blob: 052a2c0ec5fbecaa2e6d0aebffc8289bdc43f465 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Tim Shimmin87c199c2006-06-09 14:56:16 +10002 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
Nathan Scott7b718762005-11-02 14:58:39 +11003 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_sb.h"
26#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_mount.h"
28#include "xfs_error.h"
29#include "xfs_bmap_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc_btree.h"
31#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110034#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "xfs_ialloc.h"
37#include "xfs_log_priv.h"
38#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "xfs_log_recover.h"
40#include "xfs_extfree_item.h"
41#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include "xfs_quota.h"
43#include "xfs_rw.h"
Christoph Hellwig43355092008-03-27 18:01:08 +110044#include "xfs_utils.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000045#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47STATIC int xlog_find_zeroed(xlog_t *, xfs_daddr_t *);
48STATIC int xlog_clear_stale_blocks(xlog_t *, xfs_lsn_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#if defined(DEBUG)
50STATIC void xlog_recover_check_summary(xlog_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#else
52#define xlog_recover_check_summary(log)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#endif
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +000056 * This structure is used during recovery to record the buf log items which
57 * have been canceled and should not be replayed.
58 */
59struct xfs_buf_cancel {
60 xfs_daddr_t bc_blkno;
61 uint bc_len;
62 int bc_refcount;
63 struct list_head bc_list;
64};
65
66/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * Sector aligned buffer routines for buffer create/read/write/access
68 */
69
Alex Elderff30a622010-04-13 15:22:58 +100070/*
71 * Verify the given count of basic blocks is valid number of blocks
72 * to specify for an operation involving the given XFS log buffer.
73 * Returns nonzero if the count is valid, 0 otherwise.
74 */
75
76static inline int
77xlog_buf_bbcount_valid(
78 xlog_t *log,
79 int bbcount)
80{
81 return bbcount > 0 && bbcount <= log->l_logBBsize;
82}
83
Alex Elder36adecf2010-04-13 15:21:13 +100084/*
85 * Allocate a buffer to hold log data. The buffer needs to be able
86 * to map to a range of nbblks basic blocks at any valid (basic
87 * block) offset within the log.
88 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +000089STATIC xfs_buf_t *
Linus Torvalds1da177e2005-04-16 15:20:36 -070090xlog_get_bp(
91 xlog_t *log,
Dave Chinner32281492009-01-22 15:37:47 +110092 int nbblks)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093{
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +020094 struct xfs_buf *bp;
95
Alex Elderff30a622010-04-13 15:22:58 +100096 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +110097 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +100098 nbblks);
99 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100100 return NULL;
101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
Alex Elder36adecf2010-04-13 15:21:13 +1000103 /*
104 * We do log I/O in units of log sectors (a power-of-2
105 * multiple of the basic block size), so we round up the
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300106 * requested size to accommodate the basic blocks required
Alex Elder36adecf2010-04-13 15:21:13 +1000107 * for complete log sectors.
108 *
109 * In addition, the buffer may be used for a non-sector-
110 * aligned block offset, in which case an I/O of the
111 * requested size could extend beyond the end of the
112 * buffer. If the requested size is only 1 basic block it
113 * will never straddle a sector boundary, so this won't be
114 * an issue. Nor will this be a problem if the log I/O is
115 * done in basic blocks (sector size 1). But otherwise we
116 * extend the buffer by one extra log sector to ensure
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300117 * there's space to accommodate this possibility.
Alex Elder36adecf2010-04-13 15:21:13 +1000118 */
Alex Elder69ce58f2010-04-20 17:09:59 +1000119 if (nbblks > 1 && log->l_sectBBsize > 1)
120 nbblks += log->l_sectBBsize;
121 nbblks = round_up(nbblks, log->l_sectBBsize);
Alex Elder36adecf2010-04-13 15:21:13 +1000122
Christoph Hellwigc8da0fa2011-07-08 14:36:25 +0200123 bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, BBTOB(nbblks), 0);
124 if (bp)
125 xfs_buf_unlock(bp);
126 return bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127}
128
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000129STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130xlog_put_bp(
131 xfs_buf_t *bp)
132{
133 xfs_buf_free(bp);
134}
135
Alex Elder48389ef2010-04-20 17:10:21 +1000136/*
137 * Return the address of the start of the given block number's data
138 * in a log buffer. The buffer covers a log sector-aligned region.
139 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100140STATIC xfs_caddr_t
141xlog_align(
142 xlog_t *log,
143 xfs_daddr_t blk_no,
144 int nbblks,
145 xfs_buf_t *bp)
146{
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000147 xfs_daddr_t offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100148
Christoph Hellwigfdc07f42010-05-10 17:28:14 +0000149 ASSERT(BBTOB(offset + nbblks) <= XFS_BUF_SIZE(bp));
150 return XFS_BUF_PTR(bp) + BBTOB(offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100151}
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154/*
155 * nbblks should be uint, but oh well. Just want to catch that 32-bit length.
156 */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100157STATIC int
158xlog_bread_noalign(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 xlog_t *log,
160 xfs_daddr_t blk_no,
161 int nbblks,
162 xfs_buf_t *bp)
163{
164 int error;
165
Alex Elderff30a622010-04-13 15:22:58 +1000166 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100167 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000168 nbblks);
169 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100170 return EFSCORRUPTED;
171 }
172
Alex Elder69ce58f2010-04-20 17:09:59 +1000173 blk_no = round_down(blk_no, log->l_sectBBsize);
174 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 ASSERT(nbblks > 0);
177 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
180 XFS_BUF_READ(bp);
181 XFS_BUF_BUSY(bp);
182 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
183 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
184
185 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000186 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000187 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 xfs_ioerror_alert("xlog_bread", log->l_mp,
189 bp, XFS_BUF_ADDR(bp));
190 return error;
191}
192
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100193STATIC int
194xlog_bread(
195 xlog_t *log,
196 xfs_daddr_t blk_no,
197 int nbblks,
198 xfs_buf_t *bp,
199 xfs_caddr_t *offset)
200{
201 int error;
202
203 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
204 if (error)
205 return error;
206
207 *offset = xlog_align(log, blk_no, nbblks, bp);
208 return 0;
209}
210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211/*
Dave Chinner44396472011-04-21 09:34:27 +0000212 * Read at an offset into the buffer. Returns with the buffer in it's original
213 * state regardless of the result of the read.
214 */
215STATIC int
216xlog_bread_offset(
217 xlog_t *log,
218 xfs_daddr_t blk_no, /* block to read from */
219 int nbblks, /* blocks to read */
220 xfs_buf_t *bp,
221 xfs_caddr_t offset)
222{
223 xfs_caddr_t orig_offset = XFS_BUF_PTR(bp);
224 int orig_len = bp->b_buffer_length;
225 int error, error2;
226
227 error = XFS_BUF_SET_PTR(bp, offset, BBTOB(nbblks));
228 if (error)
229 return error;
230
231 error = xlog_bread_noalign(log, blk_no, nbblks, bp);
232
233 /* must reset buffer pointer even on error */
234 error2 = XFS_BUF_SET_PTR(bp, orig_offset, orig_len);
235 if (error)
236 return error;
237 return error2;
238}
239
240/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 * Write out the buffer at the given block for the given number of blocks.
242 * The buffer is kept locked across the write and is returned locked.
243 * This can only be used for synchronous log writes.
244 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000245STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246xlog_bwrite(
247 xlog_t *log,
248 xfs_daddr_t blk_no,
249 int nbblks,
250 xfs_buf_t *bp)
251{
252 int error;
253
Alex Elderff30a622010-04-13 15:22:58 +1000254 if (!xlog_buf_bbcount_valid(log, nbblks)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100255 xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
Alex Elderff30a622010-04-13 15:22:58 +1000256 nbblks);
257 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
Dave Chinner32281492009-01-22 15:37:47 +1100258 return EFSCORRUPTED;
259 }
260
Alex Elder69ce58f2010-04-20 17:09:59 +1000261 blk_no = round_down(blk_no, log->l_sectBBsize);
262 nbblks = round_up(nbblks, log->l_sectBBsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
264 ASSERT(nbblks > 0);
265 ASSERT(BBTOB(nbblks) <= XFS_BUF_SIZE(bp));
266
267 XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
268 XFS_BUF_ZEROFLAGS(bp);
269 XFS_BUF_BUSY(bp);
270 XFS_BUF_HOLD(bp);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200271 xfs_buf_lock(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 XFS_BUF_SET_COUNT(bp, BBTOB(nbblks));
273 XFS_BUF_SET_TARGET(bp, log->l_mp->m_logdev_targp);
274
275 if ((error = xfs_bwrite(log->l_mp, bp)))
276 xfs_ioerror_alert("xlog_bwrite", log->l_mp,
277 bp, XFS_BUF_ADDR(bp));
278 return error;
279}
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281#ifdef DEBUG
282/*
283 * dump debug superblock and log record information
284 */
285STATIC void
286xlog_header_check_dump(
287 xfs_mount_t *mp,
288 xlog_rec_header_t *head)
289{
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100290 xfs_debug(mp, "%s: SB : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800291 __func__, &mp->m_sb.sb_uuid, XLOG_FMT);
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100292 xfs_debug(mp, " log : uuid = %pU, fmt = %d\n",
Joe Perches03daa572009-12-14 18:01:10 -0800293 &head->h_fs_uuid, be32_to_cpu(head->h_fmt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295#else
296#define xlog_header_check_dump(mp, head)
297#endif
298
299/*
300 * check log record header for recovery
301 */
302STATIC int
303xlog_header_check_recover(
304 xfs_mount_t *mp,
305 xlog_rec_header_t *head)
306{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200307 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
309 /*
310 * IRIX doesn't write the h_fmt field and leaves it zeroed
311 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
312 * a dirty log created in IRIX.
313 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200314 if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100315 xfs_warn(mp,
316 "dirty log written in incompatible format - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 xlog_header_check_dump(mp, head);
318 XFS_ERROR_REPORT("xlog_header_check_recover(1)",
319 XFS_ERRLEVEL_HIGH, mp);
320 return XFS_ERROR(EFSCORRUPTED);
321 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100322 xfs_warn(mp,
323 "dirty log entry has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 xlog_header_check_dump(mp, head);
325 XFS_ERROR_REPORT("xlog_header_check_recover(2)",
326 XFS_ERRLEVEL_HIGH, mp);
327 return XFS_ERROR(EFSCORRUPTED);
328 }
329 return 0;
330}
331
332/*
333 * read the head block of the log and check the header
334 */
335STATIC int
336xlog_header_check_mount(
337 xfs_mount_t *mp,
338 xlog_rec_header_t *head)
339{
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200340 ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342 if (uuid_is_nil(&head->h_fs_uuid)) {
343 /*
344 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
345 * h_fs_uuid is nil, we assume this log was last mounted
346 * by IRIX and continue.
347 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100348 xfs_warn(mp, "nil uuid in log - IRIX style log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 } else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100350 xfs_warn(mp, "log has mismatched uuid - can't recover");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 xlog_header_check_dump(mp, head);
352 XFS_ERROR_REPORT("xlog_header_check_mount",
353 XFS_ERRLEVEL_HIGH, mp);
354 return XFS_ERROR(EFSCORRUPTED);
355 }
356 return 0;
357}
358
359STATIC void
360xlog_recover_iodone(
361 struct xfs_buf *bp)
362{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (XFS_BUF_GETERROR(bp)) {
364 /*
365 * We're not going to bother about retrying
366 * this during recovery. One strike!
367 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 xfs_ioerror_alert("xlog_recover_iodone",
Dave Chinnerebad8612010-09-22 10:47:20 +1000369 bp->b_target->bt_mount, bp,
370 XFS_BUF_ADDR(bp));
371 xfs_force_shutdown(bp->b_target->bt_mount,
372 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200374 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +0000375 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
377
378/*
379 * This routine finds (to an approximation) the first block in the physical
380 * log which contains the given cycle. It uses a binary search algorithm.
381 * Note that the algorithm can not be perfect because the disk will not
382 * necessarily be perfect.
383 */
David Chinnera8272ce2007-11-23 16:28:09 +1100384STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385xlog_find_cycle_start(
386 xlog_t *log,
387 xfs_buf_t *bp,
388 xfs_daddr_t first_blk,
389 xfs_daddr_t *last_blk,
390 uint cycle)
391{
392 xfs_caddr_t offset;
393 xfs_daddr_t mid_blk;
Alex Eldere3bb2e32010-04-15 18:17:30 +0000394 xfs_daddr_t end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 uint mid_cycle;
396 int error;
397
Alex Eldere3bb2e32010-04-15 18:17:30 +0000398 end_blk = *last_blk;
399 mid_blk = BLK_AVG(first_blk, end_blk);
400 while (mid_blk != first_blk && mid_blk != end_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100401 error = xlog_bread(log, mid_blk, 1, bp, &offset);
402 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 return error;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000404 mid_cycle = xlog_get_cycle(offset);
Alex Eldere3bb2e32010-04-15 18:17:30 +0000405 if (mid_cycle == cycle)
406 end_blk = mid_blk; /* last_half_cycle == mid_cycle */
407 else
408 first_blk = mid_blk; /* first_half_cycle == mid_cycle */
409 mid_blk = BLK_AVG(first_blk, end_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
Alex Eldere3bb2e32010-04-15 18:17:30 +0000411 ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
412 (mid_blk == end_blk && mid_blk-1 == first_blk));
413
414 *last_blk = end_blk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 return 0;
417}
418
419/*
Alex Elder3f943d82010-04-15 18:17:34 +0000420 * Check that a range of blocks does not contain stop_on_cycle_no.
421 * Fill in *new_blk with the block offset where such a block is
422 * found, or with -1 (an invalid block number) if there is no such
423 * block in the range. The scan needs to occur from front to back
424 * and the pointer into the region must be updated since a later
425 * routine will need to perform another test.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
427STATIC int
428xlog_find_verify_cycle(
429 xlog_t *log,
430 xfs_daddr_t start_blk,
431 int nbblks,
432 uint stop_on_cycle_no,
433 xfs_daddr_t *new_blk)
434{
435 xfs_daddr_t i, j;
436 uint cycle;
437 xfs_buf_t *bp;
438 xfs_daddr_t bufblks;
439 xfs_caddr_t buf = NULL;
440 int error = 0;
441
Alex Elder6881a222010-04-13 15:22:29 +1000442 /*
443 * Greedily allocate a buffer big enough to handle the full
444 * range of basic blocks we'll be examining. If that fails,
445 * try a smaller size. We need to be able to read at least
446 * a log sector, or we're out of luck.
447 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 bufblks = 1 << ffs(nbblks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 while (!(bp = xlog_get_bp(log, bufblks))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +1000451 if (bufblks < log->l_sectBBsize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 return ENOMEM;
453 }
454
455 for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
456 int bcount;
457
458 bcount = min(bufblks, (start_blk + nbblks - i));
459
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100460 error = xlog_bread(log, i, bcount, bp, &buf);
461 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 goto out;
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 for (j = 0; j < bcount; j++) {
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000465 cycle = xlog_get_cycle(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 if (cycle == stop_on_cycle_no) {
467 *new_blk = i+j;
468 goto out;
469 }
470
471 buf += BBSIZE;
472 }
473 }
474
475 *new_blk = -1;
476
477out:
478 xlog_put_bp(bp);
479 return error;
480}
481
482/*
483 * Potentially backup over partial log record write.
484 *
485 * In the typical case, last_blk is the number of the block directly after
486 * a good log record. Therefore, we subtract one to get the block number
487 * of the last block in the given buffer. extra_bblks contains the number
488 * of blocks we would have read on a previous read. This happens when the
489 * last log record is split over the end of the physical log.
490 *
491 * extra_bblks is the number of blocks potentially verified on a previous
492 * call to this routine.
493 */
494STATIC int
495xlog_find_verify_log_record(
496 xlog_t *log,
497 xfs_daddr_t start_blk,
498 xfs_daddr_t *last_blk,
499 int extra_bblks)
500{
501 xfs_daddr_t i;
502 xfs_buf_t *bp;
503 xfs_caddr_t offset = NULL;
504 xlog_rec_header_t *head = NULL;
505 int error = 0;
506 int smallmem = 0;
507 int num_blks = *last_blk - start_blk;
508 int xhdrs;
509
510 ASSERT(start_blk != 0 || *last_blk != start_blk);
511
512 if (!(bp = xlog_get_bp(log, num_blks))) {
513 if (!(bp = xlog_get_bp(log, 1)))
514 return ENOMEM;
515 smallmem = 1;
516 } else {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100517 error = xlog_bread(log, start_blk, num_blks, bp, &offset);
518 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 offset += ((num_blks - 1) << BBSHIFT);
521 }
522
523 for (i = (*last_blk) - 1; i >= 0; i--) {
524 if (i < start_blk) {
525 /* valid log record not found */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100526 xfs_warn(log->l_mp,
527 "Log inconsistent (didn't find previous header)");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 ASSERT(0);
529 error = XFS_ERROR(EIO);
530 goto out;
531 }
532
533 if (smallmem) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100534 error = xlog_bread(log, i, 1, bp, &offset);
535 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538
539 head = (xlog_rec_header_t *)offset;
540
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200541 if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 break;
543
544 if (!smallmem)
545 offset -= BBSIZE;
546 }
547
548 /*
549 * We hit the beginning of the physical log & still no header. Return
550 * to caller. If caller can handle a return of -1, then this routine
551 * will be called again for the end of the physical log.
552 */
553 if (i == -1) {
554 error = -1;
555 goto out;
556 }
557
558 /*
559 * We have the final block of the good log (the first block
560 * of the log record _before_ the head. So we check the uuid.
561 */
562 if ((error = xlog_header_check_mount(log->l_mp, head)))
563 goto out;
564
565 /*
566 * We may have found a log record header before we expected one.
567 * last_blk will be the 1st block # with a given cycle #. We may end
568 * up reading an entire log record. In this case, we don't want to
569 * reset last_blk. Only when last_blk points in the middle of a log
570 * record do we update last_blk.
571 */
Eric Sandeen62118702008-03-06 13:44:28 +1100572 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000573 uint h_size = be32_to_cpu(head->h_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
575 xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
576 if (h_size % XLOG_HEADER_CYCLE_SIZE)
577 xhdrs++;
578 } else {
579 xhdrs = 1;
580 }
581
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000582 if (*last_blk - i + extra_bblks !=
583 BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 *last_blk = i;
585
586out:
587 xlog_put_bp(bp);
588 return error;
589}
590
591/*
592 * Head is defined to be the point of the log where the next log write
593 * write could go. This means that incomplete LR writes at the end are
594 * eliminated when calculating the head. We aren't guaranteed that previous
595 * LR have complete transactions. We only know that a cycle number of
596 * current cycle number -1 won't be present in the log if we start writing
597 * from our current block number.
598 *
599 * last_blk contains the block number of the first block with a given
600 * cycle number.
601 *
602 * Return: zero if normal, non-zero if error.
603 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000604STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605xlog_find_head(
606 xlog_t *log,
607 xfs_daddr_t *return_head_blk)
608{
609 xfs_buf_t *bp;
610 xfs_caddr_t offset;
611 xfs_daddr_t new_blk, first_blk, start_blk, last_blk, head_blk;
612 int num_scan_bblks;
613 uint first_half_cycle, last_half_cycle;
614 uint stop_on_cycle;
615 int error, log_bbnum = log->l_logBBsize;
616
617 /* Is the end of the log device zeroed? */
618 if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
619 *return_head_blk = first_blk;
620
621 /* Is the whole lot zeroed? */
622 if (!first_blk) {
623 /* Linux XFS shouldn't generate totally zeroed logs -
624 * mkfs etc write a dummy unmount record to a fresh
625 * log so we can store the uuid in there
626 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100627 xfs_warn(log->l_mp, "totally zeroed log");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629
630 return 0;
631 } else if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100632 xfs_warn(log->l_mp, "empty log check failed");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return error;
634 }
635
636 first_blk = 0; /* get cycle # of 1st block */
637 bp = xlog_get_bp(log, 1);
638 if (!bp)
639 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100640
641 error = xlog_bread(log, 0, 1, bp, &offset);
642 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100644
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000645 first_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 last_blk = head_blk = log_bbnum - 1; /* get cycle # of last block */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100648 error = xlog_bread(log, last_blk, 1, bp, &offset);
649 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100651
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000652 last_half_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 ASSERT(last_half_cycle != 0);
654
655 /*
656 * If the 1st half cycle number is equal to the last half cycle number,
657 * then the entire log is stamped with the same cycle number. In this
658 * case, head_blk can't be set to zero (which makes sense). The below
659 * math doesn't work out properly with head_blk equal to zero. Instead,
660 * we set it to log_bbnum which is an invalid block number, but this
661 * value makes the math correct. If head_blk doesn't changed through
662 * all the tests below, *head_blk is set to zero at the very end rather
663 * than log_bbnum. In a sense, log_bbnum and zero are the same block
664 * in a circular file.
665 */
666 if (first_half_cycle == last_half_cycle) {
667 /*
668 * In this case we believe that the entire log should have
669 * cycle number last_half_cycle. We need to scan backwards
670 * from the end verifying that there are no holes still
671 * containing last_half_cycle - 1. If we find such a hole,
672 * then the start of that hole will be the new head. The
673 * simple case looks like
674 * x | x ... | x - 1 | x
675 * Another case that fits this picture would be
676 * x | x + 1 | x ... | x
Nathan Scottc41564b2006-03-29 08:55:14 +1000677 * In this case the head really is somewhere at the end of the
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 * log, as one of the latest writes at the beginning was
679 * incomplete.
680 * One more case is
681 * x | x + 1 | x ... | x - 1 | x
682 * This is really the combination of the above two cases, and
683 * the head has to end up at the start of the x-1 hole at the
684 * end of the log.
685 *
686 * In the 256k log case, we will read from the beginning to the
687 * end of the log and search for cycle numbers equal to x-1.
688 * We don't worry about the x+1 blocks that we encounter,
689 * because we know that they cannot be the head since the log
690 * started with x.
691 */
692 head_blk = log_bbnum;
693 stop_on_cycle = last_half_cycle - 1;
694 } else {
695 /*
696 * In this case we want to find the first block with cycle
697 * number matching last_half_cycle. We expect the log to be
698 * some variation on
Alex Elder3f943d82010-04-15 18:17:34 +0000699 * x + 1 ... | x ... | x
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 * The first block with cycle number x (last_half_cycle) will
701 * be where the new head belongs. First we do a binary search
702 * for the first occurrence of last_half_cycle. The binary
703 * search may not be totally accurate, so then we scan back
704 * from there looking for occurrences of last_half_cycle before
705 * us. If that backwards scan wraps around the beginning of
706 * the log, then we look for occurrences of last_half_cycle - 1
707 * at the end of the log. The cases we're looking for look
708 * like
Alex Elder3f943d82010-04-15 18:17:34 +0000709 * v binary search stopped here
710 * x + 1 ... | x | x + 1 | x ... | x
711 * ^ but we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 * or
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 * <---------> less than scan distance
Alex Elder3f943d82010-04-15 18:17:34 +0000714 * x + 1 ... | x ... | x - 1 | x
715 * ^ we want to locate this spot
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
717 stop_on_cycle = last_half_cycle;
718 if ((error = xlog_find_cycle_start(log, bp, first_blk,
719 &head_blk, last_half_cycle)))
720 goto bp_err;
721 }
722
723 /*
724 * Now validate the answer. Scan back some number of maximum possible
725 * blocks and make sure each one has the expected cycle number. The
726 * maximum is determined by the total possible amount of buffering
727 * in the in-core log. The following number can be made tighter if
728 * we actually look at the block size of the filesystem.
729 */
730 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
731 if (head_blk >= num_scan_bblks) {
732 /*
733 * We are guaranteed that the entire check can be performed
734 * in one buffer.
735 */
736 start_blk = head_blk - num_scan_bblks;
737 if ((error = xlog_find_verify_cycle(log,
738 start_blk, num_scan_bblks,
739 stop_on_cycle, &new_blk)))
740 goto bp_err;
741 if (new_blk != -1)
742 head_blk = new_blk;
743 } else { /* need to read 2 parts of log */
744 /*
745 * We are going to scan backwards in the log in two parts.
746 * First we scan the physical end of the log. In this part
747 * of the log, we are looking for blocks with cycle number
748 * last_half_cycle - 1.
749 * If we find one, then we know that the log starts there, as
750 * we've found a hole that didn't get written in going around
751 * the end of the physical log. The simple case for this is
752 * x + 1 ... | x ... | x - 1 | x
753 * <---------> less than scan distance
754 * If all of the blocks at the end of the log have cycle number
755 * last_half_cycle, then we check the blocks at the start of
756 * the log looking for occurrences of last_half_cycle. If we
757 * find one, then our current estimate for the location of the
758 * first occurrence of last_half_cycle is wrong and we move
759 * back to the hole we've found. This case looks like
760 * x + 1 ... | x | x + 1 | x ...
761 * ^ binary search stopped here
762 * Another case we need to handle that only occurs in 256k
763 * logs is
764 * x + 1 ... | x ... | x+1 | x ...
765 * ^ binary search stops here
766 * In a 256k log, the scan at the end of the log will see the
767 * x + 1 blocks. We need to skip past those since that is
768 * certainly not the head of the log. By searching for
769 * last_half_cycle-1 we accomplish that.
770 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 ASSERT(head_blk <= INT_MAX &&
Alex Elder3f943d82010-04-15 18:17:34 +0000772 (xfs_daddr_t) num_scan_bblks >= head_blk);
773 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if ((error = xlog_find_verify_cycle(log, start_blk,
775 num_scan_bblks - (int)head_blk,
776 (stop_on_cycle - 1), &new_blk)))
777 goto bp_err;
778 if (new_blk != -1) {
779 head_blk = new_blk;
Alex Elder9db127e2010-04-15 18:17:26 +0000780 goto validate_head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
782
783 /*
784 * Scan beginning of log now. The last part of the physical
785 * log is good. This scan needs to verify that it doesn't find
786 * the last_half_cycle.
787 */
788 start_blk = 0;
789 ASSERT(head_blk <= INT_MAX);
790 if ((error = xlog_find_verify_cycle(log,
791 start_blk, (int)head_blk,
792 stop_on_cycle, &new_blk)))
793 goto bp_err;
794 if (new_blk != -1)
795 head_blk = new_blk;
796 }
797
Alex Elder9db127e2010-04-15 18:17:26 +0000798validate_head:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 /*
800 * Now we need to make sure head_blk is not pointing to a block in
801 * the middle of a log record.
802 */
803 num_scan_bblks = XLOG_REC_SHIFT(log);
804 if (head_blk >= num_scan_bblks) {
805 start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
806
807 /* start ptr at last block ptr before head_blk */
808 if ((error = xlog_find_verify_log_record(log, start_blk,
809 &head_blk, 0)) == -1) {
810 error = XFS_ERROR(EIO);
811 goto bp_err;
812 } else if (error)
813 goto bp_err;
814 } else {
815 start_blk = 0;
816 ASSERT(head_blk <= INT_MAX);
817 if ((error = xlog_find_verify_log_record(log, start_blk,
818 &head_blk, 0)) == -1) {
819 /* We hit the beginning of the log during our search */
Alex Elder3f943d82010-04-15 18:17:34 +0000820 start_blk = log_bbnum - (num_scan_bblks - head_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 new_blk = log_bbnum;
822 ASSERT(start_blk <= INT_MAX &&
823 (xfs_daddr_t) log_bbnum-start_blk >= 0);
824 ASSERT(head_blk <= INT_MAX);
825 if ((error = xlog_find_verify_log_record(log,
826 start_blk, &new_blk,
827 (int)head_blk)) == -1) {
828 error = XFS_ERROR(EIO);
829 goto bp_err;
830 } else if (error)
831 goto bp_err;
832 if (new_blk != log_bbnum)
833 head_blk = new_blk;
834 } else if (error)
835 goto bp_err;
836 }
837
838 xlog_put_bp(bp);
839 if (head_blk == log_bbnum)
840 *return_head_blk = 0;
841 else
842 *return_head_blk = head_blk;
843 /*
844 * When returning here, we have a good block number. Bad block
845 * means that during a previous crash, we didn't have a clean break
846 * from cycle number N to cycle number N-1. In this case, we need
847 * to find the first block with cycle number N-1.
848 */
849 return 0;
850
851 bp_err:
852 xlog_put_bp(bp);
853
854 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100855 xfs_warn(log->l_mp, "failed to find log head");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 return error;
857}
858
859/*
860 * Find the sync block number or the tail of the log.
861 *
862 * This will be the block number of the last record to have its
863 * associated buffers synced to disk. Every log record header has
864 * a sync lsn embedded in it. LSNs hold block numbers, so it is easy
865 * to get a sync block number. The only concern is to figure out which
866 * log record header to believe.
867 *
868 * The following algorithm uses the log record header with the largest
869 * lsn. The entire log record does not need to be valid. We only care
870 * that the header is valid.
871 *
872 * We could speed up search by using current head_blk buffer, but it is not
873 * available.
874 */
Eric Sandeen5d77c0d2009-11-19 15:52:00 +0000875STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876xlog_find_tail(
877 xlog_t *log,
878 xfs_daddr_t *head_blk,
Eric Sandeen65be6052006-01-11 15:34:19 +1100879 xfs_daddr_t *tail_blk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880{
881 xlog_rec_header_t *rhead;
882 xlog_op_header_t *op_head;
883 xfs_caddr_t offset = NULL;
884 xfs_buf_t *bp;
885 int error, i, found;
886 xfs_daddr_t umount_data_blk;
887 xfs_daddr_t after_umount_blk;
888 xfs_lsn_t tail_lsn;
889 int hblks;
890
891 found = 0;
892
893 /*
894 * Find previous log record
895 */
896 if ((error = xlog_find_head(log, head_blk)))
897 return error;
898
899 bp = xlog_get_bp(log, 1);
900 if (!bp)
901 return ENOMEM;
902 if (*head_blk == 0) { /* special case */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100903 error = xlog_bread(log, 0, 1, bp, &offset);
904 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000905 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100906
Christoph Hellwig03bea6f2007-10-12 10:58:05 +1000907 if (xlog_get_cycle(offset) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 *tail_blk = 0;
909 /* leave all other log inited values alone */
Alex Elder9db127e2010-04-15 18:17:26 +0000910 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 }
912 }
913
914 /*
915 * Search backwards looking for log record header block
916 */
917 ASSERT(*head_blk < INT_MAX);
918 for (i = (int)(*head_blk) - 1; i >= 0; i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100919 error = xlog_bread(log, i, 1, bp, &offset);
920 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000921 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100922
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200923 if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 found = 1;
925 break;
926 }
927 }
928 /*
929 * If we haven't found the log record header block, start looking
930 * again from the end of the physical log. XXXmiken: There should be
931 * a check here to make sure we didn't search more than N blocks in
932 * the previous code.
933 */
934 if (!found) {
935 for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100936 error = xlog_bread(log, i, 1, bp, &offset);
937 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +0000938 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +0100939
Christoph Hellwig69ef9212011-07-08 14:36:05 +0200940 if (*(__be32 *)offset ==
941 cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 found = 2;
943 break;
944 }
945 }
946 }
947 if (!found) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +1100948 xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 ASSERT(0);
950 return XFS_ERROR(EIO);
951 }
952
953 /* find blk_no of tail of log */
954 rhead = (xlog_rec_header_t *)offset;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000955 *tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 /*
958 * Reset log values according to the state of the log when we
959 * crashed. In the case where head_blk == 0, we bump curr_cycle
960 * one because the next write starts a new cycle rather than
961 * continuing the cycle of the last good log record. At this
962 * point we have guaranteed that all partial log records have been
963 * accounted for. Therefore, we know that the last good log record
964 * written was complete and ended exactly on the end boundary
965 * of the physical log.
966 */
967 log->l_prev_block = i;
968 log->l_curr_block = (int)*head_blk;
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000969 log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (found == 2)
971 log->l_curr_cycle++;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100972 atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
Dave Chinner84f3c682010-12-03 22:11:29 +1100973 atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
Dave Chinnera69ed032010-12-21 12:08:20 +1100974 xlog_assign_grant_head(&log->l_grant_reserve_head, log->l_curr_cycle,
975 BBTOB(log->l_curr_block));
976 xlog_assign_grant_head(&log->l_grant_write_head, log->l_curr_cycle,
977 BBTOB(log->l_curr_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 /*
980 * Look for unmount record. If we find it, then we know there
981 * was a clean unmount. Since 'i' could be the last block in
982 * the physical log, we convert to a log block before comparing
983 * to the head_blk.
984 *
985 * Save the current tail lsn to use to pass to
986 * xlog_clear_stale_blocks() below. We won't want to clear the
987 * unmount record if there is one, so we pass the lsn of the
988 * unmount record rather than the block after it.
989 */
Eric Sandeen62118702008-03-06 13:44:28 +1100990 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000991 int h_size = be32_to_cpu(rhead->h_size);
992 int h_version = be32_to_cpu(rhead->h_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 if ((h_version & XLOG_VERSION_2) &&
995 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
996 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
997 if (h_size % XLOG_HEADER_CYCLE_SIZE)
998 hblks++;
999 } else {
1000 hblks = 1;
1001 }
1002 } else {
1003 hblks = 1;
1004 }
1005 after_umount_blk = (i + hblks + (int)
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001006 BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001007 tail_lsn = atomic64_read(&log->l_tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (*head_blk == after_umount_blk &&
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001009 be32_to_cpu(rhead->h_num_logops) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 umount_data_blk = (i + hblks) % log->l_logBBsize;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001011 error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1012 if (error)
Alex Elder9db127e2010-04-15 18:17:26 +00001013 goto done;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 op_head = (xlog_op_header_t *)offset;
1016 if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1017 /*
1018 * Set tail and last sync so that newly written
1019 * log records will point recovery to after the
1020 * current unmount record.
1021 */
Dave Chinner1c3cb9e2010-12-21 12:28:39 +11001022 xlog_assign_atomic_lsn(&log->l_tail_lsn,
1023 log->l_curr_cycle, after_umount_blk);
1024 xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1025 log->l_curr_cycle, after_umount_blk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 *tail_blk = after_umount_blk;
David Chinner92821e22007-05-24 15:26:31 +10001027
1028 /*
1029 * Note that the unmount was clean. If the unmount
1030 * was not clean, we need to know this to rebuild the
1031 * superblock counters from the perag headers if we
1032 * have a filesystem using non-persistent counters.
1033 */
1034 log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 }
1036 }
1037
1038 /*
1039 * Make sure that there are no blocks in front of the head
1040 * with the same cycle number as the head. This can happen
1041 * because we allow multiple outstanding log writes concurrently,
1042 * and the later writes might make it out before earlier ones.
1043 *
1044 * We use the lsn from before modifying it so that we'll never
1045 * overwrite the unmount record after a clean unmount.
1046 *
1047 * Do this only if we are going to recover the filesystem
1048 *
1049 * NOTE: This used to say "if (!readonly)"
1050 * However on Linux, we can & do recover a read-only filesystem.
1051 * We only skip recovery if NORECOVERY is specified on mount,
1052 * in which case we would not be here.
1053 *
1054 * But... if the -device- itself is readonly, just skip this.
1055 * We can't recover this device anyway, so it won't matter.
1056 */
Alex Elder9db127e2010-04-15 18:17:26 +00001057 if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 error = xlog_clear_stale_blocks(log, tail_lsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Alex Elder9db127e2010-04-15 18:17:26 +00001060done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 xlog_put_bp(bp);
1062
1063 if (error)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001064 xfs_warn(log->l_mp, "failed to locate log tail");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 return error;
1066}
1067
1068/*
1069 * Is the log zeroed at all?
1070 *
1071 * The last binary search should be changed to perform an X block read
1072 * once X becomes small enough. You can then search linearly through
1073 * the X blocks. This will cut down on the number of reads we need to do.
1074 *
1075 * If the log is partially zeroed, this routine will pass back the blkno
1076 * of the first block with cycle number 0. It won't have a complete LR
1077 * preceding it.
1078 *
1079 * Return:
1080 * 0 => the log is completely written to
1081 * -1 => use *blk_no as the first block of the log
1082 * >0 => error has occurred
1083 */
David Chinnera8272ce2007-11-23 16:28:09 +11001084STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085xlog_find_zeroed(
1086 xlog_t *log,
1087 xfs_daddr_t *blk_no)
1088{
1089 xfs_buf_t *bp;
1090 xfs_caddr_t offset;
1091 uint first_cycle, last_cycle;
1092 xfs_daddr_t new_blk, last_blk, start_blk;
1093 xfs_daddr_t num_scan_bblks;
1094 int error, log_bbnum = log->l_logBBsize;
1095
Nathan Scott6fdf8cc2006-06-28 10:13:52 +10001096 *blk_no = 0;
1097
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 /* check totally zeroed log */
1099 bp = xlog_get_bp(log, 1);
1100 if (!bp)
1101 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001102 error = xlog_bread(log, 0, 1, bp, &offset);
1103 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001105
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001106 first_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 if (first_cycle == 0) { /* completely zeroed log */
1108 *blk_no = 0;
1109 xlog_put_bp(bp);
1110 return -1;
1111 }
1112
1113 /* check partially zeroed log */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001114 error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1115 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 goto bp_err;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001117
Christoph Hellwig03bea6f2007-10-12 10:58:05 +10001118 last_cycle = xlog_get_cycle(offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (last_cycle != 0) { /* log completely written to */
1120 xlog_put_bp(bp);
1121 return 0;
1122 } else if (first_cycle != 1) {
1123 /*
1124 * If the cycle of the last block is zero, the cycle of
1125 * the first block must be 1. If it's not, maybe we're
1126 * not looking at a log... Bail out.
1127 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001128 xfs_warn(log->l_mp,
1129 "Log inconsistent or not a log (last==0, first!=1)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return XFS_ERROR(EINVAL);
1131 }
1132
1133 /* we have a partially zeroed log */
1134 last_blk = log_bbnum-1;
1135 if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1136 goto bp_err;
1137
1138 /*
1139 * Validate the answer. Because there is no way to guarantee that
1140 * the entire log is made up of log records which are the same size,
1141 * we scan over the defined maximum blocks. At this point, the maximum
1142 * is not chosen to mean anything special. XXXmiken
1143 */
1144 num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1145 ASSERT(num_scan_bblks <= INT_MAX);
1146
1147 if (last_blk < num_scan_bblks)
1148 num_scan_bblks = last_blk;
1149 start_blk = last_blk - num_scan_bblks;
1150
1151 /*
1152 * We search for any instances of cycle number 0 that occur before
1153 * our current estimate of the head. What we're trying to detect is
1154 * 1 ... | 0 | 1 | 0...
1155 * ^ binary search ends here
1156 */
1157 if ((error = xlog_find_verify_cycle(log, start_blk,
1158 (int)num_scan_bblks, 0, &new_blk)))
1159 goto bp_err;
1160 if (new_blk != -1)
1161 last_blk = new_blk;
1162
1163 /*
1164 * Potentially backup over partial log record write. We don't need
1165 * to search the end of the log because we know it is zero.
1166 */
1167 if ((error = xlog_find_verify_log_record(log, start_blk,
1168 &last_blk, 0)) == -1) {
1169 error = XFS_ERROR(EIO);
1170 goto bp_err;
1171 } else if (error)
1172 goto bp_err;
1173
1174 *blk_no = last_blk;
1175bp_err:
1176 xlog_put_bp(bp);
1177 if (error)
1178 return error;
1179 return -1;
1180}
1181
1182/*
1183 * These are simple subroutines used by xlog_clear_stale_blocks() below
1184 * to initialize a buffer full of empty log record headers and write
1185 * them into the log.
1186 */
1187STATIC void
1188xlog_add_record(
1189 xlog_t *log,
1190 xfs_caddr_t buf,
1191 int cycle,
1192 int block,
1193 int tail_cycle,
1194 int tail_block)
1195{
1196 xlog_rec_header_t *recp = (xlog_rec_header_t *)buf;
1197
1198 memset(buf, 0, BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001199 recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1200 recp->h_cycle = cpu_to_be32(cycle);
1201 recp->h_version = cpu_to_be32(
Eric Sandeen62118702008-03-06 13:44:28 +11001202 xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10001203 recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1204 recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1205 recp->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1207}
1208
1209STATIC int
1210xlog_write_log_records(
1211 xlog_t *log,
1212 int cycle,
1213 int start_block,
1214 int blocks,
1215 int tail_cycle,
1216 int tail_block)
1217{
1218 xfs_caddr_t offset;
1219 xfs_buf_t *bp;
1220 int balign, ealign;
Alex Elder69ce58f2010-04-20 17:09:59 +10001221 int sectbb = log->l_sectBBsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 int end_block = start_block + blocks;
1223 int bufblks;
1224 int error = 0;
1225 int i, j = 0;
1226
Alex Elder6881a222010-04-13 15:22:29 +10001227 /*
1228 * Greedily allocate a buffer big enough to handle the full
1229 * range of basic blocks to be written. If that fails, try
1230 * a smaller size. We need to be able to write at least a
1231 * log sector, or we're out of luck.
1232 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 bufblks = 1 << ffs(blocks);
1234 while (!(bp = xlog_get_bp(log, bufblks))) {
1235 bufblks >>= 1;
Alex Elder69ce58f2010-04-20 17:09:59 +10001236 if (bufblks < sectbb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 return ENOMEM;
1238 }
1239
1240 /* We may need to do a read at the start to fill in part of
1241 * the buffer in the starting sector not covered by the first
1242 * write below.
1243 */
Alex Elder5c17f532010-04-13 15:22:48 +10001244 balign = round_down(start_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (balign != start_block) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001246 error = xlog_bread_noalign(log, start_block, 1, bp);
1247 if (error)
1248 goto out_put_bp;
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 j = start_block - balign;
1251 }
1252
1253 for (i = start_block; i < end_block; i += bufblks) {
1254 int bcount, endcount;
1255
1256 bcount = min(bufblks, end_block - start_block);
1257 endcount = bcount - j;
1258
1259 /* We may need to do a read at the end to fill in part of
1260 * the buffer in the final sector not covered by the write.
1261 * If this is the same sector as the above read, skip it.
1262 */
Alex Elder5c17f532010-04-13 15:22:48 +10001263 ealign = round_down(end_block, sectbb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (j == 0 && (start_block + endcount > ealign)) {
Dave Chinner44396472011-04-21 09:34:27 +00001265 offset = XFS_BUF_PTR(bp) + BBTOB(ealign - start_block);
1266 error = xlog_bread_offset(log, ealign, sectbb,
1267 bp, offset);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001268 if (error)
1269 break;
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 }
1272
1273 offset = xlog_align(log, start_block, endcount, bp);
1274 for (; j < endcount; j++) {
1275 xlog_add_record(log, offset, cycle, i+j,
1276 tail_cycle, tail_block);
1277 offset += BBSIZE;
1278 }
1279 error = xlog_bwrite(log, start_block, endcount, bp);
1280 if (error)
1281 break;
1282 start_block += endcount;
1283 j = 0;
1284 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01001285
1286 out_put_bp:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 xlog_put_bp(bp);
1288 return error;
1289}
1290
1291/*
1292 * This routine is called to blow away any incomplete log writes out
1293 * in front of the log head. We do this so that we won't become confused
1294 * if we come up, write only a little bit more, and then crash again.
1295 * If we leave the partial log records out there, this situation could
1296 * cause us to think those partial writes are valid blocks since they
1297 * have the current cycle number. We get rid of them by overwriting them
1298 * with empty log records with the old cycle number rather than the
1299 * current one.
1300 *
1301 * The tail lsn is passed in rather than taken from
1302 * the log so that we will not write over the unmount record after a
1303 * clean unmount in a 512 block log. Doing so would leave the log without
1304 * any valid log records in it until a new one was written. If we crashed
1305 * during that time we would not be able to recover.
1306 */
1307STATIC int
1308xlog_clear_stale_blocks(
1309 xlog_t *log,
1310 xfs_lsn_t tail_lsn)
1311{
1312 int tail_cycle, head_cycle;
1313 int tail_block, head_block;
1314 int tail_distance, max_distance;
1315 int distance;
1316 int error;
1317
1318 tail_cycle = CYCLE_LSN(tail_lsn);
1319 tail_block = BLOCK_LSN(tail_lsn);
1320 head_cycle = log->l_curr_cycle;
1321 head_block = log->l_curr_block;
1322
1323 /*
1324 * Figure out the distance between the new head of the log
1325 * and the tail. We want to write over any blocks beyond the
1326 * head that we may have written just before the crash, but
1327 * we don't want to overwrite the tail of the log.
1328 */
1329 if (head_cycle == tail_cycle) {
1330 /*
1331 * The tail is behind the head in the physical log,
1332 * so the distance from the head to the tail is the
1333 * distance from the head to the end of the log plus
1334 * the distance from the beginning of the log to the
1335 * tail.
1336 */
1337 if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1338 XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1339 XFS_ERRLEVEL_LOW, log->l_mp);
1340 return XFS_ERROR(EFSCORRUPTED);
1341 }
1342 tail_distance = tail_block + (log->l_logBBsize - head_block);
1343 } else {
1344 /*
1345 * The head is behind the tail in the physical log,
1346 * so the distance from the head to the tail is just
1347 * the tail block minus the head block.
1348 */
1349 if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1350 XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1351 XFS_ERRLEVEL_LOW, log->l_mp);
1352 return XFS_ERROR(EFSCORRUPTED);
1353 }
1354 tail_distance = tail_block - head_block;
1355 }
1356
1357 /*
1358 * If the head is right up against the tail, we can't clear
1359 * anything.
1360 */
1361 if (tail_distance <= 0) {
1362 ASSERT(tail_distance == 0);
1363 return 0;
1364 }
1365
1366 max_distance = XLOG_TOTAL_REC_SHIFT(log);
1367 /*
1368 * Take the smaller of the maximum amount of outstanding I/O
1369 * we could have and the distance to the tail to clear out.
1370 * We take the smaller so that we don't overwrite the tail and
1371 * we don't waste all day writing from the head to the tail
1372 * for no reason.
1373 */
1374 max_distance = MIN(max_distance, tail_distance);
1375
1376 if ((head_block + max_distance) <= log->l_logBBsize) {
1377 /*
1378 * We can stomp all the blocks we need to without
1379 * wrapping around the end of the log. Just do it
1380 * in a single write. Use the cycle number of the
1381 * current cycle minus one so that the log will look like:
1382 * n ... | n - 1 ...
1383 */
1384 error = xlog_write_log_records(log, (head_cycle - 1),
1385 head_block, max_distance, tail_cycle,
1386 tail_block);
1387 if (error)
1388 return error;
1389 } else {
1390 /*
1391 * We need to wrap around the end of the physical log in
1392 * order to clear all the blocks. Do it in two separate
1393 * I/Os. The first write should be from the head to the
1394 * end of the physical log, and it should use the current
1395 * cycle number minus one just like above.
1396 */
1397 distance = log->l_logBBsize - head_block;
1398 error = xlog_write_log_records(log, (head_cycle - 1),
1399 head_block, distance, tail_cycle,
1400 tail_block);
1401
1402 if (error)
1403 return error;
1404
1405 /*
1406 * Now write the blocks at the start of the physical log.
1407 * This writes the remainder of the blocks we want to clear.
1408 * It uses the current cycle number since we're now on the
1409 * same cycle as the head so that we get:
1410 * n ... n ... | n - 1 ...
1411 * ^^^^^ blocks we're writing
1412 */
1413 distance = max_distance - (log->l_logBBsize - head_block);
1414 error = xlog_write_log_records(log, head_cycle, 0, distance,
1415 tail_cycle, tail_block);
1416 if (error)
1417 return error;
1418 }
1419
1420 return 0;
1421}
1422
1423/******************************************************************************
1424 *
1425 * Log recover routines
1426 *
1427 ******************************************************************************
1428 */
1429
1430STATIC xlog_recover_t *
1431xlog_recover_find_tid(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001432 struct hlist_head *head,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 xlog_tid_t tid)
1434{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001435 xlog_recover_t *trans;
1436 struct hlist_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Dave Chinnerf0a76952010-01-11 11:49:57 +00001438 hlist_for_each_entry(trans, n, head, r_list) {
1439 if (trans->r_log_tid == tid)
1440 return trans;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001442 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
1445STATIC void
Dave Chinnerf0a76952010-01-11 11:49:57 +00001446xlog_recover_new_tid(
1447 struct hlist_head *head,
1448 xlog_tid_t tid,
1449 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001451 xlog_recover_t *trans;
1452
1453 trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1454 trans->r_log_tid = tid;
1455 trans->r_lsn = lsn;
1456 INIT_LIST_HEAD(&trans->r_itemq);
1457
1458 INIT_HLIST_NODE(&trans->r_list);
1459 hlist_add_head(&trans->r_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
1462STATIC void
1463xlog_recover_add_item(
Dave Chinnerf0a76952010-01-11 11:49:57 +00001464 struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
1466 xlog_recover_item_t *item;
1467
1468 item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001469 INIT_LIST_HEAD(&item->ri_list);
1470 list_add_tail(&item->ri_list, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
1472
1473STATIC int
1474xlog_recover_add_to_cont_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001475 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 xlog_recover_t *trans,
1477 xfs_caddr_t dp,
1478 int len)
1479{
1480 xlog_recover_item_t *item;
1481 xfs_caddr_t ptr, old_ptr;
1482 int old_len;
1483
Dave Chinnerf0a76952010-01-11 11:49:57 +00001484 if (list_empty(&trans->r_itemq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 /* finish copying rest of trans header */
1486 xlog_recover_add_item(&trans->r_itemq);
1487 ptr = (xfs_caddr_t) &trans->r_theader +
1488 sizeof(xfs_trans_header_t) - len;
1489 memcpy(ptr, dp, len); /* d, s, l */
1490 return 0;
1491 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001492 /* take the tail entry */
1493 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1496 old_len = item->ri_buf[item->ri_cnt-1].i_len;
1497
Christoph Hellwig760dea62005-09-02 16:56:02 +10001498 ptr = kmem_realloc(old_ptr, len+old_len, old_len, 0u);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 memcpy(&ptr[old_len], dp, len); /* d, s, l */
1500 item->ri_buf[item->ri_cnt-1].i_len += len;
1501 item->ri_buf[item->ri_cnt-1].i_addr = ptr;
Dave Chinner9abbc532010-04-13 15:06:46 +10001502 trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 return 0;
1504}
1505
1506/*
1507 * The next region to add is the start of a new region. It could be
1508 * a whole region or it could be the first part of a new region. Because
1509 * of this, the assumption here is that the type and size fields of all
1510 * format structures fit into the first 32 bits of the structure.
1511 *
1512 * This works because all regions must be 32 bit aligned. Therefore, we
1513 * either have both fields or we have neither field. In the case we have
1514 * neither field, the data part of the region is zero length. We only have
1515 * a log_op_header and can throw away the header since a new one will appear
1516 * later. If we have at least 4 bytes, then we can determine how many regions
1517 * will appear in the current log item.
1518 */
1519STATIC int
1520xlog_recover_add_to_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001521 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 xlog_recover_t *trans,
1523 xfs_caddr_t dp,
1524 int len)
1525{
1526 xfs_inode_log_format_t *in_f; /* any will do */
1527 xlog_recover_item_t *item;
1528 xfs_caddr_t ptr;
1529
1530 if (!len)
1531 return 0;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001532 if (list_empty(&trans->r_itemq)) {
David Chinner5a792c42008-10-30 17:40:09 +11001533 /* we need to catch log corruptions here */
1534 if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001535 xfs_warn(log->l_mp, "%s: bad header magic number",
1536 __func__);
David Chinner5a792c42008-10-30 17:40:09 +11001537 ASSERT(0);
1538 return XFS_ERROR(EIO);
1539 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 if (len == sizeof(xfs_trans_header_t))
1541 xlog_recover_add_item(&trans->r_itemq);
1542 memcpy(&trans->r_theader, dp, len); /* d, s, l */
1543 return 0;
1544 }
1545
1546 ptr = kmem_alloc(len, KM_SLEEP);
1547 memcpy(ptr, dp, len);
1548 in_f = (xfs_inode_log_format_t *)ptr;
1549
Dave Chinnerf0a76952010-01-11 11:49:57 +00001550 /* take the tail entry */
1551 item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1552 if (item->ri_total != 0 &&
1553 item->ri_total == item->ri_cnt) {
1554 /* tail item is in use, get a new one */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 xlog_recover_add_item(&trans->r_itemq);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001556 item = list_entry(trans->r_itemq.prev,
1557 xlog_recover_item_t, ri_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559
1560 if (item->ri_total == 0) { /* first region to be added */
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001561 if (in_f->ilf_size == 0 ||
1562 in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001563 xfs_warn(log->l_mp,
1564 "bad number of regions (%d) in inode log format",
Christoph Hellwige8fa6b42009-03-03 14:48:36 -05001565 in_f->ilf_size);
1566 ASSERT(0);
1567 return XFS_ERROR(EIO);
1568 }
1569
1570 item->ri_total = in_f->ilf_size;
1571 item->ri_buf =
1572 kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1573 KM_SLEEP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 }
1575 ASSERT(item->ri_total > item->ri_cnt);
1576 /* Description region is ri_buf[0] */
1577 item->ri_buf[item->ri_cnt].i_addr = ptr;
1578 item->ri_buf[item->ri_cnt].i_len = len;
1579 item->ri_cnt++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001580 trace_xfs_log_recover_item_add(log, trans, item, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 return 0;
1582}
1583
Dave Chinnerf0a76952010-01-11 11:49:57 +00001584/*
1585 * Sort the log items in the transaction. Cancelled buffers need
1586 * to be put first so they are processed before any items that might
1587 * modify the buffers. If they are cancelled, then the modifications
1588 * don't need to be replayed.
1589 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590STATIC int
1591xlog_recover_reorder_trans(
Dave Chinner9abbc532010-04-13 15:06:46 +10001592 struct log *log,
1593 xlog_recover_t *trans,
1594 int pass)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595{
Dave Chinnerf0a76952010-01-11 11:49:57 +00001596 xlog_recover_item_t *item, *n;
1597 LIST_HEAD(sort_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Dave Chinnerf0a76952010-01-11 11:49:57 +00001599 list_splice_init(&trans->r_itemq, &sort_list);
1600 list_for_each_entry_safe(item, n, &sort_list, ri_list) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001601 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Dave Chinnerf0a76952010-01-11 11:49:57 +00001602
1603 switch (ITEM_TYPE(item)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 case XFS_LI_BUF:
Dave Chinnerc1155412010-05-07 11:05:19 +10001605 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001606 trace_xfs_log_recover_item_reorder_head(log,
1607 trans, item, pass);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001608 list_move(&item->ri_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 break;
1610 }
1611 case XFS_LI_INODE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 case XFS_LI_DQUOT:
1613 case XFS_LI_QUOTAOFF:
1614 case XFS_LI_EFD:
1615 case XFS_LI_EFI:
Dave Chinner9abbc532010-04-13 15:06:46 +10001616 trace_xfs_log_recover_item_reorder_tail(log,
1617 trans, item, pass);
Dave Chinnerf0a76952010-01-11 11:49:57 +00001618 list_move_tail(&item->ri_list, &trans->r_itemq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 break;
1620 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001621 xfs_warn(log->l_mp,
1622 "%s: unrecognized type of log operation",
1623 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 ASSERT(0);
1625 return XFS_ERROR(EIO);
1626 }
Dave Chinnerf0a76952010-01-11 11:49:57 +00001627 }
1628 ASSERT(list_empty(&sort_list));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 return 0;
1630}
1631
1632/*
1633 * Build up the table of buf cancel records so that we don't replay
1634 * cancelled data in the second pass. For buffer records that are
1635 * not cancel records, there is nothing to do here so we just return.
1636 *
1637 * If we get a cancel record which is already in the table, this indicates
1638 * that the buffer was cancelled multiple times. In order to ensure
1639 * that during pass 2 we keep the record in the table until we reach its
1640 * last occurrence in the log, we keep a reference count in the cancel
1641 * record in the table to tell us how many times we expect to see this
1642 * record during the second pass.
1643 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001644STATIC int
1645xlog_recover_buffer_pass1(
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001646 struct log *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001647 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001649 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001650 struct list_head *bucket;
1651 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
1653 /*
1654 * If this isn't a cancel buffer item, then just return.
1655 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001656 if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
Dave Chinner9abbc532010-04-13 15:06:46 +10001657 trace_xfs_log_recover_buf_not_cancel(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001658 return 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10001659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001662 * Insert an xfs_buf_cancel record into the hash table of them.
1663 * If there is already an identical record, bump its reference count.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001665 bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1666 list_for_each_entry(bcp, bucket, bc_list) {
1667 if (bcp->bc_blkno == buf_f->blf_blkno &&
1668 bcp->bc_len == buf_f->blf_len) {
1669 bcp->bc_refcount++;
Dave Chinner9abbc532010-04-13 15:06:46 +10001670 trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001671 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001674
1675 bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1676 bcp->bc_blkno = buf_f->blf_blkno;
1677 bcp->bc_len = buf_f->blf_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 bcp->bc_refcount = 1;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001679 list_add_tail(&bcp->bc_list, bucket);
1680
Dave Chinner9abbc532010-04-13 15:06:46 +10001681 trace_xfs_log_recover_buf_cancel_add(log, buf_f);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00001682 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683}
1684
1685/*
1686 * Check to see whether the buffer being recovered has a corresponding
1687 * entry in the buffer cancel record table. If it does then return 1
1688 * so that it will be cancelled, otherwise return 0. If the buffer is
Dave Chinnerc1155412010-05-07 11:05:19 +10001689 * actually a buffer cancel item (XFS_BLF_CANCEL is set), then decrement
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 * the refcount on the entry in the table and remove it from the table
1691 * if this is the last reference.
1692 *
1693 * We remove the cancel record from the table when we encounter its
1694 * last occurrence in the log so that if the same buffer is re-used
1695 * again after its last cancellation we actually replay the changes
1696 * made at that point.
1697 */
1698STATIC int
1699xlog_check_buffer_cancelled(
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001700 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 xfs_daddr_t blkno,
1702 uint len,
1703 ushort flags)
1704{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001705 struct list_head *bucket;
1706 struct xfs_buf_cancel *bcp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
1708 if (log->l_buf_cancel_table == NULL) {
1709 /*
1710 * There is nothing in the table built in pass one,
1711 * so this buffer must not be cancelled.
1712 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001713 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return 0;
1715 }
1716
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001717 /*
1718 * Search for an entry in the cancel table that matches our buffer.
1719 */
1720 bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1721 list_for_each_entry(bcp, bucket, bc_list) {
1722 if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1723 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 }
1725
1726 /*
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001727 * We didn't find a corresponding entry in the table, so return 0 so
1728 * that the buffer is NOT cancelled.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 */
Dave Chinnerc1155412010-05-07 11:05:19 +10001730 ASSERT(!(flags & XFS_BLF_CANCEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 return 0;
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00001732
1733found:
1734 /*
1735 * We've go a match, so return 1 so that the recovery of this buffer
1736 * is cancelled. If this buffer is actually a buffer cancel log
1737 * item, then decrement the refcount on the one in the table and
1738 * remove it if this is the last reference.
1739 */
1740 if (flags & XFS_BLF_CANCEL) {
1741 if (--bcp->bc_refcount == 0) {
1742 list_del(&bcp->bc_list);
1743 kmem_free(bcp);
1744 }
1745 }
1746 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747}
1748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749/*
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001750 * Perform recovery for a buffer full of inodes. In these buffers, the only
1751 * data which should be recovered is that which corresponds to the
1752 * di_next_unlinked pointers in the on disk inode structures. The rest of the
1753 * data for the inodes is always logged through the inodes themselves rather
1754 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 *
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001756 * The only time when buffers full of inodes are fully recovered is when the
1757 * buffer is full of newly allocated inodes. In this case the buffer will
1758 * not be marked as an inode buffer and so will be sent to
1759 * xlog_recover_do_reg_buffer() below during recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 */
1761STATIC int
1762xlog_recover_do_inode_buffer(
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001763 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001765 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 xfs_buf_log_format_t *buf_f)
1767{
1768 int i;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001769 int item_index = 0;
1770 int bit = 0;
1771 int nbits = 0;
1772 int reg_buf_offset = 0;
1773 int reg_buf_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 int next_unlinked_offset;
1775 int inodes_per_buf;
1776 xfs_agino_t *logged_nextp;
1777 xfs_agino_t *buffer_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Dave Chinner9abbc532010-04-13 15:06:46 +10001779 trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1780
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
1782 for (i = 0; i < inodes_per_buf; i++) {
1783 next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1784 offsetof(xfs_dinode_t, di_next_unlinked);
1785
1786 while (next_unlinked_offset >=
1787 (reg_buf_offset + reg_buf_bytes)) {
1788 /*
1789 * The next di_next_unlinked field is beyond
1790 * the current logged region. Find the next
1791 * logged region that contains or is beyond
1792 * the current di_next_unlinked field.
1793 */
1794 bit += nbits;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001795 bit = xfs_next_bit(buf_f->blf_data_map,
1796 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 /*
1799 * If there are no more logged regions in the
1800 * buffer, then we're done.
1801 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001802 if (bit == -1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001805 nbits = xfs_contig_bits(buf_f->blf_data_map,
1806 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 ASSERT(nbits > 0);
Dave Chinnerc1155412010-05-07 11:05:19 +10001808 reg_buf_offset = bit << XFS_BLF_SHIFT;
1809 reg_buf_bytes = nbits << XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 item_index++;
1811 }
1812
1813 /*
1814 * If the current logged region starts after the current
1815 * di_next_unlinked field, then move on to the next
1816 * di_next_unlinked field.
1817 */
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001818 if (next_unlinked_offset < reg_buf_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
1821 ASSERT(item->ri_buf[item_index].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001822 ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));
1824
1825 /*
1826 * The current logged region contains a copy of the
1827 * current di_next_unlinked field. Extract its value
1828 * and copy it to the buffer copy.
1829 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10001830 logged_nextp = item->ri_buf[item_index].i_addr +
1831 next_unlinked_offset - reg_buf_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 if (unlikely(*logged_nextp == 0)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001833 xfs_alert(mp,
1834 "Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1835 "Trying to replay bad (0) inode di_next_unlinked field.",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 item, bp);
1837 XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1838 XFS_ERRLEVEL_LOW, mp);
1839 return XFS_ERROR(EFSCORRUPTED);
1840 }
1841
1842 buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1843 next_unlinked_offset);
Tim Shimmin87c199c2006-06-09 14:56:16 +10001844 *buffer_nextp = *logged_nextp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 }
1846
1847 return 0;
1848}
1849
1850/*
1851 * Perform a 'normal' buffer recovery. Each logged region of the
1852 * buffer should be copied over the corresponding region in the
1853 * given buffer. The bitmap in the buf log format structure indicates
1854 * where to place the logged data.
1855 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856STATIC void
1857xlog_recover_do_reg_buffer(
Dave Chinner9abbc532010-04-13 15:06:46 +10001858 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 xlog_recover_item_t *item,
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001860 struct xfs_buf *bp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 xfs_buf_log_format_t *buf_f)
1862{
1863 int i;
1864 int bit;
1865 int nbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 int error;
1867
Dave Chinner9abbc532010-04-13 15:06:46 +10001868 trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 bit = 0;
1871 i = 1; /* 0 is the buf format structure */
1872 while (1) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001873 bit = xfs_next_bit(buf_f->blf_data_map,
1874 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 if (bit == -1)
1876 break;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00001877 nbits = xfs_contig_bits(buf_f->blf_data_map,
1878 buf_f->blf_map_size, bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 ASSERT(nbits > 0);
Christoph Hellwig4b809162007-08-16 15:37:36 +10001880 ASSERT(item->ri_buf[i].i_addr != NULL);
Dave Chinnerc1155412010-05-07 11:05:19 +10001881 ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 ASSERT(XFS_BUF_COUNT(bp) >=
Dave Chinnerc1155412010-05-07 11:05:19 +10001883 ((uint)bit << XFS_BLF_SHIFT)+(nbits<<XFS_BLF_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
1885 /*
1886 * Do a sanity check if this is a dquot buffer. Just checking
1887 * the first dquot in the buffer should do. XXXThis is
1888 * probably a good thing to do for other buf types also.
1889 */
1890 error = 0;
Nathan Scottc8ad20f2005-06-21 15:38:48 +10001891 if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10001892 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001893 if (item->ri_buf[i].i_addr == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001894 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001895 "XFS: NULL dquot in %s.", __func__);
1896 goto next;
1897 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00001898 if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001899 xfs_alert(mp,
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001900 "XFS: dquot too small (%d) in %s.",
1901 item->ri_buf[i].i_len, __func__);
1902 goto next;
1903 }
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001904 error = xfs_qm_dqcheck(mp, item->ri_buf[i].i_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 -1, 0, XFS_QMOPT_DOWARN,
1906 "dquot_buf_recover");
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001907 if (error)
1908 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 }
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001910
1911 memcpy(xfs_buf_offset(bp,
Dave Chinnerc1155412010-05-07 11:05:19 +10001912 (uint)bit << XFS_BLF_SHIFT), /* dest */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001913 item->ri_buf[i].i_addr, /* source */
Dave Chinnerc1155412010-05-07 11:05:19 +10001914 nbits<<XFS_BLF_SHIFT); /* length */
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02001915 next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 i++;
1917 bit += nbits;
1918 }
1919
1920 /* Shouldn't be any more regions */
1921 ASSERT(i == item->ri_total);
1922}
1923
1924/*
1925 * Do some primitive error checking on ondisk dquot data structures.
1926 */
1927int
1928xfs_qm_dqcheck(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001929 struct xfs_mount *mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 xfs_disk_dquot_t *ddq,
1931 xfs_dqid_t id,
1932 uint type, /* used only when IO_dorepair is true */
1933 uint flags,
1934 char *str)
1935{
1936 xfs_dqblk_t *d = (xfs_dqblk_t *)ddq;
1937 int errs = 0;
1938
1939 /*
1940 * We can encounter an uninitialized dquot buffer for 2 reasons:
1941 * 1. If we crash while deleting the quotainode(s), and those blks got
1942 * used for user data. This is because we take the path of regular
1943 * file deletion; however, the size field of quotainodes is never
1944 * updated, so all the tricks that we play in itruncate_finish
1945 * don't quite matter.
1946 *
1947 * 2. We don't play the quota buffers when there's a quotaoff logitem.
1948 * But the allocation will be replayed so we'll end up with an
1949 * uninitialized quota block.
1950 *
1951 * This is all fine; things are still consistent, and we haven't lost
1952 * any quota information. Just don't complain about bad dquot blks.
1953 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02001954 if (ddq->d_magic != cpu_to_be16(XFS_DQUOT_MAGIC)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001956 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 "%s : XFS dquot ID 0x%x, magic 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001958 str, id, be16_to_cpu(ddq->d_magic), XFS_DQUOT_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 errs++;
1960 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11001961 if (ddq->d_version != XFS_DQUOT_VERSION) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001963 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 "%s : XFS dquot ID 0x%x, version 0x%x != 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001965 str, id, ddq->d_version, XFS_DQUOT_VERSION);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 errs++;
1967 }
1968
Christoph Hellwig1149d962005-11-02 15:01:12 +11001969 if (ddq->d_flags != XFS_DQ_USER &&
1970 ddq->d_flags != XFS_DQ_PROJ &&
1971 ddq->d_flags != XFS_DQ_GROUP) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001973 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 "%s : XFS dquot ID 0x%x, unknown flags 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001975 str, id, ddq->d_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 errs++;
1977 }
1978
Christoph Hellwig1149d962005-11-02 15:01:12 +11001979 if (id != -1 && id != be32_to_cpu(ddq->d_id)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001981 xfs_alert(mp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 "%s : ondisk-dquot 0x%p, ID mismatch: "
1983 "0x%x expected, found id 0x%x",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001984 str, ddq, id, be32_to_cpu(ddq->d_id));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 errs++;
1986 }
1987
1988 if (!errs && ddq->d_id) {
Christoph Hellwig1149d962005-11-02 15:01:12 +11001989 if (ddq->d_blk_softlimit &&
1990 be64_to_cpu(ddq->d_bcount) >=
1991 be64_to_cpu(ddq->d_blk_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 if (!ddq->d_btimer) {
1993 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11001994 xfs_alert(mp,
1995 "%s : Dquot ID 0x%x (0x%p) BLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11001996 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 errs++;
1998 }
1999 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002000 if (ddq->d_ino_softlimit &&
2001 be64_to_cpu(ddq->d_icount) >=
2002 be64_to_cpu(ddq->d_ino_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 if (!ddq->d_itimer) {
2004 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002005 xfs_alert(mp,
2006 "%s : Dquot ID 0x%x (0x%p) INODE TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002007 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 errs++;
2009 }
2010 }
Christoph Hellwig1149d962005-11-02 15:01:12 +11002011 if (ddq->d_rtb_softlimit &&
2012 be64_to_cpu(ddq->d_rtbcount) >=
2013 be64_to_cpu(ddq->d_rtb_softlimit)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 if (!ddq->d_rtbtimer) {
2015 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002016 xfs_alert(mp,
2017 "%s : Dquot ID 0x%x (0x%p) RTBLK TIMER NOT STARTED",
Christoph Hellwig1149d962005-11-02 15:01:12 +11002018 str, (int)be32_to_cpu(ddq->d_id), ddq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 errs++;
2020 }
2021 }
2022 }
2023
2024 if (!errs || !(flags & XFS_QMOPT_DQREPAIR))
2025 return errs;
2026
2027 if (flags & XFS_QMOPT_DOWARN)
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002028 xfs_notice(mp, "Re-initializing dquot ID 0x%x", id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 /*
2031 * Typically, a repair is only requested by quotacheck.
2032 */
2033 ASSERT(id != -1);
2034 ASSERT(flags & XFS_QMOPT_DQREPAIR);
2035 memset(d, 0, sizeof(xfs_dqblk_t));
Christoph Hellwig1149d962005-11-02 15:01:12 +11002036
2037 d->dd_diskdq.d_magic = cpu_to_be16(XFS_DQUOT_MAGIC);
2038 d->dd_diskdq.d_version = XFS_DQUOT_VERSION;
2039 d->dd_diskdq.d_flags = type;
2040 d->dd_diskdq.d_id = cpu_to_be32(id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
2042 return errs;
2043}
2044
2045/*
2046 * Perform a dquot buffer recovery.
2047 * Simple algorithm: if we have found a QUOTAOFF logitem of the same type
2048 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2049 * Else, treat it as a regular buffer and do recovery.
2050 */
2051STATIC void
2052xlog_recover_do_dquot_buffer(
2053 xfs_mount_t *mp,
2054 xlog_t *log,
2055 xlog_recover_item_t *item,
2056 xfs_buf_t *bp,
2057 xfs_buf_log_format_t *buf_f)
2058{
2059 uint type;
2060
Dave Chinner9abbc532010-04-13 15:06:46 +10002061 trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 /*
2064 * Filesystems are required to send in quota flags at mount time.
2065 */
2066 if (mp->m_qflags == 0) {
2067 return;
2068 }
2069
2070 type = 0;
Dave Chinnerc1155412010-05-07 11:05:19 +10002071 if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 type |= XFS_DQ_USER;
Dave Chinnerc1155412010-05-07 11:05:19 +10002073 if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
Nathan Scottc8ad20f2005-06-21 15:38:48 +10002074 type |= XFS_DQ_PROJ;
Dave Chinnerc1155412010-05-07 11:05:19 +10002075 if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 type |= XFS_DQ_GROUP;
2077 /*
2078 * This type of quotas was turned off, so ignore this buffer
2079 */
2080 if (log->l_quotaoffs_flag & type)
2081 return;
2082
Dave Chinner9abbc532010-04-13 15:06:46 +10002083 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
2086/*
2087 * This routine replays a modification made to a buffer at runtime.
2088 * There are actually two types of buffer, regular and inode, which
2089 * are handled differently. Inode buffers are handled differently
2090 * in that we only recover a specific set of data from them, namely
2091 * the inode di_next_unlinked fields. This is because all other inode
2092 * data is actually logged via inode records and any data we replay
2093 * here which overlaps that may be stale.
2094 *
2095 * When meta-data buffers are freed at run time we log a buffer item
Dave Chinnerc1155412010-05-07 11:05:19 +10002096 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 * of the buffer in the log should not be replayed at recovery time.
2098 * This is so that if the blocks covered by the buffer are reused for
2099 * file data before we crash we don't end up replaying old, freed
2100 * meta-data into a user's file.
2101 *
2102 * To handle the cancellation of buffer log items, we make two passes
2103 * over the log during recovery. During the first we build a table of
2104 * those buffers which have been cancelled, and during the second we
2105 * only replay those buffers which do not have corresponding cancel
2106 * records in the table. See xlog_recover_do_buffer_pass[1,2] above
2107 * for more details on the implementation of the table of cancel records.
2108 */
2109STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002110xlog_recover_buffer_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002112 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113{
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002114 xfs_buf_log_format_t *buf_f = item->ri_buf[0].i_addr;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002115 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 xfs_buf_t *bp;
2117 int error;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002118 uint buf_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002120 /*
2121 * In this pass we only want to recover all the buffers which have
2122 * not been cancelled and are not cancellation buffers themselves.
2123 */
2124 if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2125 buf_f->blf_len, buf_f->blf_flags)) {
2126 trace_xfs_log_recover_buf_cancel(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 }
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002129
Dave Chinner9abbc532010-04-13 15:06:46 +10002130 trace_xfs_log_recover_buf_recover(log, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002132 buf_flags = XBF_LOCK;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002133 if (!(buf_f->blf_flags & XFS_BLF_INODE_BUF))
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002134 buf_flags |= XBF_MAPPED;
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002135
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002136 bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2137 buf_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (XFS_BUF_ISERROR(bp)) {
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002139 xfs_ioerror_alert("xlog_recover_do..(read#1)", mp,
2140 bp, buf_f->blf_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 error = XFS_BUF_GETERROR(bp);
2142 xfs_buf_relse(bp);
2143 return error;
2144 }
2145
2146 error = 0;
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002147 if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
Christoph Hellwige2714bf2010-12-01 22:06:21 +00002149 } else if (buf_f->blf_flags &
Dave Chinnerc1155412010-05-07 11:05:19 +10002150 (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2152 } else {
Dave Chinner9abbc532010-04-13 15:06:46 +10002153 xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 }
2155 if (error)
2156 return XFS_ERROR(error);
2157
2158 /*
2159 * Perform delayed write on the buffer. Asynchronous writes will be
2160 * slower when taking into account all the buffers to be flushed.
2161 *
2162 * Also make sure that only inode buffers with good sizes stay in
2163 * the buffer cache. The kernel moves inodes in buffers of 1 block
2164 * or XFS_INODE_CLUSTER_SIZE bytes, whichever is bigger. The inode
2165 * buffers in the log can be a different size if the log was generated
2166 * by an older kernel using unclustered inode buffers or a newer kernel
2167 * running with a different inode cluster size. Regardless, if the
2168 * the inode buffer size isn't MAX(blocksize, XFS_INODE_CLUSTER_SIZE)
2169 * for *our* value of XFS_INODE_CLUSTER_SIZE, then we need to keep
2170 * the buffer out of the buffer cache so that the buffer won't
2171 * overlap with future reads of those inodes.
2172 */
2173 if (XFS_DINODE_MAGIC ==
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002174 be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 (XFS_BUF_COUNT(bp) != MAX(log->l_mp->m_sb.sb_blocksize,
2176 (__uint32_t)XFS_INODE_CLUSTER_SIZE(log->l_mp)))) {
2177 XFS_BUF_STALE(bp);
2178 error = xfs_bwrite(mp, bp);
2179 } else {
Dave Chinnerebad8612010-09-22 10:47:20 +10002180 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002181 bp->b_iodone = xlog_recover_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 xfs_bdwrite(mp, bp);
2183 }
2184
2185 return (error);
2186}
2187
2188STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002189xlog_recover_inode_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002191 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192{
2193 xfs_inode_log_format_t *in_f;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002194 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 xfs_buf_t *bp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 xfs_dinode_t *dip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 int len;
2198 xfs_caddr_t src;
2199 xfs_caddr_t dest;
2200 int error;
2201 int attr_index;
2202 uint fields;
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002203 xfs_icdinode_t *dicp;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002204 int need_free = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
Tim Shimmin6d192a92006-06-09 14:55:38 +10002206 if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002207 in_f = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002208 } else {
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002209 in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002210 need_free = 1;
2211 error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2212 if (error)
2213 goto error;
2214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216 /*
2217 * Inode buffers can be freed, look out for it,
2218 * and do not replay the inode.
2219 */
Christoph Hellwiga1941892008-11-28 14:23:40 +11002220 if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2221 in_f->ilf_len, 0)) {
Tim Shimmin6d192a92006-06-09 14:55:38 +10002222 error = 0;
Dave Chinner9abbc532010-04-13 15:06:46 +10002223 trace_xfs_log_recover_inode_cancel(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002224 goto error;
2225 }
Dave Chinner9abbc532010-04-13 15:06:46 +10002226 trace_xfs_log_recover_inode_recover(log, in_f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227
Christoph Hellwig6ad112b2009-11-24 18:02:23 +00002228 bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len,
Christoph Hellwig0cadda12010-01-19 09:56:44 +00002229 XBF_LOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 if (XFS_BUF_ISERROR(bp)) {
2231 xfs_ioerror_alert("xlog_recover_do..(read#2)", mp,
Christoph Hellwiga1941892008-11-28 14:23:40 +11002232 bp, in_f->ilf_blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 error = XFS_BUF_GETERROR(bp);
2234 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002235 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
2237 error = 0;
2238 ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
Christoph Hellwiga1941892008-11-28 14:23:40 +11002239 dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 /*
2242 * Make sure the place we're flushing out to really looks
2243 * like an inode!
2244 */
Christoph Hellwig69ef9212011-07-08 14:36:05 +02002245 if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002247 xfs_alert(mp,
2248 "%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2249 __func__, dip, bp, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002250 XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002252 error = EFSCORRUPTED;
2253 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 }
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002255 dicp = item->ri_buf[1].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2257 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002258 xfs_alert(mp,
2259 "%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2260 __func__, item, in_f->ilf_ino);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002261 XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 XFS_ERRLEVEL_LOW, mp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002263 error = EFSCORRUPTED;
2264 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 }
2266
2267 /* Skip replay when the on disk inode is newer than the log one */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002268 if (dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 /*
2270 * Deal with the wrap case, DI_MAX_FLUSH is less
2271 * than smaller numbers
2272 */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002273 if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
Christoph Hellwig347d1c02007-08-28 13:57:51 +10002274 dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 /* do nothing */
2276 } else {
2277 xfs_buf_relse(bp);
Dave Chinner9abbc532010-04-13 15:06:46 +10002278 trace_xfs_log_recover_inode_skip(log, in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002279 error = 0;
2280 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 }
2282 }
2283 /* Take the opportunity to reset the flush iteration count */
2284 dicp->di_flushiter = 0;
2285
Al Viroabbede12011-07-26 02:31:30 -04002286 if (unlikely(S_ISREG(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2288 (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002289 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 XFS_ERRLEVEL_LOW, mp, dicp);
2291 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002292 xfs_alert(mp,
2293 "%s: Bad regular inode log record, rec ptr 0x%p, "
2294 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2295 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002296 error = EFSCORRUPTED;
2297 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 }
Al Viroabbede12011-07-26 02:31:30 -04002299 } else if (unlikely(S_ISDIR(dicp->di_mode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2301 (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2302 (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002303 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 XFS_ERRLEVEL_LOW, mp, dicp);
2305 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002306 xfs_alert(mp,
2307 "%s: Bad dir inode log record, rec ptr 0x%p, "
2308 "ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2309 __func__, item, dip, bp, in_f->ilf_ino);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002310 error = EFSCORRUPTED;
2311 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 }
2313 }
2314 if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002315 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 XFS_ERRLEVEL_LOW, mp, dicp);
2317 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002318 xfs_alert(mp,
2319 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2320 "dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2321 __func__, item, dip, bp, in_f->ilf_ino,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 dicp->di_nextents + dicp->di_anextents,
2323 dicp->di_nblocks);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002324 error = EFSCORRUPTED;
2325 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 }
2327 if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002328 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 XFS_ERRLEVEL_LOW, mp, dicp);
2330 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002331 xfs_alert(mp,
2332 "%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2333 "dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002334 item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002335 error = EFSCORRUPTED;
2336 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 }
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002338 if (unlikely(item->ri_buf[1].i_len > sizeof(struct xfs_icdinode))) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002339 XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 XFS_ERRLEVEL_LOW, mp, dicp);
2341 xfs_buf_relse(bp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002342 xfs_alert(mp,
2343 "%s: Bad inode log record length %d, rec ptr 0x%p",
2344 __func__, item->ri_buf[1].i_len, item);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002345 error = EFSCORRUPTED;
2346 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 }
2348
2349 /* The core is in in-core format */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002350 xfs_dinode_to_disk(dip, item->ri_buf[1].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351
2352 /* the rest is in on-disk format */
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002353 if (item->ri_buf[1].i_len > sizeof(struct xfs_icdinode)) {
2354 memcpy((xfs_caddr_t) dip + sizeof(struct xfs_icdinode),
2355 item->ri_buf[1].i_addr + sizeof(struct xfs_icdinode),
2356 item->ri_buf[1].i_len - sizeof(struct xfs_icdinode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 }
2358
2359 fields = in_f->ilf_fields;
2360 switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2361 case XFS_ILOG_DEV:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002362 xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 break;
2364 case XFS_ILOG_UUID:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002365 memcpy(XFS_DFORK_DPTR(dip),
2366 &in_f->ilf_u.ilfu_uuid,
2367 sizeof(uuid_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 break;
2369 }
2370
2371 if (in_f->ilf_size == 2)
2372 goto write_inode_buffer;
2373 len = item->ri_buf[2].i_len;
2374 src = item->ri_buf[2].i_addr;
2375 ASSERT(in_f->ilf_size <= 4);
2376 ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2377 ASSERT(!(fields & XFS_ILOG_DFORK) ||
2378 (len == in_f->ilf_dsize));
2379
2380 switch (fields & XFS_ILOG_DFORK) {
2381 case XFS_ILOG_DDATA:
2382 case XFS_ILOG_DEXT:
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002383 memcpy(XFS_DFORK_DPTR(dip), src, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 break;
2385
2386 case XFS_ILOG_DBROOT:
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002387 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
Christoph Hellwig81591fe2008-11-28 14:23:39 +11002388 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 XFS_DFORK_DSIZE(dip, mp));
2390 break;
2391
2392 default:
2393 /*
2394 * There are no data fork flags set.
2395 */
2396 ASSERT((fields & XFS_ILOG_DFORK) == 0);
2397 break;
2398 }
2399
2400 /*
2401 * If we logged any attribute data, recover it. There may or
2402 * may not have been any other non-core data logged in this
2403 * transaction.
2404 */
2405 if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2406 if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2407 attr_index = 3;
2408 } else {
2409 attr_index = 2;
2410 }
2411 len = item->ri_buf[attr_index].i_len;
2412 src = item->ri_buf[attr_index].i_addr;
2413 ASSERT(len == in_f->ilf_asize);
2414
2415 switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2416 case XFS_ILOG_ADATA:
2417 case XFS_ILOG_AEXT:
2418 dest = XFS_DFORK_APTR(dip);
2419 ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2420 memcpy(dest, src, len);
2421 break;
2422
2423 case XFS_ILOG_ABROOT:
2424 dest = XFS_DFORK_APTR(dip);
Christoph Hellwig7cc95a82008-10-30 17:14:34 +11002425 xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2426 len, (xfs_bmdr_block_t*)dest,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 XFS_DFORK_ASIZE(dip, mp));
2428 break;
2429
2430 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002431 xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 ASSERT(0);
2433 xfs_buf_relse(bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002434 error = EIO;
2435 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 }
2437 }
2438
2439write_inode_buffer:
Dave Chinnerebad8612010-09-22 10:47:20 +10002440 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002441 bp->b_iodone = xlog_recover_iodone;
Christoph Hellwigdd0bbad2009-03-16 08:19:59 +01002442 xfs_bdwrite(mp, bp);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002443error:
2444 if (need_free)
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002445 kmem_free(in_f);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002446 return XFS_ERROR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447}
2448
2449/*
2450 * Recover QUOTAOFF records. We simply make a note of it in the xlog_t
2451 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2452 * of that type.
2453 */
2454STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002455xlog_recover_quotaoff_pass1(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002457 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002459 xfs_qoff_logformat_t *qoff_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 ASSERT(qoff_f);
2461
2462 /*
2463 * The logitem format's flag tells us if this was user quotaoff,
Nathan Scott77a7cce2006-01-11 15:35:57 +11002464 * group/project quotaoff or both.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 */
2466 if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2467 log->l_quotaoffs_flag |= XFS_DQ_USER;
Nathan Scott77a7cce2006-01-11 15:35:57 +11002468 if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2469 log->l_quotaoffs_flag |= XFS_DQ_PROJ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2471 log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2472
2473 return (0);
2474}
2475
2476/*
2477 * Recover a dquot record
2478 */
2479STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002480xlog_recover_dquot_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002482 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002484 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 xfs_buf_t *bp;
2486 struct xfs_disk_dquot *ddq, *recddq;
2487 int error;
2488 xfs_dq_logformat_t *dq_f;
2489 uint type;
2490
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491
2492 /*
2493 * Filesystems are required to send in quota flags at mount time.
2494 */
2495 if (mp->m_qflags == 0)
2496 return (0);
2497
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002498 recddq = item->ri_buf[1].i_addr;
2499 if (recddq == NULL) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002500 xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002501 return XFS_ERROR(EIO);
2502 }
Jan Rekorajski8ec6dba2009-11-16 11:57:02 +00002503 if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002504 xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
Christoph Hellwig0c5e1ce2009-06-08 15:33:21 +02002505 item->ri_buf[1].i_len, __func__);
2506 return XFS_ERROR(EIO);
2507 }
2508
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 /*
2510 * This type of quotas was turned off, so ignore this record.
2511 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002512 type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 ASSERT(type);
2514 if (log->l_quotaoffs_flag & type)
2515 return (0);
2516
2517 /*
2518 * At this point we know that quota was _not_ turned off.
2519 * Since the mount flags are not indicating to us otherwise, this
2520 * must mean that quota is on, and the dquot needs to be replayed.
2521 * Remember that we may not have fully recovered the superblock yet,
2522 * so we can't do the usual trick of looking at the SB quota bits.
2523 *
2524 * The other possibility, of course, is that the quota subsystem was
2525 * removed since the last mount - ENOSYS.
2526 */
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002527 dq_f = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 ASSERT(dq_f);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002529 error = xfs_qm_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2530 "xlog_recover_dquot_pass2 (log copy)");
2531 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 ASSERT(dq_f->qlf_len == 1);
2534
2535 error = xfs_read_buf(mp, mp->m_ddev_targp,
2536 dq_f->qlf_blkno,
2537 XFS_FSB_TO_BB(mp, dq_f->qlf_len),
2538 0, &bp);
2539 if (error) {
2540 xfs_ioerror_alert("xlog_recover_do..(read#3)", mp,
2541 bp, dq_f->qlf_blkno);
2542 return error;
2543 }
2544 ASSERT(bp);
2545 ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
2546
2547 /*
2548 * At least the magic num portion should be on disk because this
2549 * was among a chunk of dquots created earlier, and we did some
2550 * minimal initialization then.
2551 */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002552 error = xfs_qm_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
2553 "xlog_recover_dquot_pass2");
2554 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 xfs_buf_relse(bp);
2556 return XFS_ERROR(EIO);
2557 }
2558
2559 memcpy(ddq, recddq, item->ri_buf[1].i_len);
2560
2561 ASSERT(dq_f->qlf_size == 2);
Dave Chinnerebad8612010-09-22 10:47:20 +10002562 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02002563 bp->b_iodone = xlog_recover_iodone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 xfs_bdwrite(mp, bp);
2565
2566 return (0);
2567}
2568
2569/*
2570 * This routine is called to create an in-core extent free intent
2571 * item from the efi format structure which was logged on disk.
2572 * It allocates an in-core efi, copies the extents from the format
2573 * structure into it, and adds the efi to the AIL with the given
2574 * LSN.
2575 */
Tim Shimmin6d192a92006-06-09 14:55:38 +10002576STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002577xlog_recover_efi_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 xlog_t *log,
2579 xlog_recover_item_t *item,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002580 xfs_lsn_t lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581{
Tim Shimmin6d192a92006-06-09 14:55:38 +10002582 int error;
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002583 xfs_mount_t *mp = log->l_mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 xfs_efi_log_item_t *efip;
2585 xfs_efi_log_format_t *efi_formatp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002587 efi_formatp = item->ri_buf[0].i_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002590 if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
2591 &(efip->efi_format)))) {
2592 xfs_efi_item_free(efip);
2593 return error;
2594 }
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002595 atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
David Chinnera9c21c12008-10-30 17:39:35 +11002597 spin_lock(&log->l_ailp->xa_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 /*
David Chinner783a2f62008-10-30 17:39:58 +11002599 * xfs_trans_ail_update() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 */
Dave Chinnere6059942010-12-20 12:34:26 +11002601 xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
Tim Shimmin6d192a92006-06-09 14:55:38 +10002602 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603}
2604
2605
2606/*
2607 * This routine is called when an efd format structure is found in
2608 * a committed transaction in the log. It's purpose is to cancel
2609 * the corresponding efi if it was still in the log. To do this
2610 * it searches the AIL for the efi with an id equal to that in the
2611 * efd format structure. If we find it, we remove the efi from the
2612 * AIL and free it.
2613 */
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002614STATIC int
2615xlog_recover_efd_pass2(
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 xlog_t *log,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002617 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 xfs_efd_log_format_t *efd_formatp;
2620 xfs_efi_log_item_t *efip = NULL;
2621 xfs_log_item_t *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 __uint64_t efi_id;
David Chinner27d8d5f2008-10-30 17:38:39 +11002623 struct xfs_ail_cursor cur;
David Chinner783a2f62008-10-30 17:39:58 +11002624 struct xfs_ail *ailp = log->l_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +10002626 efd_formatp = item->ri_buf[0].i_addr;
Tim Shimmin6d192a92006-06-09 14:55:38 +10002627 ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
2628 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
2629 (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
2630 ((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 efi_id = efd_formatp->efd_efi_id;
2632
2633 /*
2634 * Search for the efi with the id in the efd format structure
2635 * in the AIL.
2636 */
David Chinnera9c21c12008-10-30 17:39:35 +11002637 spin_lock(&ailp->xa_lock);
2638 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 while (lip != NULL) {
2640 if (lip->li_type == XFS_LI_EFI) {
2641 efip = (xfs_efi_log_item_t *)lip;
2642 if (efip->efi_format.efi_id == efi_id) {
2643 /*
David Chinner783a2f62008-10-30 17:39:58 +11002644 * xfs_trans_ail_delete() drops the
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 * AIL lock.
2646 */
David Chinner783a2f62008-10-30 17:39:58 +11002647 xfs_trans_ail_delete(ailp, lip);
David Chinner8ae2c0f2007-11-23 16:28:17 +11002648 xfs_efi_item_free(efip);
David Chinnera9c21c12008-10-30 17:39:35 +11002649 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11002650 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 }
2652 }
David Chinnera9c21c12008-10-30 17:39:35 +11002653 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 }
David Chinnera9c21c12008-10-30 17:39:35 +11002655 xfs_trans_ail_cursor_done(ailp, &cur);
2656 spin_unlock(&ailp->xa_lock);
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002657
2658 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659}
2660
2661/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 * Free up any resources allocated by the transaction
2663 *
2664 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2665 */
2666STATIC void
2667xlog_recover_free_trans(
Christoph Hellwigd0450942010-12-01 22:06:23 +00002668 struct xlog_recover *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669{
Dave Chinnerf0a76952010-01-11 11:49:57 +00002670 xlog_recover_item_t *item, *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 int i;
2672
Dave Chinnerf0a76952010-01-11 11:49:57 +00002673 list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2674 /* Free the regions in the item. */
2675 list_del(&item->ri_list);
2676 for (i = 0; i < item->ri_cnt; i++)
2677 kmem_free(item->ri_buf[i].i_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 /* Free the item itself */
Dave Chinnerf0a76952010-01-11 11:49:57 +00002679 kmem_free(item->ri_buf);
2680 kmem_free(item);
2681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 /* Free the transaction recover structure */
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10002683 kmem_free(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684}
2685
2686STATIC int
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002687xlog_recover_commit_pass1(
Christoph Hellwigd0450942010-12-01 22:06:23 +00002688 struct log *log,
2689 struct xlog_recover *trans,
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002690 xlog_recover_item_t *item)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691{
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002692 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002693
2694 switch (ITEM_TYPE(item)) {
2695 case XFS_LI_BUF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002696 return xlog_recover_buffer_pass1(log, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002697 case XFS_LI_QUOTAOFF:
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002698 return xlog_recover_quotaoff_pass1(log, item);
2699 case XFS_LI_INODE:
2700 case XFS_LI_EFI:
2701 case XFS_LI_EFD:
2702 case XFS_LI_DQUOT:
2703 /* nothing to do in pass 1 */
2704 return 0;
Christoph Hellwigd0450942010-12-01 22:06:23 +00002705 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002706 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2707 __func__, ITEM_TYPE(item));
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002708 ASSERT(0);
2709 return XFS_ERROR(EIO);
2710 }
2711}
2712
2713STATIC int
2714xlog_recover_commit_pass2(
2715 struct log *log,
2716 struct xlog_recover *trans,
2717 xlog_recover_item_t *item)
2718{
2719 trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
2720
2721 switch (ITEM_TYPE(item)) {
2722 case XFS_LI_BUF:
2723 return xlog_recover_buffer_pass2(log, item);
2724 case XFS_LI_INODE:
2725 return xlog_recover_inode_pass2(log, item);
2726 case XFS_LI_EFI:
2727 return xlog_recover_efi_pass2(log, item, trans->r_lsn);
2728 case XFS_LI_EFD:
2729 return xlog_recover_efd_pass2(log, item);
2730 case XFS_LI_DQUOT:
2731 return xlog_recover_dquot_pass2(log, item);
2732 case XFS_LI_QUOTAOFF:
2733 /* nothing to do in pass2 */
2734 return 0;
2735 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002736 xfs_warn(log->l_mp, "%s: invalid item type (%d)",
2737 __func__, ITEM_TYPE(item));
Christoph Hellwigd0450942010-12-01 22:06:23 +00002738 ASSERT(0);
2739 return XFS_ERROR(EIO);
2740 }
2741}
2742
2743/*
2744 * Perform the transaction.
2745 *
2746 * If the transaction modifies a buffer or inode, do it now. Otherwise,
2747 * EFIs and EFDs get queued up by adding entries into the AIL for them.
2748 */
2749STATIC int
2750xlog_recover_commit_trans(
2751 struct log *log,
2752 struct xlog_recover *trans,
2753 int pass)
2754{
2755 int error = 0;
2756 xlog_recover_item_t *item;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
Dave Chinnerf0a76952010-01-11 11:49:57 +00002758 hlist_del(&trans->r_list);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002759
2760 error = xlog_recover_reorder_trans(log, trans, pass);
2761 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 return error;
Christoph Hellwigd0450942010-12-01 22:06:23 +00002763
2764 list_for_each_entry(item, &trans->r_itemq, ri_list) {
Christoph Hellwigc9f71f52010-12-01 22:06:24 +00002765 if (pass == XLOG_RECOVER_PASS1)
2766 error = xlog_recover_commit_pass1(log, trans, item);
2767 else
2768 error = xlog_recover_commit_pass2(log, trans, item);
Christoph Hellwigd0450942010-12-01 22:06:23 +00002769 if (error)
2770 return error;
2771 }
2772
2773 xlog_recover_free_trans(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 return 0;
2775}
2776
2777STATIC int
2778xlog_recover_unmount_trans(
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002779 struct log *log,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 xlog_recover_t *trans)
2781{
2782 /* Do nothing now */
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002783 xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 return 0;
2785}
2786
2787/*
2788 * There are two valid states of the r_state field. 0 indicates that the
2789 * transaction structure is in a normal state. We have either seen the
2790 * start of the transaction or the last operation we added was not a partial
2791 * operation. If the last operation we added to the transaction was a
2792 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2793 *
2794 * NOTE: skip LRs with 0 data length.
2795 */
2796STATIC int
2797xlog_recover_process_data(
2798 xlog_t *log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00002799 struct hlist_head rhash[],
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 xlog_rec_header_t *rhead,
2801 xfs_caddr_t dp,
2802 int pass)
2803{
2804 xfs_caddr_t lp;
2805 int num_logops;
2806 xlog_op_header_t *ohead;
2807 xlog_recover_t *trans;
2808 xlog_tid_t tid;
2809 int error;
2810 unsigned long hash;
2811 uint flags;
2812
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002813 lp = dp + be32_to_cpu(rhead->h_len);
2814 num_logops = be32_to_cpu(rhead->h_num_logops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816 /* check the log format matches our own - else we can't recover */
2817 if (xlog_header_check_recover(log->l_mp, rhead))
2818 return (XFS_ERROR(EIO));
2819
2820 while ((dp < lp) && num_logops) {
2821 ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
2822 ohead = (xlog_op_header_t *)dp;
2823 dp += sizeof(xlog_op_header_t);
2824 if (ohead->oh_clientid != XFS_TRANSACTION &&
2825 ohead->oh_clientid != XFS_LOG) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002826 xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2827 __func__, ohead->oh_clientid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828 ASSERT(0);
2829 return (XFS_ERROR(EIO));
2830 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002831 tid = be32_to_cpu(ohead->oh_tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 hash = XLOG_RHASH(tid);
Dave Chinnerf0a76952010-01-11 11:49:57 +00002833 trans = xlog_recover_find_tid(&rhash[hash], tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002834 if (trans == NULL) { /* not found; add new tid */
2835 if (ohead->oh_flags & XLOG_START_TRANS)
2836 xlog_recover_new_tid(&rhash[hash], tid,
Christoph Hellwigb53e6752007-10-12 10:59:34 +10002837 be64_to_cpu(rhead->h_lsn));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 } else {
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11002839 if (dp + be32_to_cpu(ohead->oh_len) > lp) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002840 xfs_warn(log->l_mp, "%s: bad length 0x%x",
2841 __func__, be32_to_cpu(ohead->oh_len));
Lachlan McIlroy9742bb92008-01-10 16:43:36 +11002842 WARN_ON(1);
2843 return (XFS_ERROR(EIO));
2844 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 flags = ohead->oh_flags & ~XLOG_END_TRANS;
2846 if (flags & XLOG_WAS_CONT_TRANS)
2847 flags &= ~XLOG_CONTINUE_TRANS;
2848 switch (flags) {
2849 case XLOG_COMMIT_TRANS:
2850 error = xlog_recover_commit_trans(log,
Dave Chinnerf0a76952010-01-11 11:49:57 +00002851 trans, pass);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 break;
2853 case XLOG_UNMOUNT_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002854 error = xlog_recover_unmount_trans(log, trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 break;
2856 case XLOG_WAS_CONT_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10002857 error = xlog_recover_add_to_cont_trans(log,
2858 trans, dp,
2859 be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 break;
2861 case XLOG_START_TRANS:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002862 xfs_warn(log->l_mp, "%s: bad transaction",
2863 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 ASSERT(0);
2865 error = XFS_ERROR(EIO);
2866 break;
2867 case 0:
2868 case XLOG_CONTINUE_TRANS:
Dave Chinner9abbc532010-04-13 15:06:46 +10002869 error = xlog_recover_add_to_trans(log, trans,
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002870 dp, be32_to_cpu(ohead->oh_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 break;
2872 default:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11002873 xfs_warn(log->l_mp, "%s: bad flag 0x%x",
2874 __func__, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 ASSERT(0);
2876 error = XFS_ERROR(EIO);
2877 break;
2878 }
2879 if (error)
2880 return error;
2881 }
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +10002882 dp += be32_to_cpu(ohead->oh_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 num_logops--;
2884 }
2885 return 0;
2886}
2887
2888/*
2889 * Process an extent free intent item that was recovered from
2890 * the log. We need to free the extents that it describes.
2891 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10002892STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893xlog_recover_process_efi(
2894 xfs_mount_t *mp,
2895 xfs_efi_log_item_t *efip)
2896{
2897 xfs_efd_log_item_t *efdp;
2898 xfs_trans_t *tp;
2899 int i;
David Chinner3c1e2bb2008-04-10 12:21:11 +10002900 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901 xfs_extent_t *extp;
2902 xfs_fsblock_t startblock_fsb;
2903
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002904 ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905
2906 /*
2907 * First check the validity of the extents described by the
2908 * EFI. If any are bad, then assume that all are bad and
2909 * just toss the EFI.
2910 */
2911 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2912 extp = &(efip->efi_format.efi_extents[i]);
2913 startblock_fsb = XFS_BB_TO_FSB(mp,
2914 XFS_FSB_TO_DADDR(mp, extp->ext_start));
2915 if ((startblock_fsb == 0) ||
2916 (extp->ext_len == 0) ||
2917 (startblock_fsb >= mp->m_sb.sb_dblocks) ||
2918 (extp->ext_len >= mp->m_sb.sb_agblocks)) {
2919 /*
2920 * This will pull the EFI from the AIL and
2921 * free the memory associated with it.
2922 */
2923 xfs_efi_release(efip, efip->efi_format.efi_nextents);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002924 return XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 }
2926 }
2927
2928 tp = xfs_trans_alloc(mp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002929 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 0, 0);
David Chinnerfc6149d2008-04-10 12:21:53 +10002930 if (error)
2931 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
2933
2934 for (i = 0; i < efip->efi_format.efi_nextents; i++) {
2935 extp = &(efip->efi_format.efi_extents[i]);
David Chinnerfc6149d2008-04-10 12:21:53 +10002936 error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
2937 if (error)
2938 goto abort_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
2940 extp->ext_len);
2941 }
2942
Dave Chinnerb199c8a2010-12-20 11:59:49 +11002943 set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
David Chinnere5720ee2008-04-10 12:21:18 +10002944 error = xfs_trans_commit(tp, 0);
David Chinner3c1e2bb2008-04-10 12:21:11 +10002945 return error;
David Chinnerfc6149d2008-04-10 12:21:53 +10002946
2947abort_error:
2948 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
2949 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950}
2951
2952/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 * When this is called, all of the EFIs which did not have
2954 * corresponding EFDs should be in the AIL. What we do now
2955 * is free the extents associated with each one.
2956 *
2957 * Since we process the EFIs in normal transactions, they
2958 * will be removed at some point after the commit. This prevents
2959 * us from just walking down the list processing each one.
2960 * We'll use a flag in the EFI to skip those that we've already
2961 * processed and use the AIL iteration mechanism's generation
2962 * count to try to speed this up at least a bit.
2963 *
2964 * When we start, we know that the EFIs are the only things in
2965 * the AIL. As we process them, however, other items are added
2966 * to the AIL. Since everything added to the AIL must come after
2967 * everything already in the AIL, we stop processing as soon as
2968 * we see something other than an EFI in the AIL.
2969 */
David Chinner3c1e2bb2008-04-10 12:21:11 +10002970STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971xlog_recover_process_efis(
2972 xlog_t *log)
2973{
2974 xfs_log_item_t *lip;
2975 xfs_efi_log_item_t *efip;
David Chinner3c1e2bb2008-04-10 12:21:11 +10002976 int error = 0;
David Chinner27d8d5f2008-10-30 17:38:39 +11002977 struct xfs_ail_cursor cur;
David Chinnera9c21c12008-10-30 17:39:35 +11002978 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979
David Chinnera9c21c12008-10-30 17:39:35 +11002980 ailp = log->l_ailp;
2981 spin_lock(&ailp->xa_lock);
2982 lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 while (lip != NULL) {
2984 /*
2985 * We're done when we see something other than an EFI.
David Chinner27d8d5f2008-10-30 17:38:39 +11002986 * There should be no EFIs left in the AIL now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 */
2988 if (lip->li_type != XFS_LI_EFI) {
David Chinner27d8d5f2008-10-30 17:38:39 +11002989#ifdef DEBUG
David Chinnera9c21c12008-10-30 17:39:35 +11002990 for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
David Chinner27d8d5f2008-10-30 17:38:39 +11002991 ASSERT(lip->li_type != XFS_LI_EFI);
2992#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 break;
2994 }
2995
2996 /*
2997 * Skip EFIs that we've already processed.
2998 */
2999 efip = (xfs_efi_log_item_t *)lip;
Dave Chinnerb199c8a2010-12-20 11:59:49 +11003000 if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
David Chinnera9c21c12008-10-30 17:39:35 +11003001 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 continue;
3003 }
3004
David Chinnera9c21c12008-10-30 17:39:35 +11003005 spin_unlock(&ailp->xa_lock);
3006 error = xlog_recover_process_efi(log->l_mp, efip);
3007 spin_lock(&ailp->xa_lock);
David Chinner27d8d5f2008-10-30 17:38:39 +11003008 if (error)
3009 goto out;
David Chinnera9c21c12008-10-30 17:39:35 +11003010 lip = xfs_trans_ail_cursor_next(ailp, &cur);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 }
David Chinner27d8d5f2008-10-30 17:38:39 +11003012out:
David Chinnera9c21c12008-10-30 17:39:35 +11003013 xfs_trans_ail_cursor_done(ailp, &cur);
3014 spin_unlock(&ailp->xa_lock);
David Chinner3c1e2bb2008-04-10 12:21:11 +10003015 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016}
3017
3018/*
3019 * This routine performs a transaction to null out a bad inode pointer
3020 * in an agi unlinked inode hash bucket.
3021 */
3022STATIC void
3023xlog_recover_clear_agi_bucket(
3024 xfs_mount_t *mp,
3025 xfs_agnumber_t agno,
3026 int bucket)
3027{
3028 xfs_trans_t *tp;
3029 xfs_agi_t *agi;
3030 xfs_buf_t *agibp;
3031 int offset;
3032 int error;
3033
3034 tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003035 error = xfs_trans_reserve(tp, 0, XFS_CLEAR_AGI_BUCKET_LOG_RES(mp),
3036 0, 0, 0);
David Chinnere5720ee2008-04-10 12:21:18 +10003037 if (error)
3038 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003040 error = xfs_read_agi(mp, tp, agno, &agibp);
3041 if (error)
David Chinnere5720ee2008-04-10 12:21:18 +10003042 goto out_abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003044 agi = XFS_BUF_TO_AGI(agibp);
Christoph Hellwig16259e72005-11-02 15:11:25 +11003045 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 offset = offsetof(xfs_agi_t, agi_unlinked) +
3047 (sizeof(xfs_agino_t) * bucket);
3048 xfs_trans_log_buf(tp, agibp, offset,
3049 (offset + sizeof(xfs_agino_t) - 1));
3050
David Chinnere5720ee2008-04-10 12:21:18 +10003051 error = xfs_trans_commit(tp, 0);
3052 if (error)
3053 goto out_error;
3054 return;
3055
3056out_abort:
3057 xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3058out_error:
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003059 xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
David Chinnere5720ee2008-04-10 12:21:18 +10003060 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061}
3062
Christoph Hellwig23fac502008-11-28 14:23:40 +11003063STATIC xfs_agino_t
3064xlog_recover_process_one_iunlink(
3065 struct xfs_mount *mp,
3066 xfs_agnumber_t agno,
3067 xfs_agino_t agino,
3068 int bucket)
3069{
3070 struct xfs_buf *ibp;
3071 struct xfs_dinode *dip;
3072 struct xfs_inode *ip;
3073 xfs_ino_t ino;
3074 int error;
3075
3076 ino = XFS_AGINO_TO_INO(mp, agno, agino);
Dave Chinner7b6259e2010-06-24 11:35:17 +10003077 error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003078 if (error)
3079 goto fail;
3080
3081 /*
3082 * Get the on disk inode to find the next inode in the bucket.
3083 */
Christoph Hellwig0cadda12010-01-19 09:56:44 +00003084 error = xfs_itobp(mp, NULL, ip, &dip, &ibp, XBF_LOCK);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003085 if (error)
Christoph Hellwig0e446672008-11-28 14:23:42 +11003086 goto fail_iput;
Christoph Hellwig23fac502008-11-28 14:23:40 +11003087
Christoph Hellwig23fac502008-11-28 14:23:40 +11003088 ASSERT(ip->i_d.di_nlink == 0);
Christoph Hellwig0e446672008-11-28 14:23:42 +11003089 ASSERT(ip->i_d.di_mode != 0);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003090
3091 /* setup for the next pass */
3092 agino = be32_to_cpu(dip->di_next_unlinked);
3093 xfs_buf_relse(ibp);
3094
3095 /*
3096 * Prevent any DMAPI event from being sent when the reference on
3097 * the inode is dropped.
3098 */
3099 ip->i_d.di_dmevmask = 0;
3100
Christoph Hellwig0e446672008-11-28 14:23:42 +11003101 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003102 return agino;
3103
Christoph Hellwig0e446672008-11-28 14:23:42 +11003104 fail_iput:
3105 IRELE(ip);
Christoph Hellwig23fac502008-11-28 14:23:40 +11003106 fail:
3107 /*
3108 * We can't read in the inode this bucket points to, or this inode
3109 * is messed up. Just ditch this bucket of inodes. We will lose
3110 * some inodes and space, but at least we won't hang.
3111 *
3112 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3113 * clear the inode pointer in the bucket.
3114 */
3115 xlog_recover_clear_agi_bucket(mp, agno, bucket);
3116 return NULLAGINO;
3117}
3118
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119/*
3120 * xlog_iunlink_recover
3121 *
3122 * This is called during recovery to process any inodes which
3123 * we unlinked but not freed when the system crashed. These
3124 * inodes will be on the lists in the AGI blocks. What we do
3125 * here is scan all the AGIs and fully truncate and free any
3126 * inodes found on the lists. Each inode is removed from the
3127 * lists when it has been fully truncated and is freed. The
3128 * freeing of the inode and its removal from the list must be
3129 * atomic.
3130 */
Eric Sandeend96f8f82009-07-02 00:09:33 -05003131STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132xlog_recover_process_iunlinks(
3133 xlog_t *log)
3134{
3135 xfs_mount_t *mp;
3136 xfs_agnumber_t agno;
3137 xfs_agi_t *agi;
3138 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003139 xfs_agino_t agino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 int bucket;
3141 int error;
3142 uint mp_dmevmask;
3143
3144 mp = log->l_mp;
3145
3146 /*
3147 * Prevent any DMAPI event from being sent while in this function.
3148 */
3149 mp_dmevmask = mp->m_dmevmask;
3150 mp->m_dmevmask = 0;
3151
3152 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3153 /*
3154 * Find the agi for this ag.
3155 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003156 error = xfs_read_agi(mp, NULL, agno, &agibp);
3157 if (error) {
3158 /*
3159 * AGI is b0rked. Don't process it.
3160 *
3161 * We should probably mark the filesystem as corrupt
3162 * after we've recovered all the ag's we can....
3163 */
3164 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 }
3166 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167
3168 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
Christoph Hellwig16259e72005-11-02 15:11:25 +11003169 agino = be32_to_cpu(agi->agi_unlinked[bucket]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 while (agino != NULLAGINO) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 /*
3172 * Release the agi buffer so that it can
3173 * be acquired in the normal course of the
3174 * transaction to truncate and free the inode.
3175 */
3176 xfs_buf_relse(agibp);
3177
Christoph Hellwig23fac502008-11-28 14:23:40 +11003178 agino = xlog_recover_process_one_iunlink(mp,
3179 agno, agino, bucket);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180
3181 /*
3182 * Reacquire the agibuffer and continue around
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003183 * the loop. This should never fail as we know
3184 * the buffer was good earlier on.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 */
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003186 error = xfs_read_agi(mp, NULL, agno, &agibp);
3187 ASSERT(error == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 }
3190 }
3191
3192 /*
3193 * Release the buffer for the current agi so we can
3194 * go on to the next one.
3195 */
3196 xfs_buf_relse(agibp);
3197 }
3198
3199 mp->m_dmevmask = mp_dmevmask;
3200}
3201
3202
3203#ifdef DEBUG
3204STATIC void
3205xlog_pack_data_checksum(
3206 xlog_t *log,
3207 xlog_in_core_t *iclog,
3208 int size)
3209{
3210 int i;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003211 __be32 *up;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 uint chksum = 0;
3213
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003214 up = (__be32 *)iclog->ic_datap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 /* divide length by 4 to get # words */
3216 for (i = 0; i < (size >> 2); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003217 chksum ^= be32_to_cpu(*up);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218 up++;
3219 }
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003220 iclog->ic_header.h_chksum = cpu_to_be32(chksum);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221}
3222#else
3223#define xlog_pack_data_checksum(log, iclog, size)
3224#endif
3225
3226/*
3227 * Stamp cycle number in every block
3228 */
3229void
3230xlog_pack_data(
3231 xlog_t *log,
3232 xlog_in_core_t *iclog,
3233 int roundoff)
3234{
3235 int i, j, k;
3236 int size = iclog->ic_offset + roundoff;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003237 __be32 cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 xfs_caddr_t dp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239
3240 xlog_pack_data_checksum(log, iclog, size);
3241
3242 cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);
3243
3244 dp = iclog->ic_datap;
3245 for (i = 0; i < BTOBB(size) &&
3246 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003247 iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
3248 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 dp += BBSIZE;
3250 }
3251
Eric Sandeen62118702008-03-06 13:44:28 +11003252 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003253 xlog_in_core_2_t *xhdr = iclog->ic_data;
3254
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 for ( ; i < BTOBB(size); i++) {
3256 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3257 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003258 xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
3259 *(__be32 *)dp = cycle_lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 dp += BBSIZE;
3261 }
3262
3263 for (i = 1; i < log->l_iclog_heads; i++) {
3264 xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
3265 }
3266 }
3267}
3268
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269STATIC void
3270xlog_unpack_data(
3271 xlog_rec_header_t *rhead,
3272 xfs_caddr_t dp,
3273 xlog_t *log)
3274{
3275 int i, j, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003277 for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003279 *(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 dp += BBSIZE;
3281 }
3282
Eric Sandeen62118702008-03-06 13:44:28 +11003283 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Christoph Hellwigb28708d2008-11-28 14:23:38 +11003284 xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003285 for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3287 k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003288 *(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 dp += BBSIZE;
3290 }
3291 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292}
3293
3294STATIC int
3295xlog_valid_rec_header(
3296 xlog_t *log,
3297 xlog_rec_header_t *rhead,
3298 xfs_daddr_t blkno)
3299{
3300 int hlen;
3301
Christoph Hellwig69ef9212011-07-08 14:36:05 +02003302 if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
3304 XFS_ERRLEVEL_LOW, log->l_mp);
3305 return XFS_ERROR(EFSCORRUPTED);
3306 }
3307 if (unlikely(
3308 (!rhead->h_version ||
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003309 (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003310 xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
Harvey Harrison34a622b2008-04-10 12:19:21 +10003311 __func__, be32_to_cpu(rhead->h_version));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 return XFS_ERROR(EIO);
3313 }
3314
3315 /* LR body must have data or it wouldn't have been written */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003316 hlen = be32_to_cpu(rhead->h_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
3318 XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
3319 XFS_ERRLEVEL_LOW, log->l_mp);
3320 return XFS_ERROR(EFSCORRUPTED);
3321 }
3322 if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
3323 XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
3324 XFS_ERRLEVEL_LOW, log->l_mp);
3325 return XFS_ERROR(EFSCORRUPTED);
3326 }
3327 return 0;
3328}
3329
3330/*
3331 * Read the log from tail to head and process the log records found.
3332 * Handle the two cases where the tail and head are in the same cycle
3333 * and where the active portion of the log wraps around the end of
3334 * the physical log separately. The pass parameter is passed through
3335 * to the routines called to process the data and is not looked at
3336 * here.
3337 */
3338STATIC int
3339xlog_do_recovery_pass(
3340 xlog_t *log,
3341 xfs_daddr_t head_blk,
3342 xfs_daddr_t tail_blk,
3343 int pass)
3344{
3345 xlog_rec_header_t *rhead;
3346 xfs_daddr_t blk_no;
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003347 xfs_caddr_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 xfs_buf_t *hbp, *dbp;
3349 int error = 0, h_size;
3350 int bblks, split_bblks;
3351 int hblks, split_hblks, wrapped_hblks;
Dave Chinnerf0a76952010-01-11 11:49:57 +00003352 struct hlist_head rhash[XLOG_RHASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353
3354 ASSERT(head_blk != tail_blk);
3355
3356 /*
3357 * Read the header of the tail block and get the iclog buffer size from
3358 * h_size. Use this to tell how many sectors make up the log header.
3359 */
Eric Sandeen62118702008-03-06 13:44:28 +11003360 if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361 /*
3362 * When using variable length iclogs, read first sector of
3363 * iclog header and extract the header size from it. Get a
3364 * new hbp that is the correct size.
3365 */
3366 hbp = xlog_get_bp(log, 1);
3367 if (!hbp)
3368 return ENOMEM;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003369
3370 error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3371 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 goto bread_err1;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003373
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 rhead = (xlog_rec_header_t *)offset;
3375 error = xlog_valid_rec_header(log, rhead, tail_blk);
3376 if (error)
3377 goto bread_err1;
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003378 h_size = be32_to_cpu(rhead->h_size);
3379 if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380 (h_size > XLOG_HEADER_CYCLE_SIZE)) {
3381 hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
3382 if (h_size % XLOG_HEADER_CYCLE_SIZE)
3383 hblks++;
3384 xlog_put_bp(hbp);
3385 hbp = xlog_get_bp(log, hblks);
3386 } else {
3387 hblks = 1;
3388 }
3389 } else {
Alex Elder69ce58f2010-04-20 17:09:59 +10003390 ASSERT(log->l_sectBBsize == 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 hblks = 1;
3392 hbp = xlog_get_bp(log, 1);
3393 h_size = XLOG_BIG_RECORD_BSIZE;
3394 }
3395
3396 if (!hbp)
3397 return ENOMEM;
3398 dbp = xlog_get_bp(log, BTOBB(h_size));
3399 if (!dbp) {
3400 xlog_put_bp(hbp);
3401 return ENOMEM;
3402 }
3403
3404 memset(rhash, 0, sizeof(rhash));
3405 if (tail_blk <= head_blk) {
3406 for (blk_no = tail_blk; blk_no < head_blk; ) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003407 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3408 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003410
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 rhead = (xlog_rec_header_t *)offset;
3412 error = xlog_valid_rec_header(log, rhead, blk_no);
3413 if (error)
3414 goto bread_err2;
3415
3416 /* blocks in data section */
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003417 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003418 error = xlog_bread(log, blk_no + hblks, bblks, dbp,
3419 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 if (error)
3421 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003422
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423 xlog_unpack_data(rhead, offset, log);
3424 if ((error = xlog_recover_process_data(log,
3425 rhash, rhead, offset, pass)))
3426 goto bread_err2;
3427 blk_no += bblks + hblks;
3428 }
3429 } else {
3430 /*
3431 * Perform recovery around the end of the physical log.
3432 * When the head is not on the same cycle number as the tail,
3433 * we can't do a sequential recovery as above.
3434 */
3435 blk_no = tail_blk;
3436 while (blk_no < log->l_logBBsize) {
3437 /*
3438 * Check for header wrapping around physical end-of-log
3439 */
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003440 offset = XFS_BUF_PTR(hbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441 split_hblks = 0;
3442 wrapped_hblks = 0;
3443 if (blk_no + hblks <= log->l_logBBsize) {
3444 /* Read header in one read */
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003445 error = xlog_bread(log, blk_no, hblks, hbp,
3446 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 if (error)
3448 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 } else {
3450 /* This LR is split across physical log end */
3451 if (blk_no != log->l_logBBsize) {
3452 /* some data before physical log end */
3453 ASSERT(blk_no <= INT_MAX);
3454 split_hblks = log->l_logBBsize - (int)blk_no;
3455 ASSERT(split_hblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003456 error = xlog_bread(log, blk_no,
3457 split_hblks, hbp,
3458 &offset);
3459 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003462
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463 /*
3464 * Note: this black magic still works with
3465 * large sector sizes (non-512) only because:
3466 * - we increased the buffer size originally
3467 * by 1 sector giving us enough extra space
3468 * for the second read;
3469 * - the log start is guaranteed to be sector
3470 * aligned;
3471 * - we read the log end (LR header start)
3472 * _first_, then the log start (LR header end)
3473 * - order is important.
3474 */
David Chinner234f56a2008-04-10 12:24:24 +10003475 wrapped_hblks = hblks - split_hblks;
Dave Chinner44396472011-04-21 09:34:27 +00003476 error = xlog_bread_offset(log, 0,
3477 wrapped_hblks, hbp,
3478 offset + BBTOB(split_hblks));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479 if (error)
3480 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 }
3482 rhead = (xlog_rec_header_t *)offset;
3483 error = xlog_valid_rec_header(log, rhead,
3484 split_hblks ? blk_no : 0);
3485 if (error)
3486 goto bread_err2;
3487
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003488 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 blk_no += hblks;
3490
3491 /* Read in data for log record */
3492 if (blk_no + bblks <= log->l_logBBsize) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003493 error = xlog_bread(log, blk_no, bblks, dbp,
3494 &offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 if (error)
3496 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 } else {
3498 /* This log record is split across the
3499 * physical end of log */
Andy Polingfc5bc4c2009-11-03 17:26:47 +00003500 offset = XFS_BUF_PTR(dbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003501 split_bblks = 0;
3502 if (blk_no != log->l_logBBsize) {
3503 /* some data is before the physical
3504 * end of log */
3505 ASSERT(!wrapped_hblks);
3506 ASSERT(blk_no <= INT_MAX);
3507 split_bblks =
3508 log->l_logBBsize - (int)blk_no;
3509 ASSERT(split_bblks > 0);
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003510 error = xlog_bread(log, blk_no,
3511 split_bblks, dbp,
3512 &offset);
3513 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 }
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003516
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 /*
3518 * Note: this black magic still works with
3519 * large sector sizes (non-512) only because:
3520 * - we increased the buffer size originally
3521 * by 1 sector giving us enough extra space
3522 * for the second read;
3523 * - the log start is guaranteed to be sector
3524 * aligned;
3525 * - we read the log end (LR header start)
3526 * _first_, then the log start (LR header end)
3527 * - order is important.
3528 */
Dave Chinner44396472011-04-21 09:34:27 +00003529 error = xlog_bread_offset(log, 0,
3530 bblks - split_bblks, hbp,
3531 offset + BBTOB(split_bblks));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003532 if (error)
3533 goto bread_err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 }
3535 xlog_unpack_data(rhead, offset, log);
3536 if ((error = xlog_recover_process_data(log, rhash,
3537 rhead, offset, pass)))
3538 goto bread_err2;
3539 blk_no += bblks;
3540 }
3541
3542 ASSERT(blk_no >= log->l_logBBsize);
3543 blk_no -= log->l_logBBsize;
3544
3545 /* read first part of physical log */
3546 while (blk_no < head_blk) {
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003547 error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3548 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003550
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 rhead = (xlog_rec_header_t *)offset;
3552 error = xlog_valid_rec_header(log, rhead, blk_no);
3553 if (error)
3554 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003555
Christoph Hellwigb53e6752007-10-12 10:59:34 +10003556 bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003557 error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3558 &offset);
3559 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560 goto bread_err2;
Christoph Hellwig076e6ac2009-03-16 08:24:13 +01003561
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 xlog_unpack_data(rhead, offset, log);
3563 if ((error = xlog_recover_process_data(log, rhash,
3564 rhead, offset, pass)))
3565 goto bread_err2;
3566 blk_no += bblks + hblks;
3567 }
3568 }
3569
3570 bread_err2:
3571 xlog_put_bp(dbp);
3572 bread_err1:
3573 xlog_put_bp(hbp);
3574 return error;
3575}
3576
3577/*
3578 * Do the recovery of the log. We actually do this in two phases.
3579 * The two passes are necessary in order to implement the function
3580 * of cancelling a record written into the log. The first pass
3581 * determines those things which have been cancelled, and the
3582 * second pass replays log items normally except for those which
3583 * have been cancelled. The handling of the replay and cancellations
3584 * takes place in the log item type specific routines.
3585 *
3586 * The table of items which have cancel records in the log is allocated
3587 * and freed at this level, since only here do we know when all of
3588 * the log recovery has been completed.
3589 */
3590STATIC int
3591xlog_do_log_recovery(
3592 xlog_t *log,
3593 xfs_daddr_t head_blk,
3594 xfs_daddr_t tail_blk)
3595{
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003596 int error, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597
3598 ASSERT(head_blk != tail_blk);
3599
3600 /*
3601 * First do a pass to find all of the cancelled buf log items.
3602 * Store them in the buf_cancel_table for use in the second pass.
3603 */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003604 log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
3605 sizeof(struct list_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 KM_SLEEP);
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003607 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
3608 INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
3609
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3611 XLOG_RECOVER_PASS1);
3612 if (error != 0) {
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003613 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 log->l_buf_cancel_table = NULL;
3615 return error;
3616 }
3617 /*
3618 * Then do a second pass to actually recover the items in the log.
3619 * When it is complete free the table of buf cancel items.
3620 */
3621 error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3622 XLOG_RECOVER_PASS2);
3623#ifdef DEBUG
Tim Shimmin6d192a92006-06-09 14:55:38 +10003624 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625 int i;
3626
3627 for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
Christoph Hellwigd5689ea2010-12-01 22:06:22 +00003628 ASSERT(list_empty(&log->l_buf_cancel_table[i]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 }
3630#endif /* DEBUG */
3631
Denys Vlasenkof0e2d932008-05-19 16:31:57 +10003632 kmem_free(log->l_buf_cancel_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633 log->l_buf_cancel_table = NULL;
3634
3635 return error;
3636}
3637
3638/*
3639 * Do the actual recovery
3640 */
3641STATIC int
3642xlog_do_recover(
3643 xlog_t *log,
3644 xfs_daddr_t head_blk,
3645 xfs_daddr_t tail_blk)
3646{
3647 int error;
3648 xfs_buf_t *bp;
3649 xfs_sb_t *sbp;
3650
3651 /*
3652 * First replay the images in the log.
3653 */
3654 error = xlog_do_log_recovery(log, head_blk, tail_blk);
3655 if (error) {
3656 return error;
3657 }
3658
3659 XFS_bflush(log->l_mp->m_ddev_targp);
3660
3661 /*
3662 * If IO errors happened during recovery, bail out.
3663 */
3664 if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
3665 return (EIO);
3666 }
3667
3668 /*
3669 * We now update the tail_lsn since much of the recovery has completed
3670 * and there may be space available to use. If there were no extent
3671 * or iunlinks, we can free up the entire log and set the tail_lsn to
3672 * be the last_sync_lsn. This was set in xlog_find_tail to be the
3673 * lsn of the last known good LR on disk. If there are extent frees
3674 * or iunlinks they will have some entries in the AIL; so we look at
3675 * the AIL to determine how to set the tail_lsn.
3676 */
3677 xlog_assign_tail_lsn(log->l_mp);
3678
3679 /*
3680 * Now that we've finished replaying all buffer and inode
3681 * updates, re-read in the superblock.
3682 */
3683 bp = xfs_getsb(log->l_mp, 0);
3684 XFS_BUF_UNDONE(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003685 ASSERT(!(XFS_BUF_ISWRITE(bp)));
3686 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003687 XFS_BUF_READ(bp);
Lachlan McIlroybebf9632007-10-15 13:18:02 +10003688 XFS_BUF_UNASYNC(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 xfsbdstrat(log->l_mp, bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00003690 error = xfs_buf_iowait(bp);
David Chinnerd64e31a2008-04-10 12:22:17 +10003691 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692 xfs_ioerror_alert("xlog_do_recover",
3693 log->l_mp, bp, XFS_BUF_ADDR(bp));
3694 ASSERT(0);
3695 xfs_buf_relse(bp);
3696 return error;
3697 }
3698
3699 /* Convert superblock from on-disk format */
3700 sbp = &log->l_mp->m_sb;
Christoph Hellwig2bdf7cd2007-08-28 13:58:06 +10003701 xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
Eric Sandeen62118702008-03-06 13:44:28 +11003703 ASSERT(xfs_sb_good_version(sbp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 xfs_buf_relse(bp);
3705
Lachlan McIlroy5478eea2007-02-10 18:36:29 +11003706 /* We've re-read the superblock so re-initialize per-cpu counters */
3707 xfs_icsb_reinit_counters(log->l_mp);
3708
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 xlog_recover_check_summary(log);
3710
3711 /* Normal transactions can now occur */
3712 log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
3713 return 0;
3714}
3715
3716/*
3717 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3718 *
3719 * Return error or zero.
3720 */
3721int
3722xlog_recover(
Eric Sandeen65be6052006-01-11 15:34:19 +11003723 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003724{
3725 xfs_daddr_t head_blk, tail_blk;
3726 int error;
3727
3728 /* find the tail of the log */
Eric Sandeen65be6052006-01-11 15:34:19 +11003729 if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730 return error;
3731
3732 if (tail_blk != head_blk) {
3733 /* There used to be a comment here:
3734 *
3735 * disallow recovery on read-only mounts. note -- mount
3736 * checks for ENOSPC and turns it into an intelligent
3737 * error message.
3738 * ...but this is no longer true. Now, unless you specify
3739 * NORECOVERY (in which case this function would never be
3740 * called), we just go ahead and recover. We do this all
3741 * under the vfs layer, so we can get away with it unless
3742 * the device itself is read-only, in which case we fail.
3743 */
Utako Kusaka3a02ee12007-05-08 13:50:06 +10003744 if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003745 return error;
3746 }
3747
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003748 xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3749 log->l_mp->m_logname ? log->l_mp->m_logname
3750 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751
3752 error = xlog_do_recover(log, head_blk, tail_blk);
3753 log->l_flags |= XLOG_RECOVERY_NEEDED;
3754 }
3755 return error;
3756}
3757
3758/*
3759 * In the first part of recovery we replay inodes and buffers and build
3760 * up the list of extent free items which need to be processed. Here
3761 * we process the extent free items and clean up the on disk unlinked
3762 * inode lists. This is separated from the first part of recovery so
3763 * that the root and real-time bitmap inodes can be read in from disk in
3764 * between the two stages. This is necessary so that we can free space
3765 * in the real-time portion of the file system.
3766 */
3767int
3768xlog_recover_finish(
Christoph Hellwig42490232008-08-13 16:49:32 +10003769 xlog_t *log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770{
3771 /*
3772 * Now we're ready to do the transactions needed for the
3773 * rest of recovery. Start with completing all the extent
3774 * free intent records and then process the unlinked inode
3775 * lists. At this point, we essentially run in normal mode
3776 * except that we're still performing recovery actions
3777 * rather than accepting new requests.
3778 */
3779 if (log->l_flags & XLOG_RECOVERY_NEEDED) {
David Chinner3c1e2bb2008-04-10 12:21:11 +10003780 int error;
3781 error = xlog_recover_process_efis(log);
3782 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003783 xfs_alert(log->l_mp, "Failed to recover EFIs");
David Chinner3c1e2bb2008-04-10 12:21:11 +10003784 return error;
3785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786 /*
3787 * Sync the log to get all the EFIs out of the AIL.
3788 * This isn't absolutely necessary, but it helps in
3789 * case the unlink transactions would have problems
3790 * pushing the EFIs out of the way.
3791 */
Christoph Hellwiga14a3482010-01-19 09:56:46 +00003792 xfs_log_force(log->l_mp, XFS_LOG_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793
Christoph Hellwig42490232008-08-13 16:49:32 +10003794 xlog_recover_process_iunlinks(log);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795
3796 xlog_recover_check_summary(log);
3797
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003798 xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
3799 log->l_mp->m_logname ? log->l_mp->m_logname
3800 : "internal");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 log->l_flags &= ~XLOG_RECOVERY_NEEDED;
3802 } else {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003803 xfs_info(log->l_mp, "Ending clean mount");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804 }
3805 return 0;
3806}
3807
3808
3809#if defined(DEBUG)
3810/*
3811 * Read all of the agf and agi counters and check that they
3812 * are consistent with the superblock counters.
3813 */
3814void
3815xlog_recover_check_summary(
3816 xlog_t *log)
3817{
3818 xfs_mount_t *mp;
3819 xfs_agf_t *agfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 xfs_buf_t *agfbp;
3821 xfs_buf_t *agibp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003822 xfs_agnumber_t agno;
3823 __uint64_t freeblks;
3824 __uint64_t itotal;
3825 __uint64_t ifree;
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003826 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827
3828 mp = log->l_mp;
3829
3830 freeblks = 0LL;
3831 itotal = 0LL;
3832 ifree = 0LL;
3833 for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
From: Christoph Hellwig48056212008-11-28 14:23:38 +11003834 error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
3835 if (error) {
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003836 xfs_alert(mp, "%s agf read failed agno %d error %d",
3837 __func__, agno, error);
From: Christoph Hellwig48056212008-11-28 14:23:38 +11003838 } else {
3839 agfp = XFS_BUF_TO_AGF(agfbp);
3840 freeblks += be32_to_cpu(agfp->agf_freeblks) +
3841 be32_to_cpu(agfp->agf_flcount);
3842 xfs_buf_relse(agfbp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003845 error = xfs_read_agi(mp, NULL, agno, &agibp);
Dave Chinnera0fa2b62011-03-07 10:01:35 +11003846 if (error) {
3847 xfs_alert(mp, "%s agi read failed agno %d error %d",
3848 __func__, agno, error);
3849 } else {
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003850 struct xfs_agi *agi = XFS_BUF_TO_AGI(agibp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851
Christoph Hellwig5e1be0f2008-11-28 14:23:37 +11003852 itotal += be32_to_cpu(agi->agi_count);
3853 ifree += be32_to_cpu(agi->agi_freecount);
3854 xfs_buf_relse(agibp);
3855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857}
3858#endif /* DEBUG */