2971
2972 /* If the head iclog is not active nor dirty, we just attach
2973 * ourselves to the head and go to sleep.
2974 */
2975 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2976 iclog->ic_state == XLOG_STATE_DIRTY) {
2977 /*
2978 * If the head is dirty or (active and empty), then
2979 * we need to look at the previous iclog. If the previous
2980 * iclog is active or dirty we are done. There is nothing
2981 * to sync out. Otherwise, we attach ourselves to the
2982 * previous iclog and go to sleep.
2983 */
2984 if (iclog->ic_state == XLOG_STATE_DIRTY ||
David Chinner155cc6b2008-03-06 13:44:14 +11002985 (atomic_read(&iclog->ic_refcnt) == 0
2986 && iclog->ic_offset == 0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 iclog = iclog->ic_prev;
2988 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2989 iclog->ic_state == XLOG_STATE_DIRTY)
2990 goto no_sleep;
2991 else
2992 goto maybe_sleep;
2993 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11002994 if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 /* We are the only one with access to this
2996 * iclog. Flush it out now. There should
2997 * be a roundoff of zero to show that someone
2998 * has already taken care of the roundoff from
2999 * the previous sync.
3000 */
David Chinner155cc6b2008-03-06 13:44:14 +11003001 atomic_inc(&iclog->ic_refcnt);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003002 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003004 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
3006 if (xlog_state_release_iclog(log, iclog))
3007 return XFS_ERROR(EIO);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003008 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003009 spin_lock(&log->l_icloglock);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003010 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 iclog->ic_state != XLOG_STATE_DIRTY)
3012 goto maybe_sleep;
3013 else
3014 goto no_sleep;
3015 } else {
3016 /* Someone else is writing to this iclog.
3017 * Use its call to flush out the data. However,
3018 * the other thread may not force out this LR,
3019 * so we mark it WANT_SYNC.
3020 */
3021 xlog_state_switch_iclogs(log, iclog, 0);
3022 goto maybe_sleep;
3023 }
3024 }
3025 }
3026
3027 /* By the time we come around again, the iclog could've been filled
3028 * which would give it another lsn. If we have a new lsn, just
3029 * return because the relevant data has been flushed.
3030 */
3031maybe_sleep:
3032 if (flags & XFS_LOG_SYNC) {
3033 /*
3034 * We must check if we're shutting down here, before
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003035 * we wait, while we're holding the l_icloglock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 * Then we check again after waking up, in case our
3037 * sleep was disturbed by a bad news.
3038 */
3039 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003040 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 return XFS_ERROR(EIO);
3042 }
3043 XFS_STATS_INC(xs_log_force_sleep);
David Chinner12017fa2008-08-13 16:34:31 +10003044 sv_wait(&iclog->ic_force_wait, PINOD, &log->l_icloglock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 /*
3046 * No need to grab the log lock here since we're
3047 * only deciding whether or not to return EIO
3048 * and the memory read should be atomic.
3049 */
3050 if (iclog->ic_state & XLOG_STATE_IOERROR)
3051 return XFS_ERROR(EIO);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003052 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054 } else {
3055
3056no_sleep:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003057 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 }
3059 return 0;
3060} /* xlog_state_sync_all */
3061
3062
3063/*
3064 * Used by code which implements synchronous log forces.
3065 *
3066 * Find in-core log with lsn.
3067 * If it is in the DIRTY state, just return.
3068 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3069 * state and go to sleep or return.
3070 * If it is in any other state, go to sleep or return.
3071 *
3072 * If filesystem activity goes to zero, the iclog will get flushed only by
3073 * bdflush().
3074 */
David Chinnera8272ce2007-11-23 16:28:09 +11003075STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076xlog_state_sync(xlog_t *log,
3077 xfs_lsn_t lsn,
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003078 uint flags,
3079 int *log_flushed)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080{
3081 xlog_in_core_t *iclog;
3082 int already_slept = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
3084try_again:
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003085 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 iclog = log->l_iclog;
3087
3088 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003089 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 return XFS_ERROR(EIO);
3091 }
3092
3093 do {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003094 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3095 iclog = iclog->ic_next;
3096 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 }
3098
3099 if (iclog->ic_state == XLOG_STATE_DIRTY) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003100 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 return 0;
3102 }
3103
3104 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3105 /*
3106 * We sleep here if we haven't already slept (e.g.
3107 * this is the first time we've looked at the correct
3108 * iclog buf) and the buffer before us is going to
3109 * be sync'ed. The reason for this is that if we
3110 * are doing sync transactions here, by waiting for
3111 * the previous I/O to complete, we can allow a few
3112 * more transactions into this iclog before we close
3113 * it down.
3114 *
3115 * Otherwise, we mark the buffer WANT_SYNC, and bump
3116 * up the refcnt so we can release the log (which drops
3117 * the ref count). The state switch keeps new transaction
3118 * commits from using this buffer. When the current commits
3119 * finish writing into the buffer, the refcount will drop to
3120 * zero and the buffer will go out then.
3121 */
3122 if (!already_slept &&
3123 (iclog->ic_prev->ic_state & (XLOG_STATE_WANT_SYNC |
3124 XLOG_STATE_SYNCING))) {
3125 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3126 XFS_STATS_INC(xs_log_force_sleep);
David Chinner12017fa2008-08-13 16:34:31 +10003127 sv_wait(&iclog->ic_prev->ic_write_wait, PSWP,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 &log->l_icloglock, s);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003129 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 already_slept = 1;
3131 goto try_again;
3132 } else {
David Chinner155cc6b2008-03-06 13:44:14 +11003133 atomic_inc(&iclog->ic_refcnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003135 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 if (xlog_state_release_iclog(log, iclog))
3137 return XFS_ERROR(EIO);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003138 *log_flushed = 1;
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003139 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 }
3141 }
3142
3143 if ((flags & XFS_LOG_SYNC) && /* sleep */
3144 !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3145
3146 /*
David Chinner12017fa2008-08-13 16:34:31 +10003147 * Don't wait on completion if we know that we've
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 * gotten a log write error.
3149 */
3150 if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003151 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 return XFS_ERROR(EIO);
3153 }
3154 XFS_STATS_INC(xs_log_force_sleep);
David Chinner12017fa2008-08-13 16:34:31 +10003155 sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 /*
3157 * No need to grab the log lock here since we're
3158 * only deciding whether or not to return EIO
3159 * and the memory read should be atomic.
3160 */
3161 if (iclog->ic_state & XLOG_STATE_IOERROR)
3162 return XFS_ERROR(EIO);
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003163 *log_flushed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 } else { /* just return */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003165 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 }
3167 return 0;
3168
3169 } while (iclog != log->l_iclog);
3170
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003171 spin_unlock(&log->l_icloglock);
Jesper Juhl014c2542006-01-15 02:37:08 +01003172 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173} /* xlog_state_sync */
3174
3175
3176/*
3177 * Called when we want to mark the current iclog as being ready to sync to
3178 * disk.
3179 */
David Chinnera8272ce2007-11-23 16:28:09 +11003180STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3182{
Christoph Hellwiga8914f32009-08-10 11:32:44 -03003183 assert_spin_locked(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184
3185 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3186 xlog_state_switch_iclogs(log, iclog, 0);
3187 } else {
3188 ASSERT(iclog->ic_state &
3189 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3190 }
Christoph Hellwig39e2def2008-12-03 12:20:28 +01003191}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192
3193
3194/*****************************************************************************
3195 *
3196 * TICKET functions
3197 *
3198 *****************************************************************************
3199 */
3200
3201/*
Malcolm Parsons9da096f2009-03-29 09:55:42 +02003202 * Free a used ticket when its refcount falls to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003204void
3205xfs_log_ticket_put(
3206 xlog_ticket_t *ticket)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207{
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003208 ASSERT(atomic_read(&ticket->t_ref) > 0);
3209 if (atomic_dec_and_test(&ticket->t_ref)) {
3210 sv_destroy(&ticket->t_wait);
3211 kmem_zone_free(xfs_log_ticket_zone, ticket);
3212 }
3213}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003215xlog_ticket_t *
3216xfs_log_ticket_get(
3217 xlog_ticket_t *ticket)
3218{
3219 ASSERT(atomic_read(&ticket->t_ref) > 0);
3220 atomic_inc(&ticket->t_ref);
3221 return ticket;
3222}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223
3224/*
David Chinnereb01c9c2008-04-10 12:18:46 +10003225 * Allocate and initialise a new log ticket.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 */
David Chinnera8272ce2007-11-23 16:28:09 +11003227STATIC xlog_ticket_t *
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003228xlog_ticket_alloc(xlog_t *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 int unit_bytes,
3230 int cnt,
3231 char client,
3232 uint xflags)
3233{
3234 xlog_ticket_t *tic;
3235 uint num_headers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236
David Chinnereb01c9c2008-04-10 12:18:46 +10003237 tic = kmem_zone_zalloc(xfs_log_ticket_zone, KM_SLEEP|KM_MAYFAIL);
3238 if (!tic)
3239 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240
3241 /*
3242 * Permanent reservations have up to 'cnt'-1 active log operations
3243 * in the log. A unit in this case is the amount of space for one
3244 * of these log operations. Normal reservations have a cnt of 1
3245 * and their unit amount is the total amount of space required.
3246 *
3247 * The following lines of code account for non-transaction data
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003248 * which occupy space in the on-disk log.
3249 *
3250 * Normal form of a transaction is:
3251 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3252 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3253 *
3254 * We need to account for all the leadup data and trailer data
3255 * around the transaction data.
3256 * And then we need to account for the worst case in terms of using
3257 * more space.
3258 * The worst case will happen if:
3259 * - the placement of the transaction happens to be such that the
3260 * roundoff is at its maximum
3261 * - the transaction data is synced before the commit record is synced
3262 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3263 * Therefore the commit record is in its own Log Record.
3264 * This can happen as the commit record is called with its
3265 * own region to xlog_write().
3266 * This then means that in the worst case, roundoff can happen for
3267 * the commit-rec as well.
3268 * The commit-rec is smaller than padding in this scenario and so it is
3269 * not added separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 */
3271
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003272 /* for trans header */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 unit_bytes += sizeof(xlog_op_header_t);
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003274 unit_bytes += sizeof(xfs_trans_header_t);
3275
3276 /* for start-rec */
3277 unit_bytes += sizeof(xlog_op_header_t);
3278
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 /* for LR headers */
3280 num_headers = ((unit_bytes + log->l_iclog_size-1) >> log->l_iclog_size_log);
3281 unit_bytes += log->l_iclog_hsize * num_headers;
3282
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003283 /* for commit-rec LR header - note: padding will subsume the ophdr */
3284 unit_bytes += log->l_iclog_hsize;
3285
3286 /* for split-recs - ophdrs added when data split over LRs */
3287 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3288
3289 /* for roundoff padding for transaction data and one for commit record */
Eric Sandeen62118702008-03-06 13:44:28 +11003290 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Tim Shimmin32fb9b52005-09-02 16:41:43 +10003291 log->l_mp->m_sb.sb_logsunit > 1) {
3292 /* log su roundoff */
3293 unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
3294 } else {
3295 /* BB roundoff */
3296 unit_bytes += 2*BBSIZE;
3297 }
3298
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003299 atomic_set(&tic->t_ref, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 tic->t_unit_res = unit_bytes;
3301 tic->t_curr_res = unit_bytes;
3302 tic->t_cnt = cnt;
3303 tic->t_ocnt = cnt;
3304 tic->t_tid = (xlog_tid_t)((__psint_t)tic & 0xffffffff);
3305 tic->t_clientid = client;
3306 tic->t_flags = XLOG_TIC_INITED;
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003307 tic->t_trans_type = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 if (xflags & XFS_LOG_PERM_RESERV)
3309 tic->t_flags |= XLOG_TIC_PERM_RESERV;
David Chinner12017fa2008-08-13 16:34:31 +10003310 sv_init(&(tic->t_wait), SV_DEFAULT, "logtick");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311
Christoph Hellwig0adba532007-08-30 17:21:46 +10003312 xlog_tic_reset_res(tic);
Tim Shimmin7e9c6392005-09-02 16:42:05 +10003313
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314 return tic;
Dave Chinnercc09c0d2008-11-17 17:37:10 +11003315}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316
3317
3318/******************************************************************************
3319 *
3320 * Log debug routines
3321 *
3322 ******************************************************************************
3323 */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003324#if defined(DEBUG)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325/*
3326 * Make sure that the destination ptr is within the valid data region of
3327 * one of the iclogs. This uses backup pointers stored in a different
3328 * part of the log in case we trash the log structure.
3329 */
3330void
3331xlog_verify_dest_ptr(xlog_t *log,
3332 __psint_t ptr)
3333{
3334 int i;
3335 int good_ptr = 0;
3336
3337 for (i=0; i < log->l_iclog_bufs; i++) {
3338 if (ptr >= (__psint_t)log->l_iclog_bak[i] &&
3339 ptr <= (__psint_t)log->l_iclog_bak[i]+log->l_iclog_size)
3340 good_ptr++;
3341 }
3342 if (! good_ptr)
3343 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3344} /* xlog_verify_dest_ptr */
3345
3346STATIC void
3347xlog_verify_grant_head(xlog_t *log, int equals)
3348{
3349 if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) {
3350 if (equals)
3351 ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes);
3352 else
3353 ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes);
3354 } else {
3355 ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle);
3356 ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes);
3357 }
3358} /* xlog_verify_grant_head */
3359
3360/* check if it will fit */
3361STATIC void
3362xlog_verify_tail_lsn(xlog_t *log,
3363 xlog_in_core_t *iclog,
3364 xfs_lsn_t tail_lsn)
3365{
3366 int blocks;
3367
3368 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3369 blocks =
3370 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3371 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3372 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3373 } else {
3374 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3375
3376 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3377 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3378
3379 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3380 if (blocks < BTOBB(iclog->ic_offset) + 1)
3381 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3382 }
3383} /* xlog_verify_tail_lsn */
3384
3385/*
3386 * Perform a number of checks on the iclog before writing to disk.
3387 *
3388 * 1. Make sure the iclogs are still circular
3389 * 2. Make sure we have a good magic number
3390 * 3. Make sure we don't have magic numbers in the data
3391 * 4. Check fields of each log operation header for:
3392 * A. Valid client identifier
3393 * B. tid ptr value falls in valid ptr space (user space code)
3394 * C. Length in log record header is correct according to the
3395 * individual operation headers within record.
3396 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3397 * log, check the preceding blocks of the physical log to make sure all
3398 * the cycle numbers agree with the current cycle number.
3399 */
3400STATIC void
3401xlog_verify_iclog(xlog_t *log,
3402 xlog_in_core_t *iclog,
3403 int count,
3404 boolean_t syncing)
3405{
3406 xlog_op_header_t *ophead;
3407 xlog_in_core_t *icptr;
3408 xlog_in_core_2_t *xhdr;
3409 xfs_caddr_t ptr;
3410 xfs_caddr_t base_ptr;
3411 __psint_t field_offset;
3412 __uint8_t clientid;
3413 int len, i, j, k, op_len;
3414 int idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
3416 /* check validity of iclog pointers */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003417 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 icptr = log->l_iclog;
3419 for (i=0; i < log->l_iclog_bufs; i++) {
Christoph Hellwig4b809162007-08-16 15:37:36 +10003420 if (icptr == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421 xlog_panic("xlog_verify_iclog: invalid ptr");
3422 icptr = icptr->ic_next;
3423 }
3424 if (icptr != log->l_iclog)
3425 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003426 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
3428 /* check log magic numbers */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003429 if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 xlog_panic("xlog_verify_iclog: invalid magic num");
3431
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003432 ptr = (xfs_caddr_t) &iclog->ic_header;
3433 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434 ptr += BBSIZE) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003435 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 xlog_panic("xlog_verify_iclog: unexpected magic num");
3437 }
3438
3439 /* check fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003440 len = be32_to_cpu(iclog->ic_header.h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 ptr = iclog->ic_datap;
3442 base_ptr = ptr;
3443 ophead = (xlog_op_header_t *)ptr;
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003444 xhdr = iclog->ic_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445 for (i = 0; i < len; i++) {
3446 ophead = (xlog_op_header_t *)ptr;
3447
3448 /* clientid is only 1 byte */
3449 field_offset = (__psint_t)
3450 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3451 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3452 clientid = ophead->oh_clientid;
3453 } else {
3454 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3455 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3456 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3457 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003458 clientid = xlog_get_client_id(
3459 xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 } else {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10003461 clientid = xlog_get_client_id(
3462 iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 }
3464 }
3465 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
Christoph Hellwigda1650a2005-11-02 10:21:35 +11003466 cmn_err(CE_WARN, "xlog_verify_iclog: "
3467 "invalid clientid %d op 0x%p offset 0x%lx",
3468 clientid, ophead, (unsigned long)field_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469
3470 /* check length */
3471 field_offset = (__psint_t)
3472 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3473 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10003474 op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 } else {
3476 idx = BTOBBT((__psint_t)&ophead->oh_len -
3477 (__psint_t)iclog->ic_datap);
3478 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3479 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3480 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003481 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 } else {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003483 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484 }
3485 }
3486 ptr += sizeof(xlog_op_header_t) + op_len;
3487 }
3488} /* xlog_verify_iclog */
Nathan Scottcfcbbbd2005-11-02 15:12:04 +11003489#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490
3491/*
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003492 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 */
3494STATIC int
3495xlog_state_ioerror(
3496 xlog_t *log)
3497{
3498 xlog_in_core_t *iclog, *ic;
3499
3500 iclog = log->l_iclog;
3501 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3502 /*
3503 * Mark all the incore logs IOERROR.
3504 * From now on, no log flushes will result.
3505 */
3506 ic = iclog;
3507 do {
3508 ic->ic_state = XLOG_STATE_IOERROR;
3509 ic = ic->ic_next;
3510 } while (ic != iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003511 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 }
3513 /*
3514 * Return non-zero, if state transition has already happened.
3515 */
Jesper Juhl014c2542006-01-15 02:37:08 +01003516 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517}
3518
3519/*
3520 * This is called from xfs_force_shutdown, when we're forcibly
3521 * shutting down the filesystem, typically because of an IO error.
3522 * Our main objectives here are to make sure that:
3523 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3524 * parties to find out, 'atomically'.
3525 * b. those who're sleeping on log reservations, pinned objects and
3526 * other resources get woken up, and be told the bad news.
3527 * c. nothing new gets queued up after (a) and (b) are done.
3528 * d. if !logerror, flush the iclogs to disk, then seal them off
3529 * for business.
3530 */
3531int
3532xfs_log_force_umount(
3533 struct xfs_mount *mp,
3534 int logerror)
3535{
3536 xlog_ticket_t *tic;
3537 xlog_t *log;
3538 int retval;
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003539 int dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
3541 log = mp->m_log;
3542
3543 /*
3544 * If this happens during log recovery, don't worry about
3545 * locking; the log isn't open for business yet.
3546 */
3547 if (!log ||
3548 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3549 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003550 if (mp->m_sb_bp)
3551 XFS_BUF_DONE(mp->m_sb_bp);
Jesper Juhl014c2542006-01-15 02:37:08 +01003552 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553 }
3554
3555 /*
3556 * Somebody could've already done the hard work for us.
3557 * No need to get locks for this.
3558 */
3559 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3560 ASSERT(XLOG_FORCED_SHUTDOWN(log));
Jesper Juhl014c2542006-01-15 02:37:08 +01003561 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 }
3563 retval = 0;
3564 /*
3565 * We must hold both the GRANT lock and the LOG lock,
3566 * before we mark the filesystem SHUTDOWN and wake
3567 * everybody up to tell the bad news.
3568 */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003569 spin_lock(&log->l_icloglock);
David Chinner6b1d1a72008-04-10 12:19:02 +10003570 spin_lock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
Christoph Hellwigbac8dca2008-11-28 14:23:31 +11003572 if (mp->m_sb_bp)
3573 XFS_BUF_DONE(mp->m_sb_bp);
3574
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 /*
3576 * This flag is sort of redundant because of the mount flag, but
3577 * it's good to maintain the separation between the log and the rest
3578 * of XFS.
3579 */
3580 log->l_flags |= XLOG_IO_ERROR;
3581
3582 /*
3583 * If we hit a log error, we want to mark all the iclogs IOERROR
3584 * while we're still holding the loglock.
3585 */
3586 if (logerror)
3587 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003588 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589
3590 /*
3591 * We don't want anybody waiting for log reservations
3592 * after this. That means we have to wake up everybody
3593 * queued up on reserve_headq as well as write_headq.
3594 * In addition, we make sure in xlog_{re}grant_log_space
3595 * that we don't enqueue anything once the SHUTDOWN flag
3596 * is set, and this action is protected by the GRANTLOCK.
3597 */
3598 if ((tic = log->l_reserve_headq)) {
3599 do {
David Chinner12017fa2008-08-13 16:34:31 +10003600 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601 tic = tic->t_next;
3602 } while (tic != log->l_reserve_headq);
3603 }
3604
3605 if ((tic = log->l_write_headq)) {
3606 do {
David Chinner12017fa2008-08-13 16:34:31 +10003607 sv_signal(&tic->t_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 tic = tic->t_next;
3609 } while (tic != log->l_write_headq);
3610 }
Eric Sandeenc8b5ea22007-10-11 17:37:31 +10003611 spin_unlock(&log->l_grant_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612
3613 if (! (log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
3614 ASSERT(!logerror);
3615 /*
3616 * Force the incore logs to disk before shutting the
3617 * log down completely.
3618 */
Christoph Hellwigf538d4d2005-11-02 10:26:59 +11003619 xlog_state_sync_all(log, XFS_LOG_FORCE|XFS_LOG_SYNC, &dummy);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003620 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 retval = xlog_state_ioerror(log);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003622 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 }
3624 /*
3625 * Wake up everybody waiting on xfs_log_force.
3626 * Callback all log item committed functions as if the
3627 * log writes were completed.
3628 */
3629 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3630
3631#ifdef XFSERRORDEBUG
3632 {
3633 xlog_in_core_t *iclog;
3634
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003635 spin_lock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 iclog = log->l_iclog;
3637 do {
3638 ASSERT(iclog->ic_callback == 0);
3639 iclog = iclog->ic_next;
3640 } while (iclog != log->l_iclog);
Eric Sandeenb22cd72c2007-10-11 17:37:10 +10003641 spin_unlock(&log->l_icloglock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642 }
3643#endif
3644 /* return non-zero if log IOERROR transition had already happened */
Jesper Juhl014c2542006-01-15 02:37:08 +01003645 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646}
3647
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10003648STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649xlog_iclogs_empty(xlog_t *log)
3650{
3651 xlog_in_core_t *iclog;
3652
3653 iclog = log->l_iclog;
3654 do {
3655 /* endianness does not matter here, zero is zero in
3656 * any language.
3657 */
3658 if (iclog->ic_header.h_num_logops)
Jesper Juhl014c2542006-01-15 02:37:08 +01003659 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 iclog = iclog->ic_next;
3661 } while (iclog != log->l_iclog);
Jesper Juhl014c2542006-01-15 02:37:08 +01003662 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663}