]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
a844f451 NS |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" | |
a844f451 | 27 | #include "xfs_dir2.h" |
1da177e4 LT |
28 | #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" | |
a844f451 | 33 | #include "xfs_bmap_btree.h" |
1da177e4 | 34 | #include "xfs_alloc_btree.h" |
a844f451 | 35 | #include "xfs_ialloc_btree.h" |
1da177e4 | 36 | #include "xfs_log_recover.h" |
1da177e4 | 37 | #include "xfs_trans_priv.h" |
a844f451 NS |
38 | #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" | |
0b1b213f | 43 | #include "xfs_trace.h" |
1da177e4 | 44 | |
eb01c9cd | 45 | kmem_zone_t *xfs_log_ticket_zone; |
1da177e4 | 46 | |
1da177e4 | 47 | /* Local miscellaneous function prototypes */ |
55b66332 | 48 | STATIC int xlog_commit_record(struct log *log, struct xlog_ticket *ticket, |
1da177e4 LT |
49 | xlog_in_core_t **, xfs_lsn_t *); |
50 | STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp, | |
51 | xfs_buftarg_t *log_target, | |
52 | xfs_daddr_t blk_offset, | |
53 | int num_bblks); | |
54 | STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes); | |
55 | STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog); | |
c41564b5 | 56 | STATIC void xlog_dealloc_log(xlog_t *log); |
55b66332 DC |
57 | STATIC int xlog_write(struct log *log, struct xfs_log_vec *log_vector, |
58 | struct xlog_ticket *tic, xfs_lsn_t *start_lsn, | |
59 | xlog_in_core_t **commit_iclog, uint flags); | |
1da177e4 LT |
60 | |
61 | /* local state machine functions */ | |
62 | STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); | |
63 | STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog); | |
64 | STATIC int xlog_state_get_iclog_space(xlog_t *log, | |
65 | int len, | |
66 | xlog_in_core_t **iclog, | |
67 | xlog_ticket_t *ticket, | |
68 | int *continued_write, | |
69 | int *logoffsetp); | |
1da177e4 LT |
70 | STATIC int xlog_state_release_iclog(xlog_t *log, |
71 | xlog_in_core_t *iclog); | |
72 | STATIC void xlog_state_switch_iclogs(xlog_t *log, | |
73 | xlog_in_core_t *iclog, | |
74 | int eventual_size); | |
1da177e4 LT |
75 | STATIC void xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog); |
76 | ||
77 | /* local functions to manipulate grant head */ | |
78 | STATIC int xlog_grant_log_space(xlog_t *log, | |
79 | xlog_ticket_t *xtic); | |
80 | STATIC void xlog_grant_push_ail(xfs_mount_t *mp, | |
81 | int need_bytes); | |
82 | STATIC void xlog_regrant_reserve_log_space(xlog_t *log, | |
83 | xlog_ticket_t *ticket); | |
84 | STATIC int xlog_regrant_write_log_space(xlog_t *log, | |
85 | xlog_ticket_t *ticket); | |
86 | STATIC void xlog_ungrant_log_space(xlog_t *log, | |
87 | xlog_ticket_t *ticket); | |
88 | ||
89 | ||
90 | /* local ticket functions */ | |
3383ca57 DC |
91 | STATIC xlog_ticket_t *xlog_ticket_alloc(xlog_t *log, int unit_bytes, int count, |
92 | char clientid, uint flags, | |
93 | int alloc_flags); | |
1da177e4 | 94 | |
cfcbbbd0 | 95 | #if defined(DEBUG) |
e6b1f273 | 96 | STATIC void xlog_verify_dest_ptr(xlog_t *log, char *ptr); |
1da177e4 LT |
97 | STATIC void xlog_verify_grant_head(xlog_t *log, int equals); |
98 | STATIC void xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog, | |
99 | int count, boolean_t syncing); | |
100 | STATIC void xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog, | |
101 | xfs_lsn_t tail_lsn); | |
102 | #else | |
103 | #define xlog_verify_dest_ptr(a,b) | |
104 | #define xlog_verify_grant_head(a,b) | |
105 | #define xlog_verify_iclog(a,b,c,d) | |
106 | #define xlog_verify_tail_lsn(a,b,c) | |
107 | #endif | |
108 | ||
ba0f32d4 | 109 | STATIC int xlog_iclogs_empty(xlog_t *log); |
1da177e4 | 110 | |
dd954c69 CH |
111 | |
112 | static void | |
113 | xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic) | |
114 | { | |
115 | if (*qp) { | |
116 | tic->t_next = (*qp); | |
117 | tic->t_prev = (*qp)->t_prev; | |
118 | (*qp)->t_prev->t_next = tic; | |
119 | (*qp)->t_prev = tic; | |
120 | } else { | |
121 | tic->t_prev = tic->t_next = tic; | |
122 | *qp = tic; | |
123 | } | |
124 | ||
125 | tic->t_flags |= XLOG_TIC_IN_Q; | |
126 | } | |
127 | ||
128 | static void | |
129 | xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic) | |
130 | { | |
131 | if (tic == tic->t_next) { | |
132 | *qp = NULL; | |
133 | } else { | |
134 | *qp = tic->t_next; | |
135 | tic->t_next->t_prev = tic->t_prev; | |
136 | tic->t_prev->t_next = tic->t_next; | |
137 | } | |
138 | ||
139 | tic->t_next = tic->t_prev = NULL; | |
140 | tic->t_flags &= ~XLOG_TIC_IN_Q; | |
141 | } | |
142 | ||
143 | static void | |
144 | xlog_grant_sub_space(struct log *log, int bytes) | |
145 | { | |
146 | log->l_grant_write_bytes -= bytes; | |
147 | if (log->l_grant_write_bytes < 0) { | |
148 | log->l_grant_write_bytes += log->l_logsize; | |
149 | log->l_grant_write_cycle--; | |
150 | } | |
151 | ||
152 | log->l_grant_reserve_bytes -= bytes; | |
153 | if ((log)->l_grant_reserve_bytes < 0) { | |
154 | log->l_grant_reserve_bytes += log->l_logsize; | |
155 | log->l_grant_reserve_cycle--; | |
156 | } | |
157 | ||
158 | } | |
159 | ||
160 | static void | |
161 | xlog_grant_add_space_write(struct log *log, int bytes) | |
162 | { | |
d729eae8 MN |
163 | int tmp = log->l_logsize - log->l_grant_write_bytes; |
164 | if (tmp > bytes) | |
165 | log->l_grant_write_bytes += bytes; | |
166 | else { | |
dd954c69 | 167 | log->l_grant_write_cycle++; |
d729eae8 | 168 | log->l_grant_write_bytes = bytes - tmp; |
dd954c69 CH |
169 | } |
170 | } | |
171 | ||
172 | static void | |
173 | xlog_grant_add_space_reserve(struct log *log, int bytes) | |
174 | { | |
d729eae8 MN |
175 | int tmp = log->l_logsize - log->l_grant_reserve_bytes; |
176 | if (tmp > bytes) | |
177 | log->l_grant_reserve_bytes += bytes; | |
178 | else { | |
dd954c69 | 179 | log->l_grant_reserve_cycle++; |
d729eae8 | 180 | log->l_grant_reserve_bytes = bytes - tmp; |
dd954c69 CH |
181 | } |
182 | } | |
183 | ||
184 | static inline void | |
185 | xlog_grant_add_space(struct log *log, int bytes) | |
186 | { | |
187 | xlog_grant_add_space_write(log, bytes); | |
188 | xlog_grant_add_space_reserve(log, bytes); | |
189 | } | |
190 | ||
0adba536 CH |
191 | static void |
192 | xlog_tic_reset_res(xlog_ticket_t *tic) | |
193 | { | |
194 | tic->t_res_num = 0; | |
195 | tic->t_res_arr_sum = 0; | |
196 | tic->t_res_num_ophdrs = 0; | |
197 | } | |
198 | ||
199 | static void | |
200 | xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) | |
201 | { | |
202 | if (tic->t_res_num == XLOG_TIC_LEN_MAX) { | |
203 | /* add to overflow and start again */ | |
204 | tic->t_res_o_flow += tic->t_res_arr_sum; | |
205 | tic->t_res_num = 0; | |
206 | tic->t_res_arr_sum = 0; | |
207 | } | |
208 | ||
209 | tic->t_res_arr[tic->t_res_num].r_len = len; | |
210 | tic->t_res_arr[tic->t_res_num].r_type = type; | |
211 | tic->t_res_arr_sum += len; | |
212 | tic->t_res_num++; | |
213 | } | |
dd954c69 | 214 | |
1da177e4 LT |
215 | /* |
216 | * NOTES: | |
217 | * | |
218 | * 1. currblock field gets updated at startup and after in-core logs | |
219 | * marked as with WANT_SYNC. | |
220 | */ | |
221 | ||
222 | /* | |
223 | * This routine is called when a user of a log manager ticket is done with | |
224 | * the reservation. If the ticket was ever used, then a commit record for | |
225 | * the associated transaction is written out as a log operation header with | |
226 | * no data. The flag XLOG_TIC_INITED is set when the first write occurs with | |
227 | * a given ticket. If the ticket was one with a permanent reservation, then | |
228 | * a few operations are done differently. Permanent reservation tickets by | |
229 | * default don't release the reservation. They just commit the current | |
230 | * transaction with the belief that the reservation is still needed. A flag | |
231 | * must be passed in before permanent reservations are actually released. | |
232 | * When these type of tickets are not released, they need to be set into | |
233 | * the inited state again. By doing this, a start record will be written | |
234 | * out when the next write occurs. | |
235 | */ | |
236 | xfs_lsn_t | |
35a8a72f CH |
237 | xfs_log_done( |
238 | struct xfs_mount *mp, | |
239 | struct xlog_ticket *ticket, | |
240 | struct xlog_in_core **iclog, | |
241 | uint flags) | |
1da177e4 | 242 | { |
35a8a72f CH |
243 | struct log *log = mp->m_log; |
244 | xfs_lsn_t lsn = 0; | |
1da177e4 | 245 | |
1da177e4 LT |
246 | if (XLOG_FORCED_SHUTDOWN(log) || |
247 | /* | |
248 | * If nothing was ever written, don't write out commit record. | |
249 | * If we get an error, just continue and give back the log ticket. | |
250 | */ | |
251 | (((ticket->t_flags & XLOG_TIC_INITED) == 0) && | |
55b66332 | 252 | (xlog_commit_record(log, ticket, iclog, &lsn)))) { |
1da177e4 LT |
253 | lsn = (xfs_lsn_t) -1; |
254 | if (ticket->t_flags & XLOG_TIC_PERM_RESERV) { | |
255 | flags |= XFS_LOG_REL_PERM_RESERV; | |
256 | } | |
257 | } | |
258 | ||
259 | ||
260 | if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 || | |
261 | (flags & XFS_LOG_REL_PERM_RESERV)) { | |
0b1b213f CH |
262 | trace_xfs_log_done_nonperm(log, ticket); |
263 | ||
1da177e4 | 264 | /* |
c41564b5 | 265 | * Release ticket if not permanent reservation or a specific |
1da177e4 LT |
266 | * request has been made to release a permanent reservation. |
267 | */ | |
268 | xlog_ungrant_log_space(log, ticket); | |
cc09c0dc | 269 | xfs_log_ticket_put(ticket); |
1da177e4 | 270 | } else { |
0b1b213f CH |
271 | trace_xfs_log_done_perm(log, ticket); |
272 | ||
1da177e4 | 273 | xlog_regrant_reserve_log_space(log, ticket); |
c6a7b0f8 LM |
274 | /* If this ticket was a permanent reservation and we aren't |
275 | * trying to release it, reset the inited flags; so next time | |
276 | * we write, a start record will be written out. | |
277 | */ | |
1da177e4 | 278 | ticket->t_flags |= XLOG_TIC_INITED; |
c6a7b0f8 | 279 | } |
1da177e4 LT |
280 | |
281 | return lsn; | |
35a8a72f | 282 | } |
1da177e4 | 283 | |
1da177e4 LT |
284 | /* |
285 | * Attaches a new iclog I/O completion callback routine during | |
286 | * transaction commit. If the log is in error state, a non-zero | |
287 | * return code is handed back and the caller is responsible for | |
288 | * executing the callback at an appropriate time. | |
289 | */ | |
290 | int | |
35a8a72f CH |
291 | xfs_log_notify( |
292 | struct xfs_mount *mp, | |
293 | struct xlog_in_core *iclog, | |
294 | xfs_log_callback_t *cb) | |
1da177e4 | 295 | { |
b22cd72c | 296 | int abortflg; |
1da177e4 | 297 | |
114d23aa | 298 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
299 | abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); |
300 | if (!abortflg) { | |
301 | ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || | |
302 | (iclog->ic_state == XLOG_STATE_WANT_SYNC)); | |
303 | cb->cb_next = NULL; | |
304 | *(iclog->ic_callback_tail) = cb; | |
305 | iclog->ic_callback_tail = &(cb->cb_next); | |
306 | } | |
114d23aa | 307 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 | 308 | return abortflg; |
35a8a72f | 309 | } |
1da177e4 LT |
310 | |
311 | int | |
35a8a72f CH |
312 | xfs_log_release_iclog( |
313 | struct xfs_mount *mp, | |
314 | struct xlog_in_core *iclog) | |
1da177e4 | 315 | { |
35a8a72f | 316 | if (xlog_state_release_iclog(mp->m_log, iclog)) { |
7d04a335 | 317 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 318 | return EIO; |
1da177e4 LT |
319 | } |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
324 | /* | |
325 | * 1. Reserve an amount of on-disk log space and return a ticket corresponding | |
326 | * to the reservation. | |
327 | * 2. Potentially, push buffers at tail of log to disk. | |
328 | * | |
329 | * Each reservation is going to reserve extra space for a log record header. | |
330 | * When writes happen to the on-disk log, we don't subtract the length of the | |
331 | * log record header from any reservation. By wasting space in each | |
332 | * reservation, we prevent over allocation problems. | |
333 | */ | |
334 | int | |
35a8a72f CH |
335 | xfs_log_reserve( |
336 | struct xfs_mount *mp, | |
337 | int unit_bytes, | |
338 | int cnt, | |
339 | struct xlog_ticket **ticket, | |
340 | __uint8_t client, | |
341 | uint flags, | |
342 | uint t_type) | |
1da177e4 | 343 | { |
35a8a72f CH |
344 | struct log *log = mp->m_log; |
345 | struct xlog_ticket *internal_ticket; | |
346 | int retval = 0; | |
1da177e4 | 347 | |
1da177e4 LT |
348 | ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); |
349 | ASSERT((flags & XFS_LOG_NOSLEEP) == 0); | |
350 | ||
351 | if (XLOG_FORCED_SHUTDOWN(log)) | |
352 | return XFS_ERROR(EIO); | |
353 | ||
354 | XFS_STATS_INC(xs_try_logspace); | |
355 | ||
0b1b213f | 356 | |
1da177e4 LT |
357 | if (*ticket != NULL) { |
358 | ASSERT(flags & XFS_LOG_PERM_RESERV); | |
35a8a72f | 359 | internal_ticket = *ticket; |
0b1b213f | 360 | |
524ee36f DC |
361 | /* |
362 | * this is a new transaction on the ticket, so we need to | |
363 | * change the transaction ID so that the next transaction has a | |
364 | * different TID in the log. Just add one to the existing tid | |
365 | * so that we can see chains of rolling transactions in the log | |
366 | * easily. | |
367 | */ | |
368 | internal_ticket->t_tid++; | |
369 | ||
0b1b213f CH |
370 | trace_xfs_log_reserve(log, internal_ticket); |
371 | ||
1da177e4 LT |
372 | xlog_grant_push_ail(mp, internal_ticket->t_unit_res); |
373 | retval = xlog_regrant_write_log_space(log, internal_ticket); | |
374 | } else { | |
375 | /* may sleep if need to allocate more tickets */ | |
cc09c0dc | 376 | internal_ticket = xlog_ticket_alloc(log, unit_bytes, cnt, |
3383ca57 DC |
377 | client, flags, |
378 | KM_SLEEP|KM_MAYFAIL); | |
eb01c9cd DC |
379 | if (!internal_ticket) |
380 | return XFS_ERROR(ENOMEM); | |
7e9c6396 | 381 | internal_ticket->t_trans_type = t_type; |
1da177e4 | 382 | *ticket = internal_ticket; |
0b1b213f CH |
383 | |
384 | trace_xfs_log_reserve(log, internal_ticket); | |
385 | ||
1da177e4 LT |
386 | xlog_grant_push_ail(mp, |
387 | (internal_ticket->t_unit_res * | |
388 | internal_ticket->t_cnt)); | |
389 | retval = xlog_grant_log_space(log, internal_ticket); | |
390 | } | |
391 | ||
392 | return retval; | |
393 | } /* xfs_log_reserve */ | |
394 | ||
395 | ||
396 | /* | |
397 | * Mount a log filesystem | |
398 | * | |
399 | * mp - ubiquitous xfs mount point structure | |
400 | * log_target - buftarg of on-disk log device | |
401 | * blk_offset - Start block # where block size is 512 bytes (BBSIZE) | |
402 | * num_bblocks - Number of BBSIZE blocks in on-disk log | |
403 | * | |
404 | * Return error or zero. | |
405 | */ | |
406 | int | |
249a8c11 DC |
407 | xfs_log_mount( |
408 | xfs_mount_t *mp, | |
409 | xfs_buftarg_t *log_target, | |
410 | xfs_daddr_t blk_offset, | |
411 | int num_bblks) | |
1da177e4 | 412 | { |
249a8c11 DC |
413 | int error; |
414 | ||
1da177e4 LT |
415 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) |
416 | cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname); | |
417 | else { | |
418 | cmn_err(CE_NOTE, | |
419 | "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.", | |
420 | mp->m_fsname); | |
bd186aa9 | 421 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
422 | } |
423 | ||
424 | mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); | |
a6cb767e DC |
425 | if (IS_ERR(mp->m_log)) { |
426 | error = -PTR_ERR(mp->m_log); | |
644c3567 DC |
427 | goto out; |
428 | } | |
1da177e4 | 429 | |
249a8c11 DC |
430 | /* |
431 | * Initialize the AIL now we have a log. | |
432 | */ | |
249a8c11 DC |
433 | error = xfs_trans_ail_init(mp); |
434 | if (error) { | |
435 | cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error); | |
26430752 | 436 | goto out_free_log; |
249a8c11 | 437 | } |
a9c21c1b | 438 | mp->m_log->l_ailp = mp->m_ail; |
249a8c11 | 439 | |
1da177e4 LT |
440 | /* |
441 | * skip log recovery on a norecovery mount. pretend it all | |
442 | * just worked. | |
443 | */ | |
444 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { | |
249a8c11 | 445 | int readonly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
446 | |
447 | if (readonly) | |
bd186aa9 | 448 | mp->m_flags &= ~XFS_MOUNT_RDONLY; |
1da177e4 | 449 | |
65be6054 | 450 | error = xlog_recover(mp->m_log); |
1da177e4 LT |
451 | |
452 | if (readonly) | |
bd186aa9 | 453 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 LT |
454 | if (error) { |
455 | cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error); | |
26430752 | 456 | goto out_destroy_ail; |
1da177e4 LT |
457 | } |
458 | } | |
459 | ||
460 | /* Normal transactions can now occur */ | |
461 | mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; | |
462 | ||
1da177e4 | 463 | return 0; |
26430752 CH |
464 | |
465 | out_destroy_ail: | |
466 | xfs_trans_ail_destroy(mp); | |
467 | out_free_log: | |
468 | xlog_dealloc_log(mp->m_log); | |
644c3567 | 469 | out: |
249a8c11 | 470 | return error; |
26430752 | 471 | } |
1da177e4 LT |
472 | |
473 | /* | |
474 | * Finish the recovery of the file system. This is separate from | |
475 | * the xfs_log_mount() call, because it depends on the code in | |
476 | * xfs_mountfs() to read in the root and real-time bitmap inodes | |
477 | * between calling xfs_log_mount() and here. | |
478 | * | |
479 | * mp - ubiquitous xfs mount point structure | |
480 | */ | |
481 | int | |
4249023a | 482 | xfs_log_mount_finish(xfs_mount_t *mp) |
1da177e4 LT |
483 | { |
484 | int error; | |
485 | ||
486 | if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) | |
4249023a | 487 | error = xlog_recover_finish(mp->m_log); |
1da177e4 LT |
488 | else { |
489 | error = 0; | |
bd186aa9 | 490 | ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
491 | } |
492 | ||
493 | return error; | |
494 | } | |
495 | ||
1da177e4 LT |
496 | /* |
497 | * Final log writes as part of unmount. | |
498 | * | |
499 | * Mark the filesystem clean as unmount happens. Note that during relocation | |
500 | * this routine needs to be executed as part of source-bag while the | |
501 | * deallocation must not be done until source-end. | |
502 | */ | |
503 | ||
504 | /* | |
505 | * Unmount record used to have a string "Unmount filesystem--" in the | |
506 | * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). | |
507 | * We just write the magic number now since that particular field isn't | |
508 | * currently architecture converted and "nUmount" is a bit foo. | |
509 | * As far as I know, there weren't any dependencies on the old behaviour. | |
510 | */ | |
511 | ||
512 | int | |
513 | xfs_log_unmount_write(xfs_mount_t *mp) | |
514 | { | |
515 | xlog_t *log = mp->m_log; | |
516 | xlog_in_core_t *iclog; | |
517 | #ifdef DEBUG | |
518 | xlog_in_core_t *first_iclog; | |
519 | #endif | |
35a8a72f | 520 | xlog_ticket_t *tic = NULL; |
1da177e4 LT |
521 | xfs_lsn_t lsn; |
522 | int error; | |
1da177e4 | 523 | |
1da177e4 LT |
524 | /* |
525 | * Don't write out unmount record on read-only mounts. | |
526 | * Or, if we are doing a forced umount (typically because of IO errors). | |
527 | */ | |
bd186aa9 | 528 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
529 | return 0; |
530 | ||
a14a348b | 531 | error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
b911ca04 | 532 | ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); |
1da177e4 LT |
533 | |
534 | #ifdef DEBUG | |
535 | first_iclog = iclog = log->l_iclog; | |
536 | do { | |
537 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
538 | ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE); | |
539 | ASSERT(iclog->ic_offset == 0); | |
540 | } | |
541 | iclog = iclog->ic_next; | |
542 | } while (iclog != first_iclog); | |
543 | #endif | |
544 | if (! (XLOG_FORCED_SHUTDOWN(log))) { | |
955e47ad TS |
545 | error = xfs_log_reserve(mp, 600, 1, &tic, |
546 | XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE); | |
1da177e4 | 547 | if (!error) { |
55b66332 DC |
548 | /* the data section must be 32 bit size aligned */ |
549 | struct { | |
550 | __uint16_t magic; | |
551 | __uint16_t pad1; | |
552 | __uint32_t pad2; /* may as well make it 64 bits */ | |
553 | } magic = { | |
554 | .magic = XLOG_UNMOUNT_TYPE, | |
555 | }; | |
556 | struct xfs_log_iovec reg = { | |
557 | .i_addr = (void *)&magic, | |
558 | .i_len = sizeof(magic), | |
559 | .i_type = XLOG_REG_TYPE_UNMOUNT, | |
560 | }; | |
561 | struct xfs_log_vec vec = { | |
562 | .lv_niovecs = 1, | |
563 | .lv_iovecp = ®, | |
564 | }; | |
565 | ||
1da177e4 | 566 | /* remove inited flag */ |
55b66332 DC |
567 | tic->t_flags = 0; |
568 | error = xlog_write(log, &vec, tic, &lsn, | |
1da177e4 LT |
569 | NULL, XLOG_UNMOUNT_TRANS); |
570 | /* | |
571 | * At this point, we're umounting anyway, | |
572 | * so there's no point in transitioning log state | |
573 | * to IOERROR. Just continue... | |
574 | */ | |
575 | } | |
576 | ||
577 | if (error) { | |
578 | xfs_fs_cmn_err(CE_ALERT, mp, | |
579 | "xfs_log_unmount: unmount record failed"); | |
580 | } | |
581 | ||
582 | ||
b22cd72c | 583 | spin_lock(&log->l_icloglock); |
1da177e4 | 584 | iclog = log->l_iclog; |
155cc6b7 | 585 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 586 | xlog_state_want_sync(log, iclog); |
39e2defe | 587 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 588 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 589 | |
b22cd72c | 590 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
591 | if (!(iclog->ic_state == XLOG_STATE_ACTIVE || |
592 | iclog->ic_state == XLOG_STATE_DIRTY)) { | |
593 | if (!XLOG_FORCED_SHUTDOWN(log)) { | |
12017faf | 594 | sv_wait(&iclog->ic_force_wait, PMEM, |
1da177e4 LT |
595 | &log->l_icloglock, s); |
596 | } else { | |
b22cd72c | 597 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
598 | } |
599 | } else { | |
b22cd72c | 600 | spin_unlock(&log->l_icloglock); |
1da177e4 | 601 | } |
955e47ad | 602 | if (tic) { |
0b1b213f | 603 | trace_xfs_log_umount_write(log, tic); |
955e47ad | 604 | xlog_ungrant_log_space(log, tic); |
cc09c0dc | 605 | xfs_log_ticket_put(tic); |
955e47ad | 606 | } |
1da177e4 LT |
607 | } else { |
608 | /* | |
609 | * We're already in forced_shutdown mode, couldn't | |
610 | * even attempt to write out the unmount transaction. | |
611 | * | |
612 | * Go through the motions of sync'ing and releasing | |
613 | * the iclog, even though no I/O will actually happen, | |
c41564b5 | 614 | * we need to wait for other log I/Os that may already |
1da177e4 LT |
615 | * be in progress. Do this as a separate section of |
616 | * code so we'll know if we ever get stuck here that | |
617 | * we're in this odd situation of trying to unmount | |
618 | * a file system that went into forced_shutdown as | |
619 | * the result of an unmount.. | |
620 | */ | |
b22cd72c | 621 | spin_lock(&log->l_icloglock); |
1da177e4 | 622 | iclog = log->l_iclog; |
155cc6b7 | 623 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 LT |
624 | |
625 | xlog_state_want_sync(log, iclog); | |
39e2defe | 626 | spin_unlock(&log->l_icloglock); |
1bb7d6b5 | 627 | error = xlog_state_release_iclog(log, iclog); |
1da177e4 | 628 | |
b22cd72c | 629 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
630 | |
631 | if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE | |
632 | || iclog->ic_state == XLOG_STATE_DIRTY | |
633 | || iclog->ic_state == XLOG_STATE_IOERROR) ) { | |
634 | ||
12017faf | 635 | sv_wait(&iclog->ic_force_wait, PMEM, |
1da177e4 LT |
636 | &log->l_icloglock, s); |
637 | } else { | |
b22cd72c | 638 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
639 | } |
640 | } | |
641 | ||
1bb7d6b5 | 642 | return error; |
1da177e4 LT |
643 | } /* xfs_log_unmount_write */ |
644 | ||
645 | /* | |
646 | * Deallocate log structures for unmount/relocation. | |
249a8c11 DC |
647 | * |
648 | * We need to stop the aild from running before we destroy | |
649 | * and deallocate the log as the aild references the log. | |
1da177e4 LT |
650 | */ |
651 | void | |
21b699c8 | 652 | xfs_log_unmount(xfs_mount_t *mp) |
1da177e4 | 653 | { |
249a8c11 | 654 | xfs_trans_ail_destroy(mp); |
c41564b5 | 655 | xlog_dealloc_log(mp->m_log); |
1da177e4 LT |
656 | } |
657 | ||
43f5efc5 DC |
658 | void |
659 | xfs_log_item_init( | |
660 | struct xfs_mount *mp, | |
661 | struct xfs_log_item *item, | |
662 | int type, | |
663 | struct xfs_item_ops *ops) | |
664 | { | |
665 | item->li_mountp = mp; | |
666 | item->li_ailp = mp->m_ail; | |
667 | item->li_type = type; | |
668 | item->li_ops = ops; | |
669 | } | |
670 | ||
1da177e4 LT |
671 | /* |
672 | * Write region vectors to log. The write happens using the space reservation | |
673 | * of the ticket (tic). It is not a requirement that all writes for a given | |
9b9fc2b7 DC |
674 | * transaction occur with one call to xfs_log_write(). However, it is important |
675 | * to note that the transaction reservation code makes an assumption about the | |
676 | * number of log headers a transaction requires that may be violated if you | |
677 | * don't pass all the transaction vectors in one call.... | |
1da177e4 LT |
678 | */ |
679 | int | |
35a8a72f CH |
680 | xfs_log_write( |
681 | struct xfs_mount *mp, | |
682 | struct xfs_log_iovec reg[], | |
683 | int nentries, | |
684 | struct xlog_ticket *tic, | |
685 | xfs_lsn_t *start_lsn) | |
1da177e4 | 686 | { |
35a8a72f CH |
687 | struct log *log = mp->m_log; |
688 | int error; | |
55b66332 DC |
689 | struct xfs_log_vec vec = { |
690 | .lv_niovecs = nentries, | |
691 | .lv_iovecp = reg, | |
692 | }; | |
1da177e4 | 693 | |
1da177e4 LT |
694 | if (XLOG_FORCED_SHUTDOWN(log)) |
695 | return XFS_ERROR(EIO); | |
696 | ||
55b66332 | 697 | error = xlog_write(log, &vec, tic, start_lsn, NULL, 0); |
35a8a72f | 698 | if (error) |
7d04a335 | 699 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 700 | return error; |
35a8a72f | 701 | } |
1da177e4 LT |
702 | |
703 | void | |
704 | xfs_log_move_tail(xfs_mount_t *mp, | |
705 | xfs_lsn_t tail_lsn) | |
706 | { | |
707 | xlog_ticket_t *tic; | |
708 | xlog_t *log = mp->m_log; | |
709 | int need_bytes, free_bytes, cycle, bytes; | |
1da177e4 | 710 | |
1da177e4 LT |
711 | if (XLOG_FORCED_SHUTDOWN(log)) |
712 | return; | |
1da177e4 LT |
713 | |
714 | if (tail_lsn == 0) { | |
715 | /* needed since sync_lsn is 64 bits */ | |
b22cd72c | 716 | spin_lock(&log->l_icloglock); |
1da177e4 | 717 | tail_lsn = log->l_last_sync_lsn; |
b22cd72c | 718 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
719 | } |
720 | ||
c8b5ea28 | 721 | spin_lock(&log->l_grant_lock); |
1da177e4 LT |
722 | |
723 | /* Also an invalid lsn. 1 implies that we aren't passing in a valid | |
724 | * tail_lsn. | |
725 | */ | |
726 | if (tail_lsn != 1) { | |
727 | log->l_tail_lsn = tail_lsn; | |
728 | } | |
729 | ||
730 | if ((tic = log->l_write_headq)) { | |
731 | #ifdef DEBUG | |
732 | if (log->l_flags & XLOG_ACTIVE_RECOVERY) | |
733 | panic("Recovery problem"); | |
734 | #endif | |
735 | cycle = log->l_grant_write_cycle; | |
736 | bytes = log->l_grant_write_bytes; | |
737 | free_bytes = xlog_space_left(log, cycle, bytes); | |
738 | do { | |
739 | ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); | |
740 | ||
741 | if (free_bytes < tic->t_unit_res && tail_lsn != 1) | |
742 | break; | |
743 | tail_lsn = 0; | |
744 | free_bytes -= tic->t_unit_res; | |
12017faf | 745 | sv_signal(&tic->t_wait); |
1da177e4 LT |
746 | tic = tic->t_next; |
747 | } while (tic != log->l_write_headq); | |
748 | } | |
749 | if ((tic = log->l_reserve_headq)) { | |
750 | #ifdef DEBUG | |
751 | if (log->l_flags & XLOG_ACTIVE_RECOVERY) | |
752 | panic("Recovery problem"); | |
753 | #endif | |
754 | cycle = log->l_grant_reserve_cycle; | |
755 | bytes = log->l_grant_reserve_bytes; | |
756 | free_bytes = xlog_space_left(log, cycle, bytes); | |
757 | do { | |
758 | if (tic->t_flags & XLOG_TIC_PERM_RESERV) | |
759 | need_bytes = tic->t_unit_res*tic->t_cnt; | |
760 | else | |
761 | need_bytes = tic->t_unit_res; | |
762 | if (free_bytes < need_bytes && tail_lsn != 1) | |
763 | break; | |
764 | tail_lsn = 0; | |
765 | free_bytes -= need_bytes; | |
12017faf | 766 | sv_signal(&tic->t_wait); |
1da177e4 LT |
767 | tic = tic->t_next; |
768 | } while (tic != log->l_reserve_headq); | |
769 | } | |
c8b5ea28 | 770 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
771 | } /* xfs_log_move_tail */ |
772 | ||
773 | /* | |
774 | * Determine if we have a transaction that has gone to disk | |
b6f8dd49 DC |
775 | * that needs to be covered. To begin the transition to the idle state |
776 | * firstly the log needs to be idle (no AIL and nothing in the iclogs). | |
777 | * If we are then in a state where covering is needed, the caller is informed | |
778 | * that dummy transactions are required to move the log into the idle state. | |
779 | * | |
780 | * Because this is called as part of the sync process, we should also indicate | |
781 | * that dummy transactions should be issued in anything but the covered or | |
782 | * idle states. This ensures that the log tail is accurately reflected in | |
783 | * the log at the end of the sync, hence if a crash occurrs avoids replay | |
784 | * of transactions where the metadata is already on disk. | |
1da177e4 LT |
785 | */ |
786 | int | |
787 | xfs_log_need_covered(xfs_mount_t *mp) | |
788 | { | |
27d8d5fe | 789 | int needed = 0; |
1da177e4 | 790 | xlog_t *log = mp->m_log; |
1da177e4 | 791 | |
92821e2b | 792 | if (!xfs_fs_writable(mp)) |
1da177e4 LT |
793 | return 0; |
794 | ||
b22cd72c | 795 | spin_lock(&log->l_icloglock); |
b6f8dd49 DC |
796 | switch (log->l_covered_state) { |
797 | case XLOG_STATE_COVER_DONE: | |
798 | case XLOG_STATE_COVER_DONE2: | |
799 | case XLOG_STATE_COVER_IDLE: | |
800 | break; | |
801 | case XLOG_STATE_COVER_NEED: | |
802 | case XLOG_STATE_COVER_NEED2: | |
803 | if (!xfs_trans_ail_tail(log->l_ailp) && | |
804 | xlog_iclogs_empty(log)) { | |
805 | if (log->l_covered_state == XLOG_STATE_COVER_NEED) | |
806 | log->l_covered_state = XLOG_STATE_COVER_DONE; | |
807 | else | |
808 | log->l_covered_state = XLOG_STATE_COVER_DONE2; | |
1da177e4 | 809 | } |
b6f8dd49 DC |
810 | /* FALLTHRU */ |
811 | default: | |
1da177e4 | 812 | needed = 1; |
b6f8dd49 | 813 | break; |
1da177e4 | 814 | } |
b22cd72c | 815 | spin_unlock(&log->l_icloglock); |
014c2544 | 816 | return needed; |
1da177e4 LT |
817 | } |
818 | ||
819 | /****************************************************************************** | |
820 | * | |
821 | * local routines | |
822 | * | |
823 | ****************************************************************************** | |
824 | */ | |
825 | ||
826 | /* xfs_trans_tail_ail returns 0 when there is nothing in the list. | |
827 | * The log manager must keep track of the last LR which was committed | |
828 | * to disk. The lsn of this LR will become the new tail_lsn whenever | |
829 | * xfs_trans_tail_ail returns 0. If we don't do this, we run into | |
830 | * the situation where stuff could be written into the log but nothing | |
831 | * was ever in the AIL when asked. Eventually, we panic since the | |
832 | * tail hits the head. | |
833 | * | |
834 | * We may be holding the log iclog lock upon entering this routine. | |
835 | */ | |
836 | xfs_lsn_t | |
837 | xlog_assign_tail_lsn(xfs_mount_t *mp) | |
838 | { | |
839 | xfs_lsn_t tail_lsn; | |
1da177e4 LT |
840 | xlog_t *log = mp->m_log; |
841 | ||
5b00f14f | 842 | tail_lsn = xfs_trans_ail_tail(mp->m_ail); |
c8b5ea28 | 843 | spin_lock(&log->l_grant_lock); |
1da177e4 LT |
844 | if (tail_lsn != 0) { |
845 | log->l_tail_lsn = tail_lsn; | |
846 | } else { | |
847 | tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn; | |
848 | } | |
c8b5ea28 | 849 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
850 | |
851 | return tail_lsn; | |
852 | } /* xlog_assign_tail_lsn */ | |
853 | ||
854 | ||
855 | /* | |
856 | * Return the space in the log between the tail and the head. The head | |
857 | * is passed in the cycle/bytes formal parms. In the special case where | |
858 | * the reserve head has wrapped passed the tail, this calculation is no | |
859 | * longer valid. In this case, just return 0 which means there is no space | |
860 | * in the log. This works for all places where this function is called | |
861 | * with the reserve head. Of course, if the write head were to ever | |
862 | * wrap the tail, we should blow up. Rather than catch this case here, | |
863 | * we depend on other ASSERTions in other parts of the code. XXXmiken | |
864 | * | |
865 | * This code also handles the case where the reservation head is behind | |
866 | * the tail. The details of this case are described below, but the end | |
867 | * result is that we return the size of the log as the amount of space left. | |
868 | */ | |
a8272ce0 | 869 | STATIC int |
1da177e4 LT |
870 | xlog_space_left(xlog_t *log, int cycle, int bytes) |
871 | { | |
872 | int free_bytes; | |
873 | int tail_bytes; | |
874 | int tail_cycle; | |
875 | ||
876 | tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn)); | |
877 | tail_cycle = CYCLE_LSN(log->l_tail_lsn); | |
878 | if ((tail_cycle == cycle) && (bytes >= tail_bytes)) { | |
879 | free_bytes = log->l_logsize - (bytes - tail_bytes); | |
880 | } else if ((tail_cycle + 1) < cycle) { | |
881 | return 0; | |
882 | } else if (tail_cycle < cycle) { | |
883 | ASSERT(tail_cycle == (cycle - 1)); | |
884 | free_bytes = tail_bytes - bytes; | |
885 | } else { | |
886 | /* | |
887 | * The reservation head is behind the tail. | |
888 | * In this case we just want to return the size of the | |
889 | * log as the amount of space left. | |
890 | */ | |
891 | xfs_fs_cmn_err(CE_ALERT, log->l_mp, | |
892 | "xlog_space_left: head behind tail\n" | |
893 | " tail_cycle = %d, tail_bytes = %d\n" | |
894 | " GH cycle = %d, GH bytes = %d", | |
895 | tail_cycle, tail_bytes, cycle, bytes); | |
896 | ASSERT(0); | |
897 | free_bytes = log->l_logsize; | |
898 | } | |
899 | return free_bytes; | |
900 | } /* xlog_space_left */ | |
901 | ||
902 | ||
903 | /* | |
904 | * Log function which is called when an io completes. | |
905 | * | |
906 | * The log manager needs its own routine, in order to control what | |
907 | * happens with the buffer after the write completes. | |
908 | */ | |
909 | void | |
910 | xlog_iodone(xfs_buf_t *bp) | |
911 | { | |
912 | xlog_in_core_t *iclog; | |
913 | xlog_t *l; | |
914 | int aborted; | |
915 | ||
916 | iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *); | |
917 | ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2); | |
918 | XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); | |
919 | aborted = 0; | |
1da177e4 LT |
920 | l = iclog->ic_log; |
921 | ||
0bfefc46 | 922 | /* |
73f6aa4d CH |
923 | * If the _XFS_BARRIER_FAILED flag was set by a lower |
924 | * layer, it means the underlying device no longer supports | |
0bfefc46 DC |
925 | * barrier I/O. Warn loudly and turn off barriers. |
926 | */ | |
73f6aa4d CH |
927 | if (bp->b_flags & _XFS_BARRIER_FAILED) { |
928 | bp->b_flags &= ~_XFS_BARRIER_FAILED; | |
0bfefc46 DC |
929 | l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER; |
930 | xfs_fs_cmn_err(CE_WARN, l->l_mp, | |
931 | "xlog_iodone: Barriers are no longer supported" | |
932 | " by device. Disabling barriers\n"); | |
0bfefc46 DC |
933 | } |
934 | ||
1da177e4 LT |
935 | /* |
936 | * Race to shutdown the filesystem if we see an error. | |
937 | */ | |
938 | if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp, | |
939 | XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) { | |
940 | xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp)); | |
941 | XFS_BUF_STALE(bp); | |
7d04a335 | 942 | xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 LT |
943 | /* |
944 | * This flag will be propagated to the trans-committed | |
945 | * callback routines to let them know that the log-commit | |
946 | * didn't succeed. | |
947 | */ | |
948 | aborted = XFS_LI_ABORTED; | |
949 | } else if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
950 | aborted = XFS_LI_ABORTED; | |
951 | } | |
3db296f3 DC |
952 | |
953 | /* log I/O is always issued ASYNC */ | |
954 | ASSERT(XFS_BUF_ISASYNC(bp)); | |
1da177e4 | 955 | xlog_state_done_syncing(iclog, aborted); |
3db296f3 DC |
956 | /* |
957 | * do not reference the buffer (bp) here as we could race | |
958 | * with it being freed after writing the unmount record to the | |
959 | * log. | |
960 | */ | |
961 | ||
1da177e4 LT |
962 | } /* xlog_iodone */ |
963 | ||
1da177e4 LT |
964 | /* |
965 | * Return size of each in-core log record buffer. | |
966 | * | |
9da096fd | 967 | * All machines get 8 x 32kB buffers by default, unless tuned otherwise. |
1da177e4 LT |
968 | * |
969 | * If the filesystem blocksize is too large, we may need to choose a | |
970 | * larger size since the directory code currently logs entire blocks. | |
971 | */ | |
972 | ||
973 | STATIC void | |
974 | xlog_get_iclog_buffer_size(xfs_mount_t *mp, | |
975 | xlog_t *log) | |
976 | { | |
977 | int size; | |
978 | int xhdrs; | |
979 | ||
1cb51258 ES |
980 | if (mp->m_logbufs <= 0) |
981 | log->l_iclog_bufs = XLOG_MAX_ICLOGS; | |
982 | else | |
cfcbbbd0 | 983 | log->l_iclog_bufs = mp->m_logbufs; |
1da177e4 LT |
984 | |
985 | /* | |
986 | * Buffer size passed in from mount system call. | |
987 | */ | |
cfcbbbd0 | 988 | if (mp->m_logbsize > 0) { |
1da177e4 LT |
989 | size = log->l_iclog_size = mp->m_logbsize; |
990 | log->l_iclog_size_log = 0; | |
991 | while (size != 1) { | |
992 | log->l_iclog_size_log++; | |
993 | size >>= 1; | |
994 | } | |
995 | ||
62118709 | 996 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
9da096fd MP |
997 | /* # headers = size / 32k |
998 | * one header holds cycles from 32k of data | |
1da177e4 LT |
999 | */ |
1000 | ||
1001 | xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; | |
1002 | if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE) | |
1003 | xhdrs++; | |
1004 | log->l_iclog_hsize = xhdrs << BBSHIFT; | |
1005 | log->l_iclog_heads = xhdrs; | |
1006 | } else { | |
1007 | ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE); | |
1008 | log->l_iclog_hsize = BBSIZE; | |
1009 | log->l_iclog_heads = 1; | |
1010 | } | |
cfcbbbd0 | 1011 | goto done; |
1da177e4 LT |
1012 | } |
1013 | ||
9da096fd | 1014 | /* All machines use 32kB buffers by default. */ |
1cb51258 ES |
1015 | log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; |
1016 | log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; | |
1da177e4 LT |
1017 | |
1018 | /* the default log size is 16k or 32k which is one header sector */ | |
1019 | log->l_iclog_hsize = BBSIZE; | |
1020 | log->l_iclog_heads = 1; | |
1021 | ||
7153f8ba CH |
1022 | done: |
1023 | /* are we being asked to make the sizes selected above visible? */ | |
cfcbbbd0 NS |
1024 | if (mp->m_logbufs == 0) |
1025 | mp->m_logbufs = log->l_iclog_bufs; | |
1026 | if (mp->m_logbsize == 0) | |
1027 | mp->m_logbsize = log->l_iclog_size; | |
1da177e4 LT |
1028 | } /* xlog_get_iclog_buffer_size */ |
1029 | ||
1030 | ||
1031 | /* | |
1032 | * This routine initializes some of the log structure for a given mount point. | |
1033 | * Its primary purpose is to fill in enough, so recovery can occur. However, | |
1034 | * some other stuff may be filled in too. | |
1035 | */ | |
1036 | STATIC xlog_t * | |
1037 | xlog_alloc_log(xfs_mount_t *mp, | |
1038 | xfs_buftarg_t *log_target, | |
1039 | xfs_daddr_t blk_offset, | |
1040 | int num_bblks) | |
1041 | { | |
1042 | xlog_t *log; | |
1043 | xlog_rec_header_t *head; | |
1044 | xlog_in_core_t **iclogp; | |
1045 | xlog_in_core_t *iclog, *prev_iclog=NULL; | |
1046 | xfs_buf_t *bp; | |
1047 | int i; | |
1048 | int iclogsize; | |
a6cb767e | 1049 | int error = ENOMEM; |
69ce58f0 | 1050 | uint log2_size = 0; |
1da177e4 | 1051 | |
644c3567 | 1052 | log = kmem_zalloc(sizeof(xlog_t), KM_MAYFAIL); |
a6cb767e DC |
1053 | if (!log) { |
1054 | xlog_warn("XFS: Log allocation failed: No memory!"); | |
1055 | goto out; | |
1056 | } | |
1da177e4 LT |
1057 | |
1058 | log->l_mp = mp; | |
1059 | log->l_targ = log_target; | |
1060 | log->l_logsize = BBTOB(num_bblks); | |
1061 | log->l_logBBstart = blk_offset; | |
1062 | log->l_logBBsize = num_bblks; | |
1063 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
1064 | log->l_flags |= XLOG_ACTIVE_RECOVERY; | |
1065 | ||
1066 | log->l_prev_block = -1; | |
03bea6fe | 1067 | log->l_tail_lsn = xlog_assign_lsn(1, 0); |
1da177e4 LT |
1068 | /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ |
1069 | log->l_last_sync_lsn = log->l_tail_lsn; | |
1070 | log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ | |
1071 | log->l_grant_reserve_cycle = 1; | |
1072 | log->l_grant_write_cycle = 1; | |
1073 | ||
a6cb767e | 1074 | error = EFSCORRUPTED; |
62118709 | 1075 | if (xfs_sb_version_hassector(&mp->m_sb)) { |
69ce58f0 AE |
1076 | log2_size = mp->m_sb.sb_logsectlog; |
1077 | if (log2_size < BBSHIFT) { | |
1078 | xlog_warn("XFS: Log sector size too small " | |
1079 | "(0x%x < 0x%x)", log2_size, BBSHIFT); | |
a6cb767e DC |
1080 | goto out_free_log; |
1081 | } | |
1082 | ||
69ce58f0 AE |
1083 | log2_size -= BBSHIFT; |
1084 | if (log2_size > mp->m_sectbb_log) { | |
1085 | xlog_warn("XFS: Log sector size too large " | |
1086 | "(0x%x > 0x%x)", log2_size, mp->m_sectbb_log); | |
a6cb767e DC |
1087 | goto out_free_log; |
1088 | } | |
69ce58f0 AE |
1089 | |
1090 | /* for larger sector sizes, must have v2 or external log */ | |
1091 | if (log2_size && log->l_logBBstart > 0 && | |
1092 | !xfs_sb_version_haslogv2(&mp->m_sb)) { | |
1093 | ||
1094 | xlog_warn("XFS: log sector size (0x%x) invalid " | |
1095 | "for configuration.", log2_size); | |
a6cb767e DC |
1096 | goto out_free_log; |
1097 | } | |
1da177e4 | 1098 | } |
69ce58f0 | 1099 | log->l_sectBBsize = 1 << log2_size; |
1da177e4 LT |
1100 | |
1101 | xlog_get_iclog_buffer_size(mp, log); | |
1102 | ||
a6cb767e | 1103 | error = ENOMEM; |
1da177e4 | 1104 | bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp); |
644c3567 DC |
1105 | if (!bp) |
1106 | goto out_free_log; | |
1da177e4 | 1107 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone); |
1da177e4 LT |
1108 | XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); |
1109 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
1110 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
1111 | log->l_xbuf = bp; | |
1112 | ||
007c61c6 ES |
1113 | spin_lock_init(&log->l_icloglock); |
1114 | spin_lock_init(&log->l_grant_lock); | |
d748c623 | 1115 | sv_init(&log->l_flush_wait, 0, "flush_wait"); |
1da177e4 LT |
1116 | |
1117 | /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */ | |
1118 | ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0); | |
1119 | ||
1120 | iclogp = &log->l_iclog; | |
1121 | /* | |
1122 | * The amount of memory to allocate for the iclog structure is | |
1123 | * rather funky due to the way the structure is defined. It is | |
1124 | * done this way so that we can use different sizes for machines | |
1125 | * with different amounts of memory. See the definition of | |
1126 | * xlog_in_core_t in xfs_log_priv.h for details. | |
1127 | */ | |
1128 | iclogsize = log->l_iclog_size; | |
1129 | ASSERT(log->l_iclog_size >= 4096); | |
1130 | for (i=0; i < log->l_iclog_bufs; i++) { | |
644c3567 DC |
1131 | *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); |
1132 | if (!*iclogp) | |
1133 | goto out_free_iclog; | |
1134 | ||
1da177e4 | 1135 | iclog = *iclogp; |
1da177e4 LT |
1136 | iclog->ic_prev = prev_iclog; |
1137 | prev_iclog = iclog; | |
1fa40b01 CH |
1138 | |
1139 | bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp); | |
644c3567 DC |
1140 | if (!bp) |
1141 | goto out_free_iclog; | |
1fa40b01 CH |
1142 | if (!XFS_BUF_CPSEMA(bp)) |
1143 | ASSERT(0); | |
1144 | XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone); | |
1fa40b01 CH |
1145 | XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); |
1146 | iclog->ic_bp = bp; | |
b28708d6 | 1147 | iclog->ic_data = bp->b_addr; |
4679b2d3 | 1148 | #ifdef DEBUG |
1da177e4 | 1149 | log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); |
4679b2d3 | 1150 | #endif |
1da177e4 LT |
1151 | head = &iclog->ic_header; |
1152 | memset(head, 0, sizeof(xlog_rec_header_t)); | |
b53e675d CH |
1153 | head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); |
1154 | head->h_version = cpu_to_be32( | |
62118709 | 1155 | xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); |
b53e675d | 1156 | head->h_size = cpu_to_be32(log->l_iclog_size); |
1da177e4 | 1157 | /* new fields */ |
b53e675d | 1158 | head->h_fmt = cpu_to_be32(XLOG_FMT); |
1da177e4 LT |
1159 | memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); |
1160 | ||
1da177e4 LT |
1161 | iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize; |
1162 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
1163 | iclog->ic_log = log; | |
114d23aa DC |
1164 | atomic_set(&iclog->ic_refcnt, 0); |
1165 | spin_lock_init(&iclog->ic_callback_lock); | |
1da177e4 | 1166 | iclog->ic_callback_tail = &(iclog->ic_callback); |
b28708d6 | 1167 | iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; |
1da177e4 LT |
1168 | |
1169 | ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp)); | |
1170 | ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0); | |
12017faf DC |
1171 | sv_init(&iclog->ic_force_wait, SV_DEFAULT, "iclog-force"); |
1172 | sv_init(&iclog->ic_write_wait, SV_DEFAULT, "iclog-write"); | |
1da177e4 LT |
1173 | |
1174 | iclogp = &iclog->ic_next; | |
1175 | } | |
1176 | *iclogp = log->l_iclog; /* complete ring */ | |
1177 | log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ | |
1178 | ||
1179 | return log; | |
644c3567 DC |
1180 | |
1181 | out_free_iclog: | |
1182 | for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { | |
1183 | prev_iclog = iclog->ic_next; | |
1184 | if (iclog->ic_bp) { | |
1185 | sv_destroy(&iclog->ic_force_wait); | |
1186 | sv_destroy(&iclog->ic_write_wait); | |
1187 | xfs_buf_free(iclog->ic_bp); | |
644c3567 DC |
1188 | } |
1189 | kmem_free(iclog); | |
1190 | } | |
1191 | spinlock_destroy(&log->l_icloglock); | |
1192 | spinlock_destroy(&log->l_grant_lock); | |
644c3567 DC |
1193 | xfs_buf_free(log->l_xbuf); |
1194 | out_free_log: | |
1195 | kmem_free(log); | |
a6cb767e DC |
1196 | out: |
1197 | return ERR_PTR(-error); | |
1da177e4 LT |
1198 | } /* xlog_alloc_log */ |
1199 | ||
1200 | ||
1201 | /* | |
1202 | * Write out the commit record of a transaction associated with the given | |
1203 | * ticket. Return the lsn of the commit record. | |
1204 | */ | |
1205 | STATIC int | |
55b66332 DC |
1206 | xlog_commit_record( |
1207 | struct log *log, | |
1208 | struct xlog_ticket *ticket, | |
1209 | struct xlog_in_core **iclog, | |
1210 | xfs_lsn_t *commitlsnp) | |
1da177e4 | 1211 | { |
55b66332 DC |
1212 | struct xfs_mount *mp = log->l_mp; |
1213 | int error; | |
1214 | struct xfs_log_iovec reg = { | |
1215 | .i_addr = NULL, | |
1216 | .i_len = 0, | |
1217 | .i_type = XLOG_REG_TYPE_COMMIT, | |
1218 | }; | |
1219 | struct xfs_log_vec vec = { | |
1220 | .lv_niovecs = 1, | |
1221 | .lv_iovecp = ®, | |
1222 | }; | |
1da177e4 LT |
1223 | |
1224 | ASSERT_ALWAYS(iclog); | |
55b66332 DC |
1225 | error = xlog_write(log, &vec, ticket, commitlsnp, iclog, |
1226 | XLOG_COMMIT_TRANS); | |
1227 | if (error) | |
7d04a335 | 1228 | xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); |
014c2544 | 1229 | return error; |
55b66332 | 1230 | } |
1da177e4 LT |
1231 | |
1232 | /* | |
1233 | * Push on the buffer cache code if we ever use more than 75% of the on-disk | |
1234 | * log space. This code pushes on the lsn which would supposedly free up | |
1235 | * the 25% which we want to leave free. We may need to adopt a policy which | |
1236 | * pushes on an lsn which is further along in the log once we reach the high | |
1237 | * water mark. In this manner, we would be creating a low water mark. | |
1238 | */ | |
a8272ce0 | 1239 | STATIC void |
1da177e4 LT |
1240 | xlog_grant_push_ail(xfs_mount_t *mp, |
1241 | int need_bytes) | |
1242 | { | |
1243 | xlog_t *log = mp->m_log; /* pointer to the log */ | |
1244 | xfs_lsn_t tail_lsn; /* lsn of the log tail */ | |
1245 | xfs_lsn_t threshold_lsn = 0; /* lsn we'd like to be at */ | |
1246 | int free_blocks; /* free blocks left to write to */ | |
1247 | int free_bytes; /* free bytes left to write to */ | |
1248 | int threshold_block; /* block in lsn we'd like to be at */ | |
1249 | int threshold_cycle; /* lsn cycle we'd like to be at */ | |
1250 | int free_threshold; | |
1da177e4 LT |
1251 | |
1252 | ASSERT(BTOBB(need_bytes) < log->l_logBBsize); | |
1253 | ||
c8b5ea28 | 1254 | spin_lock(&log->l_grant_lock); |
1da177e4 LT |
1255 | free_bytes = xlog_space_left(log, |
1256 | log->l_grant_reserve_cycle, | |
1257 | log->l_grant_reserve_bytes); | |
1258 | tail_lsn = log->l_tail_lsn; | |
1259 | free_blocks = BTOBBT(free_bytes); | |
1260 | ||
1261 | /* | |
1262 | * Set the threshold for the minimum number of free blocks in the | |
1263 | * log to the maximum of what the caller needs, one quarter of the | |
1264 | * log, and 256 blocks. | |
1265 | */ | |
1266 | free_threshold = BTOBB(need_bytes); | |
1267 | free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); | |
1268 | free_threshold = MAX(free_threshold, 256); | |
1269 | if (free_blocks < free_threshold) { | |
1270 | threshold_block = BLOCK_LSN(tail_lsn) + free_threshold; | |
1271 | threshold_cycle = CYCLE_LSN(tail_lsn); | |
1272 | if (threshold_block >= log->l_logBBsize) { | |
1273 | threshold_block -= log->l_logBBsize; | |
1274 | threshold_cycle += 1; | |
1275 | } | |
03bea6fe | 1276 | threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block); |
1da177e4 LT |
1277 | |
1278 | /* Don't pass in an lsn greater than the lsn of the last | |
1279 | * log record known to be on disk. | |
1280 | */ | |
1281 | if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0) | |
1282 | threshold_lsn = log->l_last_sync_lsn; | |
1283 | } | |
c8b5ea28 | 1284 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
1285 | |
1286 | /* | |
1287 | * Get the transaction layer to kick the dirty buffers out to | |
1288 | * disk asynchronously. No point in trying to do this if | |
1289 | * the filesystem is shutting down. | |
1290 | */ | |
1291 | if (threshold_lsn && | |
1292 | !XLOG_FORCED_SHUTDOWN(log)) | |
783a2f65 | 1293 | xfs_trans_ail_push(log->l_ailp, threshold_lsn); |
1da177e4 LT |
1294 | } /* xlog_grant_push_ail */ |
1295 | ||
873ff550 CH |
1296 | /* |
1297 | * The bdstrat callback function for log bufs. This gives us a central | |
1298 | * place to trap bufs in case we get hit by a log I/O error and need to | |
1299 | * shutdown. Actually, in practice, even when we didn't get a log error, | |
1300 | * we transition the iclogs to IOERROR state *after* flushing all existing | |
1301 | * iclogs to disk. This is because we don't want anymore new transactions to be | |
1302 | * started or completed afterwards. | |
1303 | */ | |
1304 | STATIC int | |
1305 | xlog_bdstrat( | |
1306 | struct xfs_buf *bp) | |
1307 | { | |
1308 | struct xlog_in_core *iclog; | |
1309 | ||
1310 | iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *); | |
1311 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
1312 | XFS_BUF_ERROR(bp, EIO); | |
1313 | XFS_BUF_STALE(bp); | |
1314 | xfs_biodone(bp); | |
1315 | /* | |
1316 | * It would seem logical to return EIO here, but we rely on | |
1317 | * the log state machine to propagate I/O errors instead of | |
1318 | * doing it here. | |
1319 | */ | |
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | bp->b_flags |= _XBF_RUN_QUEUES; | |
1324 | xfs_buf_iorequest(bp); | |
1325 | return 0; | |
1326 | } | |
1da177e4 LT |
1327 | |
1328 | /* | |
1329 | * Flush out the in-core log (iclog) to the on-disk log in an asynchronous | |
1330 | * fashion. Previously, we should have moved the current iclog | |
1331 | * ptr in the log to point to the next available iclog. This allows further | |
1332 | * write to continue while this code syncs out an iclog ready to go. | |
1333 | * Before an in-core log can be written out, the data section must be scanned | |
1334 | * to save away the 1st word of each BBSIZE block into the header. We replace | |
1335 | * it with the current cycle count. Each BBSIZE block is tagged with the | |
1336 | * cycle count because there in an implicit assumption that drives will | |
1337 | * guarantee that entire 512 byte blocks get written at once. In other words, | |
1338 | * we can't have part of a 512 byte block written and part not written. By | |
1339 | * tagging each block, we will know which blocks are valid when recovering | |
1340 | * after an unclean shutdown. | |
1341 | * | |
1342 | * This routine is single threaded on the iclog. No other thread can be in | |
1343 | * this routine with the same iclog. Changing contents of iclog can there- | |
1344 | * fore be done without grabbing the state machine lock. Updating the global | |
1345 | * log will require grabbing the lock though. | |
1346 | * | |
1347 | * The entire log manager uses a logical block numbering scheme. Only | |
1348 | * log_sync (and then only bwrite()) know about the fact that the log may | |
1349 | * not start with block zero on a given device. The log block start offset | |
1350 | * is added immediately before calling bwrite(). | |
1351 | */ | |
1352 | ||
a8272ce0 | 1353 | STATIC int |
1da177e4 LT |
1354 | xlog_sync(xlog_t *log, |
1355 | xlog_in_core_t *iclog) | |
1356 | { | |
1357 | xfs_caddr_t dptr; /* pointer to byte sized element */ | |
1358 | xfs_buf_t *bp; | |
b53e675d | 1359 | int i; |
1da177e4 LT |
1360 | uint count; /* byte count of bwrite */ |
1361 | uint count_init; /* initial count before roundup */ | |
1362 | int roundoff; /* roundoff to BB or stripe */ | |
1363 | int split = 0; /* split write into two regions */ | |
1364 | int error; | |
62118709 | 1365 | int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); |
1da177e4 LT |
1366 | |
1367 | XFS_STATS_INC(xs_log_writes); | |
155cc6b7 | 1368 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
1369 | |
1370 | /* Add for LR header */ | |
1371 | count_init = log->l_iclog_hsize + iclog->ic_offset; | |
1372 | ||
1373 | /* Round out the log write size */ | |
1374 | if (v2 && log->l_mp->m_sb.sb_logsunit > 1) { | |
1375 | /* we have a v2 stripe unit to use */ | |
1376 | count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init)); | |
1377 | } else { | |
1378 | count = BBTOB(BTOBB(count_init)); | |
1379 | } | |
1380 | roundoff = count - count_init; | |
1381 | ASSERT(roundoff >= 0); | |
1382 | ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && | |
1383 | roundoff < log->l_mp->m_sb.sb_logsunit) | |
1384 | || | |
1385 | (log->l_mp->m_sb.sb_logsunit <= 1 && | |
1386 | roundoff < BBTOB(1))); | |
1387 | ||
1388 | /* move grant heads by roundoff in sync */ | |
c8b5ea28 | 1389 | spin_lock(&log->l_grant_lock); |
dd954c69 | 1390 | xlog_grant_add_space(log, roundoff); |
c8b5ea28 | 1391 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
1392 | |
1393 | /* put cycle number in every block */ | |
1394 | xlog_pack_data(log, iclog, roundoff); | |
1395 | ||
1396 | /* real byte length */ | |
1397 | if (v2) { | |
b53e675d CH |
1398 | iclog->ic_header.h_len = |
1399 | cpu_to_be32(iclog->ic_offset + roundoff); | |
1da177e4 | 1400 | } else { |
b53e675d CH |
1401 | iclog->ic_header.h_len = |
1402 | cpu_to_be32(iclog->ic_offset); | |
1da177e4 LT |
1403 | } |
1404 | ||
f5faad79 | 1405 | bp = iclog->ic_bp; |
1da177e4 LT |
1406 | ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1); |
1407 | XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2); | |
b53e675d | 1408 | XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); |
1da177e4 LT |
1409 | |
1410 | XFS_STATS_ADD(xs_log_blocks, BTOBB(count)); | |
1411 | ||
1412 | /* Do we need to split this write into 2 parts? */ | |
1413 | if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { | |
1414 | split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); | |
1415 | count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); | |
1416 | iclog->ic_bwritecnt = 2; /* split into 2 writes */ | |
1417 | } else { | |
1418 | iclog->ic_bwritecnt = 1; | |
1419 | } | |
511105b3 | 1420 | XFS_BUF_SET_COUNT(bp, count); |
1da177e4 | 1421 | XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */ |
f5faad79 | 1422 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 LT |
1423 | XFS_BUF_BUSY(bp); |
1424 | XFS_BUF_ASYNC(bp); | |
2ee1abad | 1425 | bp->b_flags |= XBF_LOG_BUFFER; |
1da177e4 | 1426 | /* |
f538d4da | 1427 | * Do an ordered write for the log block. |
f5faad79 | 1428 | * Its unnecessary to flush the first split block in the log wrap case. |
1da177e4 | 1429 | */ |
f5faad79 | 1430 | if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER)) |
f538d4da | 1431 | XFS_BUF_ORDERED(bp); |
1da177e4 LT |
1432 | |
1433 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1434 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1435 | ||
1436 | xlog_verify_iclog(log, iclog, count, B_TRUE); | |
1437 | ||
1438 | /* account for log which doesn't start at block #0 */ | |
1439 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); | |
1440 | /* | |
1441 | * Don't call xfs_bwrite here. We do log-syncs even when the filesystem | |
1442 | * is shutting down. | |
1443 | */ | |
1444 | XFS_BUF_WRITE(bp); | |
1445 | ||
873ff550 | 1446 | if ((error = xlog_bdstrat(bp))) { |
1da177e4 LT |
1447 | xfs_ioerror_alert("xlog_sync", log->l_mp, bp, |
1448 | XFS_BUF_ADDR(bp)); | |
014c2544 | 1449 | return error; |
1da177e4 LT |
1450 | } |
1451 | if (split) { | |
f5faad79 | 1452 | bp = iclog->ic_log->l_xbuf; |
1da177e4 LT |
1453 | ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == |
1454 | (unsigned long)1); | |
1455 | XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2); | |
1456 | XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ | |
1457 | XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+ | |
1458 | (__psint_t)count), split); | |
1459 | XFS_BUF_SET_FSPRIVATE(bp, iclog); | |
f5faad79 | 1460 | XFS_BUF_ZEROFLAGS(bp); |
1da177e4 LT |
1461 | XFS_BUF_BUSY(bp); |
1462 | XFS_BUF_ASYNC(bp); | |
2ee1abad | 1463 | bp->b_flags |= XBF_LOG_BUFFER; |
f538d4da CH |
1464 | if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) |
1465 | XFS_BUF_ORDERED(bp); | |
1da177e4 LT |
1466 | dptr = XFS_BUF_PTR(bp); |
1467 | /* | |
1468 | * Bump the cycle numbers at the start of each block | |
1469 | * since this part of the buffer is at the start of | |
1470 | * a new cycle. Watch out for the header magic number | |
1471 | * case, though. | |
1472 | */ | |
b53e675d | 1473 | for (i = 0; i < split; i += BBSIZE) { |
413d57c9 | 1474 | be32_add_cpu((__be32 *)dptr, 1); |
b53e675d | 1475 | if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM) |
413d57c9 | 1476 | be32_add_cpu((__be32 *)dptr, 1); |
1da177e4 LT |
1477 | dptr += BBSIZE; |
1478 | } | |
1479 | ||
1480 | ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); | |
1481 | ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); | |
1482 | ||
c41564b5 | 1483 | /* account for internal log which doesn't start at block #0 */ |
1da177e4 LT |
1484 | XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); |
1485 | XFS_BUF_WRITE(bp); | |
873ff550 | 1486 | if ((error = xlog_bdstrat(bp))) { |
1da177e4 LT |
1487 | xfs_ioerror_alert("xlog_sync (split)", log->l_mp, |
1488 | bp, XFS_BUF_ADDR(bp)); | |
014c2544 | 1489 | return error; |
1da177e4 LT |
1490 | } |
1491 | } | |
014c2544 | 1492 | return 0; |
1da177e4 LT |
1493 | } /* xlog_sync */ |
1494 | ||
1495 | ||
1496 | /* | |
c41564b5 | 1497 | * Deallocate a log structure |
1da177e4 | 1498 | */ |
a8272ce0 | 1499 | STATIC void |
c41564b5 | 1500 | xlog_dealloc_log(xlog_t *log) |
1da177e4 LT |
1501 | { |
1502 | xlog_in_core_t *iclog, *next_iclog; | |
1da177e4 LT |
1503 | int i; |
1504 | ||
1da177e4 LT |
1505 | iclog = log->l_iclog; |
1506 | for (i=0; i<log->l_iclog_bufs; i++) { | |
12017faf DC |
1507 | sv_destroy(&iclog->ic_force_wait); |
1508 | sv_destroy(&iclog->ic_write_wait); | |
1da177e4 | 1509 | xfs_buf_free(iclog->ic_bp); |
1da177e4 | 1510 | next_iclog = iclog->ic_next; |
f0e2d93c | 1511 | kmem_free(iclog); |
1da177e4 LT |
1512 | iclog = next_iclog; |
1513 | } | |
1da177e4 LT |
1514 | spinlock_destroy(&log->l_icloglock); |
1515 | spinlock_destroy(&log->l_grant_lock); | |
1516 | ||
1da177e4 | 1517 | xfs_buf_free(log->l_xbuf); |
1da177e4 | 1518 | log->l_mp->m_log = NULL; |
f0e2d93c | 1519 | kmem_free(log); |
c41564b5 | 1520 | } /* xlog_dealloc_log */ |
1da177e4 LT |
1521 | |
1522 | /* | |
1523 | * Update counters atomically now that memcpy is done. | |
1524 | */ | |
1525 | /* ARGSUSED */ | |
1526 | static inline void | |
1527 | xlog_state_finish_copy(xlog_t *log, | |
1528 | xlog_in_core_t *iclog, | |
1529 | int record_cnt, | |
1530 | int copy_bytes) | |
1531 | { | |
b22cd72c | 1532 | spin_lock(&log->l_icloglock); |
1da177e4 | 1533 | |
413d57c9 | 1534 | be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt); |
1da177e4 LT |
1535 | iclog->ic_offset += copy_bytes; |
1536 | ||
b22cd72c | 1537 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
1538 | } /* xlog_state_finish_copy */ |
1539 | ||
1540 | ||
1541 | ||
1542 | ||
7e9c6396 TS |
1543 | /* |
1544 | * print out info relating to regions written which consume | |
1545 | * the reservation | |
1546 | */ | |
7e9c6396 TS |
1547 | STATIC void |
1548 | xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket) | |
1549 | { | |
1550 | uint i; | |
1551 | uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); | |
1552 | ||
1553 | /* match with XLOG_REG_TYPE_* in xfs_log.h */ | |
1554 | static char *res_type_str[XLOG_REG_TYPE_MAX] = { | |
1555 | "bformat", | |
1556 | "bchunk", | |
1557 | "efi_format", | |
1558 | "efd_format", | |
1559 | "iformat", | |
1560 | "icore", | |
1561 | "iext", | |
1562 | "ibroot", | |
1563 | "ilocal", | |
1564 | "iattr_ext", | |
1565 | "iattr_broot", | |
1566 | "iattr_local", | |
1567 | "qformat", | |
1568 | "dquot", | |
1569 | "quotaoff", | |
1570 | "LR header", | |
1571 | "unmount", | |
1572 | "commit", | |
1573 | "trans header" | |
1574 | }; | |
1575 | static char *trans_type_str[XFS_TRANS_TYPE_MAX] = { | |
1576 | "SETATTR_NOT_SIZE", | |
1577 | "SETATTR_SIZE", | |
1578 | "INACTIVE", | |
1579 | "CREATE", | |
1580 | "CREATE_TRUNC", | |
1581 | "TRUNCATE_FILE", | |
1582 | "REMOVE", | |
1583 | "LINK", | |
1584 | "RENAME", | |
1585 | "MKDIR", | |
1586 | "RMDIR", | |
1587 | "SYMLINK", | |
1588 | "SET_DMATTRS", | |
1589 | "GROWFS", | |
1590 | "STRAT_WRITE", | |
1591 | "DIOSTRAT", | |
1592 | "WRITE_SYNC", | |
1593 | "WRITEID", | |
1594 | "ADDAFORK", | |
1595 | "ATTRINVAL", | |
1596 | "ATRUNCATE", | |
1597 | "ATTR_SET", | |
1598 | "ATTR_RM", | |
1599 | "ATTR_FLAG", | |
1600 | "CLEAR_AGI_BUCKET", | |
1601 | "QM_SBCHANGE", | |
1602 | "DUMMY1", | |
1603 | "DUMMY2", | |
1604 | "QM_QUOTAOFF", | |
1605 | "QM_DQALLOC", | |
1606 | "QM_SETQLIM", | |
1607 | "QM_DQCLUSTER", | |
1608 | "QM_QINOCREATE", | |
1609 | "QM_QUOTAOFF_END", | |
1610 | "SB_UNIT", | |
1611 | "FSYNC_TS", | |
1612 | "GROWFSRT_ALLOC", | |
1613 | "GROWFSRT_ZERO", | |
1614 | "GROWFSRT_FREE", | |
1615 | "SWAPEXT" | |
1616 | }; | |
1617 | ||
1618 | xfs_fs_cmn_err(CE_WARN, mp, | |
1619 | "xfs_log_write: reservation summary:\n" | |
1620 | " trans type = %s (%u)\n" | |
1621 | " unit res = %d bytes\n" | |
1622 | " current res = %d bytes\n" | |
1623 | " total reg = %u bytes (o/flow = %u bytes)\n" | |
1624 | " ophdrs = %u (ophdr space = %u bytes)\n" | |
1625 | " ophdr + reg = %u bytes\n" | |
1626 | " num regions = %u\n", | |
1627 | ((ticket->t_trans_type <= 0 || | |
1628 | ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? | |
1629 | "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), | |
1630 | ticket->t_trans_type, | |
1631 | ticket->t_unit_res, | |
1632 | ticket->t_curr_res, | |
1633 | ticket->t_res_arr_sum, ticket->t_res_o_flow, | |
1634 | ticket->t_res_num_ophdrs, ophdr_spc, | |
1635 | ticket->t_res_arr_sum + | |
1259845d | 1636 | ticket->t_res_o_flow + ophdr_spc, |
7e9c6396 TS |
1637 | ticket->t_res_num); |
1638 | ||
1639 | for (i = 0; i < ticket->t_res_num; i++) { | |
1259845d | 1640 | uint r_type = ticket->t_res_arr[i].r_type; |
7e9c6396 TS |
1641 | cmn_err(CE_WARN, |
1642 | "region[%u]: %s - %u bytes\n", | |
1643 | i, | |
1644 | ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? | |
1645 | "bad-rtype" : res_type_str[r_type-1]), | |
1646 | ticket->t_res_arr[i].r_len); | |
1647 | } | |
169a7b07 DC |
1648 | |
1649 | xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp, | |
1650 | "xfs_log_write: reservation ran out. Need to up reservation"); | |
1651 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
7e9c6396 | 1652 | } |
7e9c6396 | 1653 | |
b5203cd0 DC |
1654 | /* |
1655 | * Calculate the potential space needed by the log vector. Each region gets | |
1656 | * its own xlog_op_header_t and may need to be double word aligned. | |
1657 | */ | |
1658 | static int | |
1659 | xlog_write_calc_vec_length( | |
1660 | struct xlog_ticket *ticket, | |
55b66332 | 1661 | struct xfs_log_vec *log_vector) |
b5203cd0 | 1662 | { |
55b66332 | 1663 | struct xfs_log_vec *lv; |
b5203cd0 DC |
1664 | int headers = 0; |
1665 | int len = 0; | |
1666 | int i; | |
1667 | ||
1668 | /* acct for start rec of xact */ | |
1669 | if (ticket->t_flags & XLOG_TIC_INITED) | |
1670 | headers++; | |
1671 | ||
55b66332 DC |
1672 | for (lv = log_vector; lv; lv = lv->lv_next) { |
1673 | headers += lv->lv_niovecs; | |
1674 | ||
1675 | for (i = 0; i < lv->lv_niovecs; i++) { | |
1676 | struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; | |
b5203cd0 | 1677 | |
55b66332 DC |
1678 | len += vecp->i_len; |
1679 | xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); | |
1680 | } | |
b5203cd0 DC |
1681 | } |
1682 | ||
1683 | ticket->t_res_num_ophdrs += headers; | |
1684 | len += headers * sizeof(struct xlog_op_header); | |
1685 | ||
1686 | return len; | |
1687 | } | |
1688 | ||
1689 | /* | |
1690 | * If first write for transaction, insert start record We can't be trying to | |
1691 | * commit if we are inited. We can't have any "partial_copy" if we are inited. | |
1692 | */ | |
1693 | static int | |
1694 | xlog_write_start_rec( | |
e6b1f273 | 1695 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
1696 | struct xlog_ticket *ticket) |
1697 | { | |
b5203cd0 DC |
1698 | if (!(ticket->t_flags & XLOG_TIC_INITED)) |
1699 | return 0; | |
1700 | ||
1701 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); | |
1702 | ophdr->oh_clientid = ticket->t_clientid; | |
1703 | ophdr->oh_len = 0; | |
1704 | ophdr->oh_flags = XLOG_START_TRANS; | |
1705 | ophdr->oh_res2 = 0; | |
1706 | ||
1707 | ticket->t_flags &= ~XLOG_TIC_INITED; | |
1708 | ||
1709 | return sizeof(struct xlog_op_header); | |
1710 | } | |
1711 | ||
1712 | static xlog_op_header_t * | |
1713 | xlog_write_setup_ophdr( | |
1714 | struct log *log, | |
e6b1f273 | 1715 | struct xlog_op_header *ophdr, |
b5203cd0 DC |
1716 | struct xlog_ticket *ticket, |
1717 | uint flags) | |
1718 | { | |
b5203cd0 DC |
1719 | ophdr->oh_tid = cpu_to_be32(ticket->t_tid); |
1720 | ophdr->oh_clientid = ticket->t_clientid; | |
1721 | ophdr->oh_res2 = 0; | |
1722 | ||
1723 | /* are we copying a commit or unmount record? */ | |
1724 | ophdr->oh_flags = flags; | |
1725 | ||
1726 | /* | |
1727 | * We've seen logs corrupted with bad transaction client ids. This | |
1728 | * makes sure that XFS doesn't generate them on. Turn this into an EIO | |
1729 | * and shut down the filesystem. | |
1730 | */ | |
1731 | switch (ophdr->oh_clientid) { | |
1732 | case XFS_TRANSACTION: | |
1733 | case XFS_VOLUME: | |
1734 | case XFS_LOG: | |
1735 | break; | |
1736 | default: | |
1737 | xfs_fs_cmn_err(CE_WARN, log->l_mp, | |
1738 | "Bad XFS transaction clientid 0x%x in ticket 0x%p", | |
1739 | ophdr->oh_clientid, ticket); | |
1740 | return NULL; | |
1741 | } | |
1742 | ||
1743 | return ophdr; | |
1744 | } | |
1745 | ||
1746 | /* | |
1747 | * Set up the parameters of the region copy into the log. This has | |
1748 | * to handle region write split across multiple log buffers - this | |
1749 | * state is kept external to this function so that this code can | |
1750 | * can be written in an obvious, self documenting manner. | |
1751 | */ | |
1752 | static int | |
1753 | xlog_write_setup_copy( | |
1754 | struct xlog_ticket *ticket, | |
1755 | struct xlog_op_header *ophdr, | |
1756 | int space_available, | |
1757 | int space_required, | |
1758 | int *copy_off, | |
1759 | int *copy_len, | |
1760 | int *last_was_partial_copy, | |
1761 | int *bytes_consumed) | |
1762 | { | |
1763 | int still_to_copy; | |
1764 | ||
1765 | still_to_copy = space_required - *bytes_consumed; | |
1766 | *copy_off = *bytes_consumed; | |
1767 | ||
1768 | if (still_to_copy <= space_available) { | |
1769 | /* write of region completes here */ | |
1770 | *copy_len = still_to_copy; | |
1771 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
1772 | if (*last_was_partial_copy) | |
1773 | ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); | |
1774 | *last_was_partial_copy = 0; | |
1775 | *bytes_consumed = 0; | |
1776 | return 0; | |
1777 | } | |
1778 | ||
1779 | /* partial write of region, needs extra log op header reservation */ | |
1780 | *copy_len = space_available; | |
1781 | ophdr->oh_len = cpu_to_be32(*copy_len); | |
1782 | ophdr->oh_flags |= XLOG_CONTINUE_TRANS; | |
1783 | if (*last_was_partial_copy) | |
1784 | ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; | |
1785 | *bytes_consumed += *copy_len; | |
1786 | (*last_was_partial_copy)++; | |
1787 | ||
1788 | /* account for new log op header */ | |
1789 | ticket->t_curr_res -= sizeof(struct xlog_op_header); | |
1790 | ticket->t_res_num_ophdrs++; | |
1791 | ||
1792 | return sizeof(struct xlog_op_header); | |
1793 | } | |
1794 | ||
1795 | static int | |
1796 | xlog_write_copy_finish( | |
1797 | struct log *log, | |
1798 | struct xlog_in_core *iclog, | |
1799 | uint flags, | |
1800 | int *record_cnt, | |
1801 | int *data_cnt, | |
1802 | int *partial_copy, | |
1803 | int *partial_copy_len, | |
1804 | int log_offset, | |
1805 | struct xlog_in_core **commit_iclog) | |
1806 | { | |
1807 | if (*partial_copy) { | |
1808 | /* | |
1809 | * This iclog has already been marked WANT_SYNC by | |
1810 | * xlog_state_get_iclog_space. | |
1811 | */ | |
1812 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
1813 | *record_cnt = 0; | |
1814 | *data_cnt = 0; | |
1815 | return xlog_state_release_iclog(log, iclog); | |
1816 | } | |
1817 | ||
1818 | *partial_copy = 0; | |
1819 | *partial_copy_len = 0; | |
1820 | ||
1821 | if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { | |
1822 | /* no more space in this iclog - push it. */ | |
1823 | xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); | |
1824 | *record_cnt = 0; | |
1825 | *data_cnt = 0; | |
1826 | ||
1827 | spin_lock(&log->l_icloglock); | |
1828 | xlog_state_want_sync(log, iclog); | |
1829 | spin_unlock(&log->l_icloglock); | |
1830 | ||
1831 | if (!commit_iclog) | |
1832 | return xlog_state_release_iclog(log, iclog); | |
1833 | ASSERT(flags & XLOG_COMMIT_TRANS); | |
1834 | *commit_iclog = iclog; | |
1835 | } | |
1836 | ||
1837 | return 0; | |
1838 | } | |
1839 | ||
1da177e4 LT |
1840 | /* |
1841 | * Write some region out to in-core log | |
1842 | * | |
1843 | * This will be called when writing externally provided regions or when | |
1844 | * writing out a commit record for a given transaction. | |
1845 | * | |
1846 | * General algorithm: | |
1847 | * 1. Find total length of this write. This may include adding to the | |
1848 | * lengths passed in. | |
1849 | * 2. Check whether we violate the tickets reservation. | |
1850 | * 3. While writing to this iclog | |
1851 | * A. Reserve as much space in this iclog as can get | |
1852 | * B. If this is first write, save away start lsn | |
1853 | * C. While writing this region: | |
1854 | * 1. If first write of transaction, write start record | |
1855 | * 2. Write log operation header (header per region) | |
1856 | * 3. Find out if we can fit entire region into this iclog | |
1857 | * 4. Potentially, verify destination memcpy ptr | |
1858 | * 5. Memcpy (partial) region | |
1859 | * 6. If partial copy, release iclog; otherwise, continue | |
1860 | * copying more regions into current iclog | |
1861 | * 4. Mark want sync bit (in simulation mode) | |
1862 | * 5. Release iclog for potential flush to on-disk log. | |
1863 | * | |
1864 | * ERRORS: | |
1865 | * 1. Panic if reservation is overrun. This should never happen since | |
1866 | * reservation amounts are generated internal to the filesystem. | |
1867 | * NOTES: | |
1868 | * 1. Tickets are single threaded data structures. | |
1869 | * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the | |
1870 | * syncing routine. When a single log_write region needs to span | |
1871 | * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set | |
1872 | * on all log operation writes which don't contain the end of the | |
1873 | * region. The XLOG_END_TRANS bit is used for the in-core log | |
1874 | * operation which contains the end of the continued log_write region. | |
1875 | * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog, | |
1876 | * we don't really know exactly how much space will be used. As a result, | |
1877 | * we don't update ic_offset until the end when we know exactly how many | |
1878 | * bytes have been written out. | |
1879 | */ | |
a8272ce0 | 1880 | STATIC int |
35a8a72f | 1881 | xlog_write( |
55b66332 DC |
1882 | struct log *log, |
1883 | struct xfs_log_vec *log_vector, | |
35a8a72f CH |
1884 | struct xlog_ticket *ticket, |
1885 | xfs_lsn_t *start_lsn, | |
1886 | struct xlog_in_core **commit_iclog, | |
1887 | uint flags) | |
1da177e4 | 1888 | { |
99428ad0 | 1889 | struct xlog_in_core *iclog = NULL; |
55b66332 DC |
1890 | struct xfs_log_iovec *vecp; |
1891 | struct xfs_log_vec *lv; | |
99428ad0 CH |
1892 | int len; |
1893 | int index; | |
1894 | int partial_copy = 0; | |
1895 | int partial_copy_len = 0; | |
1896 | int contwr = 0; | |
1897 | int record_cnt = 0; | |
1898 | int data_cnt = 0; | |
1899 | int error; | |
1900 | ||
1901 | *start_lsn = 0; | |
1902 | ||
55b66332 | 1903 | len = xlog_write_calc_vec_length(ticket, log_vector); |
169a7b07 | 1904 | if (ticket->t_curr_res < len) |
55b66332 | 1905 | xlog_print_tic_res(log->l_mp, ticket); |
1da177e4 LT |
1906 | ticket->t_curr_res -= len; |
1907 | ||
55b66332 DC |
1908 | index = 0; |
1909 | lv = log_vector; | |
1910 | vecp = lv->lv_iovecp; | |
1911 | while (lv && index < lv->lv_niovecs) { | |
e6b1f273 | 1912 | void *ptr; |
99428ad0 | 1913 | int log_offset; |
1da177e4 | 1914 | |
99428ad0 CH |
1915 | error = xlog_state_get_iclog_space(log, len, &iclog, ticket, |
1916 | &contwr, &log_offset); | |
1917 | if (error) | |
1918 | return error; | |
1da177e4 | 1919 | |
99428ad0 | 1920 | ASSERT(log_offset <= iclog->ic_size - 1); |
e6b1f273 | 1921 | ptr = iclog->ic_datap + log_offset; |
1da177e4 | 1922 | |
99428ad0 CH |
1923 | /* start_lsn is the first lsn written to. That's all we need. */ |
1924 | if (!*start_lsn) | |
1925 | *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); | |
b5203cd0 | 1926 | |
99428ad0 CH |
1927 | /* |
1928 | * This loop writes out as many regions as can fit in the amount | |
1929 | * of space which was allocated by xlog_state_get_iclog_space(). | |
1930 | */ | |
55b66332 DC |
1931 | while (lv && index < lv->lv_niovecs) { |
1932 | struct xfs_log_iovec *reg = &vecp[index]; | |
99428ad0 CH |
1933 | struct xlog_op_header *ophdr; |
1934 | int start_rec_copy; | |
1935 | int copy_len; | |
1936 | int copy_off; | |
1937 | ||
55b66332 | 1938 | ASSERT(reg->i_len % sizeof(__int32_t) == 0); |
e6b1f273 | 1939 | ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0); |
99428ad0 CH |
1940 | |
1941 | start_rec_copy = xlog_write_start_rec(ptr, ticket); | |
1942 | if (start_rec_copy) { | |
1943 | record_cnt++; | |
e6b1f273 | 1944 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
1945 | start_rec_copy); |
1946 | } | |
b5203cd0 | 1947 | |
99428ad0 CH |
1948 | ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); |
1949 | if (!ophdr) | |
1950 | return XFS_ERROR(EIO); | |
1951 | ||
e6b1f273 | 1952 | xlog_write_adv_cnt(&ptr, &len, &log_offset, |
99428ad0 CH |
1953 | sizeof(struct xlog_op_header)); |
1954 | ||
1955 | len += xlog_write_setup_copy(ticket, ophdr, | |
1956 | iclog->ic_size-log_offset, | |
55b66332 | 1957 | reg->i_len, |
99428ad0 CH |
1958 | ©_off, ©_len, |
1959 | &partial_copy, | |
1960 | &partial_copy_len); | |
1961 | xlog_verify_dest_ptr(log, ptr); | |
1962 | ||
1963 | /* copy region */ | |
1964 | ASSERT(copy_len >= 0); | |
e6b1f273 CH |
1965 | memcpy(ptr, reg->i_addr + copy_off, copy_len); |
1966 | xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len); | |
99428ad0 CH |
1967 | |
1968 | copy_len += start_rec_copy + sizeof(xlog_op_header_t); | |
1969 | record_cnt++; | |
1970 | data_cnt += contwr ? copy_len : 0; | |
1971 | ||
1972 | error = xlog_write_copy_finish(log, iclog, flags, | |
1973 | &record_cnt, &data_cnt, | |
1974 | &partial_copy, | |
1975 | &partial_copy_len, | |
1976 | log_offset, | |
1977 | commit_iclog); | |
1978 | if (error) | |
1979 | return error; | |
1980 | ||
1981 | /* | |
1982 | * if we had a partial copy, we need to get more iclog | |
1983 | * space but we don't want to increment the region | |
1984 | * index because there is still more is this region to | |
1985 | * write. | |
1986 | * | |
1987 | * If we completed writing this region, and we flushed | |
1988 | * the iclog (indicated by resetting of the record | |
1989 | * count), then we also need to get more log space. If | |
1990 | * this was the last record, though, we are done and | |
1991 | * can just return. | |
1992 | */ | |
1993 | if (partial_copy) | |
1994 | break; | |
1995 | ||
55b66332 DC |
1996 | if (++index == lv->lv_niovecs) { |
1997 | lv = lv->lv_next; | |
1998 | index = 0; | |
1999 | if (lv) | |
2000 | vecp = lv->lv_iovecp; | |
2001 | } | |
99428ad0 | 2002 | if (record_cnt == 0) { |
55b66332 | 2003 | if (!lv) |
99428ad0 CH |
2004 | return 0; |
2005 | break; | |
2006 | } | |
2007 | } | |
2008 | } | |
2009 | ||
2010 | ASSERT(len == 0); | |
2011 | ||
2012 | xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); | |
2013 | if (!commit_iclog) | |
2014 | return xlog_state_release_iclog(log, iclog); | |
1da177e4 | 2015 | |
1da177e4 LT |
2016 | ASSERT(flags & XLOG_COMMIT_TRANS); |
2017 | *commit_iclog = iclog; | |
2018 | return 0; | |
99428ad0 | 2019 | } |
1da177e4 LT |
2020 | |
2021 | ||
2022 | /***************************************************************************** | |
2023 | * | |
2024 | * State Machine functions | |
2025 | * | |
2026 | ***************************************************************************** | |
2027 | */ | |
2028 | ||
2029 | /* Clean iclogs starting from the head. This ordering must be | |
2030 | * maintained, so an iclog doesn't become ACTIVE beyond one that | |
2031 | * is SYNCING. This is also required to maintain the notion that we use | |
12017faf | 2032 | * a ordered wait queue to hold off would be writers to the log when every |
1da177e4 LT |
2033 | * iclog is trying to sync to disk. |
2034 | * | |
2035 | * State Change: DIRTY -> ACTIVE | |
2036 | */ | |
ba0f32d4 | 2037 | STATIC void |
1da177e4 LT |
2038 | xlog_state_clean_log(xlog_t *log) |
2039 | { | |
2040 | xlog_in_core_t *iclog; | |
2041 | int changed = 0; | |
2042 | ||
2043 | iclog = log->l_iclog; | |
2044 | do { | |
2045 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
2046 | iclog->ic_state = XLOG_STATE_ACTIVE; | |
2047 | iclog->ic_offset = 0; | |
114d23aa | 2048 | ASSERT(iclog->ic_callback == NULL); |
1da177e4 LT |
2049 | /* |
2050 | * If the number of ops in this iclog indicate it just | |
2051 | * contains the dummy transaction, we can | |
2052 | * change state into IDLE (the second time around). | |
2053 | * Otherwise we should change the state into | |
2054 | * NEED a dummy. | |
2055 | * We don't need to cover the dummy. | |
2056 | */ | |
2057 | if (!changed && | |
b53e675d CH |
2058 | (be32_to_cpu(iclog->ic_header.h_num_logops) == |
2059 | XLOG_COVER_OPS)) { | |
1da177e4 LT |
2060 | changed = 1; |
2061 | } else { | |
2062 | /* | |
2063 | * We have two dirty iclogs so start over | |
2064 | * This could also be num of ops indicates | |
2065 | * this is not the dummy going out. | |
2066 | */ | |
2067 | changed = 2; | |
2068 | } | |
2069 | iclog->ic_header.h_num_logops = 0; | |
2070 | memset(iclog->ic_header.h_cycle_data, 0, | |
2071 | sizeof(iclog->ic_header.h_cycle_data)); | |
2072 | iclog->ic_header.h_lsn = 0; | |
2073 | } else if (iclog->ic_state == XLOG_STATE_ACTIVE) | |
2074 | /* do nothing */; | |
2075 | else | |
2076 | break; /* stop cleaning */ | |
2077 | iclog = iclog->ic_next; | |
2078 | } while (iclog != log->l_iclog); | |
2079 | ||
2080 | /* log is locked when we are called */ | |
2081 | /* | |
2082 | * Change state for the dummy log recording. | |
2083 | * We usually go to NEED. But we go to NEED2 if the changed indicates | |
2084 | * we are done writing the dummy record. | |
2085 | * If we are done with the second dummy recored (DONE2), then | |
2086 | * we go to IDLE. | |
2087 | */ | |
2088 | if (changed) { | |
2089 | switch (log->l_covered_state) { | |
2090 | case XLOG_STATE_COVER_IDLE: | |
2091 | case XLOG_STATE_COVER_NEED: | |
2092 | case XLOG_STATE_COVER_NEED2: | |
2093 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2094 | break; | |
2095 | ||
2096 | case XLOG_STATE_COVER_DONE: | |
2097 | if (changed == 1) | |
2098 | log->l_covered_state = XLOG_STATE_COVER_NEED2; | |
2099 | else | |
2100 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2101 | break; | |
2102 | ||
2103 | case XLOG_STATE_COVER_DONE2: | |
2104 | if (changed == 1) | |
2105 | log->l_covered_state = XLOG_STATE_COVER_IDLE; | |
2106 | else | |
2107 | log->l_covered_state = XLOG_STATE_COVER_NEED; | |
2108 | break; | |
2109 | ||
2110 | default: | |
2111 | ASSERT(0); | |
2112 | } | |
2113 | } | |
2114 | } /* xlog_state_clean_log */ | |
2115 | ||
2116 | STATIC xfs_lsn_t | |
2117 | xlog_get_lowest_lsn( | |
2118 | xlog_t *log) | |
2119 | { | |
2120 | xlog_in_core_t *lsn_log; | |
2121 | xfs_lsn_t lowest_lsn, lsn; | |
2122 | ||
2123 | lsn_log = log->l_iclog; | |
2124 | lowest_lsn = 0; | |
2125 | do { | |
2126 | if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) { | |
b53e675d | 2127 | lsn = be64_to_cpu(lsn_log->ic_header.h_lsn); |
1da177e4 LT |
2128 | if ((lsn && !lowest_lsn) || |
2129 | (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) { | |
2130 | lowest_lsn = lsn; | |
2131 | } | |
2132 | } | |
2133 | lsn_log = lsn_log->ic_next; | |
2134 | } while (lsn_log != log->l_iclog); | |
014c2544 | 2135 | return lowest_lsn; |
1da177e4 LT |
2136 | } |
2137 | ||
2138 | ||
2139 | STATIC void | |
2140 | xlog_state_do_callback( | |
2141 | xlog_t *log, | |
2142 | int aborted, | |
2143 | xlog_in_core_t *ciclog) | |
2144 | { | |
2145 | xlog_in_core_t *iclog; | |
2146 | xlog_in_core_t *first_iclog; /* used to know when we've | |
2147 | * processed all iclogs once */ | |
2148 | xfs_log_callback_t *cb, *cb_next; | |
2149 | int flushcnt = 0; | |
2150 | xfs_lsn_t lowest_lsn; | |
2151 | int ioerrors; /* counter: iclogs with errors */ | |
2152 | int loopdidcallbacks; /* flag: inner loop did callbacks*/ | |
2153 | int funcdidcallbacks; /* flag: function did callbacks */ | |
2154 | int repeats; /* for issuing console warnings if | |
2155 | * looping too many times */ | |
d748c623 | 2156 | int wake = 0; |
1da177e4 | 2157 | |
b22cd72c | 2158 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2159 | first_iclog = iclog = log->l_iclog; |
2160 | ioerrors = 0; | |
2161 | funcdidcallbacks = 0; | |
2162 | repeats = 0; | |
2163 | ||
2164 | do { | |
2165 | /* | |
2166 | * Scan all iclogs starting with the one pointed to by the | |
2167 | * log. Reset this starting point each time the log is | |
2168 | * unlocked (during callbacks). | |
2169 | * | |
2170 | * Keep looping through iclogs until one full pass is made | |
2171 | * without running any callbacks. | |
2172 | */ | |
2173 | first_iclog = log->l_iclog; | |
2174 | iclog = log->l_iclog; | |
2175 | loopdidcallbacks = 0; | |
2176 | repeats++; | |
2177 | ||
2178 | do { | |
2179 | ||
2180 | /* skip all iclogs in the ACTIVE & DIRTY states */ | |
2181 | if (iclog->ic_state & | |
2182 | (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) { | |
2183 | iclog = iclog->ic_next; | |
2184 | continue; | |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * Between marking a filesystem SHUTDOWN and stopping | |
2189 | * the log, we do flush all iclogs to disk (if there | |
2190 | * wasn't a log I/O error). So, we do want things to | |
2191 | * go smoothly in case of just a SHUTDOWN w/o a | |
2192 | * LOG_IO_ERROR. | |
2193 | */ | |
2194 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) { | |
2195 | /* | |
2196 | * Can only perform callbacks in order. Since | |
2197 | * this iclog is not in the DONE_SYNC/ | |
2198 | * DO_CALLBACK state, we skip the rest and | |
2199 | * just try to clean up. If we set our iclog | |
2200 | * to DO_CALLBACK, we will not process it when | |
2201 | * we retry since a previous iclog is in the | |
2202 | * CALLBACK and the state cannot change since | |
b22cd72c | 2203 | * we are holding the l_icloglock. |
1da177e4 LT |
2204 | */ |
2205 | if (!(iclog->ic_state & | |
2206 | (XLOG_STATE_DONE_SYNC | | |
2207 | XLOG_STATE_DO_CALLBACK))) { | |
2208 | if (ciclog && (ciclog->ic_state == | |
2209 | XLOG_STATE_DONE_SYNC)) { | |
2210 | ciclog->ic_state = XLOG_STATE_DO_CALLBACK; | |
2211 | } | |
2212 | break; | |
2213 | } | |
2214 | /* | |
2215 | * We now have an iclog that is in either the | |
2216 | * DO_CALLBACK or DONE_SYNC states. The other | |
2217 | * states (WANT_SYNC, SYNCING, or CALLBACK were | |
2218 | * caught by the above if and are going to | |
2219 | * clean (i.e. we aren't doing their callbacks) | |
2220 | * see the above if. | |
2221 | */ | |
2222 | ||
2223 | /* | |
2224 | * We will do one more check here to see if we | |
2225 | * have chased our tail around. | |
2226 | */ | |
2227 | ||
2228 | lowest_lsn = xlog_get_lowest_lsn(log); | |
b53e675d CH |
2229 | if (lowest_lsn && |
2230 | XFS_LSN_CMP(lowest_lsn, | |
2231 | be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { | |
1da177e4 LT |
2232 | iclog = iclog->ic_next; |
2233 | continue; /* Leave this iclog for | |
2234 | * another thread */ | |
2235 | } | |
2236 | ||
2237 | iclog->ic_state = XLOG_STATE_CALLBACK; | |
2238 | ||
b22cd72c | 2239 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2240 | |
2241 | /* l_last_sync_lsn field protected by | |
c8b5ea28 | 2242 | * l_grant_lock. Don't worry about iclog's lsn. |
1da177e4 LT |
2243 | * No one else can be here except us. |
2244 | */ | |
c8b5ea28 | 2245 | spin_lock(&log->l_grant_lock); |
b53e675d CH |
2246 | ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn, |
2247 | be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); | |
2248 | log->l_last_sync_lsn = | |
2249 | be64_to_cpu(iclog->ic_header.h_lsn); | |
c8b5ea28 | 2250 | spin_unlock(&log->l_grant_lock); |
1da177e4 | 2251 | |
1da177e4 | 2252 | } else { |
114d23aa | 2253 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2254 | ioerrors++; |
2255 | } | |
1da177e4 | 2256 | |
114d23aa DC |
2257 | /* |
2258 | * Keep processing entries in the callback list until | |
2259 | * we come around and it is empty. We need to | |
2260 | * atomically see that the list is empty and change the | |
2261 | * state to DIRTY so that we don't miss any more | |
2262 | * callbacks being added. | |
2263 | */ | |
2264 | spin_lock(&iclog->ic_callback_lock); | |
2265 | cb = iclog->ic_callback; | |
4b80916b | 2266 | while (cb) { |
1da177e4 LT |
2267 | iclog->ic_callback_tail = &(iclog->ic_callback); |
2268 | iclog->ic_callback = NULL; | |
114d23aa | 2269 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2270 | |
2271 | /* perform callbacks in the order given */ | |
4b80916b | 2272 | for (; cb; cb = cb_next) { |
1da177e4 LT |
2273 | cb_next = cb->cb_next; |
2274 | cb->cb_func(cb->cb_arg, aborted); | |
2275 | } | |
114d23aa | 2276 | spin_lock(&iclog->ic_callback_lock); |
1da177e4 LT |
2277 | cb = iclog->ic_callback; |
2278 | } | |
2279 | ||
2280 | loopdidcallbacks++; | |
2281 | funcdidcallbacks++; | |
2282 | ||
114d23aa | 2283 | spin_lock(&log->l_icloglock); |
4b80916b | 2284 | ASSERT(iclog->ic_callback == NULL); |
114d23aa | 2285 | spin_unlock(&iclog->ic_callback_lock); |
1da177e4 LT |
2286 | if (!(iclog->ic_state & XLOG_STATE_IOERROR)) |
2287 | iclog->ic_state = XLOG_STATE_DIRTY; | |
2288 | ||
2289 | /* | |
2290 | * Transition from DIRTY to ACTIVE if applicable. | |
2291 | * NOP if STATE_IOERROR. | |
2292 | */ | |
2293 | xlog_state_clean_log(log); | |
2294 | ||
2295 | /* wake up threads waiting in xfs_log_force() */ | |
12017faf | 2296 | sv_broadcast(&iclog->ic_force_wait); |
1da177e4 LT |
2297 | |
2298 | iclog = iclog->ic_next; | |
2299 | } while (first_iclog != iclog); | |
a3c6685e NS |
2300 | |
2301 | if (repeats > 5000) { | |
2302 | flushcnt += repeats; | |
2303 | repeats = 0; | |
1da177e4 | 2304 | xfs_fs_cmn_err(CE_WARN, log->l_mp, |
a3c6685e | 2305 | "%s: possible infinite loop (%d iterations)", |
34a622b2 | 2306 | __func__, flushcnt); |
1da177e4 LT |
2307 | } |
2308 | } while (!ioerrors && loopdidcallbacks); | |
2309 | ||
2310 | /* | |
2311 | * make one last gasp attempt to see if iclogs are being left in | |
2312 | * limbo.. | |
2313 | */ | |
2314 | #ifdef DEBUG | |
2315 | if (funcdidcallbacks) { | |
2316 | first_iclog = iclog = log->l_iclog; | |
2317 | do { | |
2318 | ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK); | |
2319 | /* | |
2320 | * Terminate the loop if iclogs are found in states | |
2321 | * which will cause other threads to clean up iclogs. | |
2322 | * | |
2323 | * SYNCING - i/o completion will go through logs | |
2324 | * DONE_SYNC - interrupt thread should be waiting for | |
b22cd72c | 2325 | * l_icloglock |
1da177e4 LT |
2326 | * IOERROR - give up hope all ye who enter here |
2327 | */ | |
2328 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC || | |
2329 | iclog->ic_state == XLOG_STATE_SYNCING || | |
2330 | iclog->ic_state == XLOG_STATE_DONE_SYNC || | |
2331 | iclog->ic_state == XLOG_STATE_IOERROR ) | |
2332 | break; | |
2333 | iclog = iclog->ic_next; | |
2334 | } while (first_iclog != iclog); | |
2335 | } | |
2336 | #endif | |
2337 | ||
d748c623 MW |
2338 | if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) |
2339 | wake = 1; | |
b22cd72c | 2340 | spin_unlock(&log->l_icloglock); |
d748c623 MW |
2341 | |
2342 | if (wake) | |
2343 | sv_broadcast(&log->l_flush_wait); | |
2344 | } | |
1da177e4 LT |
2345 | |
2346 | ||
2347 | /* | |
2348 | * Finish transitioning this iclog to the dirty state. | |
2349 | * | |
2350 | * Make sure that we completely execute this routine only when this is | |
2351 | * the last call to the iclog. There is a good chance that iclog flushes, | |
2352 | * when we reach the end of the physical log, get turned into 2 separate | |
2353 | * calls to bwrite. Hence, one iclog flush could generate two calls to this | |
2354 | * routine. By using the reference count bwritecnt, we guarantee that only | |
2355 | * the second completion goes through. | |
2356 | * | |
2357 | * Callbacks could take time, so they are done outside the scope of the | |
12017faf | 2358 | * global state machine log lock. |
1da177e4 | 2359 | */ |
a8272ce0 | 2360 | STATIC void |
1da177e4 LT |
2361 | xlog_state_done_syncing( |
2362 | xlog_in_core_t *iclog, | |
2363 | int aborted) | |
2364 | { | |
2365 | xlog_t *log = iclog->ic_log; | |
1da177e4 | 2366 | |
b22cd72c | 2367 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
2368 | |
2369 | ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || | |
2370 | iclog->ic_state == XLOG_STATE_IOERROR); | |
155cc6b7 | 2371 | ASSERT(atomic_read(&iclog->ic_refcnt) == 0); |
1da177e4 LT |
2372 | ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); |
2373 | ||
2374 | ||
2375 | /* | |
2376 | * If we got an error, either on the first buffer, or in the case of | |
2377 | * split log writes, on the second, we mark ALL iclogs STATE_IOERROR, | |
2378 | * and none should ever be attempted to be written to disk | |
2379 | * again. | |
2380 | */ | |
2381 | if (iclog->ic_state != XLOG_STATE_IOERROR) { | |
2382 | if (--iclog->ic_bwritecnt == 1) { | |
b22cd72c | 2383 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2384 | return; |
2385 | } | |
2386 | iclog->ic_state = XLOG_STATE_DONE_SYNC; | |
2387 | } | |
2388 | ||
2389 | /* | |
2390 | * Someone could be sleeping prior to writing out the next | |
2391 | * iclog buffer, we wake them all, one will get to do the | |
2392 | * I/O, the others get to wait for the result. | |
2393 | */ | |
12017faf | 2394 | sv_broadcast(&iclog->ic_write_wait); |
b22cd72c | 2395 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2396 | xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ |
2397 | } /* xlog_state_done_syncing */ | |
2398 | ||
2399 | ||
2400 | /* | |
2401 | * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must | |
12017faf DC |
2402 | * sleep. We wait on the flush queue on the head iclog as that should be |
2403 | * the first iclog to complete flushing. Hence if all iclogs are syncing, | |
2404 | * we will wait here and all new writes will sleep until a sync completes. | |
1da177e4 LT |
2405 | * |
2406 | * The in-core logs are used in a circular fashion. They are not used | |
2407 | * out-of-order even when an iclog past the head is free. | |
2408 | * | |
2409 | * return: | |
2410 | * * log_offset where xlog_write() can start writing into the in-core | |
2411 | * log's data space. | |
2412 | * * in-core log pointer to which xlog_write() should write. | |
2413 | * * boolean indicating this is a continued write to an in-core log. | |
2414 | * If this is the last write, then the in-core log's offset field | |
2415 | * needs to be incremented, depending on the amount of data which | |
2416 | * is copied. | |
2417 | */ | |
a8272ce0 | 2418 | STATIC int |
1da177e4 LT |
2419 | xlog_state_get_iclog_space(xlog_t *log, |
2420 | int len, | |
2421 | xlog_in_core_t **iclogp, | |
2422 | xlog_ticket_t *ticket, | |
2423 | int *continued_write, | |
2424 | int *logoffsetp) | |
2425 | { | |
1da177e4 LT |
2426 | int log_offset; |
2427 | xlog_rec_header_t *head; | |
2428 | xlog_in_core_t *iclog; | |
2429 | int error; | |
2430 | ||
2431 | restart: | |
b22cd72c | 2432 | spin_lock(&log->l_icloglock); |
1da177e4 | 2433 | if (XLOG_FORCED_SHUTDOWN(log)) { |
b22cd72c | 2434 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2435 | return XFS_ERROR(EIO); |
2436 | } | |
2437 | ||
2438 | iclog = log->l_iclog; | |
d748c623 | 2439 | if (iclog->ic_state != XLOG_STATE_ACTIVE) { |
1da177e4 | 2440 | XFS_STATS_INC(xs_log_noiclogs); |
d748c623 MW |
2441 | |
2442 | /* Wait for log writes to have flushed */ | |
2443 | sv_wait(&log->l_flush_wait, 0, &log->l_icloglock, 0); | |
1da177e4 LT |
2444 | goto restart; |
2445 | } | |
d748c623 | 2446 | |
1da177e4 LT |
2447 | head = &iclog->ic_header; |
2448 | ||
155cc6b7 | 2449 | atomic_inc(&iclog->ic_refcnt); /* prevents sync */ |
1da177e4 LT |
2450 | log_offset = iclog->ic_offset; |
2451 | ||
2452 | /* On the 1st write to an iclog, figure out lsn. This works | |
2453 | * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are | |
2454 | * committing to. If the offset is set, that's how many blocks | |
2455 | * must be written. | |
2456 | */ | |
2457 | if (log_offset == 0) { | |
2458 | ticket->t_curr_res -= log->l_iclog_hsize; | |
0adba536 | 2459 | xlog_tic_add_region(ticket, |
7e9c6396 TS |
2460 | log->l_iclog_hsize, |
2461 | XLOG_REG_TYPE_LRHEADER); | |
b53e675d CH |
2462 | head->h_cycle = cpu_to_be32(log->l_curr_cycle); |
2463 | head->h_lsn = cpu_to_be64( | |
03bea6fe | 2464 | xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block)); |
1da177e4 LT |
2465 | ASSERT(log->l_curr_block >= 0); |
2466 | } | |
2467 | ||
2468 | /* If there is enough room to write everything, then do it. Otherwise, | |
2469 | * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC | |
2470 | * bit is on, so this will get flushed out. Don't update ic_offset | |
2471 | * until you know exactly how many bytes get copied. Therefore, wait | |
2472 | * until later to update ic_offset. | |
2473 | * | |
2474 | * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's | |
2475 | * can fit into remaining data section. | |
2476 | */ | |
2477 | if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { | |
2478 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2479 | ||
49641f1a DC |
2480 | /* |
2481 | * If I'm the only one writing to this iclog, sync it to disk. | |
2482 | * We need to do an atomic compare and decrement here to avoid | |
2483 | * racing with concurrent atomic_dec_and_lock() calls in | |
2484 | * xlog_state_release_iclog() when there is more than one | |
2485 | * reference to the iclog. | |
2486 | */ | |
2487 | if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { | |
2488 | /* we are the only one */ | |
b22cd72c | 2489 | spin_unlock(&log->l_icloglock); |
49641f1a DC |
2490 | error = xlog_state_release_iclog(log, iclog); |
2491 | if (error) | |
014c2544 | 2492 | return error; |
1da177e4 | 2493 | } else { |
b22cd72c | 2494 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2495 | } |
2496 | goto restart; | |
2497 | } | |
2498 | ||
2499 | /* Do we have enough room to write the full amount in the remainder | |
2500 | * of this iclog? Or must we continue a write on the next iclog and | |
2501 | * mark this iclog as completely taken? In the case where we switch | |
2502 | * iclogs (to mark it taken), this particular iclog will release/sync | |
2503 | * to disk in xlog_write(). | |
2504 | */ | |
2505 | if (len <= iclog->ic_size - iclog->ic_offset) { | |
2506 | *continued_write = 0; | |
2507 | iclog->ic_offset += len; | |
2508 | } else { | |
2509 | *continued_write = 1; | |
2510 | xlog_state_switch_iclogs(log, iclog, iclog->ic_size); | |
2511 | } | |
2512 | *iclogp = iclog; | |
2513 | ||
2514 | ASSERT(iclog->ic_offset <= iclog->ic_size); | |
b22cd72c | 2515 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2516 | |
2517 | *logoffsetp = log_offset; | |
2518 | return 0; | |
2519 | } /* xlog_state_get_iclog_space */ | |
2520 | ||
2521 | /* | |
2522 | * Atomically get the log space required for a log ticket. | |
2523 | * | |
2524 | * Once a ticket gets put onto the reserveq, it will only return after | |
2525 | * the needed reservation is satisfied. | |
2526 | */ | |
2527 | STATIC int | |
2528 | xlog_grant_log_space(xlog_t *log, | |
2529 | xlog_ticket_t *tic) | |
2530 | { | |
2531 | int free_bytes; | |
2532 | int need_bytes; | |
1da177e4 LT |
2533 | #ifdef DEBUG |
2534 | xfs_lsn_t tail_lsn; | |
2535 | #endif | |
2536 | ||
2537 | ||
2538 | #ifdef DEBUG | |
2539 | if (log->l_flags & XLOG_ACTIVE_RECOVERY) | |
2540 | panic("grant Recovery problem"); | |
2541 | #endif | |
2542 | ||
2543 | /* Is there space or do we need to sleep? */ | |
c8b5ea28 | 2544 | spin_lock(&log->l_grant_lock); |
0b1b213f CH |
2545 | |
2546 | trace_xfs_log_grant_enter(log, tic); | |
1da177e4 LT |
2547 | |
2548 | /* something is already sleeping; insert new transaction at end */ | |
2549 | if (log->l_reserve_headq) { | |
dd954c69 | 2550 | xlog_ins_ticketq(&log->l_reserve_headq, tic); |
0b1b213f CH |
2551 | |
2552 | trace_xfs_log_grant_sleep1(log, tic); | |
2553 | ||
1da177e4 LT |
2554 | /* |
2555 | * Gotta check this before going to sleep, while we're | |
2556 | * holding the grant lock. | |
2557 | */ | |
2558 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2559 | goto error_return; | |
2560 | ||
2561 | XFS_STATS_INC(xs_sleep_logspace); | |
12017faf | 2562 | sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s); |
1da177e4 LT |
2563 | /* |
2564 | * If we got an error, and the filesystem is shutting down, | |
2565 | * we'll catch it down below. So just continue... | |
2566 | */ | |
0b1b213f | 2567 | trace_xfs_log_grant_wake1(log, tic); |
c8b5ea28 | 2568 | spin_lock(&log->l_grant_lock); |
1da177e4 LT |
2569 | } |
2570 | if (tic->t_flags & XFS_LOG_PERM_RESERV) | |
2571 | need_bytes = tic->t_unit_res*tic->t_ocnt; | |
2572 | else | |
2573 | need_bytes = tic->t_unit_res; | |
2574 | ||
2575 | redo: | |
2576 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2577 | goto error_return; | |
2578 | ||
2579 | free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle, | |
2580 | log->l_grant_reserve_bytes); | |
2581 | if (free_bytes < need_bytes) { | |
2582 | if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) | |
dd954c69 | 2583 | xlog_ins_ticketq(&log->l_reserve_headq, tic); |
0b1b213f CH |
2584 | |
2585 | trace_xfs_log_grant_sleep2(log, tic); | |
2586 | ||
9d7fef74 DC |
2587 | spin_unlock(&log->l_grant_lock); |
2588 | xlog_grant_push_ail(log->l_mp, need_bytes); | |
2589 | spin_lock(&log->l_grant_lock); | |
2590 | ||
1da177e4 | 2591 | XFS_STATS_INC(xs_sleep_logspace); |
12017faf | 2592 | sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s); |
1da177e4 | 2593 | |
9d7fef74 DC |
2594 | spin_lock(&log->l_grant_lock); |
2595 | if (XLOG_FORCED_SHUTDOWN(log)) | |
1da177e4 | 2596 | goto error_return; |
1da177e4 | 2597 | |
0b1b213f CH |
2598 | trace_xfs_log_grant_wake2(log, tic); |
2599 | ||
1da177e4 LT |
2600 | goto redo; |
2601 | } else if (tic->t_flags & XLOG_TIC_IN_Q) | |
dd954c69 | 2602 | xlog_del_ticketq(&log->l_reserve_headq, tic); |
1da177e4 LT |
2603 | |
2604 | /* we've got enough space */ | |
dd954c69 | 2605 | xlog_grant_add_space(log, need_bytes); |
1da177e4 LT |
2606 | #ifdef DEBUG |
2607 | tail_lsn = log->l_tail_lsn; | |
2608 | /* | |
2609 | * Check to make sure the grant write head didn't just over lap the | |
2610 | * tail. If the cycles are the same, we can't be overlapping. | |
2611 | * Otherwise, make sure that the cycles differ by exactly one and | |
2612 | * check the byte count. | |
2613 | */ | |
2614 | if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) { | |
2615 | ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn)); | |
2616 | ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn))); | |
2617 | } | |
2618 | #endif | |
0b1b213f | 2619 | trace_xfs_log_grant_exit(log, tic); |
1da177e4 | 2620 | xlog_verify_grant_head(log, 1); |
c8b5ea28 | 2621 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
2622 | return 0; |
2623 | ||
2624 | error_return: | |
2625 | if (tic->t_flags & XLOG_TIC_IN_Q) | |
dd954c69 | 2626 | xlog_del_ticketq(&log->l_reserve_headq, tic); |
0b1b213f CH |
2627 | |
2628 | trace_xfs_log_grant_error(log, tic); | |
2629 | ||
1da177e4 LT |
2630 | /* |
2631 | * If we are failing, make sure the ticket doesn't have any | |
2632 | * current reservations. We don't want to add this back when | |
2633 | * the ticket/transaction gets cancelled. | |
2634 | */ | |
2635 | tic->t_curr_res = 0; | |
2636 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
c8b5ea28 | 2637 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
2638 | return XFS_ERROR(EIO); |
2639 | } /* xlog_grant_log_space */ | |
2640 | ||
2641 | ||
2642 | /* | |
2643 | * Replenish the byte reservation required by moving the grant write head. | |
2644 | * | |
2645 | * | |
2646 | */ | |
2647 | STATIC int | |
2648 | xlog_regrant_write_log_space(xlog_t *log, | |
2649 | xlog_ticket_t *tic) | |
2650 | { | |
1da177e4 LT |
2651 | int free_bytes, need_bytes; |
2652 | xlog_ticket_t *ntic; | |
2653 | #ifdef DEBUG | |
2654 | xfs_lsn_t tail_lsn; | |
2655 | #endif | |
2656 | ||
2657 | tic->t_curr_res = tic->t_unit_res; | |
0adba536 | 2658 | xlog_tic_reset_res(tic); |
1da177e4 LT |
2659 | |
2660 | if (tic->t_cnt > 0) | |
014c2544 | 2661 | return 0; |
1da177e4 LT |
2662 | |
2663 | #ifdef DEBUG | |
2664 | if (log->l_flags & XLOG_ACTIVE_RECOVERY) | |
2665 | panic("regrant Recovery problem"); | |
2666 | #endif | |
2667 | ||
c8b5ea28 | 2668 | spin_lock(&log->l_grant_lock); |
0b1b213f CH |
2669 | |
2670 | trace_xfs_log_regrant_write_enter(log, tic); | |
1da177e4 LT |
2671 | |
2672 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2673 | goto error_return; | |
2674 | ||
2675 | /* If there are other waiters on the queue then give them a | |
2676 | * chance at logspace before us. Wake up the first waiters, | |
2677 | * if we do not wake up all the waiters then go to sleep waiting | |
2678 | * for more free space, otherwise try to get some space for | |
2679 | * this transaction. | |
2680 | */ | |
9d7fef74 | 2681 | need_bytes = tic->t_unit_res; |
1da177e4 LT |
2682 | if ((ntic = log->l_write_headq)) { |
2683 | free_bytes = xlog_space_left(log, log->l_grant_write_cycle, | |
2684 | log->l_grant_write_bytes); | |
2685 | do { | |
2686 | ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV); | |
2687 | ||
2688 | if (free_bytes < ntic->t_unit_res) | |
2689 | break; | |
2690 | free_bytes -= ntic->t_unit_res; | |
12017faf | 2691 | sv_signal(&ntic->t_wait); |
1da177e4 LT |
2692 | ntic = ntic->t_next; |
2693 | } while (ntic != log->l_write_headq); | |
2694 | ||
2695 | if (ntic != log->l_write_headq) { | |
2696 | if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) | |
dd954c69 | 2697 | xlog_ins_ticketq(&log->l_write_headq, tic); |
1da177e4 | 2698 | |
0b1b213f CH |
2699 | trace_xfs_log_regrant_write_sleep1(log, tic); |
2700 | ||
9d7fef74 DC |
2701 | spin_unlock(&log->l_grant_lock); |
2702 | xlog_grant_push_ail(log->l_mp, need_bytes); | |
2703 | spin_lock(&log->l_grant_lock); | |
2704 | ||
1da177e4 | 2705 | XFS_STATS_INC(xs_sleep_logspace); |
12017faf | 2706 | sv_wait(&tic->t_wait, PINOD|PLTWAIT, |
1da177e4 LT |
2707 | &log->l_grant_lock, s); |
2708 | ||
2709 | /* If we're shutting down, this tic is already | |
2710 | * off the queue */ | |
9d7fef74 DC |
2711 | spin_lock(&log->l_grant_lock); |
2712 | if (XLOG_FORCED_SHUTDOWN(log)) | |
1da177e4 | 2713 | goto error_return; |
1da177e4 | 2714 | |
0b1b213f | 2715 | trace_xfs_log_regrant_write_wake1(log, tic); |
1da177e4 LT |
2716 | } |
2717 | } | |
2718 | ||
1da177e4 LT |
2719 | redo: |
2720 | if (XLOG_FORCED_SHUTDOWN(log)) | |
2721 | goto error_return; | |
2722 | ||
2723 | free_bytes = xlog_space_left(log, log->l_grant_write_cycle, | |
2724 | log->l_grant_write_bytes); | |
2725 | if (free_bytes < need_bytes) { | |
2726 | if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) | |
dd954c69 | 2727 | xlog_ins_ticketq(&log->l_write_headq, tic); |
9d7fef74 DC |
2728 | spin_unlock(&log->l_grant_lock); |
2729 | xlog_grant_push_ail(log->l_mp, need_bytes); | |
2730 | spin_lock(&log->l_grant_lock); | |
2731 | ||
1da177e4 | 2732 | XFS_STATS_INC(xs_sleep_logspace); |
0b1b213f CH |
2733 | trace_xfs_log_regrant_write_sleep2(log, tic); |
2734 | ||
12017faf | 2735 | sv_wait(&tic->t_wait, PINOD|PLTWAIT, &log->l_grant_lock, s); |
1da177e4 LT |
2736 | |
2737 | /* If we're shutting down, this tic is already off the queue */ | |
9d7fef74 DC |
2738 | spin_lock(&log->l_grant_lock); |
2739 | if (XLOG_FORCED_SHUTDOWN(log)) | |
1da177e4 | 2740 | goto error_return; |
1da177e4 | 2741 | |
0b1b213f | 2742 | trace_xfs_log_regrant_write_wake2(log, tic); |
1da177e4 LT |
2743 | goto redo; |
2744 | } else if (tic->t_flags & XLOG_TIC_IN_Q) | |
dd954c69 | 2745 | xlog_del_ticketq(&log->l_write_headq, tic); |
1da177e4 | 2746 | |
dd954c69 CH |
2747 | /* we've got enough space */ |
2748 | xlog_grant_add_space_write(log, need_bytes); | |
1da177e4 LT |
2749 | #ifdef DEBUG |
2750 | tail_lsn = log->l_tail_lsn; | |
2751 | if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) { | |
2752 | ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn)); | |
2753 | ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn))); | |
2754 | } | |
2755 | #endif | |
2756 | ||
0b1b213f CH |
2757 | trace_xfs_log_regrant_write_exit(log, tic); |
2758 | ||
1da177e4 | 2759 | xlog_verify_grant_head(log, 1); |
c8b5ea28 | 2760 | spin_unlock(&log->l_grant_lock); |
014c2544 | 2761 | return 0; |
1da177e4 LT |
2762 | |
2763 | ||
2764 | error_return: | |
2765 | if (tic->t_flags & XLOG_TIC_IN_Q) | |
dd954c69 | 2766 | xlog_del_ticketq(&log->l_reserve_headq, tic); |
0b1b213f CH |
2767 | |
2768 | trace_xfs_log_regrant_write_error(log, tic); | |
2769 | ||
1da177e4 LT |
2770 | /* |
2771 | * If we are failing, make sure the ticket doesn't have any | |
2772 | * current reservations. We don't want to add this back when | |
2773 | * the ticket/transaction gets cancelled. | |
2774 | */ | |
2775 | tic->t_curr_res = 0; | |
2776 | tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ | |
c8b5ea28 | 2777 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
2778 | return XFS_ERROR(EIO); |
2779 | } /* xlog_regrant_write_log_space */ | |
2780 | ||
2781 | ||
2782 | /* The first cnt-1 times through here we don't need to | |
2783 | * move the grant write head because the permanent | |
2784 | * reservation has reserved cnt times the unit amount. | |
2785 | * Release part of current permanent unit reservation and | |
2786 | * reset current reservation to be one units worth. Also | |
2787 | * move grant reservation head forward. | |
2788 | */ | |
2789 | STATIC void | |
2790 | xlog_regrant_reserve_log_space(xlog_t *log, | |
2791 | xlog_ticket_t *ticket) | |
2792 | { | |
0b1b213f CH |
2793 | trace_xfs_log_regrant_reserve_enter(log, ticket); |
2794 | ||
1da177e4 LT |
2795 | if (ticket->t_cnt > 0) |
2796 | ticket->t_cnt--; | |
2797 | ||
c8b5ea28 | 2798 | spin_lock(&log->l_grant_lock); |
dd954c69 | 2799 | xlog_grant_sub_space(log, ticket->t_curr_res); |
1da177e4 | 2800 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2801 | xlog_tic_reset_res(ticket); |
0b1b213f CH |
2802 | |
2803 | trace_xfs_log_regrant_reserve_sub(log, ticket); | |
2804 | ||
1da177e4 LT |
2805 | xlog_verify_grant_head(log, 1); |
2806 | ||
2807 | /* just return if we still have some of the pre-reserved space */ | |
2808 | if (ticket->t_cnt > 0) { | |
c8b5ea28 | 2809 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
2810 | return; |
2811 | } | |
2812 | ||
dd954c69 | 2813 | xlog_grant_add_space_reserve(log, ticket->t_unit_res); |
0b1b213f CH |
2814 | |
2815 | trace_xfs_log_regrant_reserve_exit(log, ticket); | |
2816 | ||
1da177e4 | 2817 | xlog_verify_grant_head(log, 0); |
c8b5ea28 | 2818 | spin_unlock(&log->l_grant_lock); |
1da177e4 | 2819 | ticket->t_curr_res = ticket->t_unit_res; |
0adba536 | 2820 | xlog_tic_reset_res(ticket); |
1da177e4 LT |
2821 | } /* xlog_regrant_reserve_log_space */ |
2822 | ||
2823 | ||
2824 | /* | |
2825 | * Give back the space left from a reservation. | |
2826 | * | |
2827 | * All the information we need to make a correct determination of space left | |
2828 | * is present. For non-permanent reservations, things are quite easy. The | |
2829 | * count should have been decremented to zero. We only need to deal with the | |
2830 | * space remaining in the current reservation part of the ticket. If the | |
2831 | * ticket contains a permanent reservation, there may be left over space which | |
2832 | * needs to be released. A count of N means that N-1 refills of the current | |
2833 | * reservation can be done before we need to ask for more space. The first | |
2834 | * one goes to fill up the first current reservation. Once we run out of | |
2835 | * space, the count will stay at zero and the only space remaining will be | |
2836 | * in the current reservation field. | |
2837 | */ | |
2838 | STATIC void | |
2839 | xlog_ungrant_log_space(xlog_t *log, | |
2840 | xlog_ticket_t *ticket) | |
2841 | { | |
1da177e4 LT |
2842 | if (ticket->t_cnt > 0) |
2843 | ticket->t_cnt--; | |
2844 | ||
c8b5ea28 | 2845 | spin_lock(&log->l_grant_lock); |
0b1b213f | 2846 | trace_xfs_log_ungrant_enter(log, ticket); |
1da177e4 | 2847 | |
dd954c69 | 2848 | xlog_grant_sub_space(log, ticket->t_curr_res); |
1da177e4 | 2849 | |
0b1b213f | 2850 | trace_xfs_log_ungrant_sub(log, ticket); |
1da177e4 LT |
2851 | |
2852 | /* If this is a permanent reservation ticket, we may be able to free | |
2853 | * up more space based on the remaining count. | |
2854 | */ | |
2855 | if (ticket->t_cnt > 0) { | |
2856 | ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); | |
dd954c69 | 2857 | xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt); |
1da177e4 LT |
2858 | } |
2859 | ||
0b1b213f CH |
2860 | trace_xfs_log_ungrant_exit(log, ticket); |
2861 | ||
1da177e4 | 2862 | xlog_verify_grant_head(log, 1); |
c8b5ea28 | 2863 | spin_unlock(&log->l_grant_lock); |
1da177e4 LT |
2864 | xfs_log_move_tail(log->l_mp, 1); |
2865 | } /* xlog_ungrant_log_space */ | |
2866 | ||
2867 | ||
1da177e4 LT |
2868 | /* |
2869 | * Flush iclog to disk if this is the last reference to the given iclog and | |
2870 | * the WANT_SYNC bit is set. | |
2871 | * | |
2872 | * When this function is entered, the iclog is not necessarily in the | |
2873 | * WANT_SYNC state. It may be sitting around waiting to get filled. | |
2874 | * | |
2875 | * | |
2876 | */ | |
a8272ce0 | 2877 | STATIC int |
b589334c DC |
2878 | xlog_state_release_iclog( |
2879 | xlog_t *log, | |
2880 | xlog_in_core_t *iclog) | |
1da177e4 | 2881 | { |
1da177e4 LT |
2882 | int sync = 0; /* do we sync? */ |
2883 | ||
155cc6b7 DC |
2884 | if (iclog->ic_state & XLOG_STATE_IOERROR) |
2885 | return XFS_ERROR(EIO); | |
2886 | ||
2887 | ASSERT(atomic_read(&iclog->ic_refcnt) > 0); | |
2888 | if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) | |
2889 | return 0; | |
2890 | ||
1da177e4 | 2891 | if (iclog->ic_state & XLOG_STATE_IOERROR) { |
b22cd72c | 2892 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2893 | return XFS_ERROR(EIO); |
2894 | } | |
1da177e4 LT |
2895 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || |
2896 | iclog->ic_state == XLOG_STATE_WANT_SYNC); | |
2897 | ||
155cc6b7 | 2898 | if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { |
b589334c DC |
2899 | /* update tail before writing to iclog */ |
2900 | xlog_assign_tail_lsn(log->l_mp); | |
1da177e4 LT |
2901 | sync++; |
2902 | iclog->ic_state = XLOG_STATE_SYNCING; | |
b53e675d | 2903 | iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn); |
1da177e4 LT |
2904 | xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn); |
2905 | /* cycle incremented when incrementing curr_block */ | |
2906 | } | |
b22cd72c | 2907 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
2908 | |
2909 | /* | |
2910 | * We let the log lock go, so it's possible that we hit a log I/O | |
c41564b5 | 2911 | * error or some other SHUTDOWN condition that marks the iclog |
1da177e4 LT |
2912 | * as XLOG_STATE_IOERROR before the bwrite. However, we know that |
2913 | * this iclog has consistent data, so we ignore IOERROR | |
2914 | * flags after this point. | |
2915 | */ | |
b589334c | 2916 | if (sync) |
1da177e4 | 2917 | return xlog_sync(log, iclog); |
014c2544 | 2918 | return 0; |
1da177e4 LT |
2919 | } /* xlog_state_release_iclog */ |
2920 | ||
2921 | ||
2922 | /* | |
2923 | * This routine will mark the current iclog in the ring as WANT_SYNC | |
2924 | * and move the current iclog pointer to the next iclog in the ring. | |
2925 | * When this routine is called from xlog_state_get_iclog_space(), the | |
2926 | * exact size of the iclog has not yet been determined. All we know is | |
2927 | * that every data block. We have run out of space in this log record. | |
2928 | */ | |
2929 | STATIC void | |
2930 | xlog_state_switch_iclogs(xlog_t *log, | |
2931 | xlog_in_core_t *iclog, | |
2932 | int eventual_size) | |
2933 | { | |
2934 | ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); | |
2935 | if (!eventual_size) | |
2936 | eventual_size = iclog->ic_offset; | |
2937 | iclog->ic_state = XLOG_STATE_WANT_SYNC; | |
b53e675d | 2938 | iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block); |
1da177e4 LT |
2939 | log->l_prev_block = log->l_curr_block; |
2940 | log->l_prev_cycle = log->l_curr_cycle; | |
2941 | ||
2942 | /* roll log?: ic_offset changed later */ | |
2943 | log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); | |
2944 | ||
2945 | /* Round up to next log-sunit */ | |
62118709 | 2946 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
1da177e4 LT |
2947 | log->l_mp->m_sb.sb_logsunit > 1) { |
2948 | __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); | |
2949 | log->l_curr_block = roundup(log->l_curr_block, sunit_bb); | |
2950 | } | |
2951 | ||
2952 | if (log->l_curr_block >= log->l_logBBsize) { | |
2953 | log->l_curr_cycle++; | |
2954 | if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM) | |
2955 | log->l_curr_cycle++; | |
2956 | log->l_curr_block -= log->l_logBBsize; | |
2957 | ASSERT(log->l_curr_block >= 0); | |
2958 | } | |
2959 | ASSERT(iclog == log->l_iclog); | |
2960 | log->l_iclog = iclog->ic_next; | |
2961 | } /* xlog_state_switch_iclogs */ | |
2962 | ||
1da177e4 LT |
2963 | /* |
2964 | * Write out all data in the in-core log as of this exact moment in time. | |
2965 | * | |
2966 | * Data may be written to the in-core log during this call. However, | |
2967 | * we don't guarantee this data will be written out. A change from past | |
2968 | * implementation means this routine will *not* write out zero length LRs. | |
2969 | * | |
2970 | * Basically, we try and perform an intelligent scan of the in-core logs. | |
2971 | * If we determine there is no flushable data, we just return. There is no | |
2972 | * flushable data if: | |
2973 | * | |
2974 | * 1. the current iclog is active and has no data; the previous iclog | |
2975 | * is in the active or dirty state. | |
2976 | * 2. the current iclog is drity, and the previous iclog is in the | |
2977 | * active or dirty state. | |
2978 | * | |
12017faf | 2979 | * We may sleep if: |
1da177e4 LT |
2980 | * |
2981 | * 1. the current iclog is not in the active nor dirty state. | |
2982 | * 2. the current iclog dirty, and the previous iclog is not in the | |
2983 | * active nor dirty state. | |
2984 | * 3. the current iclog is active, and there is another thread writing | |
2985 | * to this particular iclog. | |
2986 | * 4. a) the current iclog is active and has no other writers | |
2987 | * b) when we return from flushing out this iclog, it is still | |
2988 | * not in the active nor dirty state. | |
2989 | */ | |
a14a348b CH |
2990 | int |
2991 | _xfs_log_force( | |
2992 | struct xfs_mount *mp, | |
2993 | uint flags, | |
2994 | int *log_flushed) | |
1da177e4 | 2995 | { |
a14a348b CH |
2996 | struct log *log = mp->m_log; |
2997 | struct xlog_in_core *iclog; | |
2998 | xfs_lsn_t lsn; | |
2999 | ||
3000 | XFS_STATS_INC(xs_log_force); | |
1da177e4 | 3001 | |
b22cd72c | 3002 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3003 | |
3004 | iclog = log->l_iclog; | |
3005 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3006 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3007 | return XFS_ERROR(EIO); |
3008 | } | |
3009 | ||
3010 | /* If the head iclog is not active nor dirty, we just attach | |
3011 | * ourselves to the head and go to sleep. | |
3012 | */ | |
3013 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3014 | iclog->ic_state == XLOG_STATE_DIRTY) { | |
3015 | /* | |
3016 | * If the head is dirty or (active and empty), then | |
3017 | * we need to look at the previous iclog. If the previous | |
3018 | * iclog is active or dirty we are done. There is nothing | |
3019 | * to sync out. Otherwise, we attach ourselves to the | |
3020 | * previous iclog and go to sleep. | |
3021 | */ | |
3022 | if (iclog->ic_state == XLOG_STATE_DIRTY || | |
155cc6b7 DC |
3023 | (atomic_read(&iclog->ic_refcnt) == 0 |
3024 | && iclog->ic_offset == 0)) { | |
1da177e4 LT |
3025 | iclog = iclog->ic_prev; |
3026 | if (iclog->ic_state == XLOG_STATE_ACTIVE || | |
3027 | iclog->ic_state == XLOG_STATE_DIRTY) | |
3028 | goto no_sleep; | |
3029 | else | |
3030 | goto maybe_sleep; | |
3031 | } else { | |
155cc6b7 | 3032 | if (atomic_read(&iclog->ic_refcnt) == 0) { |
1da177e4 LT |
3033 | /* We are the only one with access to this |
3034 | * iclog. Flush it out now. There should | |
3035 | * be a roundoff of zero to show that someone | |
3036 | * has already taken care of the roundoff from | |
3037 | * the previous sync. | |
3038 | */ | |
155cc6b7 | 3039 | atomic_inc(&iclog->ic_refcnt); |
b53e675d | 3040 | lsn = be64_to_cpu(iclog->ic_header.h_lsn); |
1da177e4 | 3041 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3042 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3043 | |
3044 | if (xlog_state_release_iclog(log, iclog)) | |
3045 | return XFS_ERROR(EIO); | |
a14a348b CH |
3046 | |
3047 | if (log_flushed) | |
3048 | *log_flushed = 1; | |
b22cd72c | 3049 | spin_lock(&log->l_icloglock); |
b53e675d | 3050 | if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn && |
1da177e4 LT |
3051 | iclog->ic_state != XLOG_STATE_DIRTY) |
3052 | goto maybe_sleep; | |
3053 | else | |
3054 | goto no_sleep; | |
3055 | } else { | |
3056 | /* Someone else is writing to this iclog. | |
3057 | * Use its call to flush out the data. However, | |
3058 | * the other thread may not force out this LR, | |
3059 | * so we mark it WANT_SYNC. | |
3060 | */ | |
3061 | xlog_state_switch_iclogs(log, iclog, 0); | |
3062 | goto maybe_sleep; | |
3063 | } | |
3064 | } | |
3065 | } | |
3066 | ||
3067 | /* By the time we come around again, the iclog could've been filled | |
3068 | * which would give it another lsn. If we have a new lsn, just | |
3069 | * return because the relevant data has been flushed. | |
3070 | */ | |
3071 | maybe_sleep: | |
3072 | if (flags & XFS_LOG_SYNC) { | |
3073 | /* | |
3074 | * We must check if we're shutting down here, before | |
b22cd72c | 3075 | * we wait, while we're holding the l_icloglock. |
1da177e4 LT |
3076 | * Then we check again after waking up, in case our |
3077 | * sleep was disturbed by a bad news. | |
3078 | */ | |
3079 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3080 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3081 | return XFS_ERROR(EIO); |
3082 | } | |
3083 | XFS_STATS_INC(xs_log_force_sleep); | |
12017faf | 3084 | sv_wait(&iclog->ic_force_wait, PINOD, &log->l_icloglock, s); |
1da177e4 LT |
3085 | /* |
3086 | * No need to grab the log lock here since we're | |
3087 | * only deciding whether or not to return EIO | |
3088 | * and the memory read should be atomic. | |
3089 | */ | |
3090 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3091 | return XFS_ERROR(EIO); | |
a14a348b CH |
3092 | if (log_flushed) |
3093 | *log_flushed = 1; | |
1da177e4 LT |
3094 | } else { |
3095 | ||
3096 | no_sleep: | |
b22cd72c | 3097 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3098 | } |
3099 | return 0; | |
a14a348b | 3100 | } |
1da177e4 | 3101 | |
a14a348b CH |
3102 | /* |
3103 | * Wrapper for _xfs_log_force(), to be used when caller doesn't care | |
3104 | * about errors or whether the log was flushed or not. This is the normal | |
3105 | * interface to use when trying to unpin items or move the log forward. | |
3106 | */ | |
3107 | void | |
3108 | xfs_log_force( | |
3109 | xfs_mount_t *mp, | |
3110 | uint flags) | |
3111 | { | |
3112 | int error; | |
3113 | ||
3114 | error = _xfs_log_force(mp, flags, NULL); | |
3115 | if (error) { | |
3116 | xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: " | |
3117 | "error %d returned.", error); | |
3118 | } | |
3119 | } | |
1da177e4 LT |
3120 | |
3121 | /* | |
a14a348b | 3122 | * Force the in-core log to disk for a specific LSN. |
1da177e4 LT |
3123 | * |
3124 | * Find in-core log with lsn. | |
3125 | * If it is in the DIRTY state, just return. | |
3126 | * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC | |
3127 | * state and go to sleep or return. | |
3128 | * If it is in any other state, go to sleep or return. | |
3129 | * | |
a14a348b CH |
3130 | * Synchronous forces are implemented with a signal variable. All callers |
3131 | * to force a given lsn to disk will wait on a the sv attached to the | |
3132 | * specific in-core log. When given in-core log finally completes its | |
3133 | * write to disk, that thread will wake up all threads waiting on the | |
3134 | * sv. | |
1da177e4 | 3135 | */ |
a14a348b CH |
3136 | int |
3137 | _xfs_log_force_lsn( | |
3138 | struct xfs_mount *mp, | |
3139 | xfs_lsn_t lsn, | |
3140 | uint flags, | |
3141 | int *log_flushed) | |
1da177e4 | 3142 | { |
a14a348b CH |
3143 | struct log *log = mp->m_log; |
3144 | struct xlog_in_core *iclog; | |
3145 | int already_slept = 0; | |
1da177e4 | 3146 | |
a14a348b | 3147 | ASSERT(lsn != 0); |
1da177e4 | 3148 | |
a14a348b | 3149 | XFS_STATS_INC(xs_log_force); |
1da177e4 | 3150 | |
a14a348b CH |
3151 | try_again: |
3152 | spin_lock(&log->l_icloglock); | |
3153 | iclog = log->l_iclog; | |
3154 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
b22cd72c | 3155 | spin_unlock(&log->l_icloglock); |
a14a348b | 3156 | return XFS_ERROR(EIO); |
1da177e4 LT |
3157 | } |
3158 | ||
a14a348b CH |
3159 | do { |
3160 | if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { | |
3161 | iclog = iclog->ic_next; | |
3162 | continue; | |
3163 | } | |
3164 | ||
3165 | if (iclog->ic_state == XLOG_STATE_DIRTY) { | |
3166 | spin_unlock(&log->l_icloglock); | |
3167 | return 0; | |
3168 | } | |
3169 | ||
3170 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3171 | /* | |
3172 | * We sleep here if we haven't already slept (e.g. | |
3173 | * this is the first time we've looked at the correct | |
3174 | * iclog buf) and the buffer before us is going to | |
3175 | * be sync'ed. The reason for this is that if we | |
3176 | * are doing sync transactions here, by waiting for | |
3177 | * the previous I/O to complete, we can allow a few | |
3178 | * more transactions into this iclog before we close | |
3179 | * it down. | |
3180 | * | |
3181 | * Otherwise, we mark the buffer WANT_SYNC, and bump | |
3182 | * up the refcnt so we can release the log (which | |
3183 | * drops the ref count). The state switch keeps new | |
3184 | * transaction commits from using this buffer. When | |
3185 | * the current commits finish writing into the buffer, | |
3186 | * the refcount will drop to zero and the buffer will | |
3187 | * go out then. | |
3188 | */ | |
3189 | if (!already_slept && | |
3190 | (iclog->ic_prev->ic_state & | |
3191 | (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { | |
3192 | ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); | |
3193 | ||
3194 | XFS_STATS_INC(xs_log_force_sleep); | |
3195 | ||
3196 | sv_wait(&iclog->ic_prev->ic_write_wait, | |
3197 | PSWP, &log->l_icloglock, s); | |
3198 | if (log_flushed) | |
3199 | *log_flushed = 1; | |
3200 | already_slept = 1; | |
3201 | goto try_again; | |
3202 | } | |
155cc6b7 | 3203 | atomic_inc(&iclog->ic_refcnt); |
1da177e4 | 3204 | xlog_state_switch_iclogs(log, iclog, 0); |
b22cd72c | 3205 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3206 | if (xlog_state_release_iclog(log, iclog)) |
3207 | return XFS_ERROR(EIO); | |
a14a348b CH |
3208 | if (log_flushed) |
3209 | *log_flushed = 1; | |
b22cd72c | 3210 | spin_lock(&log->l_icloglock); |
1da177e4 | 3211 | } |
1da177e4 | 3212 | |
a14a348b CH |
3213 | if ((flags & XFS_LOG_SYNC) && /* sleep */ |
3214 | !(iclog->ic_state & | |
3215 | (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { | |
3216 | /* | |
3217 | * Don't wait on completion if we know that we've | |
3218 | * gotten a log write error. | |
3219 | */ | |
3220 | if (iclog->ic_state & XLOG_STATE_IOERROR) { | |
3221 | spin_unlock(&log->l_icloglock); | |
3222 | return XFS_ERROR(EIO); | |
3223 | } | |
3224 | XFS_STATS_INC(xs_log_force_sleep); | |
3225 | sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s); | |
3226 | /* | |
3227 | * No need to grab the log lock here since we're | |
3228 | * only deciding whether or not to return EIO | |
3229 | * and the memory read should be atomic. | |
3230 | */ | |
3231 | if (iclog->ic_state & XLOG_STATE_IOERROR) | |
3232 | return XFS_ERROR(EIO); | |
1da177e4 | 3233 | |
a14a348b CH |
3234 | if (log_flushed) |
3235 | *log_flushed = 1; | |
3236 | } else { /* just return */ | |
b22cd72c | 3237 | spin_unlock(&log->l_icloglock); |
1da177e4 | 3238 | } |
1da177e4 | 3239 | |
a14a348b CH |
3240 | return 0; |
3241 | } while (iclog != log->l_iclog); | |
1da177e4 | 3242 | |
a14a348b CH |
3243 | spin_unlock(&log->l_icloglock); |
3244 | return 0; | |
3245 | } | |
3246 | ||
3247 | /* | |
3248 | * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care | |
3249 | * about errors or whether the log was flushed or not. This is the normal | |
3250 | * interface to use when trying to unpin items or move the log forward. | |
3251 | */ | |
3252 | void | |
3253 | xfs_log_force_lsn( | |
3254 | xfs_mount_t *mp, | |
3255 | xfs_lsn_t lsn, | |
3256 | uint flags) | |
3257 | { | |
3258 | int error; | |
1da177e4 | 3259 | |
a14a348b CH |
3260 | error = _xfs_log_force_lsn(mp, lsn, flags, NULL); |
3261 | if (error) { | |
3262 | xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: " | |
3263 | "error %d returned.", error); | |
3264 | } | |
3265 | } | |
1da177e4 LT |
3266 | |
3267 | /* | |
3268 | * Called when we want to mark the current iclog as being ready to sync to | |
3269 | * disk. | |
3270 | */ | |
a8272ce0 | 3271 | STATIC void |
1da177e4 LT |
3272 | xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog) |
3273 | { | |
a8914f3a | 3274 | assert_spin_locked(&log->l_icloglock); |
1da177e4 LT |
3275 | |
3276 | if (iclog->ic_state == XLOG_STATE_ACTIVE) { | |
3277 | xlog_state_switch_iclogs(log, iclog, 0); | |
3278 | } else { | |
3279 | ASSERT(iclog->ic_state & | |
3280 | (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); | |
3281 | } | |
39e2defe | 3282 | } |
1da177e4 LT |
3283 | |
3284 | ||
3285 | /***************************************************************************** | |
3286 | * | |
3287 | * TICKET functions | |
3288 | * | |
3289 | ***************************************************************************** | |
3290 | */ | |
3291 | ||
3292 | /* | |
9da096fd | 3293 | * Free a used ticket when its refcount falls to zero. |
1da177e4 | 3294 | */ |
cc09c0dc DC |
3295 | void |
3296 | xfs_log_ticket_put( | |
3297 | xlog_ticket_t *ticket) | |
1da177e4 | 3298 | { |
cc09c0dc DC |
3299 | ASSERT(atomic_read(&ticket->t_ref) > 0); |
3300 | if (atomic_dec_and_test(&ticket->t_ref)) { | |
3301 | sv_destroy(&ticket->t_wait); | |
3302 | kmem_zone_free(xfs_log_ticket_zone, ticket); | |
3303 | } | |
3304 | } | |
1da177e4 | 3305 | |
cc09c0dc DC |
3306 | xlog_ticket_t * |
3307 | xfs_log_ticket_get( | |
3308 | xlog_ticket_t *ticket) | |
3309 | { | |
3310 | ASSERT(atomic_read(&ticket->t_ref) > 0); | |
3311 | atomic_inc(&ticket->t_ref); | |
3312 | return ticket; | |
3313 | } | |
1da177e4 | 3314 | |
955833cf DC |
3315 | xlog_tid_t |
3316 | xfs_log_get_trans_ident( | |
3317 | struct xfs_trans *tp) | |
3318 | { | |
3319 | return tp->t_ticket->t_tid; | |
3320 | } | |
3321 | ||
1da177e4 | 3322 | /* |
eb01c9cd | 3323 | * Allocate and initialise a new log ticket. |
1da177e4 | 3324 | */ |
a8272ce0 | 3325 | STATIC xlog_ticket_t * |
9b9fc2b7 DC |
3326 | xlog_ticket_alloc( |
3327 | struct log *log, | |
3328 | int unit_bytes, | |
3329 | int cnt, | |
3330 | char client, | |
3383ca57 DC |
3331 | uint xflags, |
3332 | int alloc_flags) | |
1da177e4 | 3333 | { |
9b9fc2b7 | 3334 | struct xlog_ticket *tic; |
1da177e4 | 3335 | uint num_headers; |
9b9fc2b7 | 3336 | int iclog_space; |
1da177e4 | 3337 | |
3383ca57 | 3338 | tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); |
eb01c9cd DC |
3339 | if (!tic) |
3340 | return NULL; | |
1da177e4 LT |
3341 | |
3342 | /* | |
3343 | * Permanent reservations have up to 'cnt'-1 active log operations | |
3344 | * in the log. A unit in this case is the amount of space for one | |
3345 | * of these log operations. Normal reservations have a cnt of 1 | |
3346 | * and their unit amount is the total amount of space required. | |
3347 | * | |
3348 | * The following lines of code account for non-transaction data | |
32fb9b57 TS |
3349 | * which occupy space in the on-disk log. |
3350 | * | |
3351 | * Normal form of a transaction is: | |
3352 | * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph> | |
3353 | * and then there are LR hdrs, split-recs and roundoff at end of syncs. | |
3354 | * | |
3355 | * We need to account for all the leadup data and trailer data | |
3356 | * around the transaction data. | |
3357 | * And then we need to account for the worst case in terms of using | |
3358 | * more space. | |
3359 | * The worst case will happen if: | |
3360 | * - the placement of the transaction happens to be such that the | |
3361 | * roundoff is at its maximum | |
3362 | * - the transaction data is synced before the commit record is synced | |
3363 | * i.e. <transaction-data><roundoff> | <commit-rec><roundoff> | |
3364 | * Therefore the commit record is in its own Log Record. | |
3365 | * This can happen as the commit record is called with its | |
3366 | * own region to xlog_write(). | |
3367 | * This then means that in the worst case, roundoff can happen for | |
3368 | * the commit-rec as well. | |
3369 | * The commit-rec is smaller than padding in this scenario and so it is | |
3370 | * not added separately. | |
1da177e4 LT |
3371 | */ |
3372 | ||
32fb9b57 TS |
3373 | /* for trans header */ |
3374 | unit_bytes += sizeof(xlog_op_header_t); | |
3375 | unit_bytes += sizeof(xfs_trans_header_t); | |
3376 | ||
1da177e4 | 3377 | /* for start-rec */ |
32fb9b57 TS |
3378 | unit_bytes += sizeof(xlog_op_header_t); |
3379 | ||
9b9fc2b7 DC |
3380 | /* |
3381 | * for LR headers - the space for data in an iclog is the size minus | |
3382 | * the space used for the headers. If we use the iclog size, then we | |
3383 | * undercalculate the number of headers required. | |
3384 | * | |
3385 | * Furthermore - the addition of op headers for split-recs might | |
3386 | * increase the space required enough to require more log and op | |
3387 | * headers, so take that into account too. | |
3388 | * | |
3389 | * IMPORTANT: This reservation makes the assumption that if this | |
3390 | * transaction is the first in an iclog and hence has the LR headers | |
3391 | * accounted to it, then the remaining space in the iclog is | |
3392 | * exclusively for this transaction. i.e. if the transaction is larger | |
3393 | * than the iclog, it will be the only thing in that iclog. | |
3394 | * Fundamentally, this means we must pass the entire log vector to | |
3395 | * xlog_write to guarantee this. | |
3396 | */ | |
3397 | iclog_space = log->l_iclog_size - log->l_iclog_hsize; | |
3398 | num_headers = howmany(unit_bytes, iclog_space); | |
3399 | ||
3400 | /* for split-recs - ophdrs added when data split over LRs */ | |
3401 | unit_bytes += sizeof(xlog_op_header_t) * num_headers; | |
3402 | ||
3403 | /* add extra header reservations if we overrun */ | |
3404 | while (!num_headers || | |
3405 | howmany(unit_bytes, iclog_space) > num_headers) { | |
3406 | unit_bytes += sizeof(xlog_op_header_t); | |
3407 | num_headers++; | |
3408 | } | |
32fb9b57 | 3409 | unit_bytes += log->l_iclog_hsize * num_headers; |
1da177e4 | 3410 | |
32fb9b57 TS |
3411 | /* for commit-rec LR header - note: padding will subsume the ophdr */ |
3412 | unit_bytes += log->l_iclog_hsize; | |
3413 | ||
32fb9b57 | 3414 | /* for roundoff padding for transaction data and one for commit record */ |
62118709 | 3415 | if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && |
32fb9b57 | 3416 | log->l_mp->m_sb.sb_logsunit > 1) { |
1da177e4 | 3417 | /* log su roundoff */ |
32fb9b57 | 3418 | unit_bytes += 2*log->l_mp->m_sb.sb_logsunit; |
1da177e4 LT |
3419 | } else { |
3420 | /* BB roundoff */ | |
32fb9b57 | 3421 | unit_bytes += 2*BBSIZE; |
1da177e4 LT |
3422 | } |
3423 | ||
cc09c0dc | 3424 | atomic_set(&tic->t_ref, 1); |
1da177e4 LT |
3425 | tic->t_unit_res = unit_bytes; |
3426 | tic->t_curr_res = unit_bytes; | |
3427 | tic->t_cnt = cnt; | |
3428 | tic->t_ocnt = cnt; | |
f9837107 | 3429 | tic->t_tid = random32(); |
1da177e4 LT |
3430 | tic->t_clientid = client; |
3431 | tic->t_flags = XLOG_TIC_INITED; | |
7e9c6396 | 3432 | tic->t_trans_type = 0; |
1da177e4 LT |
3433 | if (xflags & XFS_LOG_PERM_RESERV) |
3434 | tic->t_flags |= XLOG_TIC_PERM_RESERV; | |
9b9fc2b7 | 3435 | sv_init(&tic->t_wait, SV_DEFAULT, "logtick"); |
1da177e4 | 3436 | |
0adba536 | 3437 | xlog_tic_reset_res(tic); |
7e9c6396 | 3438 | |
1da177e4 | 3439 | return tic; |
cc09c0dc | 3440 | } |
1da177e4 LT |
3441 | |
3442 | ||
3443 | /****************************************************************************** | |
3444 | * | |
3445 | * Log debug routines | |
3446 | * | |
3447 | ****************************************************************************** | |
3448 | */ | |
cfcbbbd0 | 3449 | #if defined(DEBUG) |
1da177e4 LT |
3450 | /* |
3451 | * Make sure that the destination ptr is within the valid data region of | |
3452 | * one of the iclogs. This uses backup pointers stored in a different | |
3453 | * part of the log in case we trash the log structure. | |
3454 | */ | |
3455 | void | |
e6b1f273 CH |
3456 | xlog_verify_dest_ptr( |
3457 | struct log *log, | |
3458 | char *ptr) | |
1da177e4 LT |
3459 | { |
3460 | int i; | |
3461 | int good_ptr = 0; | |
3462 | ||
e6b1f273 CH |
3463 | for (i = 0; i < log->l_iclog_bufs; i++) { |
3464 | if (ptr >= log->l_iclog_bak[i] && | |
3465 | ptr <= log->l_iclog_bak[i] + log->l_iclog_size) | |
1da177e4 LT |
3466 | good_ptr++; |
3467 | } | |
e6b1f273 CH |
3468 | |
3469 | if (!good_ptr) | |
1da177e4 | 3470 | xlog_panic("xlog_verify_dest_ptr: invalid ptr"); |
e6b1f273 | 3471 | } |
1da177e4 LT |
3472 | |
3473 | STATIC void | |
3474 | xlog_verify_grant_head(xlog_t *log, int equals) | |
3475 | { | |
3476 | if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) { | |
3477 | if (equals) | |
3478 | ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes); | |
3479 | else | |
3480 | ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes); | |
3481 | } else { | |
3482 | ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle); | |
3483 | ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes); | |
3484 | } | |
3485 | } /* xlog_verify_grant_head */ | |
3486 | ||
3487 | /* check if it will fit */ | |
3488 | STATIC void | |
3489 | xlog_verify_tail_lsn(xlog_t *log, | |
3490 | xlog_in_core_t *iclog, | |
3491 | xfs_lsn_t tail_lsn) | |
3492 | { | |
3493 | int blocks; | |
3494 | ||
3495 | if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) { | |
3496 | blocks = | |
3497 | log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); | |
3498 | if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) | |
3499 | xlog_panic("xlog_verify_tail_lsn: ran out of log space"); | |
3500 | } else { | |
3501 | ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); | |
3502 | ||
3503 | if (BLOCK_LSN(tail_lsn) == log->l_prev_block) | |
3504 | xlog_panic("xlog_verify_tail_lsn: tail wrapped"); | |
3505 | ||
3506 | blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; | |
3507 | if (blocks < BTOBB(iclog->ic_offset) + 1) | |
3508 | xlog_panic("xlog_verify_tail_lsn: ran out of log space"); | |
3509 | } | |
3510 | } /* xlog_verify_tail_lsn */ | |
3511 | ||
3512 | /* | |
3513 | * Perform a number of checks on the iclog before writing to disk. | |
3514 | * | |
3515 | * 1. Make sure the iclogs are still circular | |
3516 | * 2. Make sure we have a good magic number | |
3517 | * 3. Make sure we don't have magic numbers in the data | |
3518 | * 4. Check fields of each log operation header for: | |
3519 | * A. Valid client identifier | |
3520 | * B. tid ptr value falls in valid ptr space (user space code) | |
3521 | * C. Length in log record header is correct according to the | |
3522 | * individual operation headers within record. | |
3523 | * 5. When a bwrite will occur within 5 blocks of the front of the physical | |
3524 | * log, check the preceding blocks of the physical log to make sure all | |
3525 | * the cycle numbers agree with the current cycle number. | |
3526 | */ | |
3527 | STATIC void | |
3528 | xlog_verify_iclog(xlog_t *log, | |
3529 | xlog_in_core_t *iclog, | |
3530 | int count, | |
3531 | boolean_t syncing) | |
3532 | { | |
3533 | xlog_op_header_t *ophead; | |
3534 | xlog_in_core_t *icptr; | |
3535 | xlog_in_core_2_t *xhdr; | |
3536 | xfs_caddr_t ptr; | |
3537 | xfs_caddr_t base_ptr; | |
3538 | __psint_t field_offset; | |
3539 | __uint8_t clientid; | |
3540 | int len, i, j, k, op_len; | |
3541 | int idx; | |
1da177e4 LT |
3542 | |
3543 | /* check validity of iclog pointers */ | |
b22cd72c | 3544 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3545 | icptr = log->l_iclog; |
3546 | for (i=0; i < log->l_iclog_bufs; i++) { | |
4b80916b | 3547 | if (icptr == NULL) |
1da177e4 LT |
3548 | xlog_panic("xlog_verify_iclog: invalid ptr"); |
3549 | icptr = icptr->ic_next; | |
3550 | } | |
3551 | if (icptr != log->l_iclog) | |
3552 | xlog_panic("xlog_verify_iclog: corrupt iclog ring"); | |
b22cd72c | 3553 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3554 | |
3555 | /* check log magic numbers */ | |
b53e675d | 3556 | if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM) |
1da177e4 LT |
3557 | xlog_panic("xlog_verify_iclog: invalid magic num"); |
3558 | ||
b53e675d CH |
3559 | ptr = (xfs_caddr_t) &iclog->ic_header; |
3560 | for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count; | |
1da177e4 | 3561 | ptr += BBSIZE) { |
b53e675d | 3562 | if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM) |
1da177e4 LT |
3563 | xlog_panic("xlog_verify_iclog: unexpected magic num"); |
3564 | } | |
3565 | ||
3566 | /* check fields */ | |
b53e675d | 3567 | len = be32_to_cpu(iclog->ic_header.h_num_logops); |
1da177e4 LT |
3568 | ptr = iclog->ic_datap; |
3569 | base_ptr = ptr; | |
3570 | ophead = (xlog_op_header_t *)ptr; | |
b28708d6 | 3571 | xhdr = iclog->ic_data; |
1da177e4 LT |
3572 | for (i = 0; i < len; i++) { |
3573 | ophead = (xlog_op_header_t *)ptr; | |
3574 | ||
3575 | /* clientid is only 1 byte */ | |
3576 | field_offset = (__psint_t) | |
3577 | ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr); | |
3578 | if (syncing == B_FALSE || (field_offset & 0x1ff)) { | |
3579 | clientid = ophead->oh_clientid; | |
3580 | } else { | |
3581 | idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap); | |
3582 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3583 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3584 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
03bea6fe CH |
3585 | clientid = xlog_get_client_id( |
3586 | xhdr[j].hic_xheader.xh_cycle_data[k]); | |
1da177e4 | 3587 | } else { |
03bea6fe CH |
3588 | clientid = xlog_get_client_id( |
3589 | iclog->ic_header.h_cycle_data[idx]); | |
1da177e4 LT |
3590 | } |
3591 | } | |
3592 | if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) | |
da1650a5 CH |
3593 | cmn_err(CE_WARN, "xlog_verify_iclog: " |
3594 | "invalid clientid %d op 0x%p offset 0x%lx", | |
3595 | clientid, ophead, (unsigned long)field_offset); | |
1da177e4 LT |
3596 | |
3597 | /* check length */ | |
3598 | field_offset = (__psint_t) | |
3599 | ((xfs_caddr_t)&(ophead->oh_len) - base_ptr); | |
3600 | if (syncing == B_FALSE || (field_offset & 0x1ff)) { | |
67fcb7bf | 3601 | op_len = be32_to_cpu(ophead->oh_len); |
1da177e4 LT |
3602 | } else { |
3603 | idx = BTOBBT((__psint_t)&ophead->oh_len - | |
3604 | (__psint_t)iclog->ic_datap); | |
3605 | if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) { | |
3606 | j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
3607 | k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); | |
b53e675d | 3608 | op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]); |
1da177e4 | 3609 | } else { |
b53e675d | 3610 | op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]); |
1da177e4 LT |
3611 | } |
3612 | } | |
3613 | ptr += sizeof(xlog_op_header_t) + op_len; | |
3614 | } | |
3615 | } /* xlog_verify_iclog */ | |
cfcbbbd0 | 3616 | #endif |
1da177e4 LT |
3617 | |
3618 | /* | |
b22cd72c | 3619 | * Mark all iclogs IOERROR. l_icloglock is held by the caller. |
1da177e4 LT |
3620 | */ |
3621 | STATIC int | |
3622 | xlog_state_ioerror( | |
3623 | xlog_t *log) | |
3624 | { | |
3625 | xlog_in_core_t *iclog, *ic; | |
3626 | ||
3627 | iclog = log->l_iclog; | |
3628 | if (! (iclog->ic_state & XLOG_STATE_IOERROR)) { | |
3629 | /* | |
3630 | * Mark all the incore logs IOERROR. | |
3631 | * From now on, no log flushes will result. | |
3632 | */ | |
3633 | ic = iclog; | |
3634 | do { | |
3635 | ic->ic_state = XLOG_STATE_IOERROR; | |
3636 | ic = ic->ic_next; | |
3637 | } while (ic != iclog); | |
014c2544 | 3638 | return 0; |
1da177e4 LT |
3639 | } |
3640 | /* | |
3641 | * Return non-zero, if state transition has already happened. | |
3642 | */ | |
014c2544 | 3643 | return 1; |
1da177e4 LT |
3644 | } |
3645 | ||
3646 | /* | |
3647 | * This is called from xfs_force_shutdown, when we're forcibly | |
3648 | * shutting down the filesystem, typically because of an IO error. | |
3649 | * Our main objectives here are to make sure that: | |
3650 | * a. the filesystem gets marked 'SHUTDOWN' for all interested | |
3651 | * parties to find out, 'atomically'. | |
3652 | * b. those who're sleeping on log reservations, pinned objects and | |
3653 | * other resources get woken up, and be told the bad news. | |
3654 | * c. nothing new gets queued up after (a) and (b) are done. | |
3655 | * d. if !logerror, flush the iclogs to disk, then seal them off | |
3656 | * for business. | |
3657 | */ | |
3658 | int | |
3659 | xfs_log_force_umount( | |
3660 | struct xfs_mount *mp, | |
3661 | int logerror) | |
3662 | { | |
3663 | xlog_ticket_t *tic; | |
3664 | xlog_t *log; | |
3665 | int retval; | |
1da177e4 LT |
3666 | |
3667 | log = mp->m_log; | |
3668 | ||
3669 | /* | |
3670 | * If this happens during log recovery, don't worry about | |
3671 | * locking; the log isn't open for business yet. | |
3672 | */ | |
3673 | if (!log || | |
3674 | log->l_flags & XLOG_ACTIVE_RECOVERY) { | |
3675 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; | |
bac8dca9 CH |
3676 | if (mp->m_sb_bp) |
3677 | XFS_BUF_DONE(mp->m_sb_bp); | |
014c2544 | 3678 | return 0; |
1da177e4 LT |
3679 | } |
3680 | ||
3681 | /* | |
3682 | * Somebody could've already done the hard work for us. | |
3683 | * No need to get locks for this. | |
3684 | */ | |
3685 | if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) { | |
3686 | ASSERT(XLOG_FORCED_SHUTDOWN(log)); | |
014c2544 | 3687 | return 1; |
1da177e4 LT |
3688 | } |
3689 | retval = 0; | |
3690 | /* | |
3691 | * We must hold both the GRANT lock and the LOG lock, | |
3692 | * before we mark the filesystem SHUTDOWN and wake | |
3693 | * everybody up to tell the bad news. | |
3694 | */ | |
b22cd72c | 3695 | spin_lock(&log->l_icloglock); |
6b1d1a73 | 3696 | spin_lock(&log->l_grant_lock); |
1da177e4 | 3697 | mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; |
bac8dca9 CH |
3698 | if (mp->m_sb_bp) |
3699 | XFS_BUF_DONE(mp->m_sb_bp); | |
3700 | ||
1da177e4 LT |
3701 | /* |
3702 | * This flag is sort of redundant because of the mount flag, but | |
3703 | * it's good to maintain the separation between the log and the rest | |
3704 | * of XFS. | |
3705 | */ | |
3706 | log->l_flags |= XLOG_IO_ERROR; | |
3707 | ||
3708 | /* | |
3709 | * If we hit a log error, we want to mark all the iclogs IOERROR | |
3710 | * while we're still holding the loglock. | |
3711 | */ | |
3712 | if (logerror) | |
3713 | retval = xlog_state_ioerror(log); | |
b22cd72c | 3714 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3715 | |
3716 | /* | |
3717 | * We don't want anybody waiting for log reservations | |
3718 | * after this. That means we have to wake up everybody | |
3719 | * queued up on reserve_headq as well as write_headq. | |
3720 | * In addition, we make sure in xlog_{re}grant_log_space | |
3721 | * that we don't enqueue anything once the SHUTDOWN flag | |
3722 | * is set, and this action is protected by the GRANTLOCK. | |
3723 | */ | |
3724 | if ((tic = log->l_reserve_headq)) { | |
3725 | do { | |
12017faf | 3726 | sv_signal(&tic->t_wait); |
1da177e4 LT |
3727 | tic = tic->t_next; |
3728 | } while (tic != log->l_reserve_headq); | |
3729 | } | |
3730 | ||
3731 | if ((tic = log->l_write_headq)) { | |
3732 | do { | |
12017faf | 3733 | sv_signal(&tic->t_wait); |
1da177e4 LT |
3734 | tic = tic->t_next; |
3735 | } while (tic != log->l_write_headq); | |
3736 | } | |
c8b5ea28 | 3737 | spin_unlock(&log->l_grant_lock); |
1da177e4 | 3738 | |
a14a348b | 3739 | if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { |
1da177e4 LT |
3740 | ASSERT(!logerror); |
3741 | /* | |
3742 | * Force the incore logs to disk before shutting the | |
3743 | * log down completely. | |
3744 | */ | |
a14a348b CH |
3745 | _xfs_log_force(mp, XFS_LOG_SYNC, NULL); |
3746 | ||
b22cd72c | 3747 | spin_lock(&log->l_icloglock); |
1da177e4 | 3748 | retval = xlog_state_ioerror(log); |
b22cd72c | 3749 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3750 | } |
3751 | /* | |
3752 | * Wake up everybody waiting on xfs_log_force. | |
3753 | * Callback all log item committed functions as if the | |
3754 | * log writes were completed. | |
3755 | */ | |
3756 | xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); | |
3757 | ||
3758 | #ifdef XFSERRORDEBUG | |
3759 | { | |
3760 | xlog_in_core_t *iclog; | |
3761 | ||
b22cd72c | 3762 | spin_lock(&log->l_icloglock); |
1da177e4 LT |
3763 | iclog = log->l_iclog; |
3764 | do { | |
3765 | ASSERT(iclog->ic_callback == 0); | |
3766 | iclog = iclog->ic_next; | |
3767 | } while (iclog != log->l_iclog); | |
b22cd72c | 3768 | spin_unlock(&log->l_icloglock); |
1da177e4 LT |
3769 | } |
3770 | #endif | |
3771 | /* return non-zero if log IOERROR transition had already happened */ | |
014c2544 | 3772 | return retval; |
1da177e4 LT |
3773 | } |
3774 | ||
ba0f32d4 | 3775 | STATIC int |
1da177e4 LT |
3776 | xlog_iclogs_empty(xlog_t *log) |
3777 | { | |
3778 | xlog_in_core_t *iclog; | |
3779 | ||
3780 | iclog = log->l_iclog; | |
3781 | do { | |
3782 | /* endianness does not matter here, zero is zero in | |
3783 | * any language. | |
3784 | */ | |
3785 | if (iclog->ic_header.h_num_logops) | |
014c2544 | 3786 | return 0; |
1da177e4 LT |
3787 | iclog = iclog->ic_next; |
3788 | } while (iclog != log->l_iclog); | |
014c2544 | 3789 | return 1; |
1da177e4 | 3790 | } |