Linux Audio

Check our new training course

Loading...
v3.15
 
   1/*
   2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
  25#include "xfs_inum.h"
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_mount.h"
  29#include "xfs_da_format.h"
  30#include "xfs_inode.h"
  31#include "xfs_trans.h"
  32#include "xfs_log.h"
  33#include "xfs_log_priv.h"
  34#include "xfs_log_recover.h"
  35#include "xfs_inode_item.h"
  36#include "xfs_extfree_item.h"
  37#include "xfs_trans_priv.h"
  38#include "xfs_alloc.h"
  39#include "xfs_ialloc.h"
  40#include "xfs_quota.h"
  41#include "xfs_cksum.h"
  42#include "xfs_trace.h"
  43#include "xfs_icache.h"
  44#include "xfs_bmap_btree.h"
  45#include "xfs_dinode.h"
  46#include "xfs_error.h"
  47#include "xfs_dir2.h"
 
 
 
  48
  49#define BLK_AVG(blk1, blk2)	((blk1+blk2) >> 1)
  50
  51STATIC int
  52xlog_find_zeroed(
  53	struct xlog	*,
  54	xfs_daddr_t	*);
  55STATIC int
  56xlog_clear_stale_blocks(
  57	struct xlog	*,
  58	xfs_lsn_t);
  59#if defined(DEBUG)
  60STATIC void
  61xlog_recover_check_summary(
  62	struct xlog *);
  63#else
  64#define	xlog_recover_check_summary(log)
  65#endif
  66
  67/*
  68 * This structure is used during recovery to record the buf log items which
  69 * have been canceled and should not be replayed.
  70 */
  71struct xfs_buf_cancel {
  72	xfs_daddr_t		bc_blkno;
  73	uint			bc_len;
  74	int			bc_refcount;
  75	struct list_head	bc_list;
  76};
  77
  78/*
  79 * Sector aligned buffer routines for buffer create/read/write/access
  80 */
  81
  82/*
  83 * Verify the given count of basic blocks is valid number of blocks
  84 * to specify for an operation involving the given XFS log buffer.
  85 * Returns nonzero if the count is valid, 0 otherwise.
  86 */
  87
  88static inline int
  89xlog_buf_bbcount_valid(
  90	struct xlog	*log,
 
  91	int		bbcount)
  92{
  93	return bbcount > 0 && bbcount <= log->l_logBBsize;
 
 
 
 
  94}
  95
  96/*
  97 * Allocate a buffer to hold log data.  The buffer needs to be able
  98 * to map to a range of nbblks basic blocks at any valid (basic
  99 * block) offset within the log.
 100 */
 101STATIC xfs_buf_t *
 102xlog_get_bp(
 103	struct xlog	*log,
 104	int		nbblks)
 105{
 106	struct xfs_buf	*bp;
 107
 108	if (!xlog_buf_bbcount_valid(log, nbblks)) {
 
 
 109		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
 110			nbblks);
 111		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
 112		return NULL;
 113	}
 114
 115	/*
 116	 * We do log I/O in units of log sectors (a power-of-2
 117	 * multiple of the basic block size), so we round up the
 118	 * requested size to accommodate the basic blocks required
 119	 * for complete log sectors.
 120	 *
 121	 * In addition, the buffer may be used for a non-sector-
 122	 * aligned block offset, in which case an I/O of the
 123	 * requested size could extend beyond the end of the
 124	 * buffer.  If the requested size is only 1 basic block it
 125	 * will never straddle a sector boundary, so this won't be
 126	 * an issue.  Nor will this be a problem if the log I/O is
 127	 * done in basic blocks (sector size 1).  But otherwise we
 128	 * extend the buffer by one extra log sector to ensure
 129	 * there's space to accommodate this possibility.
 130	 */
 131	if (nbblks > 1 && log->l_sectBBsize > 1)
 132		nbblks += log->l_sectBBsize;
 133	nbblks = round_up(nbblks, log->l_sectBBsize);
 134
 135	bp = xfs_buf_get_uncached(log->l_mp->m_logdev_targp, nbblks, 0);
 136	if (bp)
 137		xfs_buf_unlock(bp);
 138	return bp;
 139}
 140
 141STATIC void
 142xlog_put_bp(
 143	xfs_buf_t	*bp)
 144{
 145	xfs_buf_free(bp);
 146}
 147
 148/*
 149 * Return the address of the start of the given block number's data
 150 * in a log buffer.  The buffer covers a log sector-aligned region.
 151 */
 152STATIC xfs_caddr_t
 153xlog_align(
 154	struct xlog	*log,
 155	xfs_daddr_t	blk_no,
 156	int		nbblks,
 157	struct xfs_buf	*bp)
 158{
 159	xfs_daddr_t	offset = blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1);
 160
 161	ASSERT(offset + nbblks <= bp->b_length);
 162	return bp->b_addr + BBTOB(offset);
 163}
 164
 165
 166/*
 167 * nbblks should be uint, but oh well.  Just want to catch that 32-bit length.
 168 */
 169STATIC int
 170xlog_bread_noalign(
 171	struct xlog	*log,
 172	xfs_daddr_t	blk_no,
 173	int		nbblks,
 174	struct xfs_buf	*bp)
 175{
 176	int		error;
 177
 178	if (!xlog_buf_bbcount_valid(log, nbblks)) {
 179		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
 180			nbblks);
 181		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
 182		return EFSCORRUPTED;
 183	}
 184
 185	blk_no = round_down(blk_no, log->l_sectBBsize);
 186	nbblks = round_up(nbblks, log->l_sectBBsize);
 187
 188	ASSERT(nbblks > 0);
 189	ASSERT(nbblks <= bp->b_length);
 190
 191	XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
 192	XFS_BUF_READ(bp);
 193	bp->b_io_length = nbblks;
 194	bp->b_error = 0;
 195
 196	if (XFS_FORCED_SHUTDOWN(log->l_mp))
 197		return XFS_ERROR(EIO);
 198
 199	xfs_buf_iorequest(bp);
 200	error = xfs_buf_iowait(bp);
 201	if (error)
 202		xfs_buf_ioerror_alert(bp, __func__);
 
 203	return error;
 204}
 205
 206STATIC int
 207xlog_bread(
 208	struct xlog	*log,
 209	xfs_daddr_t	blk_no,
 210	int		nbblks,
 211	struct xfs_buf	*bp,
 212	xfs_caddr_t	*offset)
 213{
 214	int		error;
 215
 216	error = xlog_bread_noalign(log, blk_no, nbblks, bp);
 217	if (error)
 218		return error;
 219
 220	*offset = xlog_align(log, blk_no, nbblks, bp);
 221	return 0;
 222}
 223
 224/*
 225 * Read at an offset into the buffer. Returns with the buffer in it's original
 226 * state regardless of the result of the read.
 227 */
 228STATIC int
 229xlog_bread_offset(
 230	struct xlog	*log,
 231	xfs_daddr_t	blk_no,		/* block to read from */
 232	int		nbblks,		/* blocks to read */
 233	struct xfs_buf	*bp,
 234	xfs_caddr_t	offset)
 235{
 236	xfs_caddr_t	orig_offset = bp->b_addr;
 237	int		orig_len = BBTOB(bp->b_length);
 238	int		error, error2;
 239
 240	error = xfs_buf_associate_memory(bp, offset, BBTOB(nbblks));
 241	if (error)
 242		return error;
 243
 244	error = xlog_bread_noalign(log, blk_no, nbblks, bp);
 245
 246	/* must reset buffer pointer even on error */
 247	error2 = xfs_buf_associate_memory(bp, orig_offset, orig_len);
 248	if (error)
 249		return error;
 250	return error2;
 251}
 252
 253/*
 254 * Write out the buffer at the given block for the given number of blocks.
 255 * The buffer is kept locked across the write and is returned locked.
 256 * This can only be used for synchronous log writes.
 257 */
 258STATIC int
 259xlog_bwrite(
 260	struct xlog	*log,
 261	xfs_daddr_t	blk_no,
 262	int		nbblks,
 263	struct xfs_buf	*bp)
 264{
 265	int		error;
 266
 267	if (!xlog_buf_bbcount_valid(log, nbblks)) {
 268		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
 269			nbblks);
 270		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
 271		return EFSCORRUPTED;
 272	}
 273
 274	blk_no = round_down(blk_no, log->l_sectBBsize);
 275	nbblks = round_up(nbblks, log->l_sectBBsize);
 276
 277	ASSERT(nbblks > 0);
 278	ASSERT(nbblks <= bp->b_length);
 279
 280	XFS_BUF_SET_ADDR(bp, log->l_logBBstart + blk_no);
 281	XFS_BUF_ZEROFLAGS(bp);
 282	xfs_buf_hold(bp);
 283	xfs_buf_lock(bp);
 284	bp->b_io_length = nbblks;
 285	bp->b_error = 0;
 286
 287	error = xfs_bwrite(bp);
 288	if (error)
 289		xfs_buf_ioerror_alert(bp, __func__);
 290	xfs_buf_relse(bp);
 291	return error;
 292}
 293
 294#ifdef DEBUG
 295/*
 296 * dump debug superblock and log record information
 297 */
 298STATIC void
 299xlog_header_check_dump(
 300	xfs_mount_t		*mp,
 301	xlog_rec_header_t	*head)
 302{
 303	xfs_debug(mp, "%s:  SB : uuid = %pU, fmt = %d",
 304		__func__, &mp->m_sb.sb_uuid, XLOG_FMT);
 305	xfs_debug(mp, "    log : uuid = %pU, fmt = %d",
 306		&head->h_fs_uuid, be32_to_cpu(head->h_fmt));
 307}
 308#else
 309#define xlog_header_check_dump(mp, head)
 310#endif
 311
 312/*
 313 * check log record header for recovery
 314 */
 315STATIC int
 316xlog_header_check_recover(
 317	xfs_mount_t		*mp,
 318	xlog_rec_header_t	*head)
 319{
 320	ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
 321
 322	/*
 323	 * IRIX doesn't write the h_fmt field and leaves it zeroed
 324	 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
 325	 * a dirty log created in IRIX.
 326	 */
 327	if (unlikely(head->h_fmt != cpu_to_be32(XLOG_FMT))) {
 328		xfs_warn(mp,
 329	"dirty log written in incompatible format - can't recover");
 330		xlog_header_check_dump(mp, head);
 331		XFS_ERROR_REPORT("xlog_header_check_recover(1)",
 332				 XFS_ERRLEVEL_HIGH, mp);
 333		return XFS_ERROR(EFSCORRUPTED);
 334	} else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
 335		xfs_warn(mp,
 336	"dirty log entry has mismatched uuid - can't recover");
 337		xlog_header_check_dump(mp, head);
 338		XFS_ERROR_REPORT("xlog_header_check_recover(2)",
 339				 XFS_ERRLEVEL_HIGH, mp);
 340		return XFS_ERROR(EFSCORRUPTED);
 341	}
 342	return 0;
 343}
 344
 345/*
 346 * read the head block of the log and check the header
 347 */
 348STATIC int
 349xlog_header_check_mount(
 350	xfs_mount_t		*mp,
 351	xlog_rec_header_t	*head)
 352{
 353	ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
 354
 355	if (uuid_is_nil(&head->h_fs_uuid)) {
 356		/*
 357		 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
 358		 * h_fs_uuid is nil, we assume this log was last mounted
 359		 * by IRIX and continue.
 360		 */
 361		xfs_warn(mp, "nil uuid in log - IRIX style log");
 362	} else if (unlikely(!uuid_equal(&mp->m_sb.sb_uuid, &head->h_fs_uuid))) {
 
 363		xfs_warn(mp, "log has mismatched uuid - can't recover");
 364		xlog_header_check_dump(mp, head);
 365		XFS_ERROR_REPORT("xlog_header_check_mount",
 366				 XFS_ERRLEVEL_HIGH, mp);
 367		return XFS_ERROR(EFSCORRUPTED);
 368	}
 369	return 0;
 370}
 371
 372STATIC void
 373xlog_recover_iodone(
 374	struct xfs_buf	*bp)
 375{
 376	if (bp->b_error) {
 377		/*
 378		 * We're not going to bother about retrying
 379		 * this during recovery. One strike!
 380		 */
 381		xfs_buf_ioerror_alert(bp, __func__);
 382		xfs_force_shutdown(bp->b_target->bt_mount,
 383					SHUTDOWN_META_IO_ERROR);
 384	}
 385	bp->b_iodone = NULL;
 386	xfs_buf_ioend(bp, 0);
 387}
 388
 389/*
 390 * This routine finds (to an approximation) the first block in the physical
 391 * log which contains the given cycle.  It uses a binary search algorithm.
 392 * Note that the algorithm can not be perfect because the disk will not
 393 * necessarily be perfect.
 394 */
 395STATIC int
 396xlog_find_cycle_start(
 397	struct xlog	*log,
 398	struct xfs_buf	*bp,
 399	xfs_daddr_t	first_blk,
 400	xfs_daddr_t	*last_blk,
 401	uint		cycle)
 402{
 403	xfs_caddr_t	offset;
 404	xfs_daddr_t	mid_blk;
 405	xfs_daddr_t	end_blk;
 406	uint		mid_cycle;
 407	int		error;
 408
 409	end_blk = *last_blk;
 410	mid_blk = BLK_AVG(first_blk, end_blk);
 411	while (mid_blk != first_blk && mid_blk != end_blk) {
 412		error = xlog_bread(log, mid_blk, 1, bp, &offset);
 413		if (error)
 414			return error;
 415		mid_cycle = xlog_get_cycle(offset);
 416		if (mid_cycle == cycle)
 417			end_blk = mid_blk;   /* last_half_cycle == mid_cycle */
 418		else
 419			first_blk = mid_blk; /* first_half_cycle == mid_cycle */
 420		mid_blk = BLK_AVG(first_blk, end_blk);
 421	}
 422	ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
 423	       (mid_blk == end_blk && mid_blk-1 == first_blk));
 424
 425	*last_blk = end_blk;
 426
 427	return 0;
 428}
 429
 430/*
 431 * Check that a range of blocks does not contain stop_on_cycle_no.
 432 * Fill in *new_blk with the block offset where such a block is
 433 * found, or with -1 (an invalid block number) if there is no such
 434 * block in the range.  The scan needs to occur from front to back
 435 * and the pointer into the region must be updated since a later
 436 * routine will need to perform another test.
 437 */
 438STATIC int
 439xlog_find_verify_cycle(
 440	struct xlog	*log,
 441	xfs_daddr_t	start_blk,
 442	int		nbblks,
 443	uint		stop_on_cycle_no,
 444	xfs_daddr_t	*new_blk)
 445{
 446	xfs_daddr_t	i, j;
 447	uint		cycle;
 448	xfs_buf_t	*bp;
 449	xfs_daddr_t	bufblks;
 450	xfs_caddr_t	buf = NULL;
 451	int		error = 0;
 452
 453	/*
 454	 * Greedily allocate a buffer big enough to handle the full
 455	 * range of basic blocks we'll be examining.  If that fails,
 456	 * try a smaller size.  We need to be able to read at least
 457	 * a log sector, or we're out of luck.
 458	 */
 459	bufblks = 1 << ffs(nbblks);
 460	while (bufblks > log->l_logBBsize)
 461		bufblks >>= 1;
 462	while (!(bp = xlog_get_bp(log, bufblks))) {
 463		bufblks >>= 1;
 464		if (bufblks < log->l_sectBBsize)
 465			return ENOMEM;
 466	}
 467
 468	for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
 469		int	bcount;
 470
 471		bcount = min(bufblks, (start_blk + nbblks - i));
 472
 473		error = xlog_bread(log, i, bcount, bp, &buf);
 474		if (error)
 475			goto out;
 476
 477		for (j = 0; j < bcount; j++) {
 478			cycle = xlog_get_cycle(buf);
 479			if (cycle == stop_on_cycle_no) {
 480				*new_blk = i+j;
 481				goto out;
 482			}
 483
 484			buf += BBSIZE;
 485		}
 486	}
 487
 488	*new_blk = -1;
 489
 490out:
 491	xlog_put_bp(bp);
 492	return error;
 493}
 494
 
 
 
 
 
 
 
 
 
 
 
 
 
 495/*
 496 * Potentially backup over partial log record write.
 497 *
 498 * In the typical case, last_blk is the number of the block directly after
 499 * a good log record.  Therefore, we subtract one to get the block number
 500 * of the last block in the given buffer.  extra_bblks contains the number
 501 * of blocks we would have read on a previous read.  This happens when the
 502 * last log record is split over the end of the physical log.
 503 *
 504 * extra_bblks is the number of blocks potentially verified on a previous
 505 * call to this routine.
 506 */
 507STATIC int
 508xlog_find_verify_log_record(
 509	struct xlog		*log,
 510	xfs_daddr_t		start_blk,
 511	xfs_daddr_t		*last_blk,
 512	int			extra_bblks)
 513{
 514	xfs_daddr_t		i;
 515	xfs_buf_t		*bp;
 516	xfs_caddr_t		offset = NULL;
 517	xlog_rec_header_t	*head = NULL;
 518	int			error = 0;
 519	int			smallmem = 0;
 520	int			num_blks = *last_blk - start_blk;
 521	int			xhdrs;
 522
 523	ASSERT(start_blk != 0 || *last_blk != start_blk);
 524
 525	if (!(bp = xlog_get_bp(log, num_blks))) {
 526		if (!(bp = xlog_get_bp(log, 1)))
 527			return ENOMEM;
 
 
 528		smallmem = 1;
 529	} else {
 530		error = xlog_bread(log, start_blk, num_blks, bp, &offset);
 531		if (error)
 532			goto out;
 533		offset += ((num_blks - 1) << BBSHIFT);
 534	}
 535
 536	for (i = (*last_blk) - 1; i >= 0; i--) {
 537		if (i < start_blk) {
 538			/* valid log record not found */
 539			xfs_warn(log->l_mp,
 540		"Log inconsistent (didn't find previous header)");
 541			ASSERT(0);
 542			error = XFS_ERROR(EIO);
 543			goto out;
 544		}
 545
 546		if (smallmem) {
 547			error = xlog_bread(log, i, 1, bp, &offset);
 548			if (error)
 549				goto out;
 550		}
 551
 552		head = (xlog_rec_header_t *)offset;
 553
 554		if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
 555			break;
 556
 557		if (!smallmem)
 558			offset -= BBSIZE;
 559	}
 560
 561	/*
 562	 * We hit the beginning of the physical log & still no header.  Return
 563	 * to caller.  If caller can handle a return of -1, then this routine
 564	 * will be called again for the end of the physical log.
 565	 */
 566	if (i == -1) {
 567		error = -1;
 568		goto out;
 569	}
 570
 571	/*
 572	 * We have the final block of the good log (the first block
 573	 * of the log record _before_ the head. So we check the uuid.
 574	 */
 575	if ((error = xlog_header_check_mount(log->l_mp, head)))
 576		goto out;
 577
 578	/*
 579	 * We may have found a log record header before we expected one.
 580	 * last_blk will be the 1st block # with a given cycle #.  We may end
 581	 * up reading an entire log record.  In this case, we don't want to
 582	 * reset last_blk.  Only when last_blk points in the middle of a log
 583	 * record do we update last_blk.
 584	 */
 585	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
 586		uint	h_size = be32_to_cpu(head->h_size);
 587
 588		xhdrs = h_size / XLOG_HEADER_CYCLE_SIZE;
 589		if (h_size % XLOG_HEADER_CYCLE_SIZE)
 590			xhdrs++;
 591	} else {
 592		xhdrs = 1;
 593	}
 594
 595	if (*last_blk - i + extra_bblks !=
 596	    BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
 597		*last_blk = i;
 598
 599out:
 600	xlog_put_bp(bp);
 601	return error;
 602}
 603
 604/*
 605 * Head is defined to be the point of the log where the next log write
 606 * could go.  This means that incomplete LR writes at the end are
 607 * eliminated when calculating the head.  We aren't guaranteed that previous
 608 * LR have complete transactions.  We only know that a cycle number of
 609 * current cycle number -1 won't be present in the log if we start writing
 610 * from our current block number.
 611 *
 612 * last_blk contains the block number of the first block with a given
 613 * cycle number.
 614 *
 615 * Return: zero if normal, non-zero if error.
 616 */
 617STATIC int
 618xlog_find_head(
 619	struct xlog	*log,
 620	xfs_daddr_t	*return_head_blk)
 621{
 622	xfs_buf_t	*bp;
 623	xfs_caddr_t	offset;
 624	xfs_daddr_t	new_blk, first_blk, start_blk, last_blk, head_blk;
 625	int		num_scan_bblks;
 626	uint		first_half_cycle, last_half_cycle;
 627	uint		stop_on_cycle;
 628	int		error, log_bbnum = log->l_logBBsize;
 629
 630	/* Is the end of the log device zeroed? */
 631	if ((error = xlog_find_zeroed(log, &first_blk)) == -1) {
 
 
 
 
 
 632		*return_head_blk = first_blk;
 633
 634		/* Is the whole lot zeroed? */
 635		if (!first_blk) {
 636			/* Linux XFS shouldn't generate totally zeroed logs -
 637			 * mkfs etc write a dummy unmount record to a fresh
 638			 * log so we can store the uuid in there
 639			 */
 640			xfs_warn(log->l_mp, "totally zeroed log");
 641		}
 642
 643		return 0;
 644	} else if (error) {
 645		xfs_warn(log->l_mp, "empty log check failed");
 646		return error;
 647	}
 648
 649	first_blk = 0;			/* get cycle # of 1st block */
 650	bp = xlog_get_bp(log, 1);
 651	if (!bp)
 652		return ENOMEM;
 653
 654	error = xlog_bread(log, 0, 1, bp, &offset);
 655	if (error)
 656		goto bp_err;
 657
 658	first_half_cycle = xlog_get_cycle(offset);
 659
 660	last_blk = head_blk = log_bbnum - 1;	/* get cycle # of last block */
 661	error = xlog_bread(log, last_blk, 1, bp, &offset);
 662	if (error)
 663		goto bp_err;
 664
 665	last_half_cycle = xlog_get_cycle(offset);
 666	ASSERT(last_half_cycle != 0);
 667
 668	/*
 669	 * If the 1st half cycle number is equal to the last half cycle number,
 670	 * then the entire log is stamped with the same cycle number.  In this
 671	 * case, head_blk can't be set to zero (which makes sense).  The below
 672	 * math doesn't work out properly with head_blk equal to zero.  Instead,
 673	 * we set it to log_bbnum which is an invalid block number, but this
 674	 * value makes the math correct.  If head_blk doesn't changed through
 675	 * all the tests below, *head_blk is set to zero at the very end rather
 676	 * than log_bbnum.  In a sense, log_bbnum and zero are the same block
 677	 * in a circular file.
 678	 */
 679	if (first_half_cycle == last_half_cycle) {
 680		/*
 681		 * In this case we believe that the entire log should have
 682		 * cycle number last_half_cycle.  We need to scan backwards
 683		 * from the end verifying that there are no holes still
 684		 * containing last_half_cycle - 1.  If we find such a hole,
 685		 * then the start of that hole will be the new head.  The
 686		 * simple case looks like
 687		 *        x | x ... | x - 1 | x
 688		 * Another case that fits this picture would be
 689		 *        x | x + 1 | x ... | x
 690		 * In this case the head really is somewhere at the end of the
 691		 * log, as one of the latest writes at the beginning was
 692		 * incomplete.
 693		 * One more case is
 694		 *        x | x + 1 | x ... | x - 1 | x
 695		 * This is really the combination of the above two cases, and
 696		 * the head has to end up at the start of the x-1 hole at the
 697		 * end of the log.
 698		 *
 699		 * In the 256k log case, we will read from the beginning to the
 700		 * end of the log and search for cycle numbers equal to x-1.
 701		 * We don't worry about the x+1 blocks that we encounter,
 702		 * because we know that they cannot be the head since the log
 703		 * started with x.
 704		 */
 705		head_blk = log_bbnum;
 706		stop_on_cycle = last_half_cycle - 1;
 707	} else {
 708		/*
 709		 * In this case we want to find the first block with cycle
 710		 * number matching last_half_cycle.  We expect the log to be
 711		 * some variation on
 712		 *        x + 1 ... | x ... | x
 713		 * The first block with cycle number x (last_half_cycle) will
 714		 * be where the new head belongs.  First we do a binary search
 715		 * for the first occurrence of last_half_cycle.  The binary
 716		 * search may not be totally accurate, so then we scan back
 717		 * from there looking for occurrences of last_half_cycle before
 718		 * us.  If that backwards scan wraps around the beginning of
 719		 * the log, then we look for occurrences of last_half_cycle - 1
 720		 * at the end of the log.  The cases we're looking for look
 721		 * like
 722		 *                               v binary search stopped here
 723		 *        x + 1 ... | x | x + 1 | x ... | x
 724		 *                   ^ but we want to locate this spot
 725		 * or
 726		 *        <---------> less than scan distance
 727		 *        x + 1 ... | x ... | x - 1 | x
 728		 *                           ^ we want to locate this spot
 729		 */
 730		stop_on_cycle = last_half_cycle;
 731		if ((error = xlog_find_cycle_start(log, bp, first_blk,
 732						&head_blk, last_half_cycle)))
 733			goto bp_err;
 
 734	}
 735
 736	/*
 737	 * Now validate the answer.  Scan back some number of maximum possible
 738	 * blocks and make sure each one has the expected cycle number.  The
 739	 * maximum is determined by the total possible amount of buffering
 740	 * in the in-core log.  The following number can be made tighter if
 741	 * we actually look at the block size of the filesystem.
 742	 */
 743	num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
 744	if (head_blk >= num_scan_bblks) {
 745		/*
 746		 * We are guaranteed that the entire check can be performed
 747		 * in one buffer.
 748		 */
 749		start_blk = head_blk - num_scan_bblks;
 750		if ((error = xlog_find_verify_cycle(log,
 751						start_blk, num_scan_bblks,
 752						stop_on_cycle, &new_blk)))
 753			goto bp_err;
 754		if (new_blk != -1)
 755			head_blk = new_blk;
 756	} else {		/* need to read 2 parts of log */
 757		/*
 758		 * We are going to scan backwards in the log in two parts.
 759		 * First we scan the physical end of the log.  In this part
 760		 * of the log, we are looking for blocks with cycle number
 761		 * last_half_cycle - 1.
 762		 * If we find one, then we know that the log starts there, as
 763		 * we've found a hole that didn't get written in going around
 764		 * the end of the physical log.  The simple case for this is
 765		 *        x + 1 ... | x ... | x - 1 | x
 766		 *        <---------> less than scan distance
 767		 * If all of the blocks at the end of the log have cycle number
 768		 * last_half_cycle, then we check the blocks at the start of
 769		 * the log looking for occurrences of last_half_cycle.  If we
 770		 * find one, then our current estimate for the location of the
 771		 * first occurrence of last_half_cycle is wrong and we move
 772		 * back to the hole we've found.  This case looks like
 773		 *        x + 1 ... | x | x + 1 | x ...
 774		 *                               ^ binary search stopped here
 775		 * Another case we need to handle that only occurs in 256k
 776		 * logs is
 777		 *        x + 1 ... | x ... | x+1 | x ...
 778		 *                   ^ binary search stops here
 779		 * In a 256k log, the scan at the end of the log will see the
 780		 * x + 1 blocks.  We need to skip past those since that is
 781		 * certainly not the head of the log.  By searching for
 782		 * last_half_cycle-1 we accomplish that.
 783		 */
 784		ASSERT(head_blk <= INT_MAX &&
 785			(xfs_daddr_t) num_scan_bblks >= head_blk);
 786		start_blk = log_bbnum - (num_scan_bblks - head_blk);
 787		if ((error = xlog_find_verify_cycle(log, start_blk,
 788					num_scan_bblks - (int)head_blk,
 789					(stop_on_cycle - 1), &new_blk)))
 790			goto bp_err;
 791		if (new_blk != -1) {
 792			head_blk = new_blk;
 793			goto validate_head;
 794		}
 795
 796		/*
 797		 * Scan beginning of log now.  The last part of the physical
 798		 * log is good.  This scan needs to verify that it doesn't find
 799		 * the last_half_cycle.
 800		 */
 801		start_blk = 0;
 802		ASSERT(head_blk <= INT_MAX);
 803		if ((error = xlog_find_verify_cycle(log,
 804					start_blk, (int)head_blk,
 805					stop_on_cycle, &new_blk)))
 806			goto bp_err;
 807		if (new_blk != -1)
 808			head_blk = new_blk;
 809	}
 810
 811validate_head:
 812	/*
 813	 * Now we need to make sure head_blk is not pointing to a block in
 814	 * the middle of a log record.
 815	 */
 816	num_scan_bblks = XLOG_REC_SHIFT(log);
 817	if (head_blk >= num_scan_bblks) {
 818		start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
 819
 820		/* start ptr at last block ptr before head_blk */
 821		if ((error = xlog_find_verify_log_record(log, start_blk,
 822							&head_blk, 0)) == -1) {
 823			error = XFS_ERROR(EIO);
 824			goto bp_err;
 825		} else if (error)
 826			goto bp_err;
 827	} else {
 828		start_blk = 0;
 829		ASSERT(head_blk <= INT_MAX);
 830		if ((error = xlog_find_verify_log_record(log, start_blk,
 831							&head_blk, 0)) == -1) {
 
 
 832			/* We hit the beginning of the log during our search */
 833			start_blk = log_bbnum - (num_scan_bblks - head_blk);
 834			new_blk = log_bbnum;
 835			ASSERT(start_blk <= INT_MAX &&
 836				(xfs_daddr_t) log_bbnum-start_blk >= 0);
 837			ASSERT(head_blk <= INT_MAX);
 838			if ((error = xlog_find_verify_log_record(log,
 839							start_blk, &new_blk,
 840							(int)head_blk)) == -1) {
 841				error = XFS_ERROR(EIO);
 842				goto bp_err;
 843			} else if (error)
 844				goto bp_err;
 845			if (new_blk != log_bbnum)
 846				head_blk = new_blk;
 847		} else if (error)
 848			goto bp_err;
 849	}
 850
 851	xlog_put_bp(bp);
 852	if (head_blk == log_bbnum)
 853		*return_head_blk = 0;
 854	else
 855		*return_head_blk = head_blk;
 856	/*
 857	 * When returning here, we have a good block number.  Bad block
 858	 * means that during a previous crash, we didn't have a clean break
 859	 * from cycle number N to cycle number N-1.  In this case, we need
 860	 * to find the first block with cycle number N-1.
 861	 */
 862	return 0;
 863
 864 bp_err:
 865	xlog_put_bp(bp);
 866
 867	if (error)
 868		xfs_warn(log->l_mp, "failed to find log head");
 869	return error;
 870}
 871
 872/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 873 * Find the sync block number or the tail of the log.
 874 *
 875 * This will be the block number of the last record to have its
 876 * associated buffers synced to disk.  Every log record header has
 877 * a sync lsn embedded in it.  LSNs hold block numbers, so it is easy
 878 * to get a sync block number.  The only concern is to figure out which
 879 * log record header to believe.
 880 *
 881 * The following algorithm uses the log record header with the largest
 882 * lsn.  The entire log record does not need to be valid.  We only care
 883 * that the header is valid.
 884 *
 885 * We could speed up search by using current head_blk buffer, but it is not
 886 * available.
 887 */
 888STATIC int
 889xlog_find_tail(
 890	struct xlog		*log,
 891	xfs_daddr_t		*head_blk,
 892	xfs_daddr_t		*tail_blk)
 893{
 894	xlog_rec_header_t	*rhead;
 895	xlog_op_header_t	*op_head;
 896	xfs_caddr_t		offset = NULL;
 897	xfs_buf_t		*bp;
 898	int			error, i, found;
 899	xfs_daddr_t		umount_data_blk;
 900	xfs_daddr_t		after_umount_blk;
 901	xfs_lsn_t		tail_lsn;
 902	int			hblks;
 903
 904	found = 0;
 905
 906	/*
 907	 * Find previous log record
 908	 */
 909	if ((error = xlog_find_head(log, head_blk)))
 910		return error;
 
 911
 912	bp = xlog_get_bp(log, 1);
 913	if (!bp)
 914		return ENOMEM;
 915	if (*head_blk == 0) {				/* special case */
 916		error = xlog_bread(log, 0, 1, bp, &offset);
 917		if (error)
 918			goto done;
 919
 920		if (xlog_get_cycle(offset) == 0) {
 921			*tail_blk = 0;
 922			/* leave all other log inited values alone */
 923			goto done;
 924		}
 925	}
 926
 927	/*
 928	 * Search backwards looking for log record header block
 929	 */
 930	ASSERT(*head_blk < INT_MAX);
 931	for (i = (int)(*head_blk) - 1; i >= 0; i--) {
 932		error = xlog_bread(log, i, 1, bp, &offset);
 933		if (error)
 934			goto done;
 935
 936		if (*(__be32 *)offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 937			found = 1;
 938			break;
 939		}
 940	}
 941	/*
 942	 * If we haven't found the log record header block, start looking
 943	 * again from the end of the physical log.  XXXmiken: There should be
 944	 * a check here to make sure we didn't search more than N blocks in
 945	 * the previous code.
 946	 */
 947	if (!found) {
 948		for (i = log->l_logBBsize - 1; i >= (int)(*head_blk); i--) {
 949			error = xlog_bread(log, i, 1, bp, &offset);
 950			if (error)
 951				goto done;
 952
 953			if (*(__be32 *)offset ==
 954			    cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 955				found = 2;
 956				break;
 957			}
 958		}
 959	}
 960	if (!found) {
 961		xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
 962		xlog_put_bp(bp);
 963		ASSERT(0);
 964		return XFS_ERROR(EIO);
 965	}
 966
 967	/* find blk_no of tail of log */
 968	rhead = (xlog_rec_header_t *)offset;
 969	*tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
 970
 971	/*
 972	 * Reset log values according to the state of the log when we
 973	 * crashed.  In the case where head_blk == 0, we bump curr_cycle
 974	 * one because the next write starts a new cycle rather than
 975	 * continuing the cycle of the last good log record.  At this
 976	 * point we have guaranteed that all partial log records have been
 977	 * accounted for.  Therefore, we know that the last good log record
 978	 * written was complete and ended exactly on the end boundary
 979	 * of the physical log.
 980	 */
 981	log->l_prev_block = i;
 982	log->l_curr_block = (int)*head_blk;
 983	log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
 984	if (found == 2)
 985		log->l_curr_cycle++;
 986	atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
 987	atomic64_set(&log->l_last_sync_lsn, be64_to_cpu(rhead->h_lsn));
 988	xlog_assign_grant_head(&log->l_reserve_head.grant, log->l_curr_cycle,
 989					BBTOB(log->l_curr_block));
 990	xlog_assign_grant_head(&log->l_write_head.grant, log->l_curr_cycle,
 991					BBTOB(log->l_curr_block));
 992
 993	/*
 994	 * Look for unmount record.  If we find it, then we know there
 995	 * was a clean unmount.  Since 'i' could be the last block in
 996	 * the physical log, we convert to a log block before comparing
 997	 * to the head_blk.
 998	 *
 999	 * Save the current tail lsn to use to pass to
1000	 * xlog_clear_stale_blocks() below.  We won't want to clear the
1001	 * unmount record if there is one, so we pass the lsn of the
1002	 * unmount record rather than the block after it.
1003	 */
1004	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
1005		int	h_size = be32_to_cpu(rhead->h_size);
1006		int	h_version = be32_to_cpu(rhead->h_version);
1007
1008		if ((h_version & XLOG_VERSION_2) &&
1009		    (h_size > XLOG_HEADER_CYCLE_SIZE)) {
1010			hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
1011			if (h_size % XLOG_HEADER_CYCLE_SIZE)
1012				hblks++;
1013		} else {
1014			hblks = 1;
1015		}
1016	} else {
1017		hblks = 1;
1018	}
1019	after_umount_blk = (i + hblks + (int)
1020		BTOBB(be32_to_cpu(rhead->h_len))) % log->l_logBBsize;
1021	tail_lsn = atomic64_read(&log->l_tail_lsn);
1022	if (*head_blk == after_umount_blk &&
1023	    be32_to_cpu(rhead->h_num_logops) == 1) {
1024		umount_data_blk = (i + hblks) % log->l_logBBsize;
1025		error = xlog_bread(log, umount_data_blk, 1, bp, &offset);
1026		if (error)
1027			goto done;
1028
1029		op_head = (xlog_op_header_t *)offset;
1030		if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1031			/*
1032			 * Set tail and last sync so that newly written
1033			 * log records will point recovery to after the
1034			 * current unmount record.
1035			 */
1036			xlog_assign_atomic_lsn(&log->l_tail_lsn,
1037					log->l_curr_cycle, after_umount_blk);
1038			xlog_assign_atomic_lsn(&log->l_last_sync_lsn,
1039					log->l_curr_cycle, after_umount_blk);
1040			*tail_blk = after_umount_blk;
1041
1042			/*
1043			 * Note that the unmount was clean. If the unmount
1044			 * was not clean, we need to know this to rebuild the
1045			 * superblock counters from the perag headers if we
1046			 * have a filesystem using non-persistent counters.
1047			 */
1048			log->l_mp->m_flags |= XFS_MOUNT_WAS_CLEAN;
1049		}
1050	}
1051
1052	/*
 
 
 
 
 
 
 
 
1053	 * Make sure that there are no blocks in front of the head
1054	 * with the same cycle number as the head.  This can happen
1055	 * because we allow multiple outstanding log writes concurrently,
1056	 * and the later writes might make it out before earlier ones.
1057	 *
1058	 * We use the lsn from before modifying it so that we'll never
1059	 * overwrite the unmount record after a clean unmount.
1060	 *
1061	 * Do this only if we are going to recover the filesystem
1062	 *
1063	 * NOTE: This used to say "if (!readonly)"
1064	 * However on Linux, we can & do recover a read-only filesystem.
1065	 * We only skip recovery if NORECOVERY is specified on mount,
1066	 * in which case we would not be here.
1067	 *
1068	 * But... if the -device- itself is readonly, just skip this.
1069	 * We can't recover this device anyway, so it won't matter.
1070	 */
1071	if (!xfs_readonly_buftarg(log->l_mp->m_logdev_targp))
1072		error = xlog_clear_stale_blocks(log, tail_lsn);
1073
1074done:
1075	xlog_put_bp(bp);
1076
1077	if (error)
1078		xfs_warn(log->l_mp, "failed to locate log tail");
1079	return error;
1080}
1081
1082/*
1083 * Is the log zeroed at all?
1084 *
1085 * The last binary search should be changed to perform an X block read
1086 * once X becomes small enough.  You can then search linearly through
1087 * the X blocks.  This will cut down on the number of reads we need to do.
1088 *
1089 * If the log is partially zeroed, this routine will pass back the blkno
1090 * of the first block with cycle number 0.  It won't have a complete LR
1091 * preceding it.
1092 *
1093 * Return:
1094 *	0  => the log is completely written to
1095 *	-1 => use *blk_no as the first block of the log
1096 *	>0 => error has occurred
1097 */
1098STATIC int
1099xlog_find_zeroed(
1100	struct xlog	*log,
1101	xfs_daddr_t	*blk_no)
1102{
1103	xfs_buf_t	*bp;
1104	xfs_caddr_t	offset;
1105	uint	        first_cycle, last_cycle;
1106	xfs_daddr_t	new_blk, last_blk, start_blk;
1107	xfs_daddr_t     num_scan_bblks;
1108	int	        error, log_bbnum = log->l_logBBsize;
 
1109
1110	*blk_no = 0;
1111
1112	/* check totally zeroed log */
1113	bp = xlog_get_bp(log, 1);
1114	if (!bp)
1115		return ENOMEM;
1116	error = xlog_bread(log, 0, 1, bp, &offset);
1117	if (error)
1118		goto bp_err;
1119
1120	first_cycle = xlog_get_cycle(offset);
1121	if (first_cycle == 0) {		/* completely zeroed log */
1122		*blk_no = 0;
1123		xlog_put_bp(bp);
1124		return -1;
1125	}
1126
1127	/* check partially zeroed log */
1128	error = xlog_bread(log, log_bbnum-1, 1, bp, &offset);
1129	if (error)
1130		goto bp_err;
1131
1132	last_cycle = xlog_get_cycle(offset);
1133	if (last_cycle != 0) {		/* log completely written to */
1134		xlog_put_bp(bp);
1135		return 0;
1136	} else if (first_cycle != 1) {
1137		/*
1138		 * If the cycle of the last block is zero, the cycle of
1139		 * the first block must be 1. If it's not, maybe we're
1140		 * not looking at a log... Bail out.
1141		 */
1142		xfs_warn(log->l_mp,
1143			"Log inconsistent or not a log (last==0, first!=1)");
1144		error = XFS_ERROR(EINVAL);
1145		goto bp_err;
1146	}
1147
1148	/* we have a partially zeroed log */
1149	last_blk = log_bbnum-1;
1150	if ((error = xlog_find_cycle_start(log, bp, 0, &last_blk, 0)))
1151		goto bp_err;
 
1152
1153	/*
1154	 * Validate the answer.  Because there is no way to guarantee that
1155	 * the entire log is made up of log records which are the same size,
1156	 * we scan over the defined maximum blocks.  At this point, the maximum
1157	 * is not chosen to mean anything special.   XXXmiken
1158	 */
1159	num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1160	ASSERT(num_scan_bblks <= INT_MAX);
1161
1162	if (last_blk < num_scan_bblks)
1163		num_scan_bblks = last_blk;
1164	start_blk = last_blk - num_scan_bblks;
1165
1166	/*
1167	 * We search for any instances of cycle number 0 that occur before
1168	 * our current estimate of the head.  What we're trying to detect is
1169	 *        1 ... | 0 | 1 | 0...
1170	 *                       ^ binary search ends here
1171	 */
1172	if ((error = xlog_find_verify_cycle(log, start_blk,
1173					 (int)num_scan_bblks, 0, &new_blk)))
1174		goto bp_err;
1175	if (new_blk != -1)
1176		last_blk = new_blk;
1177
1178	/*
1179	 * Potentially backup over partial log record write.  We don't need
1180	 * to search the end of the log because we know it is zero.
1181	 */
1182	if ((error = xlog_find_verify_log_record(log, start_blk,
1183				&last_blk, 0)) == -1) {
1184	    error = XFS_ERROR(EIO);
1185	    goto bp_err;
1186	} else if (error)
1187	    goto bp_err;
1188
1189	*blk_no = last_blk;
1190bp_err:
1191	xlog_put_bp(bp);
1192	if (error)
1193		return error;
1194	return -1;
1195}
1196
1197/*
1198 * These are simple subroutines used by xlog_clear_stale_blocks() below
1199 * to initialize a buffer full of empty log record headers and write
1200 * them into the log.
1201 */
1202STATIC void
1203xlog_add_record(
1204	struct xlog		*log,
1205	xfs_caddr_t		buf,
1206	int			cycle,
1207	int			block,
1208	int			tail_cycle,
1209	int			tail_block)
1210{
1211	xlog_rec_header_t	*recp = (xlog_rec_header_t *)buf;
1212
1213	memset(buf, 0, BBSIZE);
1214	recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1215	recp->h_cycle = cpu_to_be32(cycle);
1216	recp->h_version = cpu_to_be32(
1217			xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1218	recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1219	recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1220	recp->h_fmt = cpu_to_be32(XLOG_FMT);
1221	memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1222}
1223
1224STATIC int
1225xlog_write_log_records(
1226	struct xlog	*log,
1227	int		cycle,
1228	int		start_block,
1229	int		blocks,
1230	int		tail_cycle,
1231	int		tail_block)
1232{
1233	xfs_caddr_t	offset;
1234	xfs_buf_t	*bp;
1235	int		balign, ealign;
1236	int		sectbb = log->l_sectBBsize;
1237	int		end_block = start_block + blocks;
1238	int		bufblks;
1239	int		error = 0;
1240	int		i, j = 0;
1241
1242	/*
1243	 * Greedily allocate a buffer big enough to handle the full
1244	 * range of basic blocks to be written.  If that fails, try
1245	 * a smaller size.  We need to be able to write at least a
1246	 * log sector, or we're out of luck.
1247	 */
1248	bufblks = 1 << ffs(blocks);
1249	while (bufblks > log->l_logBBsize)
1250		bufblks >>= 1;
1251	while (!(bp = xlog_get_bp(log, bufblks))) {
1252		bufblks >>= 1;
1253		if (bufblks < sectbb)
1254			return ENOMEM;
1255	}
1256
1257	/* We may need to do a read at the start to fill in part of
1258	 * the buffer in the starting sector not covered by the first
1259	 * write below.
1260	 */
1261	balign = round_down(start_block, sectbb);
1262	if (balign != start_block) {
1263		error = xlog_bread_noalign(log, start_block, 1, bp);
1264		if (error)
1265			goto out_put_bp;
1266
1267		j = start_block - balign;
1268	}
1269
1270	for (i = start_block; i < end_block; i += bufblks) {
1271		int		bcount, endcount;
1272
1273		bcount = min(bufblks, end_block - start_block);
1274		endcount = bcount - j;
1275
1276		/* We may need to do a read at the end to fill in part of
1277		 * the buffer in the final sector not covered by the write.
1278		 * If this is the same sector as the above read, skip it.
1279		 */
1280		ealign = round_down(end_block, sectbb);
1281		if (j == 0 && (start_block + endcount > ealign)) {
1282			offset = bp->b_addr + BBTOB(ealign - start_block);
1283			error = xlog_bread_offset(log, ealign, sectbb,
1284							bp, offset);
1285			if (error)
1286				break;
1287
1288		}
1289
1290		offset = xlog_align(log, start_block, endcount, bp);
1291		for (; j < endcount; j++) {
1292			xlog_add_record(log, offset, cycle, i+j,
1293					tail_cycle, tail_block);
1294			offset += BBSIZE;
1295		}
1296		error = xlog_bwrite(log, start_block, endcount, bp);
1297		if (error)
1298			break;
1299		start_block += endcount;
1300		j = 0;
1301	}
1302
1303 out_put_bp:
1304	xlog_put_bp(bp);
1305	return error;
1306}
1307
1308/*
1309 * This routine is called to blow away any incomplete log writes out
1310 * in front of the log head.  We do this so that we won't become confused
1311 * if we come up, write only a little bit more, and then crash again.
1312 * If we leave the partial log records out there, this situation could
1313 * cause us to think those partial writes are valid blocks since they
1314 * have the current cycle number.  We get rid of them by overwriting them
1315 * with empty log records with the old cycle number rather than the
1316 * current one.
1317 *
1318 * The tail lsn is passed in rather than taken from
1319 * the log so that we will not write over the unmount record after a
1320 * clean unmount in a 512 block log.  Doing so would leave the log without
1321 * any valid log records in it until a new one was written.  If we crashed
1322 * during that time we would not be able to recover.
1323 */
1324STATIC int
1325xlog_clear_stale_blocks(
1326	struct xlog	*log,
1327	xfs_lsn_t	tail_lsn)
1328{
1329	int		tail_cycle, head_cycle;
1330	int		tail_block, head_block;
1331	int		tail_distance, max_distance;
1332	int		distance;
1333	int		error;
1334
1335	tail_cycle = CYCLE_LSN(tail_lsn);
1336	tail_block = BLOCK_LSN(tail_lsn);
1337	head_cycle = log->l_curr_cycle;
1338	head_block = log->l_curr_block;
1339
1340	/*
1341	 * Figure out the distance between the new head of the log
1342	 * and the tail.  We want to write over any blocks beyond the
1343	 * head that we may have written just before the crash, but
1344	 * we don't want to overwrite the tail of the log.
1345	 */
1346	if (head_cycle == tail_cycle) {
1347		/*
1348		 * The tail is behind the head in the physical log,
1349		 * so the distance from the head to the tail is the
1350		 * distance from the head to the end of the log plus
1351		 * the distance from the beginning of the log to the
1352		 * tail.
1353		 */
1354		if (unlikely(head_block < tail_block || head_block >= log->l_logBBsize)) {
1355			XFS_ERROR_REPORT("xlog_clear_stale_blocks(1)",
1356					 XFS_ERRLEVEL_LOW, log->l_mp);
1357			return XFS_ERROR(EFSCORRUPTED);
1358		}
1359		tail_distance = tail_block + (log->l_logBBsize - head_block);
1360	} else {
1361		/*
1362		 * The head is behind the tail in the physical log,
1363		 * so the distance from the head to the tail is just
1364		 * the tail block minus the head block.
1365		 */
1366		if (unlikely(head_block >= tail_block || head_cycle != (tail_cycle + 1))){
1367			XFS_ERROR_REPORT("xlog_clear_stale_blocks(2)",
1368					 XFS_ERRLEVEL_LOW, log->l_mp);
1369			return XFS_ERROR(EFSCORRUPTED);
1370		}
1371		tail_distance = tail_block - head_block;
1372	}
1373
1374	/*
1375	 * If the head is right up against the tail, we can't clear
1376	 * anything.
1377	 */
1378	if (tail_distance <= 0) {
1379		ASSERT(tail_distance == 0);
1380		return 0;
1381	}
1382
1383	max_distance = XLOG_TOTAL_REC_SHIFT(log);
1384	/*
1385	 * Take the smaller of the maximum amount of outstanding I/O
1386	 * we could have and the distance to the tail to clear out.
1387	 * We take the smaller so that we don't overwrite the tail and
1388	 * we don't waste all day writing from the head to the tail
1389	 * for no reason.
1390	 */
1391	max_distance = MIN(max_distance, tail_distance);
1392
1393	if ((head_block + max_distance) <= log->l_logBBsize) {
1394		/*
1395		 * We can stomp all the blocks we need to without
1396		 * wrapping around the end of the log.  Just do it
1397		 * in a single write.  Use the cycle number of the
1398		 * current cycle minus one so that the log will look like:
1399		 *     n ... | n - 1 ...
1400		 */
1401		error = xlog_write_log_records(log, (head_cycle - 1),
1402				head_block, max_distance, tail_cycle,
1403				tail_block);
1404		if (error)
1405			return error;
1406	} else {
1407		/*
1408		 * We need to wrap around the end of the physical log in
1409		 * order to clear all the blocks.  Do it in two separate
1410		 * I/Os.  The first write should be from the head to the
1411		 * end of the physical log, and it should use the current
1412		 * cycle number minus one just like above.
1413		 */
1414		distance = log->l_logBBsize - head_block;
1415		error = xlog_write_log_records(log, (head_cycle - 1),
1416				head_block, distance, tail_cycle,
1417				tail_block);
1418
1419		if (error)
1420			return error;
1421
1422		/*
1423		 * Now write the blocks at the start of the physical log.
1424		 * This writes the remainder of the blocks we want to clear.
1425		 * It uses the current cycle number since we're now on the
1426		 * same cycle as the head so that we get:
1427		 *    n ... n ... | n - 1 ...
1428		 *    ^^^^^ blocks we're writing
1429		 */
1430		distance = max_distance - (log->l_logBBsize - head_block);
1431		error = xlog_write_log_records(log, head_cycle, 0, distance,
1432				tail_cycle, tail_block);
1433		if (error)
1434			return error;
1435	}
1436
1437	return 0;
1438}
1439
1440/******************************************************************************
1441 *
1442 *		Log recover routines
1443 *
1444 ******************************************************************************
1445 */
1446
1447STATIC xlog_recover_t *
1448xlog_recover_find_tid(
1449	struct hlist_head	*head,
1450	xlog_tid_t		tid)
1451{
1452	xlog_recover_t		*trans;
1453
1454	hlist_for_each_entry(trans, head, r_list) {
1455		if (trans->r_log_tid == tid)
1456			return trans;
1457	}
1458	return NULL;
1459}
1460
1461STATIC void
1462xlog_recover_new_tid(
1463	struct hlist_head	*head,
1464	xlog_tid_t		tid,
1465	xfs_lsn_t		lsn)
1466{
1467	xlog_recover_t		*trans;
1468
1469	trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
1470	trans->r_log_tid   = tid;
1471	trans->r_lsn	   = lsn;
1472	INIT_LIST_HEAD(&trans->r_itemq);
1473
1474	INIT_HLIST_NODE(&trans->r_list);
1475	hlist_add_head(&trans->r_list, head);
1476}
 
1477
1478STATIC void
1479xlog_recover_add_item(
1480	struct list_head	*head)
1481{
1482	xlog_recover_item_t	*item;
1483
1484	item = kmem_zalloc(sizeof(xlog_recover_item_t), KM_SLEEP);
1485	INIT_LIST_HEAD(&item->ri_list);
1486	list_add_tail(&item->ri_list, head);
1487}
1488
1489STATIC int
1490xlog_recover_add_to_cont_trans(
1491	struct xlog		*log,
1492	struct xlog_recover	*trans,
1493	xfs_caddr_t		dp,
1494	int			len)
1495{
1496	xlog_recover_item_t	*item;
1497	xfs_caddr_t		ptr, old_ptr;
1498	int			old_len;
1499
1500	if (list_empty(&trans->r_itemq)) {
1501		/* finish copying rest of trans header */
1502		xlog_recover_add_item(&trans->r_itemq);
1503		ptr = (xfs_caddr_t) &trans->r_theader +
1504				sizeof(xfs_trans_header_t) - len;
1505		memcpy(ptr, dp, len); /* d, s, l */
1506		return 0;
 
1507	}
1508	/* take the tail entry */
1509	item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1510
1511	old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
1512	old_len = item->ri_buf[item->ri_cnt-1].i_len;
1513
1514	ptr = kmem_realloc(old_ptr, len+old_len, old_len, KM_SLEEP);
1515	memcpy(&ptr[old_len], dp, len); /* d, s, l */
1516	item->ri_buf[item->ri_cnt-1].i_len += len;
1517	item->ri_buf[item->ri_cnt-1].i_addr = ptr;
1518	trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
1519	return 0;
1520}
1521
1522/*
1523 * The next region to add is the start of a new region.  It could be
1524 * a whole region or it could be the first part of a new region.  Because
1525 * of this, the assumption here is that the type and size fields of all
1526 * format structures fit into the first 32 bits of the structure.
1527 *
1528 * This works because all regions must be 32 bit aligned.  Therefore, we
1529 * either have both fields or we have neither field.  In the case we have
1530 * neither field, the data part of the region is zero length.  We only have
1531 * a log_op_header and can throw away the header since a new one will appear
1532 * later.  If we have at least 4 bytes, then we can determine how many regions
1533 * will appear in the current log item.
1534 */
1535STATIC int
1536xlog_recover_add_to_trans(
1537	struct xlog		*log,
1538	struct xlog_recover	*trans,
1539	xfs_caddr_t		dp,
1540	int			len)
1541{
1542	xfs_inode_log_format_t	*in_f;			/* any will do */
1543	xlog_recover_item_t	*item;
1544	xfs_caddr_t		ptr;
1545
1546	if (!len)
1547		return 0;
1548	if (list_empty(&trans->r_itemq)) {
1549		/* we need to catch log corruptions here */
1550		if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
1551			xfs_warn(log->l_mp, "%s: bad header magic number",
1552				__func__);
1553			ASSERT(0);
1554			return XFS_ERROR(EIO);
1555		}
1556		if (len == sizeof(xfs_trans_header_t))
1557			xlog_recover_add_item(&trans->r_itemq);
1558		memcpy(&trans->r_theader, dp, len); /* d, s, l */
1559		return 0;
1560	}
1561
1562	ptr = kmem_alloc(len, KM_SLEEP);
1563	memcpy(ptr, dp, len);
1564	in_f = (xfs_inode_log_format_t *)ptr;
1565
1566	/* take the tail entry */
1567	item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
1568	if (item->ri_total != 0 &&
1569	     item->ri_total == item->ri_cnt) {
1570		/* tail item is in use, get a new one */
1571		xlog_recover_add_item(&trans->r_itemq);
1572		item = list_entry(trans->r_itemq.prev,
1573					xlog_recover_item_t, ri_list);
1574	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1575
1576	if (item->ri_total == 0) {		/* first region to be added */
1577		if (in_f->ilf_size == 0 ||
1578		    in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
1579			xfs_warn(log->l_mp,
1580		"bad number of regions (%d) in inode log format",
1581				  in_f->ilf_size);
1582			ASSERT(0);
1583			kmem_free(ptr);
1584			return XFS_ERROR(EIO);
1585		}
1586
1587		item->ri_total = in_f->ilf_size;
1588		item->ri_buf =
1589			kmem_zalloc(item->ri_total * sizeof(xfs_log_iovec_t),
1590				    KM_SLEEP);
1591	}
1592	ASSERT(item->ri_total > item->ri_cnt);
1593	/* Description region is ri_buf[0] */
1594	item->ri_buf[item->ri_cnt].i_addr = ptr;
1595	item->ri_buf[item->ri_cnt].i_len  = len;
1596	item->ri_cnt++;
1597	trace_xfs_log_recover_item_add(log, trans, item, 0);
1598	return 0;
1599}
1600
1601/*
1602 * Sort the log items in the transaction.
1603 *
1604 * The ordering constraints are defined by the inode allocation and unlink
1605 * behaviour. The rules are:
1606 *
1607 *	1. Every item is only logged once in a given transaction. Hence it
1608 *	   represents the last logged state of the item. Hence ordering is
1609 *	   dependent on the order in which operations need to be performed so
1610 *	   required initial conditions are always met.
1611 *
1612 *	2. Cancelled buffers are recorded in pass 1 in a separate table and
1613 *	   there's nothing to replay from them so we can simply cull them
1614 *	   from the transaction. However, we can't do that until after we've
1615 *	   replayed all the other items because they may be dependent on the
1616 *	   cancelled buffer and replaying the cancelled buffer can remove it
1617 *	   form the cancelled buffer table. Hence they have tobe done last.
1618 *
1619 *	3. Inode allocation buffers must be replayed before inode items that
1620 *	   read the buffer and replay changes into it. For filesystems using the
1621 *	   ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1622 *	   treated the same as inode allocation buffers as they create and
1623 *	   initialise the buffers directly.
1624 *
1625 *	4. Inode unlink buffers must be replayed after inode items are replayed.
1626 *	   This ensures that inodes are completely flushed to the inode buffer
1627 *	   in a "free" state before we remove the unlinked inode list pointer.
1628 *
1629 * Hence the ordering needs to be inode allocation buffers first, inode items
1630 * second, inode unlink buffers third and cancelled buffers last.
1631 *
1632 * But there's a problem with that - we can't tell an inode allocation buffer
1633 * apart from a regular buffer, so we can't separate them. We can, however,
1634 * tell an inode unlink buffer from the others, and so we can separate them out
1635 * from all the other buffers and move them to last.
1636 *
1637 * Hence, 4 lists, in order from head to tail:
1638 *	- buffer_list for all buffers except cancelled/inode unlink buffers
1639 *	- item_list for all non-buffer items
1640 *	- inode_buffer_list for inode unlink buffers
1641 *	- cancel_list for the cancelled buffers
1642 *
1643 * Note that we add objects to the tail of the lists so that first-to-last
1644 * ordering is preserved within the lists. Adding objects to the head of the
1645 * list means when we traverse from the head we walk them in last-to-first
1646 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1647 * but for all other items there may be specific ordering that we need to
1648 * preserve.
1649 */
1650STATIC int
1651xlog_recover_reorder_trans(
1652	struct xlog		*log,
1653	struct xlog_recover	*trans,
1654	int			pass)
1655{
1656	xlog_recover_item_t	*item, *n;
1657	int			error = 0;
1658	LIST_HEAD(sort_list);
1659	LIST_HEAD(cancel_list);
1660	LIST_HEAD(buffer_list);
1661	LIST_HEAD(inode_buffer_list);
1662	LIST_HEAD(inode_list);
1663
1664	list_splice_init(&trans->r_itemq, &sort_list);
1665	list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1666		xfs_buf_log_format_t	*buf_f = item->ri_buf[0].i_addr;
1667
1668		switch (ITEM_TYPE(item)) {
1669		case XFS_LI_ICREATE:
1670			list_move_tail(&item->ri_list, &buffer_list);
1671			break;
1672		case XFS_LI_BUF:
1673			if (buf_f->blf_flags & XFS_BLF_CANCEL) {
1674				trace_xfs_log_recover_item_reorder_head(log,
1675							trans, item, pass);
1676				list_move(&item->ri_list, &cancel_list);
1677				break;
1678			}
1679			if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
1680				list_move(&item->ri_list, &inode_buffer_list);
1681				break;
1682			}
1683			list_move_tail(&item->ri_list, &buffer_list);
1684			break;
1685		case XFS_LI_INODE:
1686		case XFS_LI_DQUOT:
1687		case XFS_LI_QUOTAOFF:
1688		case XFS_LI_EFD:
1689		case XFS_LI_EFI:
1690			trace_xfs_log_recover_item_reorder_tail(log,
1691							trans, item, pass);
1692			list_move_tail(&item->ri_list, &inode_list);
1693			break;
1694		default:
1695			xfs_warn(log->l_mp,
1696				"%s: unrecognized type of log operation",
1697				__func__);
1698			ASSERT(0);
1699			/*
1700			 * return the remaining items back to the transaction
1701			 * item list so they can be freed in caller.
1702			 */
1703			if (!list_empty(&sort_list))
1704				list_splice_init(&sort_list, &trans->r_itemq);
1705			error = XFS_ERROR(EIO);
1706			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1707		}
1708	}
1709out:
1710	ASSERT(list_empty(&sort_list));
1711	if (!list_empty(&buffer_list))
1712		list_splice(&buffer_list, &trans->r_itemq);
1713	if (!list_empty(&inode_list))
1714		list_splice_tail(&inode_list, &trans->r_itemq);
1715	if (!list_empty(&inode_buffer_list))
1716		list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1717	if (!list_empty(&cancel_list))
1718		list_splice_tail(&cancel_list, &trans->r_itemq);
1719	return error;
1720}
1721
1722/*
1723 * Build up the table of buf cancel records so that we don't replay
1724 * cancelled data in the second pass.  For buffer records that are
1725 * not cancel records, there is nothing to do here so we just return.
1726 *
1727 * If we get a cancel record which is already in the table, this indicates
1728 * that the buffer was cancelled multiple times.  In order to ensure
1729 * that during pass 2 we keep the record in the table until we reach its
1730 * last occurrence in the log, we keep a reference count in the cancel
1731 * record in the table to tell us how many times we expect to see this
1732 * record during the second pass.
1733 */
1734STATIC int
1735xlog_recover_buffer_pass1(
1736	struct xlog			*log,
1737	struct xlog_recover_item	*item)
1738{
1739	xfs_buf_log_format_t	*buf_f = item->ri_buf[0].i_addr;
1740	struct list_head	*bucket;
1741	struct xfs_buf_cancel	*bcp;
1742
1743	/*
1744	 * If this isn't a cancel buffer item, then just return.
1745	 */
1746	if (!(buf_f->blf_flags & XFS_BLF_CANCEL)) {
1747		trace_xfs_log_recover_buf_not_cancel(log, buf_f);
1748		return 0;
1749	}
1750
1751	/*
1752	 * Insert an xfs_buf_cancel record into the hash table of them.
1753	 * If there is already an identical record, bump its reference count.
1754	 */
1755	bucket = XLOG_BUF_CANCEL_BUCKET(log, buf_f->blf_blkno);
1756	list_for_each_entry(bcp, bucket, bc_list) {
1757		if (bcp->bc_blkno == buf_f->blf_blkno &&
1758		    bcp->bc_len == buf_f->blf_len) {
1759			bcp->bc_refcount++;
1760			trace_xfs_log_recover_buf_cancel_ref_inc(log, buf_f);
1761			return 0;
1762		}
1763	}
1764
1765	bcp = kmem_alloc(sizeof(struct xfs_buf_cancel), KM_SLEEP);
1766	bcp->bc_blkno = buf_f->blf_blkno;
1767	bcp->bc_len = buf_f->blf_len;
1768	bcp->bc_refcount = 1;
1769	list_add_tail(&bcp->bc_list, bucket);
1770
1771	trace_xfs_log_recover_buf_cancel_add(log, buf_f);
1772	return 0;
1773}
1774
1775/*
1776 * Check to see whether the buffer being recovered has a corresponding
1777 * entry in the buffer cancel record table. If it is, return the cancel
1778 * buffer structure to the caller.
1779 */
1780STATIC struct xfs_buf_cancel *
1781xlog_peek_buffer_cancelled(
1782	struct xlog		*log,
1783	xfs_daddr_t		blkno,
1784	uint			len,
1785	ushort			flags)
1786{
1787	struct list_head	*bucket;
1788	struct xfs_buf_cancel	*bcp;
1789
1790	if (!log->l_buf_cancel_table) {
1791		/* empty table means no cancelled buffers in the log */
1792		ASSERT(!(flags & XFS_BLF_CANCEL));
1793		return NULL;
1794	}
1795
1796	bucket = XLOG_BUF_CANCEL_BUCKET(log, blkno);
1797	list_for_each_entry(bcp, bucket, bc_list) {
1798		if (bcp->bc_blkno == blkno && bcp->bc_len == len)
1799			return bcp;
1800	}
1801
1802	/*
1803	 * We didn't find a corresponding entry in the table, so return 0 so
1804	 * that the buffer is NOT cancelled.
1805	 */
1806	ASSERT(!(flags & XFS_BLF_CANCEL));
1807	return NULL;
1808}
1809
1810/*
1811 * If the buffer is being cancelled then return 1 so that it will be cancelled,
1812 * otherwise return 0.  If the buffer is actually a buffer cancel item
1813 * (XFS_BLF_CANCEL is set), then decrement the refcount on the entry in the
1814 * table and remove it from the table if this is the last reference.
1815 *
1816 * We remove the cancel record from the table when we encounter its last
1817 * occurrence in the log so that if the same buffer is re-used again after its
1818 * last cancellation we actually replay the changes made at that point.
1819 */
1820STATIC int
1821xlog_check_buffer_cancelled(
1822	struct xlog		*log,
1823	xfs_daddr_t		blkno,
1824	uint			len,
1825	ushort			flags)
1826{
1827	struct xfs_buf_cancel	*bcp;
1828
1829	bcp = xlog_peek_buffer_cancelled(log, blkno, len, flags);
1830	if (!bcp)
1831		return 0;
1832
1833	/*
1834	 * We've go a match, so return 1 so that the recovery of this buffer
1835	 * is cancelled.  If this buffer is actually a buffer cancel log
1836	 * item, then decrement the refcount on the one in the table and
1837	 * remove it if this is the last reference.
1838	 */
1839	if (flags & XFS_BLF_CANCEL) {
1840		if (--bcp->bc_refcount == 0) {
1841			list_del(&bcp->bc_list);
1842			kmem_free(bcp);
1843		}
1844	}
1845	return 1;
1846}
1847
1848/*
1849 * Perform recovery for a buffer full of inodes.  In these buffers, the only
1850 * data which should be recovered is that which corresponds to the
1851 * di_next_unlinked pointers in the on disk inode structures.  The rest of the
1852 * data for the inodes is always logged through the inodes themselves rather
1853 * than the inode buffer and is recovered in xlog_recover_inode_pass2().
1854 *
1855 * The only time when buffers full of inodes are fully recovered is when the
1856 * buffer is full of newly allocated inodes.  In this case the buffer will
1857 * not be marked as an inode buffer and so will be sent to
1858 * xlog_recover_do_reg_buffer() below during recovery.
1859 */
1860STATIC int
1861xlog_recover_do_inode_buffer(
1862	struct xfs_mount	*mp,
1863	xlog_recover_item_t	*item,
1864	struct xfs_buf		*bp,
1865	xfs_buf_log_format_t	*buf_f)
1866{
1867	int			i;
1868	int			item_index = 0;
1869	int			bit = 0;
1870	int			nbits = 0;
1871	int			reg_buf_offset = 0;
1872	int			reg_buf_bytes = 0;
1873	int			next_unlinked_offset;
1874	int			inodes_per_buf;
1875	xfs_agino_t		*logged_nextp;
1876	xfs_agino_t		*buffer_nextp;
1877
1878	trace_xfs_log_recover_buf_inode_buf(mp->m_log, buf_f);
1879
1880	/*
1881	 * Post recovery validation only works properly on CRC enabled
1882	 * filesystems.
1883	 */
1884	if (xfs_sb_version_hascrc(&mp->m_sb))
1885		bp->b_ops = &xfs_inode_buf_ops;
1886
1887	inodes_per_buf = BBTOB(bp->b_io_length) >> mp->m_sb.sb_inodelog;
1888	for (i = 0; i < inodes_per_buf; i++) {
1889		next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
1890			offsetof(xfs_dinode_t, di_next_unlinked);
1891
1892		while (next_unlinked_offset >=
1893		       (reg_buf_offset + reg_buf_bytes)) {
1894			/*
1895			 * The next di_next_unlinked field is beyond
1896			 * the current logged region.  Find the next
1897			 * logged region that contains or is beyond
1898			 * the current di_next_unlinked field.
1899			 */
1900			bit += nbits;
1901			bit = xfs_next_bit(buf_f->blf_data_map,
1902					   buf_f->blf_map_size, bit);
1903
1904			/*
1905			 * If there are no more logged regions in the
1906			 * buffer, then we're done.
1907			 */
1908			if (bit == -1)
1909				return 0;
1910
1911			nbits = xfs_contig_bits(buf_f->blf_data_map,
1912						buf_f->blf_map_size, bit);
1913			ASSERT(nbits > 0);
1914			reg_buf_offset = bit << XFS_BLF_SHIFT;
1915			reg_buf_bytes = nbits << XFS_BLF_SHIFT;
1916			item_index++;
1917		}
1918
1919		/*
1920		 * If the current logged region starts after the current
1921		 * di_next_unlinked field, then move on to the next
1922		 * di_next_unlinked field.
1923		 */
1924		if (next_unlinked_offset < reg_buf_offset)
1925			continue;
1926
1927		ASSERT(item->ri_buf[item_index].i_addr != NULL);
1928		ASSERT((item->ri_buf[item_index].i_len % XFS_BLF_CHUNK) == 0);
1929		ASSERT((reg_buf_offset + reg_buf_bytes) <=
1930							BBTOB(bp->b_io_length));
1931
1932		/*
1933		 * The current logged region contains a copy of the
1934		 * current di_next_unlinked field.  Extract its value
1935		 * and copy it to the buffer copy.
1936		 */
1937		logged_nextp = item->ri_buf[item_index].i_addr +
1938				next_unlinked_offset - reg_buf_offset;
1939		if (unlikely(*logged_nextp == 0)) {
1940			xfs_alert(mp,
1941		"Bad inode buffer log record (ptr = 0x%p, bp = 0x%p). "
1942		"Trying to replay bad (0) inode di_next_unlinked field.",
1943				item, bp);
1944			XFS_ERROR_REPORT("xlog_recover_do_inode_buf",
1945					 XFS_ERRLEVEL_LOW, mp);
1946			return XFS_ERROR(EFSCORRUPTED);
1947		}
1948
1949		buffer_nextp = (xfs_agino_t *)xfs_buf_offset(bp,
1950					      next_unlinked_offset);
1951		*buffer_nextp = *logged_nextp;
1952
1953		/*
1954		 * If necessary, recalculate the CRC in the on-disk inode. We
1955		 * have to leave the inode in a consistent state for whoever
1956		 * reads it next....
1957		 */
1958		xfs_dinode_calc_crc(mp, (struct xfs_dinode *)
1959				xfs_buf_offset(bp, i * mp->m_sb.sb_inodesize));
1960
 
 
 
 
 
1961	}
1962
1963	return 0;
1964}
1965
1966/*
1967 * V5 filesystems know the age of the buffer on disk being recovered. We can
1968 * have newer objects on disk than we are replaying, and so for these cases we
1969 * don't want to replay the current change as that will make the buffer contents
1970 * temporarily invalid on disk.
1971 *
1972 * The magic number might not match the buffer type we are going to recover
1973 * (e.g. reallocated blocks), so we ignore the xfs_buf_log_format flags.  Hence
1974 * extract the LSN of the existing object in the buffer based on it's current
1975 * magic number.  If we don't recognise the magic number in the buffer, then
1976 * return a LSN of -1 so that the caller knows it was an unrecognised block and
1977 * so can recover the buffer.
1978 *
1979 * Note: we cannot rely solely on magic number matches to determine that the
1980 * buffer has a valid LSN - we also need to verify that it belongs to this
1981 * filesystem, so we need to extract the object's LSN and compare it to that
1982 * which we read from the superblock. If the UUIDs don't match, then we've got a
1983 * stale metadata block from an old filesystem instance that we need to recover
1984 * over the top of.
1985 */
1986static xfs_lsn_t
1987xlog_recover_get_buf_lsn(
1988	struct xfs_mount	*mp,
1989	struct xfs_buf		*bp)
 
 
1990{
1991	__uint32_t		magic32;
1992	__uint16_t		magic16;
1993	__uint16_t		magicda;
1994	void			*blk = bp->b_addr;
1995	uuid_t			*uuid;
1996	xfs_lsn_t		lsn = -1;
1997
1998	/* v4 filesystems always recover immediately */
1999	if (!xfs_sb_version_hascrc(&mp->m_sb))
2000		goto recover_immediately;
2001
2002	magic32 = be32_to_cpu(*(__be32 *)blk);
2003	switch (magic32) {
2004	case XFS_ABTB_CRC_MAGIC:
2005	case XFS_ABTC_CRC_MAGIC:
2006	case XFS_ABTB_MAGIC:
2007	case XFS_ABTC_MAGIC:
2008	case XFS_IBT_CRC_MAGIC:
2009	case XFS_IBT_MAGIC: {
2010		struct xfs_btree_block *btb = blk;
2011
2012		lsn = be64_to_cpu(btb->bb_u.s.bb_lsn);
2013		uuid = &btb->bb_u.s.bb_uuid;
2014		break;
2015	}
2016	case XFS_BMAP_CRC_MAGIC:
2017	case XFS_BMAP_MAGIC: {
2018		struct xfs_btree_block *btb = blk;
2019
2020		lsn = be64_to_cpu(btb->bb_u.l.bb_lsn);
2021		uuid = &btb->bb_u.l.bb_uuid;
2022		break;
2023	}
2024	case XFS_AGF_MAGIC:
2025		lsn = be64_to_cpu(((struct xfs_agf *)blk)->agf_lsn);
2026		uuid = &((struct xfs_agf *)blk)->agf_uuid;
2027		break;
2028	case XFS_AGFL_MAGIC:
2029		lsn = be64_to_cpu(((struct xfs_agfl *)blk)->agfl_lsn);
2030		uuid = &((struct xfs_agfl *)blk)->agfl_uuid;
2031		break;
2032	case XFS_AGI_MAGIC:
2033		lsn = be64_to_cpu(((struct xfs_agi *)blk)->agi_lsn);
2034		uuid = &((struct xfs_agi *)blk)->agi_uuid;
2035		break;
2036	case XFS_SYMLINK_MAGIC:
2037		lsn = be64_to_cpu(((struct xfs_dsymlink_hdr *)blk)->sl_lsn);
2038		uuid = &((struct xfs_dsymlink_hdr *)blk)->sl_uuid;
2039		break;
2040	case XFS_DIR3_BLOCK_MAGIC:
2041	case XFS_DIR3_DATA_MAGIC:
2042	case XFS_DIR3_FREE_MAGIC:
2043		lsn = be64_to_cpu(((struct xfs_dir3_blk_hdr *)blk)->lsn);
2044		uuid = &((struct xfs_dir3_blk_hdr *)blk)->uuid;
2045		break;
2046	case XFS_ATTR3_RMT_MAGIC:
2047		lsn = be64_to_cpu(((struct xfs_attr3_rmt_hdr *)blk)->rm_lsn);
2048		uuid = &((struct xfs_attr3_rmt_hdr *)blk)->rm_uuid;
2049		break;
2050	case XFS_SB_MAGIC:
2051		lsn = be64_to_cpu(((struct xfs_dsb *)blk)->sb_lsn);
2052		uuid = &((struct xfs_dsb *)blk)->sb_uuid;
2053		break;
2054	default:
2055		break;
2056	}
2057
2058	if (lsn != (xfs_lsn_t)-1) {
2059		if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2060			goto recover_immediately;
2061		return lsn;
2062	}
2063
2064	magicda = be16_to_cpu(((struct xfs_da_blkinfo *)blk)->magic);
2065	switch (magicda) {
2066	case XFS_DIR3_LEAF1_MAGIC:
2067	case XFS_DIR3_LEAFN_MAGIC:
2068	case XFS_DA3_NODE_MAGIC:
2069		lsn = be64_to_cpu(((struct xfs_da3_blkinfo *)blk)->lsn);
2070		uuid = &((struct xfs_da3_blkinfo *)blk)->uuid;
2071		break;
2072	default:
2073		break;
2074	}
2075
2076	if (lsn != (xfs_lsn_t)-1) {
2077		if (!uuid_equal(&mp->m_sb.sb_uuid, uuid))
2078			goto recover_immediately;
2079		return lsn;
2080	}
2081
2082	/*
2083	 * We do individual object checks on dquot and inode buffers as they
2084	 * have their own individual LSN records. Also, we could have a stale
2085	 * buffer here, so we have to at least recognise these buffer types.
2086	 *
2087	 * A notd complexity here is inode unlinked list processing - it logs
2088	 * the inode directly in the buffer, but we don't know which inodes have
2089	 * been modified, and there is no global buffer LSN. Hence we need to
2090	 * recover all inode buffer types immediately. This problem will be
2091	 * fixed by logical logging of the unlinked list modifications.
2092	 */
2093	magic16 = be16_to_cpu(*(__be16 *)blk);
2094	switch (magic16) {
2095	case XFS_DQUOT_MAGIC:
2096	case XFS_DINODE_MAGIC:
2097		goto recover_immediately;
2098	default:
2099		break;
2100	}
2101
2102	/* unknown buffer contents, recover immediately */
2103
2104recover_immediately:
2105	return (xfs_lsn_t)-1;
 
2106
2107}
 
2108
2109/*
2110 * Validate the recovered buffer is of the correct type and attach the
2111 * appropriate buffer operations to them for writeback. Magic numbers are in a
2112 * few places:
2113 *	the first 16 bits of the buffer (inode buffer, dquot buffer),
2114 *	the first 32 bits of the buffer (most blocks),
2115 *	inside a struct xfs_da_blkinfo at the start of the buffer.
2116 */
2117static void
2118xlog_recover_validate_buf_type(
2119	struct xfs_mount	*mp,
2120	struct xfs_buf		*bp,
2121	xfs_buf_log_format_t	*buf_f)
2122{
2123	struct xfs_da_blkinfo	*info = bp->b_addr;
2124	__uint32_t		magic32;
2125	__uint16_t		magic16;
2126	__uint16_t		magicda;
2127
2128	magic32 = be32_to_cpu(*(__be32 *)bp->b_addr);
2129	magic16 = be16_to_cpu(*(__be16*)bp->b_addr);
2130	magicda = be16_to_cpu(info->magic);
2131	switch (xfs_blft_from_flags(buf_f)) {
2132	case XFS_BLFT_BTREE_BUF:
2133		switch (magic32) {
2134		case XFS_ABTB_CRC_MAGIC:
2135		case XFS_ABTC_CRC_MAGIC:
2136		case XFS_ABTB_MAGIC:
2137		case XFS_ABTC_MAGIC:
2138			bp->b_ops = &xfs_allocbt_buf_ops;
2139			break;
2140		case XFS_IBT_CRC_MAGIC:
2141		case XFS_IBT_MAGIC:
2142			bp->b_ops = &xfs_inobt_buf_ops;
2143			break;
2144		case XFS_BMAP_CRC_MAGIC:
2145		case XFS_BMAP_MAGIC:
2146			bp->b_ops = &xfs_bmbt_buf_ops;
 
 
 
 
 
 
 
 
 
2147			break;
2148		default:
2149			xfs_warn(mp, "Bad btree block magic!");
2150			ASSERT(0);
2151			break;
2152		}
2153		break;
2154	case XFS_BLFT_AGF_BUF:
2155		if (magic32 != XFS_AGF_MAGIC) {
2156			xfs_warn(mp, "Bad AGF block magic!");
2157			ASSERT(0);
2158			break;
2159		}
2160		bp->b_ops = &xfs_agf_buf_ops;
2161		break;
2162	case XFS_BLFT_AGFL_BUF:
2163		if (!xfs_sb_version_hascrc(&mp->m_sb))
2164			break;
2165		if (magic32 != XFS_AGFL_MAGIC) {
2166			xfs_warn(mp, "Bad AGFL block magic!");
2167			ASSERT(0);
2168			break;
2169		}
2170		bp->b_ops = &xfs_agfl_buf_ops;
2171		break;
2172	case XFS_BLFT_AGI_BUF:
2173		if (magic32 != XFS_AGI_MAGIC) {
2174			xfs_warn(mp, "Bad AGI block magic!");
2175			ASSERT(0);
2176			break;
2177		}
2178		bp->b_ops = &xfs_agi_buf_ops;
2179		break;
2180	case XFS_BLFT_UDQUOT_BUF:
2181	case XFS_BLFT_PDQUOT_BUF:
2182	case XFS_BLFT_GDQUOT_BUF:
2183#ifdef CONFIG_XFS_QUOTA
2184		if (magic16 != XFS_DQUOT_MAGIC) {
2185			xfs_warn(mp, "Bad DQUOT block magic!");
2186			ASSERT(0);
2187			break;
2188		}
2189		bp->b_ops = &xfs_dquot_buf_ops;
2190#else
2191		xfs_alert(mp,
2192	"Trying to recover dquots without QUOTA support built in!");
2193		ASSERT(0);
2194#endif
2195		break;
2196	case XFS_BLFT_DINO_BUF:
2197		/*
2198		 * we get here with inode allocation buffers, not buffers that
2199		 * track unlinked list changes.
2200		 */
2201		if (magic16 != XFS_DINODE_MAGIC) {
2202			xfs_warn(mp, "Bad INODE block magic!");
2203			ASSERT(0);
2204			break;
2205		}
2206		bp->b_ops = &xfs_inode_buf_ops;
2207		break;
2208	case XFS_BLFT_SYMLINK_BUF:
2209		if (magic32 != XFS_SYMLINK_MAGIC) {
2210			xfs_warn(mp, "Bad symlink block magic!");
2211			ASSERT(0);
2212			break;
2213		}
2214		bp->b_ops = &xfs_symlink_buf_ops;
2215		break;
2216	case XFS_BLFT_DIR_BLOCK_BUF:
2217		if (magic32 != XFS_DIR2_BLOCK_MAGIC &&
2218		    magic32 != XFS_DIR3_BLOCK_MAGIC) {
2219			xfs_warn(mp, "Bad dir block magic!");
2220			ASSERT(0);
2221			break;
2222		}
2223		bp->b_ops = &xfs_dir3_block_buf_ops;
2224		break;
2225	case XFS_BLFT_DIR_DATA_BUF:
2226		if (magic32 != XFS_DIR2_DATA_MAGIC &&
2227		    magic32 != XFS_DIR3_DATA_MAGIC) {
2228			xfs_warn(mp, "Bad dir data magic!");
2229			ASSERT(0);
2230			break;
2231		}
2232		bp->b_ops = &xfs_dir3_data_buf_ops;
2233		break;
2234	case XFS_BLFT_DIR_FREE_BUF:
2235		if (magic32 != XFS_DIR2_FREE_MAGIC &&
2236		    magic32 != XFS_DIR3_FREE_MAGIC) {
2237			xfs_warn(mp, "Bad dir3 free magic!");
2238			ASSERT(0);
2239			break;
2240		}
2241		bp->b_ops = &xfs_dir3_free_buf_ops;
2242		break;
2243	case XFS_BLFT_DIR_LEAF1_BUF:
2244		if (magicda != XFS_DIR2_LEAF1_MAGIC &&
2245		    magicda != XFS_DIR3_LEAF1_MAGIC) {
2246			xfs_warn(mp, "Bad dir leaf1 magic!");
2247			ASSERT(0);
2248			break;
2249		}
2250		bp->b_ops = &xfs_dir3_leaf1_buf_ops;
2251		break;
2252	case XFS_BLFT_DIR_LEAFN_BUF:
2253		if (magicda != XFS_DIR2_LEAFN_MAGIC &&
2254		    magicda != XFS_DIR3_LEAFN_MAGIC) {
2255			xfs_warn(mp, "Bad dir leafn magic!");
2256			ASSERT(0);
2257			break;
2258		}
2259		bp->b_ops = &xfs_dir3_leafn_buf_ops;
2260		break;
2261	case XFS_BLFT_DA_NODE_BUF:
2262		if (magicda != XFS_DA_NODE_MAGIC &&
2263		    magicda != XFS_DA3_NODE_MAGIC) {
2264			xfs_warn(mp, "Bad da node magic!");
2265			ASSERT(0);
2266			break;
2267		}
2268		bp->b_ops = &xfs_da3_node_buf_ops;
2269		break;
2270	case XFS_BLFT_ATTR_LEAF_BUF:
2271		if (magicda != XFS_ATTR_LEAF_MAGIC &&
2272		    magicda != XFS_ATTR3_LEAF_MAGIC) {
2273			xfs_warn(mp, "Bad attr leaf magic!");
2274			ASSERT(0);
2275			break;
2276		}
2277		bp->b_ops = &xfs_attr3_leaf_buf_ops;
2278		break;
2279	case XFS_BLFT_ATTR_RMT_BUF:
2280		if (!xfs_sb_version_hascrc(&mp->m_sb))
2281			break;
2282		if (magic32 != XFS_ATTR3_RMT_MAGIC) {
2283			xfs_warn(mp, "Bad attr remote magic!");
2284			ASSERT(0);
2285			break;
2286		}
2287		bp->b_ops = &xfs_attr3_rmt_buf_ops;
2288		break;
2289	case XFS_BLFT_SB_BUF:
2290		if (magic32 != XFS_SB_MAGIC) {
2291			xfs_warn(mp, "Bad SB block magic!");
2292			ASSERT(0);
2293			break;
2294		}
2295		bp->b_ops = &xfs_sb_buf_ops;
2296		break;
2297	default:
2298		xfs_warn(mp, "Unknown buffer type %d!",
2299			 xfs_blft_from_flags(buf_f));
2300		break;
2301	}
2302}
2303
2304/*
2305 * Perform a 'normal' buffer recovery.  Each logged region of the
2306 * buffer should be copied over the corresponding region in the
2307 * given buffer.  The bitmap in the buf log format structure indicates
2308 * where to place the logged data.
2309 */
2310STATIC void
2311xlog_recover_do_reg_buffer(
2312	struct xfs_mount	*mp,
2313	xlog_recover_item_t	*item,
2314	struct xfs_buf		*bp,
2315	xfs_buf_log_format_t	*buf_f)
2316{
2317	int			i;
2318	int			bit;
2319	int			nbits;
2320	int                     error;
2321
2322	trace_xfs_log_recover_buf_reg_buf(mp->m_log, buf_f);
2323
2324	bit = 0;
2325	i = 1;  /* 0 is the buf format structure */
2326	while (1) {
2327		bit = xfs_next_bit(buf_f->blf_data_map,
2328				   buf_f->blf_map_size, bit);
2329		if (bit == -1)
2330			break;
2331		nbits = xfs_contig_bits(buf_f->blf_data_map,
2332					buf_f->blf_map_size, bit);
2333		ASSERT(nbits > 0);
2334		ASSERT(item->ri_buf[i].i_addr != NULL);
2335		ASSERT(item->ri_buf[i].i_len % XFS_BLF_CHUNK == 0);
2336		ASSERT(BBTOB(bp->b_io_length) >=
2337		       ((uint)bit << XFS_BLF_SHIFT) + (nbits << XFS_BLF_SHIFT));
2338
2339		/*
2340		 * The dirty regions logged in the buffer, even though
2341		 * contiguous, may span multiple chunks. This is because the
2342		 * dirty region may span a physical page boundary in a buffer
2343		 * and hence be split into two separate vectors for writing into
2344		 * the log. Hence we need to trim nbits back to the length of
2345		 * the current region being copied out of the log.
2346		 */
2347		if (item->ri_buf[i].i_len < (nbits << XFS_BLF_SHIFT))
2348			nbits = item->ri_buf[i].i_len >> XFS_BLF_SHIFT;
2349
2350		/*
2351		 * Do a sanity check if this is a dquot buffer. Just checking
2352		 * the first dquot in the buffer should do. XXXThis is
2353		 * probably a good thing to do for other buf types also.
2354		 */
2355		error = 0;
2356		if (buf_f->blf_flags &
2357		   (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2358			if (item->ri_buf[i].i_addr == NULL) {
2359				xfs_alert(mp,
2360					"XFS: NULL dquot in %s.", __func__);
2361				goto next;
2362			}
2363			if (item->ri_buf[i].i_len < sizeof(xfs_disk_dquot_t)) {
2364				xfs_alert(mp,
2365					"XFS: dquot too small (%d) in %s.",
2366					item->ri_buf[i].i_len, __func__);
2367				goto next;
2368			}
2369			error = xfs_dqcheck(mp, item->ri_buf[i].i_addr,
2370					       -1, 0, XFS_QMOPT_DOWARN,
2371					       "dquot_buf_recover");
2372			if (error)
2373				goto next;
2374		}
2375
2376		memcpy(xfs_buf_offset(bp,
2377			(uint)bit << XFS_BLF_SHIFT),	/* dest */
2378			item->ri_buf[i].i_addr,		/* source */
2379			nbits<<XFS_BLF_SHIFT);		/* length */
2380 next:
2381		i++;
2382		bit += nbits;
2383	}
2384
2385	/* Shouldn't be any more regions */
2386	ASSERT(i == item->ri_total);
2387
2388	/*
2389	 * We can only do post recovery validation on items on CRC enabled
2390	 * fielsystems as we need to know when the buffer was written to be able
2391	 * to determine if we should have replayed the item. If we replay old
2392	 * metadata over a newer buffer, then it will enter a temporarily
2393	 * inconsistent state resulting in verification failures. Hence for now
2394	 * just avoid the verification stage for non-crc filesystems
2395	 */
2396	if (xfs_sb_version_hascrc(&mp->m_sb))
2397		xlog_recover_validate_buf_type(mp, bp, buf_f);
2398}
2399
2400/*
2401 * Perform a dquot buffer recovery.
2402 * Simple algorithm: if we have found a QUOTAOFF log item of the same type
2403 * (ie. USR or GRP), then just toss this buffer away; don't recover it.
2404 * Else, treat it as a regular buffer and do recovery.
2405 */
2406STATIC void
2407xlog_recover_do_dquot_buffer(
2408	struct xfs_mount		*mp,
2409	struct xlog			*log,
2410	struct xlog_recover_item	*item,
2411	struct xfs_buf			*bp,
2412	struct xfs_buf_log_format	*buf_f)
2413{
2414	uint			type;
2415
2416	trace_xfs_log_recover_buf_dquot_buf(log, buf_f);
2417
2418	/*
2419	 * Filesystems are required to send in quota flags at mount time.
2420	 */
2421	if (mp->m_qflags == 0) {
2422		return;
2423	}
2424
2425	type = 0;
2426	if (buf_f->blf_flags & XFS_BLF_UDQUOT_BUF)
2427		type |= XFS_DQ_USER;
2428	if (buf_f->blf_flags & XFS_BLF_PDQUOT_BUF)
2429		type |= XFS_DQ_PROJ;
2430	if (buf_f->blf_flags & XFS_BLF_GDQUOT_BUF)
2431		type |= XFS_DQ_GROUP;
2432	/*
2433	 * This type of quotas was turned off, so ignore this buffer
2434	 */
2435	if (log->l_quotaoffs_flag & type)
2436		return;
2437
2438	xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2439}
2440
2441/*
2442 * This routine replays a modification made to a buffer at runtime.
2443 * There are actually two types of buffer, regular and inode, which
2444 * are handled differently.  Inode buffers are handled differently
2445 * in that we only recover a specific set of data from them, namely
2446 * the inode di_next_unlinked fields.  This is because all other inode
2447 * data is actually logged via inode records and any data we replay
2448 * here which overlaps that may be stale.
2449 *
2450 * When meta-data buffers are freed at run time we log a buffer item
2451 * with the XFS_BLF_CANCEL bit set to indicate that previous copies
2452 * of the buffer in the log should not be replayed at recovery time.
2453 * This is so that if the blocks covered by the buffer are reused for
2454 * file data before we crash we don't end up replaying old, freed
2455 * meta-data into a user's file.
2456 *
2457 * To handle the cancellation of buffer log items, we make two passes
2458 * over the log during recovery.  During the first we build a table of
2459 * those buffers which have been cancelled, and during the second we
2460 * only replay those buffers which do not have corresponding cancel
2461 * records in the table.  See xlog_recover_buffer_pass[1,2] above
2462 * for more details on the implementation of the table of cancel records.
2463 */
2464STATIC int
2465xlog_recover_buffer_pass2(
2466	struct xlog			*log,
2467	struct list_head		*buffer_list,
2468	struct xlog_recover_item	*item,
2469	xfs_lsn_t			current_lsn)
2470{
2471	xfs_buf_log_format_t	*buf_f = item->ri_buf[0].i_addr;
2472	xfs_mount_t		*mp = log->l_mp;
2473	xfs_buf_t		*bp;
2474	int			error;
2475	uint			buf_flags;
2476	xfs_lsn_t		lsn;
2477
2478	/*
2479	 * In this pass we only want to recover all the buffers which have
2480	 * not been cancelled and are not cancellation buffers themselves.
2481	 */
2482	if (xlog_check_buffer_cancelled(log, buf_f->blf_blkno,
2483			buf_f->blf_len, buf_f->blf_flags)) {
2484		trace_xfs_log_recover_buf_cancel(log, buf_f);
2485		return 0;
2486	}
2487
2488	trace_xfs_log_recover_buf_recover(log, buf_f);
2489
2490	buf_flags = 0;
2491	if (buf_f->blf_flags & XFS_BLF_INODE_BUF)
2492		buf_flags |= XBF_UNMAPPED;
2493
2494	bp = xfs_buf_read(mp->m_ddev_targp, buf_f->blf_blkno, buf_f->blf_len,
2495			  buf_flags, NULL);
2496	if (!bp)
2497		return XFS_ERROR(ENOMEM);
2498	error = bp->b_error;
2499	if (error) {
2500		xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#1)");
2501		goto out_release;
2502	}
2503
2504	/*
2505	 * recover the buffer only if we get an LSN from it and it's less than
2506	 * the lsn of the transaction we are replaying.
2507	 */
2508	lsn = xlog_recover_get_buf_lsn(mp, bp);
2509	if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0)
2510		goto out_release;
2511
2512	if (buf_f->blf_flags & XFS_BLF_INODE_BUF) {
2513		error = xlog_recover_do_inode_buffer(mp, item, bp, buf_f);
2514	} else if (buf_f->blf_flags &
2515		  (XFS_BLF_UDQUOT_BUF|XFS_BLF_PDQUOT_BUF|XFS_BLF_GDQUOT_BUF)) {
2516		xlog_recover_do_dquot_buffer(mp, log, item, bp, buf_f);
2517	} else {
2518		xlog_recover_do_reg_buffer(mp, item, bp, buf_f);
2519	}
2520	if (error)
2521		goto out_release;
2522
2523	/*
2524	 * Perform delayed write on the buffer.  Asynchronous writes will be
2525	 * slower when taking into account all the buffers to be flushed.
2526	 *
2527	 * Also make sure that only inode buffers with good sizes stay in
2528	 * the buffer cache.  The kernel moves inodes in buffers of 1 block
2529	 * or mp->m_inode_cluster_size bytes, whichever is bigger.  The inode
2530	 * buffers in the log can be a different size if the log was generated
2531	 * by an older kernel using unclustered inode buffers or a newer kernel
2532	 * running with a different inode cluster size.  Regardless, if the
2533	 * the inode buffer size isn't MAX(blocksize, mp->m_inode_cluster_size)
2534	 * for *our* value of mp->m_inode_cluster_size, then we need to keep
2535	 * the buffer out of the buffer cache so that the buffer won't
2536	 * overlap with future reads of those inodes.
2537	 */
2538	if (XFS_DINODE_MAGIC ==
2539	    be16_to_cpu(*((__be16 *)xfs_buf_offset(bp, 0))) &&
2540	    (BBTOB(bp->b_io_length) != MAX(log->l_mp->m_sb.sb_blocksize,
2541			(__uint32_t)log->l_mp->m_inode_cluster_size))) {
2542		xfs_buf_stale(bp);
2543		error = xfs_bwrite(bp);
2544	} else {
2545		ASSERT(bp->b_target->bt_mount == mp);
2546		bp->b_iodone = xlog_recover_iodone;
2547		xfs_buf_delwri_queue(bp, buffer_list);
2548	}
2549
2550out_release:
2551	xfs_buf_relse(bp);
2552	return error;
 
 
 
 
 
2553}
2554
2555/*
2556 * Inode fork owner changes
 
 
 
2557 *
2558 * If we have been told that we have to reparent the inode fork, it's because an
2559 * extent swap operation on a CRC enabled filesystem has been done and we are
2560 * replaying it. We need to walk the BMBT of the appropriate fork and change the
2561 * owners of it.
2562 *
2563 * The complexity here is that we don't have an inode context to work with, so
2564 * after we've replayed the inode we need to instantiate one.  This is where the
2565 * fun begins.
2566 *
2567 * We are in the middle of log recovery, so we can't run transactions. That
2568 * means we cannot use cache coherent inode instantiation via xfs_iget(), as
2569 * that will result in the corresponding iput() running the inode through
2570 * xfs_inactive(). If we've just replayed an inode core that changes the link
2571 * count to zero (i.e. it's been unlinked), then xfs_inactive() will run
2572 * transactions (bad!).
2573 *
2574 * So, to avoid this, we instantiate an inode directly from the inode core we've
2575 * just recovered. We have the buffer still locked, and all we really need to
2576 * instantiate is the inode core and the forks being modified. We can do this
2577 * manually, then run the inode btree owner change, and then tear down the
2578 * xfs_inode without having to run any transactions at all.
2579 *
2580 * Also, because we don't have a transaction context available here but need to
2581 * gather all the buffers we modify for writeback so we pass the buffer_list
2582 * instead for the operation to use.
2583 */
2584
2585STATIC int
2586xfs_recover_inode_owner_change(
2587	struct xfs_mount	*mp,
2588	struct xfs_dinode	*dip,
2589	struct xfs_inode_log_format *in_f,
2590	struct list_head	*buffer_list)
2591{
2592	struct xfs_inode	*ip;
2593	int			error;
2594
2595	ASSERT(in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER));
2596
2597	ip = xfs_inode_alloc(mp, in_f->ilf_ino);
2598	if (!ip)
2599		return ENOMEM;
2600
2601	/* instantiate the inode */
2602	xfs_dinode_from_disk(&ip->i_d, dip);
2603	ASSERT(ip->i_d.di_version >= 3);
2604
2605	error = xfs_iformat_fork(ip, dip);
2606	if (error)
2607		goto out_free_ip;
2608
2609
2610	if (in_f->ilf_fields & XFS_ILOG_DOWNER) {
2611		ASSERT(in_f->ilf_fields & XFS_ILOG_DBROOT);
2612		error = xfs_bmbt_change_owner(NULL, ip, XFS_DATA_FORK,
2613					      ip->i_ino, buffer_list);
2614		if (error)
2615			goto out_free_ip;
2616	}
2617
2618	if (in_f->ilf_fields & XFS_ILOG_AOWNER) {
2619		ASSERT(in_f->ilf_fields & XFS_ILOG_ABROOT);
2620		error = xfs_bmbt_change_owner(NULL, ip, XFS_ATTR_FORK,
2621					      ip->i_ino, buffer_list);
2622		if (error)
2623			goto out_free_ip;
2624	}
2625
2626out_free_ip:
2627	xfs_inode_free(ip);
2628	return error;
2629}
2630
2631STATIC int
2632xlog_recover_inode_pass2(
2633	struct xlog			*log,
2634	struct list_head		*buffer_list,
2635	struct xlog_recover_item	*item,
2636	xfs_lsn_t			current_lsn)
2637{
2638	xfs_inode_log_format_t	*in_f;
2639	xfs_mount_t		*mp = log->l_mp;
2640	xfs_buf_t		*bp;
2641	xfs_dinode_t		*dip;
2642	int			len;
2643	xfs_caddr_t		src;
2644	xfs_caddr_t		dest;
2645	int			error;
2646	int			attr_index;
2647	uint			fields;
2648	xfs_icdinode_t		*dicp;
2649	uint			isize;
2650	int			need_free = 0;
2651
2652	if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
2653		in_f = item->ri_buf[0].i_addr;
2654	} else {
2655		in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
2656		need_free = 1;
2657		error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
2658		if (error)
2659			goto error;
2660	}
2661
2662	/*
2663	 * Inode buffers can be freed, look out for it,
2664	 * and do not replay the inode.
2665	 */
2666	if (xlog_check_buffer_cancelled(log, in_f->ilf_blkno,
2667					in_f->ilf_len, 0)) {
2668		error = 0;
2669		trace_xfs_log_recover_inode_cancel(log, in_f);
2670		goto error;
2671	}
2672	trace_xfs_log_recover_inode_recover(log, in_f);
2673
2674	bp = xfs_buf_read(mp->m_ddev_targp, in_f->ilf_blkno, in_f->ilf_len, 0,
2675			  &xfs_inode_buf_ops);
2676	if (!bp) {
2677		error = ENOMEM;
2678		goto error;
2679	}
2680	error = bp->b_error;
2681	if (error) {
2682		xfs_buf_ioerror_alert(bp, "xlog_recover_do..(read#2)");
2683		goto out_release;
2684	}
2685	ASSERT(in_f->ilf_fields & XFS_ILOG_CORE);
2686	dip = (xfs_dinode_t *)xfs_buf_offset(bp, in_f->ilf_boffset);
2687
2688	/*
2689	 * Make sure the place we're flushing out to really looks
2690	 * like an inode!
2691	 */
2692	if (unlikely(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))) {
2693		xfs_alert(mp,
2694	"%s: Bad inode magic number, dip = 0x%p, dino bp = 0x%p, ino = %Ld",
2695			__func__, dip, bp, in_f->ilf_ino);
2696		XFS_ERROR_REPORT("xlog_recover_inode_pass2(1)",
2697				 XFS_ERRLEVEL_LOW, mp);
2698		error = EFSCORRUPTED;
2699		goto out_release;
2700	}
2701	dicp = item->ri_buf[1].i_addr;
2702	if (unlikely(dicp->di_magic != XFS_DINODE_MAGIC)) {
2703		xfs_alert(mp,
2704			"%s: Bad inode log record, rec ptr 0x%p, ino %Ld",
2705			__func__, item, in_f->ilf_ino);
2706		XFS_ERROR_REPORT("xlog_recover_inode_pass2(2)",
2707				 XFS_ERRLEVEL_LOW, mp);
2708		error = EFSCORRUPTED;
2709		goto out_release;
2710	}
2711
2712	/*
2713	 * If the inode has an LSN in it, recover the inode only if it's less
2714	 * than the lsn of the transaction we are replaying. Note: we still
2715	 * need to replay an owner change even though the inode is more recent
2716	 * than the transaction as there is no guarantee that all the btree
2717	 * blocks are more recent than this transaction, too.
2718	 */
2719	if (dip->di_version >= 3) {
2720		xfs_lsn_t	lsn = be64_to_cpu(dip->di_lsn);
2721
2722		if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
2723			trace_xfs_log_recover_inode_skip(log, in_f);
2724			error = 0;
2725			goto out_owner_change;
2726		}
2727	}
2728
2729	/*
2730	 * di_flushiter is only valid for v1/2 inodes. All changes for v3 inodes
2731	 * are transactional and if ordering is necessary we can determine that
2732	 * more accurately by the LSN field in the V3 inode core. Don't trust
2733	 * the inode versions we might be changing them here - use the
2734	 * superblock flag to determine whether we need to look at di_flushiter
2735	 * to skip replay when the on disk inode is newer than the log one
2736	 */
2737	if (!xfs_sb_version_hascrc(&mp->m_sb) &&
2738	    dicp->di_flushiter < be16_to_cpu(dip->di_flushiter)) {
2739		/*
2740		 * Deal with the wrap case, DI_MAX_FLUSH is less
2741		 * than smaller numbers
2742		 */
2743		if (be16_to_cpu(dip->di_flushiter) == DI_MAX_FLUSH &&
2744		    dicp->di_flushiter < (DI_MAX_FLUSH >> 1)) {
2745			/* do nothing */
2746		} else {
2747			trace_xfs_log_recover_inode_skip(log, in_f);
2748			error = 0;
2749			goto out_release;
2750		}
2751	}
2752
2753	/* Take the opportunity to reset the flush iteration count */
2754	dicp->di_flushiter = 0;
2755
2756	if (unlikely(S_ISREG(dicp->di_mode))) {
2757		if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2758		    (dicp->di_format != XFS_DINODE_FMT_BTREE)) {
2759			XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(3)",
2760					 XFS_ERRLEVEL_LOW, mp, dicp);
2761			xfs_alert(mp,
2762		"%s: Bad regular inode log record, rec ptr 0x%p, "
2763		"ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2764				__func__, item, dip, bp, in_f->ilf_ino);
2765			error = EFSCORRUPTED;
2766			goto out_release;
2767		}
2768	} else if (unlikely(S_ISDIR(dicp->di_mode))) {
2769		if ((dicp->di_format != XFS_DINODE_FMT_EXTENTS) &&
2770		    (dicp->di_format != XFS_DINODE_FMT_BTREE) &&
2771		    (dicp->di_format != XFS_DINODE_FMT_LOCAL)) {
2772			XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(4)",
2773					     XFS_ERRLEVEL_LOW, mp, dicp);
2774			xfs_alert(mp,
2775		"%s: Bad dir inode log record, rec ptr 0x%p, "
2776		"ino ptr = 0x%p, ino bp = 0x%p, ino %Ld",
2777				__func__, item, dip, bp, in_f->ilf_ino);
2778			error = EFSCORRUPTED;
2779			goto out_release;
2780		}
2781	}
2782	if (unlikely(dicp->di_nextents + dicp->di_anextents > dicp->di_nblocks)){
2783		XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(5)",
2784				     XFS_ERRLEVEL_LOW, mp, dicp);
2785		xfs_alert(mp,
2786	"%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2787	"dino bp 0x%p, ino %Ld, total extents = %d, nblocks = %Ld",
2788			__func__, item, dip, bp, in_f->ilf_ino,
2789			dicp->di_nextents + dicp->di_anextents,
2790			dicp->di_nblocks);
2791		error = EFSCORRUPTED;
2792		goto out_release;
2793	}
2794	if (unlikely(dicp->di_forkoff > mp->m_sb.sb_inodesize)) {
2795		XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(6)",
2796				     XFS_ERRLEVEL_LOW, mp, dicp);
2797		xfs_alert(mp,
2798	"%s: Bad inode log record, rec ptr 0x%p, dino ptr 0x%p, "
2799	"dino bp 0x%p, ino %Ld, forkoff 0x%x", __func__,
2800			item, dip, bp, in_f->ilf_ino, dicp->di_forkoff);
2801		error = EFSCORRUPTED;
2802		goto out_release;
2803	}
2804	isize = xfs_icdinode_size(dicp->di_version);
2805	if (unlikely(item->ri_buf[1].i_len > isize)) {
2806		XFS_CORRUPTION_ERROR("xlog_recover_inode_pass2(7)",
2807				     XFS_ERRLEVEL_LOW, mp, dicp);
2808		xfs_alert(mp,
2809			"%s: Bad inode log record length %d, rec ptr 0x%p",
2810			__func__, item->ri_buf[1].i_len, item);
2811		error = EFSCORRUPTED;
2812		goto out_release;
2813	}
2814
2815	/* The core is in in-core format */
2816	xfs_dinode_to_disk(dip, dicp);
2817
2818	/* the rest is in on-disk format */
2819	if (item->ri_buf[1].i_len > isize) {
2820		memcpy((char *)dip + isize,
2821			item->ri_buf[1].i_addr + isize,
2822			item->ri_buf[1].i_len - isize);
2823	}
2824
2825	fields = in_f->ilf_fields;
2826	switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
2827	case XFS_ILOG_DEV:
2828		xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
2829		break;
2830	case XFS_ILOG_UUID:
2831		memcpy(XFS_DFORK_DPTR(dip),
2832		       &in_f->ilf_u.ilfu_uuid,
2833		       sizeof(uuid_t));
2834		break;
2835	}
2836
2837	if (in_f->ilf_size == 2)
2838		goto out_owner_change;
2839	len = item->ri_buf[2].i_len;
2840	src = item->ri_buf[2].i_addr;
2841	ASSERT(in_f->ilf_size <= 4);
2842	ASSERT((in_f->ilf_size == 3) || (fields & XFS_ILOG_AFORK));
2843	ASSERT(!(fields & XFS_ILOG_DFORK) ||
2844	       (len == in_f->ilf_dsize));
2845
2846	switch (fields & XFS_ILOG_DFORK) {
2847	case XFS_ILOG_DDATA:
2848	case XFS_ILOG_DEXT:
2849		memcpy(XFS_DFORK_DPTR(dip), src, len);
2850		break;
2851
2852	case XFS_ILOG_DBROOT:
2853		xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src, len,
2854				 (xfs_bmdr_block_t *)XFS_DFORK_DPTR(dip),
2855				 XFS_DFORK_DSIZE(dip, mp));
2856		break;
2857
2858	default:
2859		/*
2860		 * There are no data fork flags set.
 
 
2861		 */
2862		ASSERT((fields & XFS_ILOG_DFORK) == 0);
2863		break;
2864	}
2865
2866	/*
2867	 * If we logged any attribute data, recover it.  There may or
2868	 * may not have been any other non-core data logged in this
2869	 * transaction.
2870	 */
2871	if (in_f->ilf_fields & XFS_ILOG_AFORK) {
2872		if (in_f->ilf_fields & XFS_ILOG_DFORK) {
2873			attr_index = 3;
2874		} else {
2875			attr_index = 2;
2876		}
2877		len = item->ri_buf[attr_index].i_len;
2878		src = item->ri_buf[attr_index].i_addr;
2879		ASSERT(len == in_f->ilf_asize);
2880
2881		switch (in_f->ilf_fields & XFS_ILOG_AFORK) {
2882		case XFS_ILOG_ADATA:
2883		case XFS_ILOG_AEXT:
2884			dest = XFS_DFORK_APTR(dip);
2885			ASSERT(len <= XFS_DFORK_ASIZE(dip, mp));
2886			memcpy(dest, src, len);
2887			break;
2888
2889		case XFS_ILOG_ABROOT:
2890			dest = XFS_DFORK_APTR(dip);
2891			xfs_bmbt_to_bmdr(mp, (struct xfs_btree_block *)src,
2892					 len, (xfs_bmdr_block_t*)dest,
2893					 XFS_DFORK_ASIZE(dip, mp));
2894			break;
2895
2896		default:
2897			xfs_warn(log->l_mp, "%s: Invalid flag", __func__);
2898			ASSERT(0);
2899			error = EIO;
2900			goto out_release;
2901		}
2902	}
2903
2904out_owner_change:
2905	if (in_f->ilf_fields & (XFS_ILOG_DOWNER|XFS_ILOG_AOWNER))
2906		error = xfs_recover_inode_owner_change(mp, dip, in_f,
2907						       buffer_list);
2908	/* re-generate the checksum. */
2909	xfs_dinode_calc_crc(log->l_mp, dip);
2910
2911	ASSERT(bp->b_target->bt_mount == mp);
2912	bp->b_iodone = xlog_recover_iodone;
2913	xfs_buf_delwri_queue(bp, buffer_list);
2914
2915out_release:
2916	xfs_buf_relse(bp);
2917error:
2918	if (need_free)
2919		kmem_free(in_f);
2920	return XFS_ERROR(error);
2921}
2922
2923/*
2924 * Recover QUOTAOFF records. We simply make a note of it in the xlog
2925 * structure, so that we know not to do any dquot item or dquot buffer recovery,
2926 * of that type.
2927 */
2928STATIC int
2929xlog_recover_quotaoff_pass1(
2930	struct xlog			*log,
2931	struct xlog_recover_item	*item)
2932{
2933	xfs_qoff_logformat_t	*qoff_f = item->ri_buf[0].i_addr;
2934	ASSERT(qoff_f);
2935
2936	/*
2937	 * The logitem format's flag tells us if this was user quotaoff,
2938	 * group/project quotaoff or both.
2939	 */
2940	if (qoff_f->qf_flags & XFS_UQUOTA_ACCT)
2941		log->l_quotaoffs_flag |= XFS_DQ_USER;
2942	if (qoff_f->qf_flags & XFS_PQUOTA_ACCT)
2943		log->l_quotaoffs_flag |= XFS_DQ_PROJ;
2944	if (qoff_f->qf_flags & XFS_GQUOTA_ACCT)
2945		log->l_quotaoffs_flag |= XFS_DQ_GROUP;
2946
2947	return (0);
2948}
2949
2950/*
2951 * Recover a dquot record
2952 */
2953STATIC int
2954xlog_recover_dquot_pass2(
2955	struct xlog			*log,
2956	struct list_head		*buffer_list,
2957	struct xlog_recover_item	*item,
2958	xfs_lsn_t			current_lsn)
2959{
2960	xfs_mount_t		*mp = log->l_mp;
2961	xfs_buf_t		*bp;
2962	struct xfs_disk_dquot	*ddq, *recddq;
2963	int			error;
2964	xfs_dq_logformat_t	*dq_f;
2965	uint			type;
2966
2967
2968	/*
2969	 * Filesystems are required to send in quota flags at mount time.
2970	 */
2971	if (mp->m_qflags == 0)
2972		return (0);
2973
2974	recddq = item->ri_buf[1].i_addr;
2975	if (recddq == NULL) {
2976		xfs_alert(log->l_mp, "NULL dquot in %s.", __func__);
2977		return XFS_ERROR(EIO);
2978	}
2979	if (item->ri_buf[1].i_len < sizeof(xfs_disk_dquot_t)) {
2980		xfs_alert(log->l_mp, "dquot too small (%d) in %s.",
2981			item->ri_buf[1].i_len, __func__);
2982		return XFS_ERROR(EIO);
2983	}
2984
2985	/*
2986	 * This type of quotas was turned off, so ignore this record.
2987	 */
2988	type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
2989	ASSERT(type);
2990	if (log->l_quotaoffs_flag & type)
2991		return (0);
2992
2993	/*
2994	 * At this point we know that quota was _not_ turned off.
2995	 * Since the mount flags are not indicating to us otherwise, this
2996	 * must mean that quota is on, and the dquot needs to be replayed.
2997	 * Remember that we may not have fully recovered the superblock yet,
2998	 * so we can't do the usual trick of looking at the SB quota bits.
2999	 *
3000	 * The other possibility, of course, is that the quota subsystem was
3001	 * removed since the last mount - ENOSYS.
3002	 */
3003	dq_f = item->ri_buf[0].i_addr;
3004	ASSERT(dq_f);
3005	error = xfs_dqcheck(mp, recddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3006			   "xlog_recover_dquot_pass2 (log copy)");
3007	if (error)
3008		return XFS_ERROR(EIO);
3009	ASSERT(dq_f->qlf_len == 1);
3010
3011	error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp, dq_f->qlf_blkno,
3012				   XFS_FSB_TO_BB(mp, dq_f->qlf_len), 0, &bp,
3013				   NULL);
3014	if (error)
3015		return error;
3016
3017	ASSERT(bp);
3018	ddq = (xfs_disk_dquot_t *)xfs_buf_offset(bp, dq_f->qlf_boffset);
3019
3020	/*
3021	 * At least the magic num portion should be on disk because this
3022	 * was among a chunk of dquots created earlier, and we did some
3023	 * minimal initialization then.
3024	 */
3025	error = xfs_dqcheck(mp, ddq, dq_f->qlf_id, 0, XFS_QMOPT_DOWARN,
3026			   "xlog_recover_dquot_pass2");
3027	if (error) {
3028		xfs_buf_relse(bp);
3029		return XFS_ERROR(EIO);
3030	}
3031
3032	/*
3033	 * If the dquot has an LSN in it, recover the dquot only if it's less
3034	 * than the lsn of the transaction we are replaying.
3035	 */
3036	if (xfs_sb_version_hascrc(&mp->m_sb)) {
3037		struct xfs_dqblk *dqb = (struct xfs_dqblk *)ddq;
3038		xfs_lsn_t	lsn = be64_to_cpu(dqb->dd_lsn);
3039
3040		if (lsn && lsn != -1 && XFS_LSN_CMP(lsn, current_lsn) >= 0) {
3041			goto out_release;
3042		}
3043	}
3044
3045	memcpy(ddq, recddq, item->ri_buf[1].i_len);
3046	if (xfs_sb_version_hascrc(&mp->m_sb)) {
3047		xfs_update_cksum((char *)ddq, sizeof(struct xfs_dqblk),
3048				 XFS_DQUOT_CRC_OFF);
3049	}
3050
3051	ASSERT(dq_f->qlf_size == 2);
3052	ASSERT(bp->b_target->bt_mount == mp);
3053	bp->b_iodone = xlog_recover_iodone;
3054	xfs_buf_delwri_queue(bp, buffer_list);
3055
3056out_release:
3057	xfs_buf_relse(bp);
3058	return 0;
3059}
3060
3061/*
3062 * This routine is called to create an in-core extent free intent
3063 * item from the efi format structure which was logged on disk.
3064 * It allocates an in-core efi, copies the extents from the format
3065 * structure into it, and adds the efi to the AIL with the given
3066 * LSN.
3067 */
3068STATIC int
3069xlog_recover_efi_pass2(
3070	struct xlog			*log,
3071	struct xlog_recover_item	*item,
3072	xfs_lsn_t			lsn)
3073{
3074	int			error;
3075	xfs_mount_t		*mp = log->l_mp;
3076	xfs_efi_log_item_t	*efip;
3077	xfs_efi_log_format_t	*efi_formatp;
3078
3079	efi_formatp = item->ri_buf[0].i_addr;
3080
3081	efip = xfs_efi_init(mp, efi_formatp->efi_nextents);
3082	if ((error = xfs_efi_copy_format(&(item->ri_buf[0]),
3083					 &(efip->efi_format)))) {
3084		xfs_efi_item_free(efip);
3085		return error;
3086	}
3087	atomic_set(&efip->efi_next_extent, efi_formatp->efi_nextents);
3088
3089	spin_lock(&log->l_ailp->xa_lock);
3090	/*
3091	 * xfs_trans_ail_update() drops the AIL lock.
3092	 */
3093	xfs_trans_ail_update(log->l_ailp, &efip->efi_item, lsn);
3094	return 0;
3095}
3096
3097
3098/*
3099 * This routine is called when an efd format structure is found in
3100 * a committed transaction in the log.  It's purpose is to cancel
3101 * the corresponding efi if it was still in the log.  To do this
3102 * it searches the AIL for the efi with an id equal to that in the
3103 * efd format structure.  If we find it, we remove the efi from the
3104 * AIL and free it.
3105 */
3106STATIC int
3107xlog_recover_efd_pass2(
3108	struct xlog			*log,
3109	struct xlog_recover_item	*item)
3110{
3111	xfs_efd_log_format_t	*efd_formatp;
3112	xfs_efi_log_item_t	*efip = NULL;
3113	xfs_log_item_t		*lip;
3114	__uint64_t		efi_id;
3115	struct xfs_ail_cursor	cur;
3116	struct xfs_ail		*ailp = log->l_ailp;
3117
3118	efd_formatp = item->ri_buf[0].i_addr;
3119	ASSERT((item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_32_t) +
3120		((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_32_t)))) ||
3121	       (item->ri_buf[0].i_len == (sizeof(xfs_efd_log_format_64_t) +
3122		((efd_formatp->efd_nextents - 1) * sizeof(xfs_extent_64_t)))));
3123	efi_id = efd_formatp->efd_efi_id;
3124
3125	/*
3126	 * Search for the efi with the id in the efd format structure
3127	 * in the AIL.
3128	 */
3129	spin_lock(&ailp->xa_lock);
3130	lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3131	while (lip != NULL) {
3132		if (lip->li_type == XFS_LI_EFI) {
3133			efip = (xfs_efi_log_item_t *)lip;
3134			if (efip->efi_format.efi_id == efi_id) {
3135				/*
3136				 * xfs_trans_ail_delete() drops the
3137				 * AIL lock.
3138				 */
3139				xfs_trans_ail_delete(ailp, lip,
3140						     SHUTDOWN_CORRUPT_INCORE);
3141				xfs_efi_item_free(efip);
3142				spin_lock(&ailp->xa_lock);
3143				break;
3144			}
3145		}
3146		lip = xfs_trans_ail_cursor_next(ailp, &cur);
3147	}
3148	xfs_trans_ail_cursor_done(ailp, &cur);
3149	spin_unlock(&ailp->xa_lock);
3150
3151	return 0;
3152}
3153
3154/*
3155 * This routine is called when an inode create format structure is found in a
3156 * committed transaction in the log.  It's purpose is to initialise the inodes
3157 * being allocated on disk. This requires us to get inode cluster buffers that
3158 * match the range to be intialised, stamped with inode templates and written
3159 * by delayed write so that subsequent modifications will hit the cached buffer
3160 * and only need writing out at the end of recovery.
3161 */
3162STATIC int
3163xlog_recover_do_icreate_pass2(
3164	struct xlog		*log,
3165	struct list_head	*buffer_list,
3166	xlog_recover_item_t	*item)
3167{
3168	struct xfs_mount	*mp = log->l_mp;
3169	struct xfs_icreate_log	*icl;
3170	xfs_agnumber_t		agno;
3171	xfs_agblock_t		agbno;
3172	unsigned int		count;
3173	unsigned int		isize;
3174	xfs_agblock_t		length;
3175
3176	icl = (struct xfs_icreate_log *)item->ri_buf[0].i_addr;
3177	if (icl->icl_type != XFS_LI_ICREATE) {
3178		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad type");
3179		return EINVAL;
3180	}
3181
3182	if (icl->icl_size != 1) {
3183		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad icl size");
3184		return EINVAL;
3185	}
3186
3187	agno = be32_to_cpu(icl->icl_ag);
3188	if (agno >= mp->m_sb.sb_agcount) {
3189		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agno");
3190		return EINVAL;
3191	}
3192	agbno = be32_to_cpu(icl->icl_agbno);
3193	if (!agbno || agbno == NULLAGBLOCK || agbno >= mp->m_sb.sb_agblocks) {
3194		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad agbno");
3195		return EINVAL;
3196	}
3197	isize = be32_to_cpu(icl->icl_isize);
3198	if (isize != mp->m_sb.sb_inodesize) {
3199		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad isize");
3200		return EINVAL;
3201	}
3202	count = be32_to_cpu(icl->icl_count);
3203	if (!count) {
3204		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count");
3205		return EINVAL;
3206	}
3207	length = be32_to_cpu(icl->icl_length);
3208	if (!length || length >= mp->m_sb.sb_agblocks) {
3209		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad length");
3210		return EINVAL;
3211	}
3212
3213	/* existing allocation is fixed value */
3214	ASSERT(count == mp->m_ialloc_inos);
3215	ASSERT(length == mp->m_ialloc_blks);
3216	if (count != mp->m_ialloc_inos ||
3217	     length != mp->m_ialloc_blks) {
3218		xfs_warn(log->l_mp, "xlog_recover_do_icreate_trans: bad count 2");
3219		return EINVAL;
3220	}
3221
3222	/*
3223	 * Inode buffers can be freed. Do not replay the inode initialisation as
3224	 * we could be overwriting something written after this inode buffer was
3225	 * cancelled.
3226	 *
3227	 * XXX: we need to iterate all buffers and only init those that are not
3228	 * cancelled. I think that a more fine grained factoring of
3229	 * xfs_ialloc_inode_init may be appropriate here to enable this to be
3230	 * done easily.
3231	 */
3232	if (xlog_check_buffer_cancelled(log,
3233			XFS_AGB_TO_DADDR(mp, agno, agbno), length, 0))
3234		return 0;
3235
3236	xfs_ialloc_inode_init(mp, NULL, buffer_list, agno, agbno, length,
3237					be32_to_cpu(icl->icl_gen));
3238	return 0;
3239}
3240
3241/*
3242 * Free up any resources allocated by the transaction
3243 *
3244 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
3245 */
3246STATIC void
3247xlog_recover_free_trans(
3248	struct xlog_recover	*trans)
3249{
3250	xlog_recover_item_t	*item, *n;
3251	int			i;
3252
 
 
3253	list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
3254		/* Free the regions in the item. */
3255		list_del(&item->ri_list);
3256		for (i = 0; i < item->ri_cnt; i++)
3257			kmem_free(item->ri_buf[i].i_addr);
3258		/* Free the item itself */
3259		kmem_free(item->ri_buf);
3260		kmem_free(item);
3261	}
3262	/* Free the transaction recover structure */
3263	kmem_free(trans);
3264}
3265
3266STATIC void
3267xlog_recover_buffer_ra_pass2(
3268	struct xlog                     *log,
3269	struct xlog_recover_item        *item)
3270{
3271	struct xfs_buf_log_format	*buf_f = item->ri_buf[0].i_addr;
3272	struct xfs_mount		*mp = log->l_mp;
3273
3274	if (xlog_peek_buffer_cancelled(log, buf_f->blf_blkno,
3275			buf_f->blf_len, buf_f->blf_flags)) {
3276		return;
3277	}
3278
3279	xfs_buf_readahead(mp->m_ddev_targp, buf_f->blf_blkno,
3280				buf_f->blf_len, NULL);
3281}
3282
3283STATIC void
3284xlog_recover_inode_ra_pass2(
3285	struct xlog                     *log,
3286	struct xlog_recover_item        *item)
3287{
3288	struct xfs_inode_log_format	ilf_buf;
3289	struct xfs_inode_log_format	*ilfp;
3290	struct xfs_mount		*mp = log->l_mp;
3291	int			error;
3292
3293	if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
3294		ilfp = item->ri_buf[0].i_addr;
3295	} else {
3296		ilfp = &ilf_buf;
3297		memset(ilfp, 0, sizeof(*ilfp));
3298		error = xfs_inode_item_format_convert(&item->ri_buf[0], ilfp);
3299		if (error)
3300			return;
3301	}
3302
3303	if (xlog_peek_buffer_cancelled(log, ilfp->ilf_blkno, ilfp->ilf_len, 0))
3304		return;
3305
3306	xfs_buf_readahead(mp->m_ddev_targp, ilfp->ilf_blkno,
3307				ilfp->ilf_len, &xfs_inode_buf_ra_ops);
3308}
3309
3310STATIC void
3311xlog_recover_dquot_ra_pass2(
3312	struct xlog			*log,
3313	struct xlog_recover_item	*item)
 
 
 
 
 
 
 
 
3314{
3315	struct xfs_mount	*mp = log->l_mp;
3316	struct xfs_disk_dquot	*recddq;
3317	struct xfs_dq_logformat	*dq_f;
3318	uint			type;
3319
3320
3321	if (mp->m_qflags == 0)
3322		return;
3323
3324	recddq = item->ri_buf[1].i_addr;
3325	if (recddq == NULL)
3326		return;
3327	if (item->ri_buf[1].i_len < sizeof(struct xfs_disk_dquot))
3328		return;
3329
3330	type = recddq->d_flags & (XFS_DQ_USER | XFS_DQ_PROJ | XFS_DQ_GROUP);
3331	ASSERT(type);
3332	if (log->l_quotaoffs_flag & type)
3333		return;
3334
3335	dq_f = item->ri_buf[0].i_addr;
3336	ASSERT(dq_f);
3337	ASSERT(dq_f->qlf_len == 1);
3338
3339	xfs_buf_readahead(mp->m_ddev_targp, dq_f->qlf_blkno,
3340			  XFS_FSB_TO_BB(mp, dq_f->qlf_len), NULL);
3341}
 
3342
3343STATIC void
3344xlog_recover_ra_pass2(
3345	struct xlog			*log,
3346	struct xlog_recover_item	*item)
3347{
3348	switch (ITEM_TYPE(item)) {
3349	case XFS_LI_BUF:
3350		xlog_recover_buffer_ra_pass2(log, item);
 
 
 
 
3351		break;
3352	case XFS_LI_INODE:
3353		xlog_recover_inode_ra_pass2(log, item);
 
 
 
3354		break;
3355	case XFS_LI_DQUOT:
3356		xlog_recover_dquot_ra_pass2(log, item);
 
 
 
 
3357		break;
3358	case XFS_LI_EFI:
3359	case XFS_LI_EFD:
3360	case XFS_LI_QUOTAOFF:
3361	default:
 
 
 
3362		break;
3363	}
 
 
 
3364}
3365
3366STATIC int
3367xlog_recover_commit_pass1(
3368	struct xlog			*log,
3369	struct xlog_recover		*trans,
3370	struct xlog_recover_item	*item)
 
 
 
 
 
 
 
3371{
3372	trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS1);
 
 
3373
3374	switch (ITEM_TYPE(item)) {
3375	case XFS_LI_BUF:
3376		return xlog_recover_buffer_pass1(log, item);
3377	case XFS_LI_QUOTAOFF:
3378		return xlog_recover_quotaoff_pass1(log, item);
3379	case XFS_LI_INODE:
3380	case XFS_LI_EFI:
3381	case XFS_LI_EFD:
3382	case XFS_LI_DQUOT:
3383	case XFS_LI_ICREATE:
3384		/* nothing to do in pass 1 */
3385		return 0;
3386	default:
3387		xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3388			__func__, ITEM_TYPE(item));
3389		ASSERT(0);
3390		return XFS_ERROR(EIO);
3391	}
3392}
3393
3394STATIC int
3395xlog_recover_commit_pass2(
3396	struct xlog			*log,
3397	struct xlog_recover		*trans,
3398	struct list_head		*buffer_list,
3399	struct xlog_recover_item	*item)
3400{
3401	trace_xfs_log_recover_item_recover(log, trans, item, XLOG_RECOVER_PASS2);
3402
3403	switch (ITEM_TYPE(item)) {
3404	case XFS_LI_BUF:
3405		return xlog_recover_buffer_pass2(log, buffer_list, item,
3406						 trans->r_lsn);
3407	case XFS_LI_INODE:
3408		return xlog_recover_inode_pass2(log, buffer_list, item,
3409						 trans->r_lsn);
3410	case XFS_LI_EFI:
3411		return xlog_recover_efi_pass2(log, item, trans->r_lsn);
3412	case XFS_LI_EFD:
3413		return xlog_recover_efd_pass2(log, item);
3414	case XFS_LI_DQUOT:
3415		return xlog_recover_dquot_pass2(log, buffer_list, item,
3416						trans->r_lsn);
3417	case XFS_LI_ICREATE:
3418		return xlog_recover_do_icreate_pass2(log, buffer_list, item);
3419	case XFS_LI_QUOTAOFF:
3420		/* nothing to do in pass2 */
3421		return 0;
3422	default:
3423		xfs_warn(log->l_mp, "%s: invalid item type (%d)",
3424			__func__, ITEM_TYPE(item));
3425		ASSERT(0);
3426		return XFS_ERROR(EIO);
3427	}
3428}
3429
3430STATIC int
3431xlog_recover_items_pass2(
3432	struct xlog                     *log,
3433	struct xlog_recover             *trans,
3434	struct list_head                *buffer_list,
3435	struct list_head                *item_list)
3436{
3437	struct xlog_recover_item	*item;
3438	int				error = 0;
3439
3440	list_for_each_entry(item, item_list, ri_list) {
3441		error = xlog_recover_commit_pass2(log, trans,
3442					  buffer_list, item);
3443		if (error)
3444			return error;
3445	}
 
 
 
 
3446
3447	return error;
 
 
 
 
3448}
3449
3450/*
3451 * Perform the transaction.
3452 *
3453 * If the transaction modifies a buffer or inode, do it now.  Otherwise,
3454 * EFIs and EFDs get queued up by adding entries into the AIL for them.
3455 */
3456STATIC int
3457xlog_recover_commit_trans(
3458	struct xlog		*log,
3459	struct xlog_recover	*trans,
3460	int			pass)
 
 
 
 
 
3461{
3462	int				error = 0;
3463	int				error2;
3464	int				items_queued = 0;
3465	struct xlog_recover_item	*item;
3466	struct xlog_recover_item	*next;
3467	LIST_HEAD			(buffer_list);
3468	LIST_HEAD			(ra_list);
3469	LIST_HEAD			(done_list);
3470
3471	#define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
3472
3473	hlist_del(&trans->r_list);
3474
3475	error = xlog_recover_reorder_trans(log, trans, pass);
3476	if (error)
3477		return error;
3478
3479	list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
3480		switch (pass) {
3481		case XLOG_RECOVER_PASS1:
3482			error = xlog_recover_commit_pass1(log, trans, item);
3483			break;
3484		case XLOG_RECOVER_PASS2:
3485			xlog_recover_ra_pass2(log, item);
3486			list_move_tail(&item->ri_list, &ra_list);
3487			items_queued++;
3488			if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
3489				error = xlog_recover_items_pass2(log, trans,
3490						&buffer_list, &ra_list);
3491				list_splice_tail_init(&ra_list, &done_list);
3492				items_queued = 0;
3493			}
3494
3495			break;
3496		default:
3497			ASSERT(0);
3498		}
3499
3500		if (error)
3501			goto out;
 
 
 
 
 
3502	}
3503
3504out:
3505	if (!list_empty(&ra_list)) {
3506		if (!error)
3507			error = xlog_recover_items_pass2(log, trans,
3508					&buffer_list, &ra_list);
3509		list_splice_tail_init(&ra_list, &done_list);
 
 
3510	}
3511
3512	if (!list_empty(&done_list))
3513		list_splice_init(&done_list, &trans->r_itemq);
3514
3515	xlog_recover_free_trans(trans);
 
3516
3517	error2 = xfs_buf_delwri_submit(&buffer_list);
3518	return error ? error : error2;
3519}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3520
3521STATIC int
3522xlog_recover_unmount_trans(
3523	struct xlog		*log,
3524	struct xlog_recover	*trans)
3525{
3526	/* Do nothing now */
3527	xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
3528	return 0;
3529}
3530
3531/*
3532 * There are two valid states of the r_state field.  0 indicates that the
3533 * transaction structure is in a normal state.  We have either seen the
3534 * start of the transaction or the last operation we added was not a partial
3535 * operation.  If the last operation we added to the transaction was a
3536 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
3537 *
3538 * NOTE: skip LRs with 0 data length.
3539 */
3540STATIC int
3541xlog_recover_process_data(
3542	struct xlog		*log,
3543	struct hlist_head	rhash[],
3544	struct xlog_rec_header	*rhead,
3545	xfs_caddr_t		dp,
3546	int			pass)
 
3547{
3548	xfs_caddr_t		lp;
 
3549	int			num_logops;
3550	xlog_op_header_t	*ohead;
3551	xlog_recover_t		*trans;
3552	xlog_tid_t		tid;
3553	int			error;
3554	unsigned long		hash;
3555	uint			flags;
3556
3557	lp = dp + be32_to_cpu(rhead->h_len);
3558	num_logops = be32_to_cpu(rhead->h_num_logops);
3559
3560	/* check the log format matches our own - else we can't recover */
3561	if (xlog_header_check_recover(log->l_mp, rhead))
3562		return (XFS_ERROR(EIO));
3563
3564	while ((dp < lp) && num_logops) {
3565		ASSERT(dp + sizeof(xlog_op_header_t) <= lp);
3566		ohead = (xlog_op_header_t *)dp;
3567		dp += sizeof(xlog_op_header_t);
3568		if (ohead->oh_clientid != XFS_TRANSACTION &&
3569		    ohead->oh_clientid != XFS_LOG) {
3570			xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
3571					__func__, ohead->oh_clientid);
3572			ASSERT(0);
3573			return (XFS_ERROR(EIO));
3574		}
3575		tid = be32_to_cpu(ohead->oh_tid);
3576		hash = XLOG_RHASH(tid);
3577		trans = xlog_recover_find_tid(&rhash[hash], tid);
3578		if (trans == NULL) {		   /* not found; add new tid */
3579			if (ohead->oh_flags & XLOG_START_TRANS)
3580				xlog_recover_new_tid(&rhash[hash], tid,
3581					be64_to_cpu(rhead->h_lsn));
3582		} else {
3583			if (dp + be32_to_cpu(ohead->oh_len) > lp) {
3584				xfs_warn(log->l_mp, "%s: bad length 0x%x",
3585					__func__, be32_to_cpu(ohead->oh_len));
3586				WARN_ON(1);
3587				return (XFS_ERROR(EIO));
3588			}
3589			flags = ohead->oh_flags & ~XLOG_END_TRANS;
3590			if (flags & XLOG_WAS_CONT_TRANS)
3591				flags &= ~XLOG_CONTINUE_TRANS;
3592			switch (flags) {
3593			case XLOG_COMMIT_TRANS:
3594				error = xlog_recover_commit_trans(log,
3595								trans, pass);
3596				break;
3597			case XLOG_UNMOUNT_TRANS:
3598				error = xlog_recover_unmount_trans(log, trans);
3599				break;
3600			case XLOG_WAS_CONT_TRANS:
3601				error = xlog_recover_add_to_cont_trans(log,
3602						trans, dp,
3603						be32_to_cpu(ohead->oh_len));
3604				break;
3605			case XLOG_START_TRANS:
3606				xfs_warn(log->l_mp, "%s: bad transaction",
3607					__func__);
3608				ASSERT(0);
3609				error = XFS_ERROR(EIO);
3610				break;
3611			case 0:
3612			case XLOG_CONTINUE_TRANS:
3613				error = xlog_recover_add_to_trans(log, trans,
3614						dp, be32_to_cpu(ohead->oh_len));
3615				break;
3616			default:
3617				xfs_warn(log->l_mp, "%s: bad flag 0x%x",
3618					__func__, flags);
3619				ASSERT(0);
3620				error = XFS_ERROR(EIO);
3621				break;
3622			}
3623			if (error) {
3624				xlog_recover_free_trans(trans);
3625				return error;
3626			}
3627		}
 
 
 
 
 
 
 
3628		dp += be32_to_cpu(ohead->oh_len);
3629		num_logops--;
3630	}
3631	return 0;
3632}
3633
3634/*
3635 * Process an extent free intent item that was recovered from
3636 * the log.  We need to free the extents that it describes.
3637 */
3638STATIC int
3639xlog_recover_process_efi(
3640	xfs_mount_t		*mp,
3641	xfs_efi_log_item_t	*efip)
3642{
3643	xfs_efd_log_item_t	*efdp;
3644	xfs_trans_t		*tp;
3645	int			i;
3646	int			error = 0;
3647	xfs_extent_t		*extp;
3648	xfs_fsblock_t		startblock_fsb;
3649
3650	ASSERT(!test_bit(XFS_EFI_RECOVERED, &efip->efi_flags));
 
 
 
 
 
 
 
 
 
 
 
 
3651
3652	/*
3653	 * First check the validity of the extents described by the
3654	 * EFI.  If any are bad, then assume that all are bad and
3655	 * just toss the EFI.
3656	 */
3657	for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3658		extp = &(efip->efi_format.efi_extents[i]);
3659		startblock_fsb = XFS_BB_TO_FSB(mp,
3660				   XFS_FSB_TO_DADDR(mp, extp->ext_start));
3661		if ((startblock_fsb == 0) ||
3662		    (extp->ext_len == 0) ||
3663		    (startblock_fsb >= mp->m_sb.sb_dblocks) ||
3664		    (extp->ext_len >= mp->m_sb.sb_agblocks)) {
3665			/*
3666			 * This will pull the EFI from the AIL and
3667			 * free the memory associated with it.
3668			 */
3669			set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3670			xfs_efi_release(efip, efip->efi_format.efi_nextents);
3671			return XFS_ERROR(EIO);
3672		}
3673	}
3674
3675	tp = xfs_trans_alloc(mp, 0);
3676	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
3677	if (error)
3678		goto abort_error;
3679	efdp = xfs_trans_get_efd(tp, efip, efip->efi_format.efi_nextents);
3680
3681	for (i = 0; i < efip->efi_format.efi_nextents; i++) {
3682		extp = &(efip->efi_format.efi_extents[i]);
3683		error = xfs_free_extent(tp, extp->ext_start, extp->ext_len);
 
 
 
 
 
3684		if (error)
3685			goto abort_error;
3686		xfs_trans_log_efd_extent(tp, efdp, extp->ext_start,
3687					 extp->ext_len);
3688	}
3689
3690	set_bit(XFS_EFI_RECOVERED, &efip->efi_flags);
3691	error = xfs_trans_commit(tp, 0);
3692	return error;
3693
3694abort_error:
3695	xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3696	return error;
 
 
 
 
 
 
 
 
 
 
3697}
3698
3699/*
3700 * When this is called, all of the EFIs which did not have
3701 * corresponding EFDs should be in the AIL.  What we do now
3702 * is free the extents associated with each one.
3703 *
3704 * Since we process the EFIs in normal transactions, they
3705 * will be removed at some point after the commit.  This prevents
3706 * us from just walking down the list processing each one.
3707 * We'll use a flag in the EFI to skip those that we've already
3708 * processed and use the AIL iteration mechanism's generation
3709 * count to try to speed this up at least a bit.
3710 *
3711 * When we start, we know that the EFIs are the only things in
3712 * the AIL.  As we process them, however, other items are added
3713 * to the AIL.  Since everything added to the AIL must come after
3714 * everything already in the AIL, we stop processing as soon as
3715 * we see something other than an EFI in the AIL.
3716 */
3717STATIC int
3718xlog_recover_process_efis(
3719	struct xlog	*log)
3720{
3721	xfs_log_item_t		*lip;
3722	xfs_efi_log_item_t	*efip;
3723	int			error = 0;
3724	struct xfs_ail_cursor	cur;
3725	struct xfs_ail		*ailp;
3726
3727	ailp = log->l_ailp;
3728	spin_lock(&ailp->xa_lock);
3729	lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
3730	while (lip != NULL) {
3731		/*
3732		 * We're done when we see something other than an EFI.
3733		 * There should be no EFIs left in the AIL now.
3734		 */
3735		if (lip->li_type != XFS_LI_EFI) {
3736#ifdef DEBUG
3737			for (; lip; lip = xfs_trans_ail_cursor_next(ailp, &cur))
3738				ASSERT(lip->li_type != XFS_LI_EFI);
3739#endif
3740			break;
3741		}
 
3742
3743		/*
3744		 * Skip EFIs that we've already processed.
 
 
3745		 */
3746		efip = (xfs_efi_log_item_t *)lip;
3747		if (test_bit(XFS_EFI_RECOVERED, &efip->efi_flags)) {
3748			lip = xfs_trans_ail_cursor_next(ailp, &cur);
3749			continue;
3750		}
3751
3752		spin_unlock(&ailp->xa_lock);
3753		error = xlog_recover_process_efi(log->l_mp, efip);
3754		spin_lock(&ailp->xa_lock);
 
 
 
 
 
 
 
 
 
3755		if (error)
3756			goto out;
3757		lip = xfs_trans_ail_cursor_next(ailp, &cur);
3758	}
3759out:
3760	xfs_trans_ail_cursor_done(ailp, &cur);
3761	spin_unlock(&ailp->xa_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3762	return error;
3763}
3764
3765/*
3766 * This routine performs a transaction to null out a bad inode pointer
3767 * in an agi unlinked inode hash bucket.
3768 */
3769STATIC void
3770xlog_recover_clear_agi_bucket(
3771	xfs_mount_t	*mp,
3772	xfs_agnumber_t	agno,
3773	int		bucket)
3774{
3775	xfs_trans_t	*tp;
3776	xfs_agi_t	*agi;
3777	xfs_buf_t	*agibp;
3778	int		offset;
3779	int		error;
3780
3781	tp = xfs_trans_alloc(mp, XFS_TRANS_CLEAR_AGI_BUCKET);
3782	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_clearagi, 0, 0);
3783	if (error)
3784		goto out_abort;
3785
3786	error = xfs_read_agi(mp, tp, agno, &agibp);
3787	if (error)
3788		goto out_abort;
3789
3790	agi = XFS_BUF_TO_AGI(agibp);
3791	agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
3792	offset = offsetof(xfs_agi_t, agi_unlinked) +
3793		 (sizeof(xfs_agino_t) * bucket);
3794	xfs_trans_log_buf(tp, agibp, offset,
3795			  (offset + sizeof(xfs_agino_t) - 1));
3796
3797	error = xfs_trans_commit(tp, 0);
3798	if (error)
3799		goto out_error;
3800	return;
3801
3802out_abort:
3803	xfs_trans_cancel(tp, XFS_TRANS_ABORT);
3804out_error:
3805	xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__, agno);
 
3806	return;
3807}
3808
3809STATIC xfs_agino_t
3810xlog_recover_process_one_iunlink(
3811	struct xfs_mount		*mp,
3812	xfs_agnumber_t			agno,
3813	xfs_agino_t			agino,
3814	int				bucket)
3815{
3816	struct xfs_buf			*ibp;
3817	struct xfs_dinode		*dip;
3818	struct xfs_inode		*ip;
3819	xfs_ino_t			ino;
3820	int				error;
3821
3822	ino = XFS_AGINO_TO_INO(mp, agno, agino);
3823	error = xfs_iget(mp, NULL, ino, 0, 0, &ip);
3824	if (error)
3825		goto fail;
 
 
3826
3827	/*
3828	 * Get the on disk inode to find the next inode in the bucket.
3829	 */
3830	error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &ibp, 0, 0);
3831	if (error)
3832		goto fail_iput;
3833
3834	ASSERT(ip->i_d.di_nlink == 0);
3835	ASSERT(ip->i_d.di_mode != 0);
 
3836
3837	/* setup for the next pass */
3838	agino = be32_to_cpu(dip->di_next_unlinked);
3839	xfs_buf_relse(ibp);
 
 
 
 
 
 
 
 
 
 
3840
3841	/*
3842	 * Prevent any DMAPI event from being sent when the reference on
3843	 * the inode is dropped.
3844	 */
3845	ip->i_d.di_dmevmask = 0;
3846
3847	IRELE(ip);
3848	return agino;
3849
3850 fail_iput:
3851	IRELE(ip);
3852 fail:
3853	/*
3854	 * We can't read in the inode this bucket points to, or this inode
3855	 * is messed up.  Just ditch this bucket of inodes.  We will lose
3856	 * some inodes and space, but at least we won't hang.
3857	 *
3858	 * Call xlog_recover_clear_agi_bucket() to perform a transaction to
3859	 * clear the inode pointer in the bucket.
3860	 */
3861	xlog_recover_clear_agi_bucket(mp, agno, bucket);
3862	return NULLAGINO;
3863}
3864
3865/*
3866 * xlog_iunlink_recover
3867 *
3868 * This is called during recovery to process any inodes which
3869 * we unlinked but not freed when the system crashed.  These
3870 * inodes will be on the lists in the AGI blocks.  What we do
3871 * here is scan all the AGIs and fully truncate and free any
3872 * inodes found on the lists.  Each inode is removed from the
3873 * lists when it has been fully truncated and is freed.  The
3874 * freeing of the inode and its removal from the list must be
3875 * atomic.
 
 
 
 
 
 
 
 
 
 
 
3876 */
3877STATIC void
3878xlog_recover_process_iunlinks(
3879	struct xlog	*log)
3880{
3881	xfs_mount_t	*mp;
3882	xfs_agnumber_t	agno;
3883	xfs_agi_t	*agi;
3884	xfs_buf_t	*agibp;
3885	xfs_agino_t	agino;
3886	int		bucket;
3887	int		error;
3888	uint		mp_dmevmask;
3889
3890	mp = log->l_mp;
 
 
 
 
 
 
 
 
 
3891
3892	/*
3893	 * Prevent any DMAPI event from being sent while in this function.
 
 
 
 
 
3894	 */
3895	mp_dmevmask = mp->m_dmevmask;
3896	mp->m_dmevmask = 0;
3897
3898	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
3899		/*
3900		 * Find the agi for this ag.
3901		 */
3902		error = xfs_read_agi(mp, NULL, agno, &agibp);
3903		if (error) {
3904			/*
3905			 * AGI is b0rked. Don't process it.
3906			 *
3907			 * We should probably mark the filesystem as corrupt
3908			 * after we've recovered all the ag's we can....
3909			 */
3910			continue;
3911		}
3912		/*
3913		 * Unlock the buffer so that it can be acquired in the normal
3914		 * course of the transaction to truncate and free each inode.
3915		 * Because we are not racing with anyone else here for the AGI
3916		 * buffer, we don't even need to hold it locked to read the
3917		 * initial unlinked bucket entries out of the buffer. We keep
3918		 * buffer reference though, so that it stays pinned in memory
3919		 * while we need the buffer.
3920		 */
3921		agi = XFS_BUF_TO_AGI(agibp);
3922		xfs_buf_unlock(agibp);
3923
3924		for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
3925			agino = be32_to_cpu(agi->agi_unlinked[bucket]);
3926			while (agino != NULLAGINO) {
3927				agino = xlog_recover_process_one_iunlink(mp,
3928							agno, agino, bucket);
3929			}
3930		}
3931		xfs_buf_rele(agibp);
3932	}
3933
3934	mp->m_dmevmask = mp_dmevmask;
3935}
3936
3937/*
3938 * Upack the log buffer data and crc check it. If the check fails, issue a
3939 * warning if and only if the CRC in the header is non-zero. This makes the
3940 * check an advisory warning, and the zero CRC check will prevent failure
3941 * warnings from being emitted when upgrading the kernel from one that does not
3942 * add CRCs by default.
3943 *
3944 * When filesystems are CRC enabled, this CRC mismatch becomes a fatal log
3945 * corruption failure
3946 */
3947STATIC int
3948xlog_unpack_data_crc(
3949	struct xlog_rec_header	*rhead,
3950	xfs_caddr_t		dp,
3951	struct xlog		*log)
3952{
3953	__le32			crc;
3954
3955	crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
3956	if (crc != rhead->h_crc) {
3957		if (rhead->h_crc || xfs_sb_version_hascrc(&log->l_mp->m_sb)) {
3958			xfs_alert(log->l_mp,
3959		"log record CRC mismatch: found 0x%x, expected 0x%x.",
3960					le32_to_cpu(rhead->h_crc),
3961					le32_to_cpu(crc));
3962			xfs_hex_dump(dp, 32);
3963		}
3964
3965		/*
3966		 * If we've detected a log record corruption, then we can't
3967		 * recover past this point. Abort recovery if we are enforcing
3968		 * CRC protection by punting an error back up the stack.
3969		 */
3970		if (xfs_sb_version_hascrc(&log->l_mp->m_sb))
3971			return EFSCORRUPTED;
3972	}
3973
3974	return 0;
 
3975}
3976
3977STATIC int
3978xlog_unpack_data(
3979	struct xlog_rec_header	*rhead,
3980	xfs_caddr_t		dp,
3981	struct xlog		*log)
3982{
3983	int			i, j, k;
3984	int			error;
3985
3986	error = xlog_unpack_data_crc(rhead, dp, log);
3987	if (error)
3988		return error;
3989
3990	for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
3991		  i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
3992		*(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
3993		dp += BBSIZE;
3994	}
3995
3996	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
3997		xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
3998		for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
3999			j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4000			k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
4001			*(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
4002			dp += BBSIZE;
4003		}
4004	}
 
4005
4006	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4007}
4008
4009STATIC int
4010xlog_valid_rec_header(
4011	struct xlog		*log,
4012	struct xlog_rec_header	*rhead,
4013	xfs_daddr_t		blkno)
 
4014{
4015	int			hlen;
4016
4017	if (unlikely(rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))) {
4018		XFS_ERROR_REPORT("xlog_valid_rec_header(1)",
4019				XFS_ERRLEVEL_LOW, log->l_mp);
4020		return XFS_ERROR(EFSCORRUPTED);
4021	}
4022	if (unlikely(
4023	    (!rhead->h_version ||
4024	    (be32_to_cpu(rhead->h_version) & (~XLOG_VERSION_OKBITS))))) {
4025		xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
4026			__func__, be32_to_cpu(rhead->h_version));
4027		return XFS_ERROR(EIO);
4028	}
4029
4030	/* LR body must have data or it wouldn't have been written */
 
 
 
4031	hlen = be32_to_cpu(rhead->h_len);
4032	if (unlikely( hlen <= 0 || hlen > INT_MAX )) {
4033		XFS_ERROR_REPORT("xlog_valid_rec_header(2)",
4034				XFS_ERRLEVEL_LOW, log->l_mp);
4035		return XFS_ERROR(EFSCORRUPTED);
4036	}
4037	if (unlikely( blkno > log->l_logBBsize || blkno > INT_MAX )) {
4038		XFS_ERROR_REPORT("xlog_valid_rec_header(3)",
4039				XFS_ERRLEVEL_LOW, log->l_mp);
4040		return XFS_ERROR(EFSCORRUPTED);
4041	}
4042	return 0;
4043}
4044
4045/*
4046 * Read the log from tail to head and process the log records found.
4047 * Handle the two cases where the tail and head are in the same cycle
4048 * and where the active portion of the log wraps around the end of
4049 * the physical log separately.  The pass parameter is passed through
4050 * to the routines called to process the data and is not looked at
4051 * here.
4052 */
4053STATIC int
4054xlog_do_recovery_pass(
4055	struct xlog		*log,
4056	xfs_daddr_t		head_blk,
4057	xfs_daddr_t		tail_blk,
4058	int			pass)
 
4059{
4060	xlog_rec_header_t	*rhead;
4061	xfs_daddr_t		blk_no;
4062	xfs_caddr_t		offset;
4063	xfs_buf_t		*hbp, *dbp;
4064	int			error = 0, h_size;
 
 
4065	int			bblks, split_bblks;
4066	int			hblks, split_hblks, wrapped_hblks;
 
4067	struct hlist_head	rhash[XLOG_RHASH_SIZE];
 
4068
4069	ASSERT(head_blk != tail_blk);
 
 
 
 
 
 
 
 
4070
4071	/*
4072	 * Read the header of the tail block and get the iclog buffer size from
4073	 * h_size.  Use this to tell how many sectors make up the log header.
4074	 */
4075	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
4076		/*
4077		 * When using variable length iclogs, read first sector of
4078		 * iclog header and extract the header size from it.  Get a
4079		 * new hbp that is the correct size.
4080		 */
4081		hbp = xlog_get_bp(log, 1);
4082		if (!hbp)
4083			return ENOMEM;
4084
4085		error = xlog_bread(log, tail_blk, 1, hbp, &offset);
4086		if (error)
4087			goto bread_err1;
4088
4089		rhead = (xlog_rec_header_t *)offset;
4090		error = xlog_valid_rec_header(log, rhead, tail_blk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4091		if (error)
4092			goto bread_err1;
4093		h_size = be32_to_cpu(rhead->h_size);
4094		if ((be32_to_cpu(rhead->h_version) & XLOG_VERSION_2) &&
4095		    (h_size > XLOG_HEADER_CYCLE_SIZE)) {
4096			hblks = h_size / XLOG_HEADER_CYCLE_SIZE;
4097			if (h_size % XLOG_HEADER_CYCLE_SIZE)
4098				hblks++;
4099			xlog_put_bp(hbp);
4100			hbp = xlog_get_bp(log, hblks);
4101		} else {
4102			hblks = 1;
 
 
 
 
 
 
4103		}
4104	} else {
4105		ASSERT(log->l_sectBBsize == 1);
4106		hblks = 1;
4107		hbp = xlog_get_bp(log, 1);
4108		h_size = XLOG_BIG_RECORD_BSIZE;
4109	}
4110
4111	if (!hbp)
4112		return ENOMEM;
4113	dbp = xlog_get_bp(log, BTOBB(h_size));
4114	if (!dbp) {
4115		xlog_put_bp(hbp);
4116		return ENOMEM;
4117	}
4118
4119	memset(rhash, 0, sizeof(rhash));
4120	if (tail_blk <= head_blk) {
4121		for (blk_no = tail_blk; blk_no < head_blk; ) {
4122			error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4123			if (error)
4124				goto bread_err2;
4125
4126			rhead = (xlog_rec_header_t *)offset;
4127			error = xlog_valid_rec_header(log, rhead, blk_no);
4128			if (error)
4129				goto bread_err2;
4130
4131			/* blocks in data section */
4132			bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4133			error = xlog_bread(log, blk_no + hblks, bblks, dbp,
4134					   &offset);
4135			if (error)
4136				goto bread_err2;
4137
4138			error = xlog_unpack_data(rhead, offset, log);
4139			if (error)
4140				goto bread_err2;
4141
4142			error = xlog_recover_process_data(log,
4143						rhash, rhead, offset, pass);
4144			if (error)
4145				goto bread_err2;
4146			blk_no += bblks + hblks;
4147		}
4148	} else {
4149		/*
4150		 * Perform recovery around the end of the physical log.
4151		 * When the head is not on the same cycle number as the tail,
4152		 * we can't do a sequential recovery as above.
4153		 */
4154		blk_no = tail_blk;
4155		while (blk_no < log->l_logBBsize) {
4156			/*
4157			 * Check for header wrapping around physical end-of-log
4158			 */
4159			offset = hbp->b_addr;
4160			split_hblks = 0;
4161			wrapped_hblks = 0;
4162			if (blk_no + hblks <= log->l_logBBsize) {
4163				/* Read header in one read */
4164				error = xlog_bread(log, blk_no, hblks, hbp,
4165						   &offset);
4166				if (error)
4167					goto bread_err2;
4168			} else {
4169				/* This LR is split across physical log end */
4170				if (blk_no != log->l_logBBsize) {
4171					/* some data before physical log end */
4172					ASSERT(blk_no <= INT_MAX);
4173					split_hblks = log->l_logBBsize - (int)blk_no;
4174					ASSERT(split_hblks > 0);
4175					error = xlog_bread(log, blk_no,
4176							   split_hblks, hbp,
4177							   &offset);
4178					if (error)
4179						goto bread_err2;
4180				}
4181
4182				/*
4183				 * Note: this black magic still works with
4184				 * large sector sizes (non-512) only because:
4185				 * - we increased the buffer size originally
4186				 *   by 1 sector giving us enough extra space
4187				 *   for the second read;
4188				 * - the log start is guaranteed to be sector
4189				 *   aligned;
4190				 * - we read the log end (LR header start)
4191				 *   _first_, then the log start (LR header end)
4192				 *   - order is important.
4193				 */
4194				wrapped_hblks = hblks - split_hblks;
4195				error = xlog_bread_offset(log, 0,
4196						wrapped_hblks, hbp,
4197						offset + BBTOB(split_hblks));
4198				if (error)
4199					goto bread_err2;
4200			}
4201			rhead = (xlog_rec_header_t *)offset;
4202			error = xlog_valid_rec_header(log, rhead,
4203						split_hblks ? blk_no : 0);
4204			if (error)
4205				goto bread_err2;
4206
4207			bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4208			blk_no += hblks;
4209
4210			/* Read in data for log record */
4211			if (blk_no + bblks <= log->l_logBBsize) {
4212				error = xlog_bread(log, blk_no, bblks, dbp,
 
 
 
 
 
 
 
 
4213						   &offset);
4214				if (error)
4215					goto bread_err2;
4216			} else {
4217				/* This log record is split across the
4218				 * physical end of log */
4219				offset = dbp->b_addr;
4220				split_bblks = 0;
4221				if (blk_no != log->l_logBBsize) {
4222					/* some data is before the physical
4223					 * end of log */
4224					ASSERT(!wrapped_hblks);
4225					ASSERT(blk_no <= INT_MAX);
4226					split_bblks =
4227						log->l_logBBsize - (int)blk_no;
4228					ASSERT(split_bblks > 0);
4229					error = xlog_bread(log, blk_no,
4230							split_bblks, dbp,
4231							&offset);
4232					if (error)
4233						goto bread_err2;
4234				}
4235
4236				/*
4237				 * Note: this black magic still works with
4238				 * large sector sizes (non-512) only because:
4239				 * - we increased the buffer size originally
4240				 *   by 1 sector giving us enough extra space
4241				 *   for the second read;
4242				 * - the log start is guaranteed to be sector
4243				 *   aligned;
4244				 * - we read the log end (LR header start)
4245				 *   _first_, then the log start (LR header end)
4246				 *   - order is important.
4247				 */
4248				error = xlog_bread_offset(log, 0,
4249						bblks - split_bblks, dbp,
4250						offset + BBTOB(split_bblks));
4251				if (error)
4252					goto bread_err2;
4253			}
4254
4255			error = xlog_unpack_data(rhead, offset, log);
 
4256			if (error)
4257				goto bread_err2;
4258
4259			error = xlog_recover_process_data(log, rhash,
4260							rhead, offset, pass);
4261			if (error)
4262				goto bread_err2;
4263			blk_no += bblks;
 
4264		}
4265
4266		ASSERT(blk_no >= log->l_logBBsize);
4267		blk_no -= log->l_logBBsize;
 
 
4268
4269		/* read first part of physical log */
4270		while (blk_no < head_blk) {
4271			error = xlog_bread(log, blk_no, hblks, hbp, &offset);
4272			if (error)
4273				goto bread_err2;
4274
4275			rhead = (xlog_rec_header_t *)offset;
4276			error = xlog_valid_rec_header(log, rhead, blk_no);
4277			if (error)
4278				goto bread_err2;
4279
4280			bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
4281			error = xlog_bread(log, blk_no+hblks, bblks, dbp,
4282					   &offset);
4283			if (error)
4284				goto bread_err2;
 
4285
4286			error = xlog_unpack_data(rhead, offset, log);
4287			if (error)
4288				goto bread_err2;
 
4289
4290			error = xlog_recover_process_data(log, rhash,
4291							rhead, offset, pass);
4292			if (error)
4293				goto bread_err2;
4294			blk_no += bblks + hblks;
4295		}
4296	}
4297
4298 bread_err2:
4299	xlog_put_bp(dbp);
4300 bread_err1:
4301	xlog_put_bp(hbp);
4302	return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4303}
4304
4305/*
4306 * Do the recovery of the log.  We actually do this in two phases.
4307 * The two passes are necessary in order to implement the function
4308 * of cancelling a record written into the log.  The first pass
4309 * determines those things which have been cancelled, and the
4310 * second pass replays log items normally except for those which
4311 * have been cancelled.  The handling of the replay and cancellations
4312 * takes place in the log item type specific routines.
4313 *
4314 * The table of items which have cancel records in the log is allocated
4315 * and freed at this level, since only here do we know when all of
4316 * the log recovery has been completed.
4317 */
4318STATIC int
4319xlog_do_log_recovery(
4320	struct xlog	*log,
4321	xfs_daddr_t	head_blk,
4322	xfs_daddr_t	tail_blk)
4323{
4324	int		error, i;
4325
4326	ASSERT(head_blk != tail_blk);
4327
4328	/*
4329	 * First do a pass to find all of the cancelled buf log items.
4330	 * Store them in the buf_cancel_table for use in the second pass.
4331	 */
4332	log->l_buf_cancel_table = kmem_zalloc(XLOG_BC_TABLE_SIZE *
4333						 sizeof(struct list_head),
4334						 KM_SLEEP);
4335	for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4336		INIT_LIST_HEAD(&log->l_buf_cancel_table[i]);
4337
4338	error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4339				      XLOG_RECOVER_PASS1);
4340	if (error != 0) {
4341		kmem_free(log->l_buf_cancel_table);
4342		log->l_buf_cancel_table = NULL;
4343		return error;
4344	}
4345	/*
4346	 * Then do a second pass to actually recover the items in the log.
4347	 * When it is complete free the table of buf cancel items.
4348	 */
4349	error = xlog_do_recovery_pass(log, head_blk, tail_blk,
4350				      XLOG_RECOVER_PASS2);
4351#ifdef DEBUG
4352	if (!error) {
4353		int	i;
4354
4355		for (i = 0; i < XLOG_BC_TABLE_SIZE; i++)
4356			ASSERT(list_empty(&log->l_buf_cancel_table[i]));
4357	}
4358#endif	/* DEBUG */
4359
4360	kmem_free(log->l_buf_cancel_table);
4361	log->l_buf_cancel_table = NULL;
4362
4363	return error;
4364}
4365
4366/*
4367 * Do the actual recovery
4368 */
4369STATIC int
4370xlog_do_recover(
4371	struct xlog	*log,
4372	xfs_daddr_t	head_blk,
4373	xfs_daddr_t	tail_blk)
4374{
4375	int		error;
4376	xfs_buf_t	*bp;
4377	xfs_sb_t	*sbp;
 
 
 
4378
4379	/*
4380	 * First replay the images in the log.
4381	 */
4382	error = xlog_do_log_recovery(log, head_blk, tail_blk);
4383	if (error)
4384		return error;
4385
4386	/*
4387	 * If IO errors happened during recovery, bail out.
4388	 */
4389	if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4390		return (EIO);
4391	}
4392
4393	/*
4394	 * We now update the tail_lsn since much of the recovery has completed
4395	 * and there may be space available to use.  If there were no extent
4396	 * or iunlinks, we can free up the entire log and set the tail_lsn to
4397	 * be the last_sync_lsn.  This was set in xlog_find_tail to be the
4398	 * lsn of the last known good LR on disk.  If there are extent frees
4399	 * or iunlinks they will have some entries in the AIL; so we look at
4400	 * the AIL to determine how to set the tail_lsn.
4401	 */
4402	xlog_assign_tail_lsn(log->l_mp);
4403
4404	/*
4405	 * Now that we've finished replaying all buffer and inode
4406	 * updates, re-read in the superblock and reverify it.
4407	 */
4408	bp = xfs_getsb(log->l_mp, 0);
4409	XFS_BUF_UNDONE(bp);
4410	ASSERT(!(XFS_BUF_ISWRITE(bp)));
4411	XFS_BUF_READ(bp);
4412	XFS_BUF_UNASYNC(bp);
4413	bp->b_ops = &xfs_sb_buf_ops;
4414
4415	if (XFS_FORCED_SHUTDOWN(log->l_mp)) {
4416		xfs_buf_relse(bp);
4417		return XFS_ERROR(EIO);
4418	}
4419
4420	xfs_buf_iorequest(bp);
4421	error = xfs_buf_iowait(bp);
 
 
 
 
 
4422	if (error) {
4423		xfs_buf_ioerror_alert(bp, __func__);
4424		ASSERT(0);
 
 
4425		xfs_buf_relse(bp);
4426		return error;
4427	}
4428
4429	/* Convert superblock from on-disk format */
4430	sbp = &log->l_mp->m_sb;
4431	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
4432	ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
4433	ASSERT(xfs_sb_good_version(sbp));
4434	xfs_buf_relse(bp);
4435
4436	/* We've re-read the superblock so re-initialize per-cpu counters */
4437	xfs_icsb_reinit_counters(log->l_mp);
4438
4439	xlog_recover_check_summary(log);
4440
4441	/* Normal transactions can now occur */
4442	log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
4443	return 0;
4444}
4445
4446/*
4447 * Perform recovery and re-initialize some log variables in xlog_find_tail.
4448 *
4449 * Return error or zero.
4450 */
4451int
4452xlog_recover(
4453	struct xlog	*log)
4454{
4455	xfs_daddr_t	head_blk, tail_blk;
4456	int		error;
4457
4458	/* find the tail of the log */
4459	if ((error = xlog_find_tail(log, &head_blk, &tail_blk)))
 
4460		return error;
4461
 
 
 
 
 
 
 
 
 
4462	if (tail_blk != head_blk) {
4463		/* There used to be a comment here:
4464		 *
4465		 * disallow recovery on read-only mounts.  note -- mount
4466		 * checks for ENOSPC and turns it into an intelligent
4467		 * error message.
4468		 * ...but this is no longer true.  Now, unless you specify
4469		 * NORECOVERY (in which case this function would never be
4470		 * called), we just go ahead and recover.  We do this all
4471		 * under the vfs layer, so we can get away with it unless
4472		 * the device itself is read-only, in which case we fail.
4473		 */
4474		if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
4475			return error;
4476		}
4477
4478		/*
4479		 * Version 5 superblock log feature mask validation. We know the
4480		 * log is dirty so check if there are any unknown log features
4481		 * in what we need to recover. If there are unknown features
4482		 * (e.g. unsupported transactions, then simply reject the
4483		 * attempt at recovery before touching anything.
4484		 */
4485		if (XFS_SB_VERSION_NUM(&log->l_mp->m_sb) == XFS_SB_VERSION_5 &&
4486		    xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
4487					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
4488			xfs_warn(log->l_mp,
4489"Superblock has unknown incompatible log features (0x%x) enabled.\n"
4490"The log can not be fully and/or safely recovered by this kernel.\n"
4491"Please recover the log on a kernel that supports the unknown features.",
4492				(log->l_mp->m_sb.sb_features_log_incompat &
4493					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
4494			return EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4495		}
4496
4497		xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
4498				log->l_mp->m_logname ? log->l_mp->m_logname
4499						     : "internal");
4500
4501		error = xlog_do_recover(log, head_blk, tail_blk);
4502		log->l_flags |= XLOG_RECOVERY_NEEDED;
4503	}
4504	return error;
4505}
4506
4507/*
4508 * In the first part of recovery we replay inodes and buffers and build
4509 * up the list of extent free items which need to be processed.  Here
4510 * we process the extent free items and clean up the on disk unlinked
4511 * inode lists.  This is separated from the first part of recovery so
4512 * that the root and real-time bitmap inodes can be read in from disk in
4513 * between the two stages.  This is necessary so that we can free space
4514 * in the real-time portion of the file system.
 
 
 
 
 
4515 */
4516int
4517xlog_recover_finish(
4518	struct xlog	*log)
4519{
4520	/*
4521	 * Now we're ready to do the transactions needed for the
4522	 * rest of recovery.  Start with completing all the extent
4523	 * free intent records and then process the unlinked inode
4524	 * lists.  At this point, we essentially run in normal mode
4525	 * except that we're still performing recovery actions
4526	 * rather than accepting new requests.
4527	 */
4528	if (log->l_flags & XLOG_RECOVERY_NEEDED) {
4529		int	error;
4530		error = xlog_recover_process_efis(log);
4531		if (error) {
4532			xfs_alert(log->l_mp, "Failed to recover EFIs");
4533			return error;
4534		}
4535		/*
4536		 * Sync the log to get all the EFIs out of the AIL.
4537		 * This isn't absolutely necessary, but it helps in
4538		 * case the unlink transactions would have problems
4539		 * pushing the EFIs out of the way.
4540		 */
4541		xfs_log_force(log->l_mp, XFS_LOG_SYNC);
 
 
 
 
 
4542
4543		xlog_recover_process_iunlinks(log);
 
 
 
 
 
4544
4545		xlog_recover_check_summary(log);
4546
4547		xfs_notice(log->l_mp, "Ending recovery (logdev: %s)",
4548				log->l_mp->m_logname ? log->l_mp->m_logname
4549						     : "internal");
4550		log->l_flags &= ~XLOG_RECOVERY_NEEDED;
4551	} else {
4552		xfs_info(log->l_mp, "Ending clean mount");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4553	}
4554	return 0;
4555}
4556
 
 
 
 
4557
4558#if defined(DEBUG)
4559/*
4560 * Read all of the agf and agi counters and check that they
4561 * are consistent with the superblock counters.
4562 */
4563void
4564xlog_recover_check_summary(
4565	struct xlog	*log)
4566{
4567	xfs_mount_t	*mp;
4568	xfs_agf_t	*agfp;
4569	xfs_buf_t	*agfbp;
4570	xfs_buf_t	*agibp;
4571	xfs_agnumber_t	agno;
4572	__uint64_t	freeblks;
4573	__uint64_t	itotal;
4574	__uint64_t	ifree;
4575	int		error;
4576
4577	mp = log->l_mp;
4578
4579	freeblks = 0LL;
4580	itotal = 0LL;
4581	ifree = 0LL;
4582	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
4583		error = xfs_read_agf(mp, NULL, agno, 0, &agfbp);
4584		if (error) {
4585			xfs_alert(mp, "%s agf read failed agno %d error %d",
4586						__func__, agno, error);
4587		} else {
4588			agfp = XFS_BUF_TO_AGF(agfbp);
4589			freeblks += be32_to_cpu(agfp->agf_freeblks) +
4590				    be32_to_cpu(agfp->agf_flcount);
4591			xfs_buf_relse(agfbp);
4592		}
4593
4594		error = xfs_read_agi(mp, NULL, agno, &agibp);
4595		if (error) {
4596			xfs_alert(mp, "%s agi read failed agno %d error %d",
4597						__func__, agno, error);
4598		} else {
4599			struct xfs_agi	*agi = XFS_BUF_TO_AGI(agibp);
4600
4601			itotal += be32_to_cpu(agi->agi_count);
4602			ifree += be32_to_cpu(agi->agi_freecount);
4603			xfs_buf_relse(agibp);
4604		}
4605	}
4606}
4607#endif /* DEBUG */
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4 * All Rights Reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_bit.h"
 
  13#include "xfs_sb.h"
 
  14#include "xfs_mount.h"
  15#include "xfs_defer.h"
  16#include "xfs_inode.h"
  17#include "xfs_trans.h"
  18#include "xfs_log.h"
  19#include "xfs_log_priv.h"
  20#include "xfs_log_recover.h"
 
 
  21#include "xfs_trans_priv.h"
  22#include "xfs_alloc.h"
  23#include "xfs_ialloc.h"
 
 
  24#include "xfs_trace.h"
  25#include "xfs_icache.h"
 
 
  26#include "xfs_error.h"
  27#include "xfs_buf_item.h"
  28#include "xfs_ag.h"
  29#include "xfs_quota.h"
  30#include "xfs_reflink.h"
  31
  32#define BLK_AVG(blk1, blk2)	((blk1+blk2) >> 1)
  33
  34STATIC int
  35xlog_find_zeroed(
  36	struct xlog	*,
  37	xfs_daddr_t	*);
  38STATIC int
  39xlog_clear_stale_blocks(
  40	struct xlog	*,
  41	xfs_lsn_t);
  42STATIC int
  43xlog_do_recovery_pass(
  44        struct xlog *, xfs_daddr_t, xfs_daddr_t, int, xfs_daddr_t *);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  45
  46/*
  47 * Sector aligned buffer routines for buffer create/read/write/access
  48 */
  49
  50/*
  51 * Verify the log-relative block number and length in basic blocks are valid for
  52 * an operation involving the given XFS log buffer. Returns true if the fields
  53 * are valid, false otherwise.
  54 */
  55static inline bool
  56xlog_verify_bno(
 
  57	struct xlog	*log,
  58	xfs_daddr_t	blk_no,
  59	int		bbcount)
  60{
  61	if (blk_no < 0 || blk_no >= log->l_logBBsize)
  62		return false;
  63	if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
  64		return false;
  65	return true;
  66}
  67
  68/*
  69 * Allocate a buffer to hold log data.  The buffer needs to be able to map to
  70 * a range of nbblks basic blocks at any valid offset within the log.
 
  71 */
  72static char *
  73xlog_alloc_buffer(
  74	struct xlog	*log,
  75	int		nbblks)
  76{
  77	/*
  78	 * Pass log block 0 since we don't have an addr yet, buffer will be
  79	 * verified on read.
  80	 */
  81	if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, 0, nbblks))) {
  82		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
  83			nbblks);
 
  84		return NULL;
  85	}
  86
  87	/*
  88	 * We do log I/O in units of log sectors (a power-of-2 multiple of the
  89	 * basic block size), so we round up the requested size to accommodate
  90	 * the basic blocks required for complete log sectors.
 
  91	 *
  92	 * In addition, the buffer may be used for a non-sector-aligned block
  93	 * offset, in which case an I/O of the requested size could extend
  94	 * beyond the end of the buffer.  If the requested size is only 1 basic
  95	 * block it will never straddle a sector boundary, so this won't be an
  96	 * issue.  Nor will this be a problem if the log I/O is done in basic
  97	 * blocks (sector size 1).  But otherwise we extend the buffer by one
  98	 * extra log sector to ensure there's space to accommodate this
  99	 * possibility.
 
 100	 */
 101	if (nbblks > 1 && log->l_sectBBsize > 1)
 102		nbblks += log->l_sectBBsize;
 103	nbblks = round_up(nbblks, log->l_sectBBsize);
 104	return kvzalloc(BBTOB(nbblks), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 
 
 
 
 
 
 
 
 
 
 
 105}
 106
 107/*
 108 * Return the address of the start of the given block number's data
 109 * in a log buffer.  The buffer covers a log sector-aligned region.
 110 */
 111static inline unsigned int
 112xlog_align(
 113	struct xlog	*log,
 114	xfs_daddr_t	blk_no)
 
 
 115{
 116	return BBTOB(blk_no & ((xfs_daddr_t)log->l_sectBBsize - 1));
 
 
 
 117}
 118
 119static int
 120xlog_do_io(
 121	struct xlog		*log,
 122	xfs_daddr_t		blk_no,
 123	unsigned int		nbblks,
 124	char			*data,
 125	enum req_op		op)
 
 
 
 126{
 127	int			error;
 128
 129	if (XFS_IS_CORRUPT(log->l_mp, !xlog_verify_bno(log, blk_no, nbblks))) {
 130		xfs_warn(log->l_mp,
 131			 "Invalid log block/length (0x%llx, 0x%x) for buffer",
 132			 blk_no, nbblks);
 133		return -EFSCORRUPTED;
 134	}
 135
 136	blk_no = round_down(blk_no, log->l_sectBBsize);
 137	nbblks = round_up(nbblks, log->l_sectBBsize);
 
 138	ASSERT(nbblks > 0);
 
 
 
 
 
 
 139
 140	error = xfs_rw_bdev(log->l_targ->bt_bdev, log->l_logBBstart + blk_no,
 141			BBTOB(nbblks), data, op);
 142	if (error && !xlog_is_shutdown(log)) {
 143		xfs_alert(log->l_mp,
 144			  "log recovery %s I/O error at daddr 0x%llx len %d error %d",
 145			  op == REQ_OP_WRITE ? "write" : "read",
 146			  blk_no, nbblks, error);
 147	}
 148	return error;
 149}
 150
 151STATIC int
 152xlog_bread_noalign(
 153	struct xlog	*log,
 154	xfs_daddr_t	blk_no,
 155	int		nbblks,
 156	char		*data)
 
 157{
 158	return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
 
 
 
 
 
 
 
 159}
 160
 
 
 
 
 161STATIC int
 162xlog_bread(
 163	struct xlog	*log,
 164	xfs_daddr_t	blk_no,
 165	int		nbblks,
 166	char		*data,
 167	char		**offset)
 168{
 169	int		error;
 
 
 
 
 
 
 
 
 170
 171	error = xlog_do_io(log, blk_no, nbblks, data, REQ_OP_READ);
 172	if (!error)
 173		*offset = data + xlog_align(log, blk_no);
 174	return error;
 
 175}
 176
 
 
 
 
 
 177STATIC int
 178xlog_bwrite(
 179	struct xlog	*log,
 180	xfs_daddr_t	blk_no,
 181	int		nbblks,
 182	char		*data)
 183{
 184	return xlog_do_io(log, blk_no, nbblks, data, REQ_OP_WRITE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 185}
 186
 187#ifdef DEBUG
 188/*
 189 * dump debug superblock and log record information
 190 */
 191STATIC void
 192xlog_header_check_dump(
 193	xfs_mount_t		*mp,
 194	xlog_rec_header_t	*head)
 195{
 196	xfs_debug(mp, "%s:  SB : uuid = %pU, fmt = %d",
 197		__func__, &mp->m_sb.sb_uuid, XLOG_FMT);
 198	xfs_debug(mp, "    log : uuid = %pU, fmt = %d",
 199		&head->h_fs_uuid, be32_to_cpu(head->h_fmt));
 200}
 201#else
 202#define xlog_header_check_dump(mp, head)
 203#endif
 204
 205/*
 206 * check log record header for recovery
 207 */
 208STATIC int
 209xlog_header_check_recover(
 210	xfs_mount_t		*mp,
 211	xlog_rec_header_t	*head)
 212{
 213	ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
 214
 215	/*
 216	 * IRIX doesn't write the h_fmt field and leaves it zeroed
 217	 * (XLOG_FMT_UNKNOWN). This stops us from trying to recover
 218	 * a dirty log created in IRIX.
 219	 */
 220	if (XFS_IS_CORRUPT(mp, head->h_fmt != cpu_to_be32(XLOG_FMT))) {
 221		xfs_warn(mp,
 222	"dirty log written in incompatible format - can't recover");
 223		xlog_header_check_dump(mp, head);
 224		return -EFSCORRUPTED;
 225	}
 226	if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
 227					   &head->h_fs_uuid))) {
 228		xfs_warn(mp,
 229	"dirty log entry has mismatched uuid - can't recover");
 230		xlog_header_check_dump(mp, head);
 231		return -EFSCORRUPTED;
 
 
 232	}
 233	return 0;
 234}
 235
 236/*
 237 * read the head block of the log and check the header
 238 */
 239STATIC int
 240xlog_header_check_mount(
 241	xfs_mount_t		*mp,
 242	xlog_rec_header_t	*head)
 243{
 244	ASSERT(head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM));
 245
 246	if (uuid_is_null(&head->h_fs_uuid)) {
 247		/*
 248		 * IRIX doesn't write the h_fs_uuid or h_fmt fields. If
 249		 * h_fs_uuid is null, we assume this log was last mounted
 250		 * by IRIX and continue.
 251		 */
 252		xfs_warn(mp, "null uuid in log - IRIX style log");
 253	} else if (XFS_IS_CORRUPT(mp, !uuid_equal(&mp->m_sb.sb_uuid,
 254						  &head->h_fs_uuid))) {
 255		xfs_warn(mp, "log has mismatched uuid - can't recover");
 256		xlog_header_check_dump(mp, head);
 257		return -EFSCORRUPTED;
 
 
 258	}
 259	return 0;
 260}
 261
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 262/*
 263 * This routine finds (to an approximation) the first block in the physical
 264 * log which contains the given cycle.  It uses a binary search algorithm.
 265 * Note that the algorithm can not be perfect because the disk will not
 266 * necessarily be perfect.
 267 */
 268STATIC int
 269xlog_find_cycle_start(
 270	struct xlog	*log,
 271	char		*buffer,
 272	xfs_daddr_t	first_blk,
 273	xfs_daddr_t	*last_blk,
 274	uint		cycle)
 275{
 276	char		*offset;
 277	xfs_daddr_t	mid_blk;
 278	xfs_daddr_t	end_blk;
 279	uint		mid_cycle;
 280	int		error;
 281
 282	end_blk = *last_blk;
 283	mid_blk = BLK_AVG(first_blk, end_blk);
 284	while (mid_blk != first_blk && mid_blk != end_blk) {
 285		error = xlog_bread(log, mid_blk, 1, buffer, &offset);
 286		if (error)
 287			return error;
 288		mid_cycle = xlog_get_cycle(offset);
 289		if (mid_cycle == cycle)
 290			end_blk = mid_blk;   /* last_half_cycle == mid_cycle */
 291		else
 292			first_blk = mid_blk; /* first_half_cycle == mid_cycle */
 293		mid_blk = BLK_AVG(first_blk, end_blk);
 294	}
 295	ASSERT((mid_blk == first_blk && mid_blk+1 == end_blk) ||
 296	       (mid_blk == end_blk && mid_blk-1 == first_blk));
 297
 298	*last_blk = end_blk;
 299
 300	return 0;
 301}
 302
 303/*
 304 * Check that a range of blocks does not contain stop_on_cycle_no.
 305 * Fill in *new_blk with the block offset where such a block is
 306 * found, or with -1 (an invalid block number) if there is no such
 307 * block in the range.  The scan needs to occur from front to back
 308 * and the pointer into the region must be updated since a later
 309 * routine will need to perform another test.
 310 */
 311STATIC int
 312xlog_find_verify_cycle(
 313	struct xlog	*log,
 314	xfs_daddr_t	start_blk,
 315	int		nbblks,
 316	uint		stop_on_cycle_no,
 317	xfs_daddr_t	*new_blk)
 318{
 319	xfs_daddr_t	i, j;
 320	uint		cycle;
 321	char		*buffer;
 322	xfs_daddr_t	bufblks;
 323	char		*buf = NULL;
 324	int		error = 0;
 325
 326	/*
 327	 * Greedily allocate a buffer big enough to handle the full
 328	 * range of basic blocks we'll be examining.  If that fails,
 329	 * try a smaller size.  We need to be able to read at least
 330	 * a log sector, or we're out of luck.
 331	 */
 332	bufblks = roundup_pow_of_two(nbblks);
 333	while (bufblks > log->l_logBBsize)
 334		bufblks >>= 1;
 335	while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
 336		bufblks >>= 1;
 337		if (bufblks < log->l_sectBBsize)
 338			return -ENOMEM;
 339	}
 340
 341	for (i = start_blk; i < start_blk + nbblks; i += bufblks) {
 342		int	bcount;
 343
 344		bcount = min(bufblks, (start_blk + nbblks - i));
 345
 346		error = xlog_bread(log, i, bcount, buffer, &buf);
 347		if (error)
 348			goto out;
 349
 350		for (j = 0; j < bcount; j++) {
 351			cycle = xlog_get_cycle(buf);
 352			if (cycle == stop_on_cycle_no) {
 353				*new_blk = i+j;
 354				goto out;
 355			}
 356
 357			buf += BBSIZE;
 358		}
 359	}
 360
 361	*new_blk = -1;
 362
 363out:
 364	kvfree(buffer);
 365	return error;
 366}
 367
 368static inline int
 369xlog_logrec_hblks(struct xlog *log, struct xlog_rec_header *rh)
 370{
 371	if (xfs_has_logv2(log->l_mp)) {
 372		int	h_size = be32_to_cpu(rh->h_size);
 373
 374		if ((be32_to_cpu(rh->h_version) & XLOG_VERSION_2) &&
 375		    h_size > XLOG_HEADER_CYCLE_SIZE)
 376			return DIV_ROUND_UP(h_size, XLOG_HEADER_CYCLE_SIZE);
 377	}
 378	return 1;
 379}
 380
 381/*
 382 * Potentially backup over partial log record write.
 383 *
 384 * In the typical case, last_blk is the number of the block directly after
 385 * a good log record.  Therefore, we subtract one to get the block number
 386 * of the last block in the given buffer.  extra_bblks contains the number
 387 * of blocks we would have read on a previous read.  This happens when the
 388 * last log record is split over the end of the physical log.
 389 *
 390 * extra_bblks is the number of blocks potentially verified on a previous
 391 * call to this routine.
 392 */
 393STATIC int
 394xlog_find_verify_log_record(
 395	struct xlog		*log,
 396	xfs_daddr_t		start_blk,
 397	xfs_daddr_t		*last_blk,
 398	int			extra_bblks)
 399{
 400	xfs_daddr_t		i;
 401	char			*buffer;
 402	char			*offset = NULL;
 403	xlog_rec_header_t	*head = NULL;
 404	int			error = 0;
 405	int			smallmem = 0;
 406	int			num_blks = *last_blk - start_blk;
 407	int			xhdrs;
 408
 409	ASSERT(start_blk != 0 || *last_blk != start_blk);
 410
 411	buffer = xlog_alloc_buffer(log, num_blks);
 412	if (!buffer) {
 413		buffer = xlog_alloc_buffer(log, 1);
 414		if (!buffer)
 415			return -ENOMEM;
 416		smallmem = 1;
 417	} else {
 418		error = xlog_bread(log, start_blk, num_blks, buffer, &offset);
 419		if (error)
 420			goto out;
 421		offset += ((num_blks - 1) << BBSHIFT);
 422	}
 423
 424	for (i = (*last_blk) - 1; i >= 0; i--) {
 425		if (i < start_blk) {
 426			/* valid log record not found */
 427			xfs_warn(log->l_mp,
 428		"Log inconsistent (didn't find previous header)");
 429			ASSERT(0);
 430			error = -EFSCORRUPTED;
 431			goto out;
 432		}
 433
 434		if (smallmem) {
 435			error = xlog_bread(log, i, 1, buffer, &offset);
 436			if (error)
 437				goto out;
 438		}
 439
 440		head = (xlog_rec_header_t *)offset;
 441
 442		if (head->h_magicno == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
 443			break;
 444
 445		if (!smallmem)
 446			offset -= BBSIZE;
 447	}
 448
 449	/*
 450	 * We hit the beginning of the physical log & still no header.  Return
 451	 * to caller.  If caller can handle a return of -1, then this routine
 452	 * will be called again for the end of the physical log.
 453	 */
 454	if (i == -1) {
 455		error = 1;
 456		goto out;
 457	}
 458
 459	/*
 460	 * We have the final block of the good log (the first block
 461	 * of the log record _before_ the head. So we check the uuid.
 462	 */
 463	if ((error = xlog_header_check_mount(log->l_mp, head)))
 464		goto out;
 465
 466	/*
 467	 * We may have found a log record header before we expected one.
 468	 * last_blk will be the 1st block # with a given cycle #.  We may end
 469	 * up reading an entire log record.  In this case, we don't want to
 470	 * reset last_blk.  Only when last_blk points in the middle of a log
 471	 * record do we update last_blk.
 472	 */
 473	xhdrs = xlog_logrec_hblks(log, head);
 
 
 
 
 
 
 
 
 474
 475	if (*last_blk - i + extra_bblks !=
 476	    BTOBB(be32_to_cpu(head->h_len)) + xhdrs)
 477		*last_blk = i;
 478
 479out:
 480	kvfree(buffer);
 481	return error;
 482}
 483
 484/*
 485 * Head is defined to be the point of the log where the next log write
 486 * could go.  This means that incomplete LR writes at the end are
 487 * eliminated when calculating the head.  We aren't guaranteed that previous
 488 * LR have complete transactions.  We only know that a cycle number of
 489 * current cycle number -1 won't be present in the log if we start writing
 490 * from our current block number.
 491 *
 492 * last_blk contains the block number of the first block with a given
 493 * cycle number.
 494 *
 495 * Return: zero if normal, non-zero if error.
 496 */
 497STATIC int
 498xlog_find_head(
 499	struct xlog	*log,
 500	xfs_daddr_t	*return_head_blk)
 501{
 502	char		*buffer;
 503	char		*offset;
 504	xfs_daddr_t	new_blk, first_blk, start_blk, last_blk, head_blk;
 505	int		num_scan_bblks;
 506	uint		first_half_cycle, last_half_cycle;
 507	uint		stop_on_cycle;
 508	int		error, log_bbnum = log->l_logBBsize;
 509
 510	/* Is the end of the log device zeroed? */
 511	error = xlog_find_zeroed(log, &first_blk);
 512	if (error < 0) {
 513		xfs_warn(log->l_mp, "empty log check failed");
 514		return error;
 515	}
 516	if (error == 1) {
 517		*return_head_blk = first_blk;
 518
 519		/* Is the whole lot zeroed? */
 520		if (!first_blk) {
 521			/* Linux XFS shouldn't generate totally zeroed logs -
 522			 * mkfs etc write a dummy unmount record to a fresh
 523			 * log so we can store the uuid in there
 524			 */
 525			xfs_warn(log->l_mp, "totally zeroed log");
 526		}
 527
 528		return 0;
 
 
 
 529	}
 530
 531	first_blk = 0;			/* get cycle # of 1st block */
 532	buffer = xlog_alloc_buffer(log, 1);
 533	if (!buffer)
 534		return -ENOMEM;
 535
 536	error = xlog_bread(log, 0, 1, buffer, &offset);
 537	if (error)
 538		goto out_free_buffer;
 539
 540	first_half_cycle = xlog_get_cycle(offset);
 541
 542	last_blk = head_blk = log_bbnum - 1;	/* get cycle # of last block */
 543	error = xlog_bread(log, last_blk, 1, buffer, &offset);
 544	if (error)
 545		goto out_free_buffer;
 546
 547	last_half_cycle = xlog_get_cycle(offset);
 548	ASSERT(last_half_cycle != 0);
 549
 550	/*
 551	 * If the 1st half cycle number is equal to the last half cycle number,
 552	 * then the entire log is stamped with the same cycle number.  In this
 553	 * case, head_blk can't be set to zero (which makes sense).  The below
 554	 * math doesn't work out properly with head_blk equal to zero.  Instead,
 555	 * we set it to log_bbnum which is an invalid block number, but this
 556	 * value makes the math correct.  If head_blk doesn't changed through
 557	 * all the tests below, *head_blk is set to zero at the very end rather
 558	 * than log_bbnum.  In a sense, log_bbnum and zero are the same block
 559	 * in a circular file.
 560	 */
 561	if (first_half_cycle == last_half_cycle) {
 562		/*
 563		 * In this case we believe that the entire log should have
 564		 * cycle number last_half_cycle.  We need to scan backwards
 565		 * from the end verifying that there are no holes still
 566		 * containing last_half_cycle - 1.  If we find such a hole,
 567		 * then the start of that hole will be the new head.  The
 568		 * simple case looks like
 569		 *        x | x ... | x - 1 | x
 570		 * Another case that fits this picture would be
 571		 *        x | x + 1 | x ... | x
 572		 * In this case the head really is somewhere at the end of the
 573		 * log, as one of the latest writes at the beginning was
 574		 * incomplete.
 575		 * One more case is
 576		 *        x | x + 1 | x ... | x - 1 | x
 577		 * This is really the combination of the above two cases, and
 578		 * the head has to end up at the start of the x-1 hole at the
 579		 * end of the log.
 580		 *
 581		 * In the 256k log case, we will read from the beginning to the
 582		 * end of the log and search for cycle numbers equal to x-1.
 583		 * We don't worry about the x+1 blocks that we encounter,
 584		 * because we know that they cannot be the head since the log
 585		 * started with x.
 586		 */
 587		head_blk = log_bbnum;
 588		stop_on_cycle = last_half_cycle - 1;
 589	} else {
 590		/*
 591		 * In this case we want to find the first block with cycle
 592		 * number matching last_half_cycle.  We expect the log to be
 593		 * some variation on
 594		 *        x + 1 ... | x ... | x
 595		 * The first block with cycle number x (last_half_cycle) will
 596		 * be where the new head belongs.  First we do a binary search
 597		 * for the first occurrence of last_half_cycle.  The binary
 598		 * search may not be totally accurate, so then we scan back
 599		 * from there looking for occurrences of last_half_cycle before
 600		 * us.  If that backwards scan wraps around the beginning of
 601		 * the log, then we look for occurrences of last_half_cycle - 1
 602		 * at the end of the log.  The cases we're looking for look
 603		 * like
 604		 *                               v binary search stopped here
 605		 *        x + 1 ... | x | x + 1 | x ... | x
 606		 *                   ^ but we want to locate this spot
 607		 * or
 608		 *        <---------> less than scan distance
 609		 *        x + 1 ... | x ... | x - 1 | x
 610		 *                           ^ we want to locate this spot
 611		 */
 612		stop_on_cycle = last_half_cycle;
 613		error = xlog_find_cycle_start(log, buffer, first_blk, &head_blk,
 614				last_half_cycle);
 615		if (error)
 616			goto out_free_buffer;
 617	}
 618
 619	/*
 620	 * Now validate the answer.  Scan back some number of maximum possible
 621	 * blocks and make sure each one has the expected cycle number.  The
 622	 * maximum is determined by the total possible amount of buffering
 623	 * in the in-core log.  The following number can be made tighter if
 624	 * we actually look at the block size of the filesystem.
 625	 */
 626	num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
 627	if (head_blk >= num_scan_bblks) {
 628		/*
 629		 * We are guaranteed that the entire check can be performed
 630		 * in one buffer.
 631		 */
 632		start_blk = head_blk - num_scan_bblks;
 633		if ((error = xlog_find_verify_cycle(log,
 634						start_blk, num_scan_bblks,
 635						stop_on_cycle, &new_blk)))
 636			goto out_free_buffer;
 637		if (new_blk != -1)
 638			head_blk = new_blk;
 639	} else {		/* need to read 2 parts of log */
 640		/*
 641		 * We are going to scan backwards in the log in two parts.
 642		 * First we scan the physical end of the log.  In this part
 643		 * of the log, we are looking for blocks with cycle number
 644		 * last_half_cycle - 1.
 645		 * If we find one, then we know that the log starts there, as
 646		 * we've found a hole that didn't get written in going around
 647		 * the end of the physical log.  The simple case for this is
 648		 *        x + 1 ... | x ... | x - 1 | x
 649		 *        <---------> less than scan distance
 650		 * If all of the blocks at the end of the log have cycle number
 651		 * last_half_cycle, then we check the blocks at the start of
 652		 * the log looking for occurrences of last_half_cycle.  If we
 653		 * find one, then our current estimate for the location of the
 654		 * first occurrence of last_half_cycle is wrong and we move
 655		 * back to the hole we've found.  This case looks like
 656		 *        x + 1 ... | x | x + 1 | x ...
 657		 *                               ^ binary search stopped here
 658		 * Another case we need to handle that only occurs in 256k
 659		 * logs is
 660		 *        x + 1 ... | x ... | x+1 | x ...
 661		 *                   ^ binary search stops here
 662		 * In a 256k log, the scan at the end of the log will see the
 663		 * x + 1 blocks.  We need to skip past those since that is
 664		 * certainly not the head of the log.  By searching for
 665		 * last_half_cycle-1 we accomplish that.
 666		 */
 667		ASSERT(head_blk <= INT_MAX &&
 668			(xfs_daddr_t) num_scan_bblks >= head_blk);
 669		start_blk = log_bbnum - (num_scan_bblks - head_blk);
 670		if ((error = xlog_find_verify_cycle(log, start_blk,
 671					num_scan_bblks - (int)head_blk,
 672					(stop_on_cycle - 1), &new_blk)))
 673			goto out_free_buffer;
 674		if (new_blk != -1) {
 675			head_blk = new_blk;
 676			goto validate_head;
 677		}
 678
 679		/*
 680		 * Scan beginning of log now.  The last part of the physical
 681		 * log is good.  This scan needs to verify that it doesn't find
 682		 * the last_half_cycle.
 683		 */
 684		start_blk = 0;
 685		ASSERT(head_blk <= INT_MAX);
 686		if ((error = xlog_find_verify_cycle(log,
 687					start_blk, (int)head_blk,
 688					stop_on_cycle, &new_blk)))
 689			goto out_free_buffer;
 690		if (new_blk != -1)
 691			head_blk = new_blk;
 692	}
 693
 694validate_head:
 695	/*
 696	 * Now we need to make sure head_blk is not pointing to a block in
 697	 * the middle of a log record.
 698	 */
 699	num_scan_bblks = XLOG_REC_SHIFT(log);
 700	if (head_blk >= num_scan_bblks) {
 701		start_blk = head_blk - num_scan_bblks; /* don't read head_blk */
 702
 703		/* start ptr at last block ptr before head_blk */
 704		error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
 705		if (error == 1)
 706			error = -EIO;
 707		if (error)
 708			goto out_free_buffer;
 
 709	} else {
 710		start_blk = 0;
 711		ASSERT(head_blk <= INT_MAX);
 712		error = xlog_find_verify_log_record(log, start_blk, &head_blk, 0);
 713		if (error < 0)
 714			goto out_free_buffer;
 715		if (error == 1) {
 716			/* We hit the beginning of the log during our search */
 717			start_blk = log_bbnum - (num_scan_bblks - head_blk);
 718			new_blk = log_bbnum;
 719			ASSERT(start_blk <= INT_MAX &&
 720				(xfs_daddr_t) log_bbnum-start_blk >= 0);
 721			ASSERT(head_blk <= INT_MAX);
 722			error = xlog_find_verify_log_record(log, start_blk,
 723							&new_blk, (int)head_blk);
 724			if (error == 1)
 725				error = -EIO;
 726			if (error)
 727				goto out_free_buffer;
 
 728			if (new_blk != log_bbnum)
 729				head_blk = new_blk;
 730		} else if (error)
 731			goto out_free_buffer;
 732	}
 733
 734	kvfree(buffer);
 735	if (head_blk == log_bbnum)
 736		*return_head_blk = 0;
 737	else
 738		*return_head_blk = head_blk;
 739	/*
 740	 * When returning here, we have a good block number.  Bad block
 741	 * means that during a previous crash, we didn't have a clean break
 742	 * from cycle number N to cycle number N-1.  In this case, we need
 743	 * to find the first block with cycle number N-1.
 744	 */
 745	return 0;
 746
 747out_free_buffer:
 748	kvfree(buffer);
 
 749	if (error)
 750		xfs_warn(log->l_mp, "failed to find log head");
 751	return error;
 752}
 753
 754/*
 755 * Seek backwards in the log for log record headers.
 756 *
 757 * Given a starting log block, walk backwards until we find the provided number
 758 * of records or hit the provided tail block. The return value is the number of
 759 * records encountered or a negative error code. The log block and buffer
 760 * pointer of the last record seen are returned in rblk and rhead respectively.
 761 */
 762STATIC int
 763xlog_rseek_logrec_hdr(
 764	struct xlog		*log,
 765	xfs_daddr_t		head_blk,
 766	xfs_daddr_t		tail_blk,
 767	int			count,
 768	char			*buffer,
 769	xfs_daddr_t		*rblk,
 770	struct xlog_rec_header	**rhead,
 771	bool			*wrapped)
 772{
 773	int			i;
 774	int			error;
 775	int			found = 0;
 776	char			*offset = NULL;
 777	xfs_daddr_t		end_blk;
 778
 779	*wrapped = false;
 780
 781	/*
 782	 * Walk backwards from the head block until we hit the tail or the first
 783	 * block in the log.
 784	 */
 785	end_blk = head_blk > tail_blk ? tail_blk : 0;
 786	for (i = (int) head_blk - 1; i >= end_blk; i--) {
 787		error = xlog_bread(log, i, 1, buffer, &offset);
 788		if (error)
 789			goto out_error;
 790
 791		if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 792			*rblk = i;
 793			*rhead = (struct xlog_rec_header *) offset;
 794			if (++found == count)
 795				break;
 796		}
 797	}
 798
 799	/*
 800	 * If we haven't hit the tail block or the log record header count,
 801	 * start looking again from the end of the physical log. Note that
 802	 * callers can pass head == tail if the tail is not yet known.
 803	 */
 804	if (tail_blk >= head_blk && found != count) {
 805		for (i = log->l_logBBsize - 1; i >= (int) tail_blk; i--) {
 806			error = xlog_bread(log, i, 1, buffer, &offset);
 807			if (error)
 808				goto out_error;
 809
 810			if (*(__be32 *)offset ==
 811			    cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 812				*wrapped = true;
 813				*rblk = i;
 814				*rhead = (struct xlog_rec_header *) offset;
 815				if (++found == count)
 816					break;
 817			}
 818		}
 819	}
 820
 821	return found;
 822
 823out_error:
 824	return error;
 825}
 826
 827/*
 828 * Seek forward in the log for log record headers.
 829 *
 830 * Given head and tail blocks, walk forward from the tail block until we find
 831 * the provided number of records or hit the head block. The return value is the
 832 * number of records encountered or a negative error code. The log block and
 833 * buffer pointer of the last record seen are returned in rblk and rhead
 834 * respectively.
 835 */
 836STATIC int
 837xlog_seek_logrec_hdr(
 838	struct xlog		*log,
 839	xfs_daddr_t		head_blk,
 840	xfs_daddr_t		tail_blk,
 841	int			count,
 842	char			*buffer,
 843	xfs_daddr_t		*rblk,
 844	struct xlog_rec_header	**rhead,
 845	bool			*wrapped)
 846{
 847	int			i;
 848	int			error;
 849	int			found = 0;
 850	char			*offset = NULL;
 851	xfs_daddr_t		end_blk;
 852
 853	*wrapped = false;
 854
 855	/*
 856	 * Walk forward from the tail block until we hit the head or the last
 857	 * block in the log.
 858	 */
 859	end_blk = head_blk > tail_blk ? head_blk : log->l_logBBsize - 1;
 860	for (i = (int) tail_blk; i <= end_blk; i++) {
 861		error = xlog_bread(log, i, 1, buffer, &offset);
 862		if (error)
 863			goto out_error;
 864
 865		if (*(__be32 *) offset == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 866			*rblk = i;
 867			*rhead = (struct xlog_rec_header *) offset;
 868			if (++found == count)
 869				break;
 870		}
 871	}
 872
 873	/*
 874	 * If we haven't hit the head block or the log record header count,
 875	 * start looking again from the start of the physical log.
 876	 */
 877	if (tail_blk > head_blk && found != count) {
 878		for (i = 0; i < (int) head_blk; i++) {
 879			error = xlog_bread(log, i, 1, buffer, &offset);
 880			if (error)
 881				goto out_error;
 882
 883			if (*(__be32 *)offset ==
 884			    cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) {
 885				*wrapped = true;
 886				*rblk = i;
 887				*rhead = (struct xlog_rec_header *) offset;
 888				if (++found == count)
 889					break;
 890			}
 891		}
 892	}
 893
 894	return found;
 895
 896out_error:
 897	return error;
 898}
 899
 900/*
 901 * Calculate distance from head to tail (i.e., unused space in the log).
 902 */
 903static inline int
 904xlog_tail_distance(
 905	struct xlog	*log,
 906	xfs_daddr_t	head_blk,
 907	xfs_daddr_t	tail_blk)
 908{
 909	if (head_blk < tail_blk)
 910		return tail_blk - head_blk;
 911
 912	return tail_blk + (log->l_logBBsize - head_blk);
 913}
 914
 915/*
 916 * Verify the log tail. This is particularly important when torn or incomplete
 917 * writes have been detected near the front of the log and the head has been
 918 * walked back accordingly.
 919 *
 920 * We also have to handle the case where the tail was pinned and the head
 921 * blocked behind the tail right before a crash. If the tail had been pushed
 922 * immediately prior to the crash and the subsequent checkpoint was only
 923 * partially written, it's possible it overwrote the last referenced tail in the
 924 * log with garbage. This is not a coherency problem because the tail must have
 925 * been pushed before it can be overwritten, but appears as log corruption to
 926 * recovery because we have no way to know the tail was updated if the
 927 * subsequent checkpoint didn't write successfully.
 928 *
 929 * Therefore, CRC check the log from tail to head. If a failure occurs and the
 930 * offending record is within max iclog bufs from the head, walk the tail
 931 * forward and retry until a valid tail is found or corruption is detected out
 932 * of the range of a possible overwrite.
 933 */
 934STATIC int
 935xlog_verify_tail(
 936	struct xlog		*log,
 937	xfs_daddr_t		head_blk,
 938	xfs_daddr_t		*tail_blk,
 939	int			hsize)
 940{
 941	struct xlog_rec_header	*thead;
 942	char			*buffer;
 943	xfs_daddr_t		first_bad;
 944	int			error = 0;
 945	bool			wrapped;
 946	xfs_daddr_t		tmp_tail;
 947	xfs_daddr_t		orig_tail = *tail_blk;
 948
 949	buffer = xlog_alloc_buffer(log, 1);
 950	if (!buffer)
 951		return -ENOMEM;
 952
 953	/*
 954	 * Make sure the tail points to a record (returns positive count on
 955	 * success).
 956	 */
 957	error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, buffer,
 958			&tmp_tail, &thead, &wrapped);
 959	if (error < 0)
 960		goto out;
 961	if (*tail_blk != tmp_tail)
 962		*tail_blk = tmp_tail;
 963
 964	/*
 965	 * Run a CRC check from the tail to the head. We can't just check
 966	 * MAX_ICLOGS records past the tail because the tail may point to stale
 967	 * blocks cleared during the search for the head/tail. These blocks are
 968	 * overwritten with zero-length records and thus record count is not a
 969	 * reliable indicator of the iclog state before a crash.
 970	 */
 971	first_bad = 0;
 972	error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
 973				      XLOG_RECOVER_CRCPASS, &first_bad);
 974	while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
 975		int	tail_distance;
 976
 977		/*
 978		 * Is corruption within range of the head? If so, retry from
 979		 * the next record. Otherwise return an error.
 980		 */
 981		tail_distance = xlog_tail_distance(log, head_blk, first_bad);
 982		if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
 983			break;
 984
 985		/* skip to the next record; returns positive count on success */
 986		error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2,
 987				buffer, &tmp_tail, &thead, &wrapped);
 988		if (error < 0)
 989			goto out;
 990
 991		*tail_blk = tmp_tail;
 992		first_bad = 0;
 993		error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
 994					      XLOG_RECOVER_CRCPASS, &first_bad);
 995	}
 996
 997	if (!error && *tail_blk != orig_tail)
 998		xfs_warn(log->l_mp,
 999		"Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
1000			 orig_tail, *tail_blk);
1001out:
1002	kvfree(buffer);
1003	return error;
1004}
1005
1006/*
1007 * Detect and trim torn writes from the head of the log.
1008 *
1009 * Storage without sector atomicity guarantees can result in torn writes in the
1010 * log in the event of a crash. Our only means to detect this scenario is via
1011 * CRC verification. While we can't always be certain that CRC verification
1012 * failure is due to a torn write vs. an unrelated corruption, we do know that
1013 * only a certain number (XLOG_MAX_ICLOGS) of log records can be written out at
1014 * one time. Therefore, CRC verify up to XLOG_MAX_ICLOGS records at the head of
1015 * the log and treat failures in this range as torn writes as a matter of
1016 * policy. In the event of CRC failure, the head is walked back to the last good
1017 * record in the log and the tail is updated from that record and verified.
1018 */
1019STATIC int
1020xlog_verify_head(
1021	struct xlog		*log,
1022	xfs_daddr_t		*head_blk,	/* in/out: unverified head */
1023	xfs_daddr_t		*tail_blk,	/* out: tail block */
1024	char			*buffer,
1025	xfs_daddr_t		*rhead_blk,	/* start blk of last record */
1026	struct xlog_rec_header	**rhead,	/* ptr to last record */
1027	bool			*wrapped)	/* last rec. wraps phys. log */
1028{
1029	struct xlog_rec_header	*tmp_rhead;
1030	char			*tmp_buffer;
1031	xfs_daddr_t		first_bad;
1032	xfs_daddr_t		tmp_rhead_blk;
1033	int			found;
1034	int			error;
1035	bool			tmp_wrapped;
1036
1037	/*
1038	 * Check the head of the log for torn writes. Search backwards from the
1039	 * head until we hit the tail or the maximum number of log record I/Os
1040	 * that could have been in flight at one time. Use a temporary buffer so
1041	 * we don't trash the rhead/buffer pointers from the caller.
1042	 */
1043	tmp_buffer = xlog_alloc_buffer(log, 1);
1044	if (!tmp_buffer)
1045		return -ENOMEM;
1046	error = xlog_rseek_logrec_hdr(log, *head_blk, *tail_blk,
1047				      XLOG_MAX_ICLOGS, tmp_buffer,
1048				      &tmp_rhead_blk, &tmp_rhead, &tmp_wrapped);
1049	kvfree(tmp_buffer);
1050	if (error < 0)
1051		return error;
1052
1053	/*
1054	 * Now run a CRC verification pass over the records starting at the
1055	 * block found above to the current head. If a CRC failure occurs, the
1056	 * log block of the first bad record is saved in first_bad.
1057	 */
1058	error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
1059				      XLOG_RECOVER_CRCPASS, &first_bad);
1060	if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
1061		/*
1062		 * We've hit a potential torn write. Reset the error and warn
1063		 * about it.
1064		 */
1065		error = 0;
1066		xfs_warn(log->l_mp,
1067"Torn write (CRC failure) detected at log block 0x%llx. Truncating head block from 0x%llx.",
1068			 first_bad, *head_blk);
1069
1070		/*
1071		 * Get the header block and buffer pointer for the last good
1072		 * record before the bad record.
1073		 *
1074		 * Note that xlog_find_tail() clears the blocks at the new head
1075		 * (i.e., the records with invalid CRC) if the cycle number
1076		 * matches the current cycle.
1077		 */
1078		found = xlog_rseek_logrec_hdr(log, first_bad, *tail_blk, 1,
1079				buffer, rhead_blk, rhead, wrapped);
1080		if (found < 0)
1081			return found;
1082		if (found == 0)		/* XXX: right thing to do here? */
1083			return -EIO;
1084
1085		/*
1086		 * Reset the head block to the starting block of the first bad
1087		 * log record and set the tail block based on the last good
1088		 * record.
1089		 *
1090		 * Bail out if the updated head/tail match as this indicates
1091		 * possible corruption outside of the acceptable
1092		 * (XLOG_MAX_ICLOGS) range. This is a job for xfs_repair...
1093		 */
1094		*head_blk = first_bad;
1095		*tail_blk = BLOCK_LSN(be64_to_cpu((*rhead)->h_tail_lsn));
1096		if (*head_blk == *tail_blk) {
1097			ASSERT(0);
1098			return 0;
1099		}
1100	}
1101	if (error)
1102		return error;
1103
1104	return xlog_verify_tail(log, *head_blk, tail_blk,
1105				be32_to_cpu((*rhead)->h_size));
1106}
1107
1108/*
1109 * We need to make sure we handle log wrapping properly, so we can't use the
1110 * calculated logbno directly. Make sure it wraps to the correct bno inside the
1111 * log.
1112 *
1113 * The log is limited to 32 bit sizes, so we use the appropriate modulus
1114 * operation here and cast it back to a 64 bit daddr on return.
1115 */
1116static inline xfs_daddr_t
1117xlog_wrap_logbno(
1118	struct xlog		*log,
1119	xfs_daddr_t		bno)
1120{
1121	int			mod;
1122
1123	div_s64_rem(bno, log->l_logBBsize, &mod);
1124	return mod;
1125}
1126
1127/*
1128 * Check whether the head of the log points to an unmount record. In other
1129 * words, determine whether the log is clean. If so, update the in-core state
1130 * appropriately.
1131 */
1132static int
1133xlog_check_unmount_rec(
1134	struct xlog		*log,
1135	xfs_daddr_t		*head_blk,
1136	xfs_daddr_t		*tail_blk,
1137	struct xlog_rec_header	*rhead,
1138	xfs_daddr_t		rhead_blk,
1139	char			*buffer,
1140	bool			*clean)
1141{
1142	struct xlog_op_header	*op_head;
1143	xfs_daddr_t		umount_data_blk;
1144	xfs_daddr_t		after_umount_blk;
1145	int			hblks;
1146	int			error;
1147	char			*offset;
1148
1149	*clean = false;
1150
1151	/*
1152	 * Look for unmount record. If we find it, then we know there was a
1153	 * clean unmount. Since 'i' could be the last block in the physical
1154	 * log, we convert to a log block before comparing to the head_blk.
1155	 *
1156	 * Save the current tail lsn to use to pass to xlog_clear_stale_blocks()
1157	 * below. We won't want to clear the unmount record if there is one, so
1158	 * we pass the lsn of the unmount record rather than the block after it.
1159	 */
1160	hblks = xlog_logrec_hblks(log, rhead);
1161	after_umount_blk = xlog_wrap_logbno(log,
1162			rhead_blk + hblks + BTOBB(be32_to_cpu(rhead->h_len)));
1163
1164	if (*head_blk == after_umount_blk &&
1165	    be32_to_cpu(rhead->h_num_logops) == 1) {
1166		umount_data_blk = xlog_wrap_logbno(log, rhead_blk + hblks);
1167		error = xlog_bread(log, umount_data_blk, 1, buffer, &offset);
1168		if (error)
1169			return error;
1170
1171		op_head = (struct xlog_op_header *)offset;
1172		if (op_head->oh_flags & XLOG_UNMOUNT_TRANS) {
1173			/*
1174			 * Set tail and last sync so that newly written log
1175			 * records will point recovery to after the current
1176			 * unmount record.
1177			 */
1178			xlog_assign_atomic_lsn(&log->l_tail_lsn,
1179					log->l_curr_cycle, after_umount_blk);
1180			log->l_ailp->ail_head_lsn =
1181					atomic64_read(&log->l_tail_lsn);
1182			*tail_blk = after_umount_blk;
1183
1184			*clean = true;
1185		}
1186	}
1187
1188	return 0;
1189}
1190
1191static void
1192xlog_set_state(
1193	struct xlog		*log,
1194	xfs_daddr_t		head_blk,
1195	struct xlog_rec_header	*rhead,
1196	xfs_daddr_t		rhead_blk,
1197	bool			bump_cycle)
1198{
1199	/*
1200	 * Reset log values according to the state of the log when we
1201	 * crashed.  In the case where head_blk == 0, we bump curr_cycle
1202	 * one because the next write starts a new cycle rather than
1203	 * continuing the cycle of the last good log record.  At this
1204	 * point we have guaranteed that all partial log records have been
1205	 * accounted for.  Therefore, we know that the last good log record
1206	 * written was complete and ended exactly on the end boundary
1207	 * of the physical log.
1208	 */
1209	log->l_prev_block = rhead_blk;
1210	log->l_curr_block = (int)head_blk;
1211	log->l_curr_cycle = be32_to_cpu(rhead->h_cycle);
1212	if (bump_cycle)
1213		log->l_curr_cycle++;
1214	atomic64_set(&log->l_tail_lsn, be64_to_cpu(rhead->h_tail_lsn));
1215	log->l_ailp->ail_head_lsn = be64_to_cpu(rhead->h_lsn);
1216}
1217
1218/*
1219 * Find the sync block number or the tail of the log.
1220 *
1221 * This will be the block number of the last record to have its
1222 * associated buffers synced to disk.  Every log record header has
1223 * a sync lsn embedded in it.  LSNs hold block numbers, so it is easy
1224 * to get a sync block number.  The only concern is to figure out which
1225 * log record header to believe.
1226 *
1227 * The following algorithm uses the log record header with the largest
1228 * lsn.  The entire log record does not need to be valid.  We only care
1229 * that the header is valid.
1230 *
1231 * We could speed up search by using current head_blk buffer, but it is not
1232 * available.
1233 */
1234STATIC int
1235xlog_find_tail(
1236	struct xlog		*log,
1237	xfs_daddr_t		*head_blk,
1238	xfs_daddr_t		*tail_blk)
1239{
1240	xlog_rec_header_t	*rhead;
1241	char			*offset = NULL;
1242	char			*buffer;
1243	int			error;
1244	xfs_daddr_t		rhead_blk;
 
 
1245	xfs_lsn_t		tail_lsn;
1246	bool			wrapped = false;
1247	bool			clean = false;
 
1248
1249	/*
1250	 * Find previous log record
1251	 */
1252	if ((error = xlog_find_head(log, head_blk)))
1253		return error;
1254	ASSERT(*head_blk < INT_MAX);
1255
1256	buffer = xlog_alloc_buffer(log, 1);
1257	if (!buffer)
1258		return -ENOMEM;
1259	if (*head_blk == 0) {				/* special case */
1260		error = xlog_bread(log, 0, 1, buffer, &offset);
1261		if (error)
1262			goto done;
1263
1264		if (xlog_get_cycle(offset) == 0) {
1265			*tail_blk = 0;
1266			/* leave all other log inited values alone */
1267			goto done;
1268		}
1269	}
1270
1271	/*
1272	 * Search backwards through the log looking for the log record header
1273	 * block. This wraps all the way back around to the head so something is
1274	 * seriously wrong if we can't find it.
1275	 */
1276	error = xlog_rseek_logrec_hdr(log, *head_blk, *head_blk, 1, buffer,
1277				      &rhead_blk, &rhead, &wrapped);
1278	if (error < 0)
1279		goto done;
1280	if (!error) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1281		xfs_warn(log->l_mp, "%s: couldn't find sync record", __func__);
1282		error = -EFSCORRUPTED;
1283		goto done;
 
1284	}
 
 
 
1285	*tail_blk = BLOCK_LSN(be64_to_cpu(rhead->h_tail_lsn));
1286
1287	/*
1288	 * Set the log state based on the current head record.
 
 
 
 
 
 
 
1289	 */
1290	xlog_set_state(log, *head_blk, rhead, rhead_blk, wrapped);
1291	tail_lsn = atomic64_read(&log->l_tail_lsn);
1292
1293	/*
1294	 * Look for an unmount record at the head of the log. This sets the log
1295	 * state to determine whether recovery is necessary.
1296	 */
1297	error = xlog_check_unmount_rec(log, head_blk, tail_blk, rhead,
1298				       rhead_blk, buffer, &clean);
1299	if (error)
1300		goto done;
1301
1302	/*
1303	 * Verify the log head if the log is not clean (e.g., we have anything
1304	 * but an unmount record at the head). This uses CRC verification to
1305	 * detect and trim torn writes. If discovered, CRC failures are
1306	 * considered torn writes and the log head is trimmed accordingly.
1307	 *
1308	 * Note that we can only run CRC verification when the log is dirty
1309	 * because there's no guarantee that the log data behind an unmount
1310	 * record is compatible with the current architecture.
1311	 */
1312	if (!clean) {
1313		xfs_daddr_t	orig_head = *head_blk;
1314
1315		error = xlog_verify_head(log, head_blk, tail_blk, buffer,
1316					 &rhead_blk, &rhead, &wrapped);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1317		if (error)
1318			goto done;
1319
1320		/* update in-core state again if the head changed */
1321		if (*head_blk != orig_head) {
1322			xlog_set_state(log, *head_blk, rhead, rhead_blk,
1323				       wrapped);
1324			tail_lsn = atomic64_read(&log->l_tail_lsn);
1325			error = xlog_check_unmount_rec(log, head_blk, tail_blk,
1326						       rhead, rhead_blk, buffer,
1327						       &clean);
1328			if (error)
1329				goto done;
 
 
 
 
 
 
 
 
 
 
1330		}
1331	}
1332
1333	/*
1334	 * Note that the unmount was clean. If the unmount was not clean, we
1335	 * need to know this to rebuild the superblock counters from the perag
1336	 * headers if we have a filesystem using non-persistent counters.
1337	 */
1338	if (clean)
1339		xfs_set_clean(log->l_mp);
1340
1341	/*
1342	 * Make sure that there are no blocks in front of the head
1343	 * with the same cycle number as the head.  This can happen
1344	 * because we allow multiple outstanding log writes concurrently,
1345	 * and the later writes might make it out before earlier ones.
1346	 *
1347	 * We use the lsn from before modifying it so that we'll never
1348	 * overwrite the unmount record after a clean unmount.
1349	 *
1350	 * Do this only if we are going to recover the filesystem
1351	 *
1352	 * NOTE: This used to say "if (!readonly)"
1353	 * However on Linux, we can & do recover a read-only filesystem.
1354	 * We only skip recovery if NORECOVERY is specified on mount,
1355	 * in which case we would not be here.
1356	 *
1357	 * But... if the -device- itself is readonly, just skip this.
1358	 * We can't recover this device anyway, so it won't matter.
1359	 */
1360	if (!xfs_readonly_buftarg(log->l_targ))
1361		error = xlog_clear_stale_blocks(log, tail_lsn);
1362
1363done:
1364	kvfree(buffer);
1365
1366	if (error)
1367		xfs_warn(log->l_mp, "failed to locate log tail");
1368	return error;
1369}
1370
1371/*
1372 * Is the log zeroed at all?
1373 *
1374 * The last binary search should be changed to perform an X block read
1375 * once X becomes small enough.  You can then search linearly through
1376 * the X blocks.  This will cut down on the number of reads we need to do.
1377 *
1378 * If the log is partially zeroed, this routine will pass back the blkno
1379 * of the first block with cycle number 0.  It won't have a complete LR
1380 * preceding it.
1381 *
1382 * Return:
1383 *	0  => the log is completely written to
1384 *	1 => use *blk_no as the first block of the log
1385 *	<0 => error has occurred
1386 */
1387STATIC int
1388xlog_find_zeroed(
1389	struct xlog	*log,
1390	xfs_daddr_t	*blk_no)
1391{
1392	char		*buffer;
1393	char		*offset;
1394	uint	        first_cycle, last_cycle;
1395	xfs_daddr_t	new_blk, last_blk, start_blk;
1396	xfs_daddr_t     num_scan_bblks;
1397	int	        error, log_bbnum = log->l_logBBsize;
1398	int		ret = 1;
1399
1400	*blk_no = 0;
1401
1402	/* check totally zeroed log */
1403	buffer = xlog_alloc_buffer(log, 1);
1404	if (!buffer)
1405		return -ENOMEM;
1406	error = xlog_bread(log, 0, 1, buffer, &offset);
1407	if (error)
1408		goto out_free_buffer;
1409
1410	first_cycle = xlog_get_cycle(offset);
1411	if (first_cycle == 0) {		/* completely zeroed log */
1412		*blk_no = 0;
1413		goto out_free_buffer;
 
1414	}
1415
1416	/* check partially zeroed log */
1417	error = xlog_bread(log, log_bbnum-1, 1, buffer, &offset);
1418	if (error)
1419		goto out_free_buffer;
1420
1421	last_cycle = xlog_get_cycle(offset);
1422	if (last_cycle != 0) {		/* log completely written to */
1423		ret = 0;
1424		goto out_free_buffer;
 
 
 
 
 
 
 
 
 
 
1425	}
1426
1427	/* we have a partially zeroed log */
1428	last_blk = log_bbnum-1;
1429	error = xlog_find_cycle_start(log, buffer, 0, &last_blk, 0);
1430	if (error)
1431		goto out_free_buffer;
1432
1433	/*
1434	 * Validate the answer.  Because there is no way to guarantee that
1435	 * the entire log is made up of log records which are the same size,
1436	 * we scan over the defined maximum blocks.  At this point, the maximum
1437	 * is not chosen to mean anything special.   XXXmiken
1438	 */
1439	num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
1440	ASSERT(num_scan_bblks <= INT_MAX);
1441
1442	if (last_blk < num_scan_bblks)
1443		num_scan_bblks = last_blk;
1444	start_blk = last_blk - num_scan_bblks;
1445
1446	/*
1447	 * We search for any instances of cycle number 0 that occur before
1448	 * our current estimate of the head.  What we're trying to detect is
1449	 *        1 ... | 0 | 1 | 0...
1450	 *                       ^ binary search ends here
1451	 */
1452	if ((error = xlog_find_verify_cycle(log, start_blk,
1453					 (int)num_scan_bblks, 0, &new_blk)))
1454		goto out_free_buffer;
1455	if (new_blk != -1)
1456		last_blk = new_blk;
1457
1458	/*
1459	 * Potentially backup over partial log record write.  We don't need
1460	 * to search the end of the log because we know it is zero.
1461	 */
1462	error = xlog_find_verify_log_record(log, start_blk, &last_blk, 0);
1463	if (error == 1)
1464		error = -EIO;
1465	if (error)
1466		goto out_free_buffer;
 
1467
1468	*blk_no = last_blk;
1469out_free_buffer:
1470	kvfree(buffer);
1471	if (error)
1472		return error;
1473	return ret;
1474}
1475
1476/*
1477 * These are simple subroutines used by xlog_clear_stale_blocks() below
1478 * to initialize a buffer full of empty log record headers and write
1479 * them into the log.
1480 */
1481STATIC void
1482xlog_add_record(
1483	struct xlog		*log,
1484	char			*buf,
1485	int			cycle,
1486	int			block,
1487	int			tail_cycle,
1488	int			tail_block)
1489{
1490	xlog_rec_header_t	*recp = (xlog_rec_header_t *)buf;
1491
1492	memset(buf, 0, BBSIZE);
1493	recp->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1494	recp->h_cycle = cpu_to_be32(cycle);
1495	recp->h_version = cpu_to_be32(
1496			xfs_has_logv2(log->l_mp) ? 2 : 1);
1497	recp->h_lsn = cpu_to_be64(xlog_assign_lsn(cycle, block));
1498	recp->h_tail_lsn = cpu_to_be64(xlog_assign_lsn(tail_cycle, tail_block));
1499	recp->h_fmt = cpu_to_be32(XLOG_FMT);
1500	memcpy(&recp->h_fs_uuid, &log->l_mp->m_sb.sb_uuid, sizeof(uuid_t));
1501}
1502
1503STATIC int
1504xlog_write_log_records(
1505	struct xlog	*log,
1506	int		cycle,
1507	int		start_block,
1508	int		blocks,
1509	int		tail_cycle,
1510	int		tail_block)
1511{
1512	char		*offset;
1513	char		*buffer;
1514	int		balign, ealign;
1515	int		sectbb = log->l_sectBBsize;
1516	int		end_block = start_block + blocks;
1517	int		bufblks;
1518	int		error = 0;
1519	int		i, j = 0;
1520
1521	/*
1522	 * Greedily allocate a buffer big enough to handle the full
1523	 * range of basic blocks to be written.  If that fails, try
1524	 * a smaller size.  We need to be able to write at least a
1525	 * log sector, or we're out of luck.
1526	 */
1527	bufblks = roundup_pow_of_two(blocks);
1528	while (bufblks > log->l_logBBsize)
1529		bufblks >>= 1;
1530	while (!(buffer = xlog_alloc_buffer(log, bufblks))) {
1531		bufblks >>= 1;
1532		if (bufblks < sectbb)
1533			return -ENOMEM;
1534	}
1535
1536	/* We may need to do a read at the start to fill in part of
1537	 * the buffer in the starting sector not covered by the first
1538	 * write below.
1539	 */
1540	balign = round_down(start_block, sectbb);
1541	if (balign != start_block) {
1542		error = xlog_bread_noalign(log, start_block, 1, buffer);
1543		if (error)
1544			goto out_free_buffer;
1545
1546		j = start_block - balign;
1547	}
1548
1549	for (i = start_block; i < end_block; i += bufblks) {
1550		int		bcount, endcount;
1551
1552		bcount = min(bufblks, end_block - start_block);
1553		endcount = bcount - j;
1554
1555		/* We may need to do a read at the end to fill in part of
1556		 * the buffer in the final sector not covered by the write.
1557		 * If this is the same sector as the above read, skip it.
1558		 */
1559		ealign = round_down(end_block, sectbb);
1560		if (j == 0 && (start_block + endcount > ealign)) {
1561			error = xlog_bread_noalign(log, ealign, sectbb,
1562					buffer + BBTOB(ealign - start_block));
 
1563			if (error)
1564				break;
1565
1566		}
1567
1568		offset = buffer + xlog_align(log, start_block);
1569		for (; j < endcount; j++) {
1570			xlog_add_record(log, offset, cycle, i+j,
1571					tail_cycle, tail_block);
1572			offset += BBSIZE;
1573		}
1574		error = xlog_bwrite(log, start_block, endcount, buffer);
1575		if (error)
1576			break;
1577		start_block += endcount;
1578		j = 0;
1579	}
1580
1581out_free_buffer:
1582	kvfree(buffer);
1583	return error;
1584}
1585
1586/*
1587 * This routine is called to blow away any incomplete log writes out
1588 * in front of the log head.  We do this so that we won't become confused
1589 * if we come up, write only a little bit more, and then crash again.
1590 * If we leave the partial log records out there, this situation could
1591 * cause us to think those partial writes are valid blocks since they
1592 * have the current cycle number.  We get rid of them by overwriting them
1593 * with empty log records with the old cycle number rather than the
1594 * current one.
1595 *
1596 * The tail lsn is passed in rather than taken from
1597 * the log so that we will not write over the unmount record after a
1598 * clean unmount in a 512 block log.  Doing so would leave the log without
1599 * any valid log records in it until a new one was written.  If we crashed
1600 * during that time we would not be able to recover.
1601 */
1602STATIC int
1603xlog_clear_stale_blocks(
1604	struct xlog	*log,
1605	xfs_lsn_t	tail_lsn)
1606{
1607	int		tail_cycle, head_cycle;
1608	int		tail_block, head_block;
1609	int		tail_distance, max_distance;
1610	int		distance;
1611	int		error;
1612
1613	tail_cycle = CYCLE_LSN(tail_lsn);
1614	tail_block = BLOCK_LSN(tail_lsn);
1615	head_cycle = log->l_curr_cycle;
1616	head_block = log->l_curr_block;
1617
1618	/*
1619	 * Figure out the distance between the new head of the log
1620	 * and the tail.  We want to write over any blocks beyond the
1621	 * head that we may have written just before the crash, but
1622	 * we don't want to overwrite the tail of the log.
1623	 */
1624	if (head_cycle == tail_cycle) {
1625		/*
1626		 * The tail is behind the head in the physical log,
1627		 * so the distance from the head to the tail is the
1628		 * distance from the head to the end of the log plus
1629		 * the distance from the beginning of the log to the
1630		 * tail.
1631		 */
1632		if (XFS_IS_CORRUPT(log->l_mp,
1633				   head_block < tail_block ||
1634				   head_block >= log->l_logBBsize))
1635			return -EFSCORRUPTED;
 
1636		tail_distance = tail_block + (log->l_logBBsize - head_block);
1637	} else {
1638		/*
1639		 * The head is behind the tail in the physical log,
1640		 * so the distance from the head to the tail is just
1641		 * the tail block minus the head block.
1642		 */
1643		if (XFS_IS_CORRUPT(log->l_mp,
1644				   head_block >= tail_block ||
1645				   head_cycle != tail_cycle + 1))
1646			return -EFSCORRUPTED;
 
1647		tail_distance = tail_block - head_block;
1648	}
1649
1650	/*
1651	 * If the head is right up against the tail, we can't clear
1652	 * anything.
1653	 */
1654	if (tail_distance <= 0) {
1655		ASSERT(tail_distance == 0);
1656		return 0;
1657	}
1658
1659	max_distance = XLOG_TOTAL_REC_SHIFT(log);
1660	/*
1661	 * Take the smaller of the maximum amount of outstanding I/O
1662	 * we could have and the distance to the tail to clear out.
1663	 * We take the smaller so that we don't overwrite the tail and
1664	 * we don't waste all day writing from the head to the tail
1665	 * for no reason.
1666	 */
1667	max_distance = min(max_distance, tail_distance);
1668
1669	if ((head_block + max_distance) <= log->l_logBBsize) {
1670		/*
1671		 * We can stomp all the blocks we need to without
1672		 * wrapping around the end of the log.  Just do it
1673		 * in a single write.  Use the cycle number of the
1674		 * current cycle minus one so that the log will look like:
1675		 *     n ... | n - 1 ...
1676		 */
1677		error = xlog_write_log_records(log, (head_cycle - 1),
1678				head_block, max_distance, tail_cycle,
1679				tail_block);
1680		if (error)
1681			return error;
1682	} else {
1683		/*
1684		 * We need to wrap around the end of the physical log in
1685		 * order to clear all the blocks.  Do it in two separate
1686		 * I/Os.  The first write should be from the head to the
1687		 * end of the physical log, and it should use the current
1688		 * cycle number minus one just like above.
1689		 */
1690		distance = log->l_logBBsize - head_block;
1691		error = xlog_write_log_records(log, (head_cycle - 1),
1692				head_block, distance, tail_cycle,
1693				tail_block);
1694
1695		if (error)
1696			return error;
1697
1698		/*
1699		 * Now write the blocks at the start of the physical log.
1700		 * This writes the remainder of the blocks we want to clear.
1701		 * It uses the current cycle number since we're now on the
1702		 * same cycle as the head so that we get:
1703		 *    n ... n ... | n - 1 ...
1704		 *    ^^^^^ blocks we're writing
1705		 */
1706		distance = max_distance - (log->l_logBBsize - head_block);
1707		error = xlog_write_log_records(log, head_cycle, 0, distance,
1708				tail_cycle, tail_block);
1709		if (error)
1710			return error;
1711	}
1712
1713	return 0;
1714}
1715
1716/*
1717 * Release the recovered intent item in the AIL that matches the given intent
1718 * type and intent id.
 
 
1719 */
1720void
1721xlog_recover_release_intent(
1722	struct xlog			*log,
1723	unsigned short			intent_type,
1724	uint64_t			intent_id)
1725{
1726	struct xfs_defer_pending	*dfp, *n;
1727
1728	list_for_each_entry_safe(dfp, n, &log->r_dfops, dfp_list) {
1729		struct xfs_log_item	*lip = dfp->dfp_intent;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1730
1731		if (lip->li_type != intent_type)
1732			continue;
1733		if (!lip->li_ops->iop_match(lip, intent_id))
1734			continue;
1735
1736		ASSERT(xlog_item_is_intent(lip));
 
 
 
 
1737
1738		xfs_defer_cancel_recovery(log->l_mp, dfp);
1739	}
 
1740}
1741
1742int
1743xlog_recover_iget(
1744	struct xfs_mount	*mp,
1745	xfs_ino_t		ino,
1746	struct xfs_inode	**ipp)
 
1747{
1748	int			error;
 
 
1749
1750	error = xfs_iget(mp, NULL, ino, 0, 0, ipp);
1751	if (error)
1752		return error;
1753
1754	error = xfs_qm_dqattach(*ipp);
1755	if (error) {
1756		xfs_irele(*ipp);
1757		return error;
1758	}
 
 
1759
1760	if (VFS_I(*ipp)->i_nlink == 0)
1761		xfs_iflags_set(*ipp, XFS_IRECOVERY);
1762
 
 
 
 
 
1763	return 0;
1764}
1765
1766/*
1767 * Get an inode so that we can recover a log operation.
 
 
 
1768 *
1769 * Log intent items that target inodes effectively contain a file handle.
1770 * Check that the generation number matches the intent item like we do for
1771 * other file handles.  Log intent items defined after this validation weakness
1772 * was identified must use this function.
 
 
1773 */
1774int
1775xlog_recover_iget_handle(
1776	struct xfs_mount	*mp,
1777	xfs_ino_t		ino,
1778	uint32_t		gen,
1779	struct xfs_inode	**ipp)
1780{
1781	struct xfs_inode	*ip;
1782	int			error;
 
1783
1784	error = xlog_recover_iget(mp, ino, &ip);
1785	if (error)
1786		return error;
1787
1788	if (VFS_I(ip)->i_generation != gen) {
1789		xfs_irele(ip);
1790		return -EFSCORRUPTED;
 
 
 
 
 
 
 
1791	}
1792
1793	*ipp = ip;
1794	return 0;
1795}
1796
1797/******************************************************************************
1798 *
1799 *		Log recover routines
1800 *
1801 ******************************************************************************
1802 */
1803static const struct xlog_recover_item_ops *xlog_recover_item_ops[] = {
1804	&xlog_buf_item_ops,
1805	&xlog_inode_item_ops,
1806	&xlog_dquot_item_ops,
1807	&xlog_quotaoff_item_ops,
1808	&xlog_icreate_item_ops,
1809	&xlog_efi_item_ops,
1810	&xlog_efd_item_ops,
1811	&xlog_rui_item_ops,
1812	&xlog_rud_item_ops,
1813	&xlog_cui_item_ops,
1814	&xlog_cud_item_ops,
1815	&xlog_bui_item_ops,
1816	&xlog_bud_item_ops,
1817	&xlog_attri_item_ops,
1818	&xlog_attrd_item_ops,
1819	&xlog_xmi_item_ops,
1820	&xlog_xmd_item_ops,
1821	&xlog_rtefi_item_ops,
1822	&xlog_rtefd_item_ops,
1823};
1824
1825static const struct xlog_recover_item_ops *
1826xlog_find_item_ops(
1827	struct xlog_recover_item		*item)
1828{
1829	unsigned int				i;
1830
1831	for (i = 0; i < ARRAY_SIZE(xlog_recover_item_ops); i++)
1832		if (ITEM_TYPE(item) == xlog_recover_item_ops[i]->item_type)
1833			return xlog_recover_item_ops[i];
 
1834
1835	return NULL;
 
 
 
 
 
 
 
 
 
 
 
1836}
1837
1838/*
1839 * Sort the log items in the transaction.
1840 *
1841 * The ordering constraints are defined by the inode allocation and unlink
1842 * behaviour. The rules are:
1843 *
1844 *	1. Every item is only logged once in a given transaction. Hence it
1845 *	   represents the last logged state of the item. Hence ordering is
1846 *	   dependent on the order in which operations need to be performed so
1847 *	   required initial conditions are always met.
1848 *
1849 *	2. Cancelled buffers are recorded in pass 1 in a separate table and
1850 *	   there's nothing to replay from them so we can simply cull them
1851 *	   from the transaction. However, we can't do that until after we've
1852 *	   replayed all the other items because they may be dependent on the
1853 *	   cancelled buffer and replaying the cancelled buffer can remove it
1854 *	   form the cancelled buffer table. Hence they have to be done last.
1855 *
1856 *	3. Inode allocation buffers must be replayed before inode items that
1857 *	   read the buffer and replay changes into it. For filesystems using the
1858 *	   ICREATE transactions, this means XFS_LI_ICREATE objects need to get
1859 *	   treated the same as inode allocation buffers as they create and
1860 *	   initialise the buffers directly.
1861 *
1862 *	4. Inode unlink buffers must be replayed after inode items are replayed.
1863 *	   This ensures that inodes are completely flushed to the inode buffer
1864 *	   in a "free" state before we remove the unlinked inode list pointer.
1865 *
1866 * Hence the ordering needs to be inode allocation buffers first, inode items
1867 * second, inode unlink buffers third and cancelled buffers last.
1868 *
1869 * But there's a problem with that - we can't tell an inode allocation buffer
1870 * apart from a regular buffer, so we can't separate them. We can, however,
1871 * tell an inode unlink buffer from the others, and so we can separate them out
1872 * from all the other buffers and move them to last.
1873 *
1874 * Hence, 4 lists, in order from head to tail:
1875 *	- buffer_list for all buffers except cancelled/inode unlink buffers
1876 *	- item_list for all non-buffer items
1877 *	- inode_buffer_list for inode unlink buffers
1878 *	- cancel_list for the cancelled buffers
1879 *
1880 * Note that we add objects to the tail of the lists so that first-to-last
1881 * ordering is preserved within the lists. Adding objects to the head of the
1882 * list means when we traverse from the head we walk them in last-to-first
1883 * order. For cancelled buffers and inode unlink buffers this doesn't matter,
1884 * but for all other items there may be specific ordering that we need to
1885 * preserve.
1886 */
1887STATIC int
1888xlog_recover_reorder_trans(
1889	struct xlog		*log,
1890	struct xlog_recover	*trans,
1891	int			pass)
1892{
1893	struct xlog_recover_item *item, *n;
1894	int			error = 0;
1895	LIST_HEAD(sort_list);
1896	LIST_HEAD(cancel_list);
1897	LIST_HEAD(buffer_list);
1898	LIST_HEAD(inode_buffer_list);
1899	LIST_HEAD(item_list);
1900
1901	list_splice_init(&trans->r_itemq, &sort_list);
1902	list_for_each_entry_safe(item, n, &sort_list, ri_list) {
1903		enum xlog_recover_reorder	fate = XLOG_REORDER_ITEM_LIST;
1904
1905		item->ri_ops = xlog_find_item_ops(item);
1906		if (!item->ri_ops) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1907			xfs_warn(log->l_mp,
1908				"%s: unrecognized type of log operation (%d)",
1909				__func__, ITEM_TYPE(item));
1910			ASSERT(0);
1911			/*
1912			 * return the remaining items back to the transaction
1913			 * item list so they can be freed in caller.
1914			 */
1915			if (!list_empty(&sort_list))
1916				list_splice_init(&sort_list, &trans->r_itemq);
1917			error = -EFSCORRUPTED;
1918			break;
1919		}
1920
1921		if (item->ri_ops->reorder)
1922			fate = item->ri_ops->reorder(item);
1923
1924		switch (fate) {
1925		case XLOG_REORDER_BUFFER_LIST:
1926			list_move_tail(&item->ri_list, &buffer_list);
1927			break;
1928		case XLOG_REORDER_CANCEL_LIST:
1929			trace_xfs_log_recover_item_reorder_head(log,
1930					trans, item, pass);
1931			list_move(&item->ri_list, &cancel_list);
1932			break;
1933		case XLOG_REORDER_INODE_BUFFER_LIST:
1934			list_move(&item->ri_list, &inode_buffer_list);
1935			break;
1936		case XLOG_REORDER_ITEM_LIST:
1937			trace_xfs_log_recover_item_reorder_tail(log,
1938							trans, item, pass);
1939			list_move_tail(&item->ri_list, &item_list);
1940			break;
1941		}
1942	}
1943
1944	ASSERT(list_empty(&sort_list));
1945	if (!list_empty(&buffer_list))
1946		list_splice(&buffer_list, &trans->r_itemq);
1947	if (!list_empty(&item_list))
1948		list_splice_tail(&item_list, &trans->r_itemq);
1949	if (!list_empty(&inode_buffer_list))
1950		list_splice_tail(&inode_buffer_list, &trans->r_itemq);
1951	if (!list_empty(&cancel_list))
1952		list_splice_tail(&cancel_list, &trans->r_itemq);
1953	return error;
1954}
1955
1956void
1957xlog_buf_readahead(
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1958	struct xlog		*log,
1959	xfs_daddr_t		blkno,
1960	uint			len,
1961	const struct xfs_buf_ops *ops)
1962{
1963	if (!xlog_is_buffer_cancelled(log, blkno, len))
1964		xfs_buf_readahead(log->l_mp->m_ddev_targp, blkno, len, ops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1965}
1966
1967/*
1968 * Create a deferred work structure for resuming and tracking the progress of a
1969 * log intent item that was found during recovery.
 
 
 
 
 
 
1970 */
1971void
1972xlog_recover_intent_item(
1973	struct xlog			*log,
1974	struct xfs_log_item		*lip,
1975	xfs_lsn_t			lsn,
1976	const struct xfs_defer_op_type	*ops)
1977{
1978	ASSERT(xlog_item_is_intent(lip));
1979
1980	xfs_defer_start_recovery(lip, &log->r_dfops, ops);
 
 
1981
1982	/*
1983	 * Insert the intent into the AIL directly and drop one reference so
1984	 * that finishing or canceling the work will drop the other.
1985	 */
1986	xfs_trans_ail_insert(log->l_ailp, lip, lsn);
1987	lip->li_ops->iop_unpin(lip, 0);
 
 
 
 
 
 
 
1988}
1989
 
 
 
 
 
 
 
 
 
 
 
 
1990STATIC int
1991xlog_recover_items_pass2(
1992	struct xlog                     *log,
1993	struct xlog_recover             *trans,
1994	struct list_head                *buffer_list,
1995	struct list_head                *item_list)
1996{
1997	struct xlog_recover_item	*item;
1998	int				error = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1999
2000	list_for_each_entry(item, item_list, ri_list) {
2001		trace_xfs_log_recover_item_recover(log, trans, item,
2002				XLOG_RECOVER_PASS2);
 
 
 
 
2003
2004		if (item->ri_ops->commit_pass2)
2005			error = item->ri_ops->commit_pass2(log, buffer_list,
2006					item, trans->r_lsn);
2007		if (error)
2008			return error;
2009	}
2010
2011	return error;
2012}
2013
2014/*
2015 * Perform the transaction.
2016 *
2017 * If the transaction modifies a buffer or inode, do it now.  Otherwise,
2018 * EFIs and EFDs get queued up by adding entries into the AIL for them.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2019 */
2020STATIC int
2021xlog_recover_commit_trans(
2022	struct xlog		*log,
2023	struct xlog_recover	*trans,
2024	int			pass,
2025	struct list_head	*buffer_list)
2026{
2027	int				error = 0;
2028	int				items_queued = 0;
2029	struct xlog_recover_item	*item;
2030	struct xlog_recover_item	*next;
2031	LIST_HEAD			(ra_list);
2032	LIST_HEAD			(done_list);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2033
2034	#define XLOG_RECOVER_COMMIT_QUEUE_MAX 100
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2035
2036	hlist_del_init(&trans->r_list);
2037
2038	error = xlog_recover_reorder_trans(log, trans, pass);
2039	if (error)
2040		return error;
2041
2042	list_for_each_entry_safe(item, next, &trans->r_itemq, ri_list) {
2043		trace_xfs_log_recover_item_recover(log, trans, item, pass);
2044
2045		switch (pass) {
2046		case XLOG_RECOVER_PASS1:
2047			if (item->ri_ops->commit_pass1)
2048				error = item->ri_ops->commit_pass1(log, item);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2049			break;
2050		case XLOG_RECOVER_PASS2:
2051			if (item->ri_ops->ra_pass2)
2052				item->ri_ops->ra_pass2(log, item);
2053			list_move_tail(&item->ri_list, &ra_list);
2054			items_queued++;
2055			if (items_queued >= XLOG_RECOVER_COMMIT_QUEUE_MAX) {
2056				error = xlog_recover_items_pass2(log, trans,
2057						buffer_list, &ra_list);
2058				list_splice_tail_init(&ra_list, &done_list);
2059				items_queued = 0;
2060			}
2061
2062			break;
2063		default:
 
 
 
 
 
 
 
 
2064			ASSERT(0);
 
 
 
 
 
 
 
 
 
 
 
2065		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2066
2067		if (error)
2068			goto out;
2069	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2070
2071out:
2072	if (!list_empty(&ra_list)) {
2073		if (!error)
2074			error = xlog_recover_items_pass2(log, trans,
2075					buffer_list, &ra_list);
2076		list_splice_tail_init(&ra_list, &done_list);
 
2077	}
2078
2079	if (!list_empty(&done_list))
2080		list_splice_init(&done_list, &trans->r_itemq);
2081
2082	return error;
 
 
 
 
 
 
 
 
 
2083}
2084
 
 
 
 
 
 
2085STATIC void
2086xlog_recover_add_item(
2087	struct list_head	*head)
 
 
 
 
2088{
2089	struct xlog_recover_item *item;
 
 
 
 
 
 
 
 
 
2090
2091	item = kzalloc(sizeof(struct xlog_recover_item),
2092			GFP_KERNEL | __GFP_NOFAIL);
2093	INIT_LIST_HEAD(&item->ri_list);
2094	list_add_tail(&item->ri_list, head);
 
 
 
 
 
 
 
 
 
 
2095}
2096
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2097STATIC int
2098xlog_recover_add_to_cont_trans(
2099	struct xlog		*log,
2100	struct xlog_recover	*trans,
2101	char			*dp,
2102	int			len)
2103{
2104	struct xlog_recover_item *item;
2105	char			*ptr, *old_ptr;
2106	int			old_len;
 
 
 
2107
2108	/*
2109	 * If the transaction is empty, the header was split across this and the
2110	 * previous record. Copy the rest of the header.
2111	 */
2112	if (list_empty(&trans->r_itemq)) {
2113		ASSERT(len <= sizeof(struct xfs_trans_header));
2114		if (len > sizeof(struct xfs_trans_header)) {
2115			xfs_warn(log->l_mp, "%s: bad header length", __func__);
2116			return -EFSCORRUPTED;
2117		}
 
2118
2119		xlog_recover_add_item(&trans->r_itemq);
2120		ptr = (char *)&trans->r_theader +
2121				sizeof(struct xfs_trans_header) - len;
2122		memcpy(ptr, dp, len);
2123		return 0;
 
 
 
 
 
 
 
2124	}
2125
2126	/* take the tail entry */
2127	item = list_entry(trans->r_itemq.prev, struct xlog_recover_item,
2128			  ri_list);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2129
2130	old_ptr = item->ri_buf[item->ri_cnt-1].i_addr;
2131	old_len = item->ri_buf[item->ri_cnt-1].i_len;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2132
2133	ptr = kvrealloc(old_ptr, len + old_len, GFP_KERNEL);
2134	if (!ptr)
2135		return -ENOMEM;
2136	memcpy(&ptr[old_len], dp, len);
2137	item->ri_buf[item->ri_cnt-1].i_len += len;
2138	item->ri_buf[item->ri_cnt-1].i_addr = ptr;
2139	trace_xfs_log_recover_item_add_cont(log, trans, item, 0);
2140	return 0;
2141}
2142
2143/*
2144 * The next region to add is the start of a new region.  It could be
2145 * a whole region or it could be the first part of a new region.  Because
2146 * of this, the assumption here is that the type and size fields of all
2147 * format structures fit into the first 32 bits of the structure.
2148 *
2149 * This works because all regions must be 32 bit aligned.  Therefore, we
2150 * either have both fields or we have neither field.  In the case we have
2151 * neither field, the data part of the region is zero length.  We only have
2152 * a log_op_header and can throw away the header since a new one will appear
2153 * later.  If we have at least 4 bytes, then we can determine how many regions
2154 * will appear in the current log item.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2155 */
 
2156STATIC int
2157xlog_recover_add_to_trans(
2158	struct xlog		*log,
2159	struct xlog_recover	*trans,
2160	char			*dp,
2161	int			len)
2162{
2163	struct xfs_inode_log_format	*in_f;			/* any will do */
2164	struct xlog_recover_item *item;
2165	char			*ptr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2166
2167	if (!len)
2168		return 0;
2169	if (list_empty(&trans->r_itemq)) {
2170		/* we need to catch log corruptions here */
2171		if (*(uint *)dp != XFS_TRANS_HEADER_MAGIC) {
2172			xfs_warn(log->l_mp, "%s: bad header magic number",
2173				__func__);
2174			ASSERT(0);
2175			return -EFSCORRUPTED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2176		}
 
 
 
 
2177
2178		if (len > sizeof(struct xfs_trans_header)) {
2179			xfs_warn(log->l_mp, "%s: bad header length", __func__);
2180			ASSERT(0);
2181			return -EFSCORRUPTED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2182		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2183
 
2184		/*
2185		 * The transaction header can be arbitrarily split across op
2186		 * records. If we don't have the whole thing here, copy what we
2187		 * do have and handle the rest in the next record.
2188		 */
2189		if (len == sizeof(struct xfs_trans_header))
2190			xlog_recover_add_item(&trans->r_itemq);
2191		memcpy(&trans->r_theader, dp, len);
2192		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2193	}
2194
2195	ptr = xlog_kvmalloc(len);
2196	memcpy(ptr, dp, len);
2197	in_f = (struct xfs_inode_log_format *)ptr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2198
2199	/* take the tail entry */
2200	item = list_entry(trans->r_itemq.prev, struct xlog_recover_item,
2201			  ri_list);
2202	if (item->ri_total != 0 &&
2203	     item->ri_total == item->ri_cnt) {
2204		/* tail item is in use, get a new one */
2205		xlog_recover_add_item(&trans->r_itemq);
2206		item = list_entry(trans->r_itemq.prev,
2207					struct xlog_recover_item, ri_list);
 
2208	}
2209
2210	if (item->ri_total == 0) {		/* first region to be added */
2211		if (in_f->ilf_size == 0 ||
2212		    in_f->ilf_size > XLOG_MAX_REGIONS_IN_ITEM) {
2213			xfs_warn(log->l_mp,
2214		"bad number of regions (%d) in inode log format",
2215				  in_f->ilf_size);
2216			ASSERT(0);
2217			kvfree(ptr);
2218			return -EFSCORRUPTED;
 
2219		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2220
2221		item->ri_total = in_f->ilf_size;
2222		item->ri_buf = kzalloc(item->ri_total * sizeof(xfs_log_iovec_t),
2223				GFP_KERNEL | __GFP_NOFAIL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2224	}
 
 
 
 
 
2225
2226	if (item->ri_total <= item->ri_cnt) {
2227		xfs_warn(log->l_mp,
2228	"log item region count (%d) overflowed size (%d)",
2229				item->ri_cnt, item->ri_total);
2230		ASSERT(0);
2231		kvfree(ptr);
2232		return -EFSCORRUPTED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2233	}
2234
2235	/* Description region is ri_buf[0] */
2236	item->ri_buf[item->ri_cnt].i_addr = ptr;
2237	item->ri_buf[item->ri_cnt].i_len  = len;
2238	item->ri_cnt++;
2239	trace_xfs_log_recover_item_add(log, trans, item, 0);
 
 
 
 
 
 
 
 
 
 
 
2240	return 0;
2241}
2242
2243/*
2244 * Free up any resources allocated by the transaction
2245 *
2246 * Remember that EFIs, EFDs, and IUNLINKs are handled later.
2247 */
2248STATIC void
2249xlog_recover_free_trans(
2250	struct xlog_recover	*trans)
2251{
2252	struct xlog_recover_item *item, *n;
2253	int			i;
2254
2255	hlist_del_init(&trans->r_list);
2256
2257	list_for_each_entry_safe(item, n, &trans->r_itemq, ri_list) {
2258		/* Free the regions in the item. */
2259		list_del(&item->ri_list);
2260		for (i = 0; i < item->ri_cnt; i++)
2261			kvfree(item->ri_buf[i].i_addr);
2262		/* Free the item itself */
2263		kfree(item->ri_buf);
2264		kfree(item);
2265	}
2266	/* Free the transaction recover structure */
2267	kfree(trans);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2268}
2269
2270/*
2271 * On error or completion, trans is freed.
2272 */
2273STATIC int
2274xlog_recovery_process_trans(
2275	struct xlog		*log,
2276	struct xlog_recover	*trans,
2277	char			*dp,
2278	unsigned int		len,
2279	unsigned int		flags,
2280	int			pass,
2281	struct list_head	*buffer_list)
2282{
2283	int			error = 0;
2284	bool			freeit = false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2285
2286	/* mask off ophdr transaction container flags */
2287	flags &= ~XLOG_END_TRANS;
2288	if (flags & XLOG_WAS_CONT_TRANS)
2289		flags &= ~XLOG_CONTINUE_TRANS;
2290
2291	/*
2292	 * Callees must not free the trans structure. We'll decide if we need to
2293	 * free it or not based on the operation being done and it's result.
2294	 */
2295	switch (flags) {
2296	/* expected flag values */
2297	case 0:
2298	case XLOG_CONTINUE_TRANS:
2299		error = xlog_recover_add_to_trans(log, trans, dp, len);
2300		break;
2301	case XLOG_WAS_CONT_TRANS:
2302		error = xlog_recover_add_to_cont_trans(log, trans, dp, len);
2303		break;
2304	case XLOG_COMMIT_TRANS:
2305		error = xlog_recover_commit_trans(log, trans, pass,
2306						  buffer_list);
2307		/* success or fail, we are now done with this transaction. */
2308		freeit = true;
2309		break;
2310
2311	/* unexpected flag values */
2312	case XLOG_UNMOUNT_TRANS:
2313		/* just skip trans */
2314		xfs_warn(log->l_mp, "%s: Unmount LR", __func__);
2315		freeit = true;
2316		break;
2317	case XLOG_START_TRANS:
 
 
2318	default:
2319		xfs_warn(log->l_mp, "%s: bad flag 0x%x", __func__, flags);
2320		ASSERT(0);
2321		error = -EFSCORRUPTED;
2322		break;
2323	}
2324	if (error || freeit)
2325		xlog_recover_free_trans(trans);
2326	return error;
2327}
2328
2329/*
2330 * Lookup the transaction recovery structure associated with the ID in the
2331 * current ophdr. If the transaction doesn't exist and the start flag is set in
2332 * the ophdr, then allocate a new transaction for future ID matches to find.
2333 * Either way, return what we found during the lookup - an existing transaction
2334 * or nothing.
2335 */
2336STATIC struct xlog_recover *
2337xlog_recover_ophdr_to_trans(
2338	struct hlist_head	rhash[],
2339	struct xlog_rec_header	*rhead,
2340	struct xlog_op_header	*ohead)
2341{
2342	struct xlog_recover	*trans;
2343	xlog_tid_t		tid;
2344	struct hlist_head	*rhp;
2345
2346	tid = be32_to_cpu(ohead->oh_tid);
2347	rhp = &rhash[XLOG_RHASH(tid)];
2348	hlist_for_each_entry(trans, rhp, r_list) {
2349		if (trans->r_log_tid == tid)
2350			return trans;
 
 
 
 
 
 
 
 
 
 
 
 
2351	}
 
2352
2353	/*
2354	 * skip over non-start transaction headers - we could be
2355	 * processing slack space before the next transaction starts
2356	 */
2357	if (!(ohead->oh_flags & XLOG_START_TRANS))
2358		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2359
2360	ASSERT(be32_to_cpu(ohead->oh_len) == 0);
 
 
 
 
 
 
 
 
2361
2362	/*
2363	 * This is a new transaction so allocate a new recovery container to
2364	 * hold the recovery ops that will follow.
2365	 */
2366	trans = kzalloc(sizeof(struct xlog_recover), GFP_KERNEL | __GFP_NOFAIL);
2367	trans->r_log_tid = tid;
2368	trans->r_lsn = be64_to_cpu(rhead->h_lsn);
2369	INIT_LIST_HEAD(&trans->r_itemq);
2370	INIT_HLIST_NODE(&trans->r_list);
2371	hlist_add_head(&trans->r_list, rhp);
2372
2373	/*
2374	 * Nothing more to do for this ophdr. Items to be added to this new
2375	 * transaction will be in subsequent ophdr containers.
2376	 */
2377	return NULL;
2378}
2379
 
 
 
 
 
 
2380STATIC int
2381xlog_recover_process_ophdr(
2382	struct xlog		*log,
2383	struct hlist_head	rhash[],
2384	struct xlog_rec_header	*rhead,
2385	struct xlog_op_header	*ohead,
2386	char			*dp,
2387	char			*end,
2388	int			pass,
2389	struct list_head	*buffer_list)
2390{
2391	struct xlog_recover	*trans;
2392	unsigned int		len;
2393	int			error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2394
2395	/* Do we understand who wrote this op? */
2396	if (ohead->oh_clientid != XFS_TRANSACTION &&
2397	    ohead->oh_clientid != XFS_LOG) {
2398		xfs_warn(log->l_mp, "%s: bad clientid 0x%x",
2399			__func__, ohead->oh_clientid);
2400		ASSERT(0);
2401		return -EFSCORRUPTED;
2402	}
2403
2404	/*
2405	 * Check the ophdr contains all the data it is supposed to contain.
2406	 */
2407	len = be32_to_cpu(ohead->oh_len);
2408	if (dp + len > end) {
2409		xfs_warn(log->l_mp, "%s: bad length 0x%x", __func__, len);
2410		WARN_ON(1);
2411		return -EFSCORRUPTED;
2412	}
2413
2414	trans = xlog_recover_ophdr_to_trans(rhash, rhead, ohead);
2415	if (!trans) {
2416		/* nothing to do, so skip over this ophdr */
2417		return 0;
2418	}
2419
2420	/*
2421	 * The recovered buffer queue is drained only once we know that all
2422	 * recovery items for the current LSN have been processed. This is
2423	 * required because:
2424	 *
2425	 * - Buffer write submission updates the metadata LSN of the buffer.
2426	 * - Log recovery skips items with a metadata LSN >= the current LSN of
2427	 *   the recovery item.
2428	 * - Separate recovery items against the same metadata buffer can share
2429	 *   a current LSN. I.e., consider that the LSN of a recovery item is
2430	 *   defined as the starting LSN of the first record in which its
2431	 *   transaction appears, that a record can hold multiple transactions,
2432	 *   and/or that a transaction can span multiple records.
2433	 *
2434	 * In other words, we are allowed to submit a buffer from log recovery
2435	 * once per current LSN. Otherwise, we may incorrectly skip recovery
2436	 * items and cause corruption.
2437	 *
2438	 * We don't know up front whether buffers are updated multiple times per
2439	 * LSN. Therefore, track the current LSN of each commit log record as it
2440	 * is processed and drain the queue when it changes. Use commit records
2441	 * because they are ordered correctly by the logging code.
2442	 */
2443	if (log->l_recovery_lsn != trans->r_lsn &&
2444	    ohead->oh_flags & XLOG_COMMIT_TRANS) {
2445		error = xfs_buf_delwri_submit(buffer_list);
2446		if (error)
2447			return error;
2448		log->l_recovery_lsn = trans->r_lsn;
2449	}
2450
2451	return xlog_recovery_process_trans(log, trans, dp, len,
2452					   ohead->oh_flags, pass, buffer_list);
 
 
 
 
 
 
2453}
2454
2455/*
2456 * There are two valid states of the r_state field.  0 indicates that the
2457 * transaction structure is in a normal state.  We have either seen the
2458 * start of the transaction or the last operation we added was not a partial
2459 * operation.  If the last operation we added to the transaction was a
2460 * partial operation, we need to mark r_state with XLOG_WAS_CONT_TRANS.
2461 *
2462 * NOTE: skip LRs with 0 data length.
2463 */
2464STATIC int
2465xlog_recover_process_data(
2466	struct xlog		*log,
2467	struct hlist_head	rhash[],
2468	struct xlog_rec_header	*rhead,
2469	char			*dp,
2470	int			pass,
2471	struct list_head	*buffer_list)
2472{
2473	struct xlog_op_header	*ohead;
2474	char			*end;
2475	int			num_logops;
 
 
 
2476	int			error;
 
 
2477
2478	end = dp + be32_to_cpu(rhead->h_len);
2479	num_logops = be32_to_cpu(rhead->h_num_logops);
2480
2481	/* check the log format matches our own - else we can't recover */
2482	if (xlog_header_check_recover(log->l_mp, rhead))
2483		return -EIO;
2484
2485	trace_xfs_log_recover_record(log, rhead, pass);
2486	while ((dp < end) && num_logops) {
2487
2488		ohead = (struct xlog_op_header *)dp;
2489		dp += sizeof(*ohead);
2490		if (dp > end) {
2491			xfs_warn(log->l_mp, "%s: op header overrun", __func__);
2492			return -EFSCORRUPTED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2493		}
2494
2495		/* errors will abort recovery */
2496		error = xlog_recover_process_ophdr(log, rhash, rhead, ohead,
2497						   dp, end, pass, buffer_list);
2498		if (error)
2499			return error;
2500
2501		dp += be32_to_cpu(ohead->oh_len);
2502		num_logops--;
2503	}
2504	return 0;
2505}
2506
2507/* Take all the collected deferred ops and finish them in order. */
2508static int
2509xlog_finish_defer_ops(
2510	struct xfs_mount	*mp,
2511	struct list_head	*capture_list)
 
 
 
2512{
2513	struct xfs_defer_capture *dfc, *next;
2514	struct xfs_trans	*tp;
 
2515	int			error = 0;
 
 
2516
2517	list_for_each_entry_safe(dfc, next, capture_list, dfc_list) {
2518		struct xfs_trans_res	resv;
2519		struct xfs_defer_resources dres;
2520
2521		/*
2522		 * Create a new transaction reservation from the captured
2523		 * information.  Set logcount to 1 to force the new transaction
2524		 * to regrant every roll so that we can make forward progress
2525		 * in recovery no matter how full the log might be.
2526		 */
2527		resv.tr_logres = dfc->dfc_logres;
2528		resv.tr_logcount = 1;
2529		resv.tr_logflags = XFS_TRANS_PERM_LOG_RES;
2530
2531		error = xfs_trans_alloc(mp, &resv, dfc->dfc_blkres,
2532				dfc->dfc_rtxres, XFS_TRANS_RESERVE, &tp);
2533		if (error) {
2534			xlog_force_shutdown(mp->m_log, SHUTDOWN_LOG_IO_ERROR);
2535			return error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2536		}
 
 
 
 
 
 
 
2537
2538		/*
2539		 * Transfer to this new transaction all the dfops we captured
2540		 * from recovering a single intent item.
2541		 */
2542		list_del_init(&dfc->dfc_list);
2543		xfs_defer_ops_continue(dfc, tp, &dres);
2544		error = xfs_trans_commit(tp);
2545		xfs_defer_resources_rele(&dres);
2546		if (error)
2547			return error;
 
 
2548	}
2549
2550	ASSERT(list_empty(capture_list));
2551	return 0;
2552}
2553
2554/* Release all the captured defer ops and capture structures in this list. */
2555static void
2556xlog_abort_defer_ops(
2557	struct xfs_mount		*mp,
2558	struct list_head		*capture_list)
2559{
2560	struct xfs_defer_capture	*dfc;
2561	struct xfs_defer_capture	*next;
2562
2563	list_for_each_entry_safe(dfc, next, capture_list, dfc_list) {
2564		list_del_init(&dfc->dfc_list);
2565		xfs_defer_ops_capture_abort(mp, dfc);
2566	}
2567}
2568
2569/*
2570 * When this is called, all of the log intent items which did not have
2571 * corresponding log done items should be in the AIL.  What we do now is update
2572 * the data structures associated with each one.
2573 *
2574 * Since we process the log intent items in normal transactions, they will be
2575 * removed at some point after the commit.  This prevents us from just walking
2576 * down the list processing each one.  We'll use a flag in the intent item to
2577 * skip those that we've already processed and use the AIL iteration mechanism's
2578 * generation count to try to speed this up at least a bit.
2579 *
2580 * When we start, we know that the intents are the only things in the AIL. As we
2581 * process them, however, other items are added to the AIL. Hence we know we
2582 * have started recovery on all the pending intents when we find an non-intent
2583 * item in the AIL.
 
 
2584 */
2585STATIC int
2586xlog_recover_process_intents(
2587	struct xlog			*log)
2588{
2589	LIST_HEAD(capture_list);
2590	struct xfs_defer_pending	*dfp, *n;
2591	int				error = 0;
2592#if defined(DEBUG) || defined(XFS_WARN)
2593	xfs_lsn_t			last_lsn;
2594
2595	last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
 
 
 
 
 
 
 
 
 
 
 
2596#endif
2597
2598	list_for_each_entry_safe(dfp, n, &log->r_dfops, dfp_list) {
2599		ASSERT(xlog_item_is_intent(dfp->dfp_intent));
2600
2601		/*
2602		 * We should never see a redo item with a LSN higher than
2603		 * the last transaction we found in the log at the start
2604		 * of recovery.
2605		 */
2606		ASSERT(XFS_LSN_CMP(last_lsn, dfp->dfp_intent->li_lsn) >= 0);
 
 
 
 
2607
2608		/*
2609		 * NOTE: If your intent processing routine can create more
2610		 * deferred ops, you /must/ attach them to the capture list in
2611		 * the recover routine or else those subsequent intents will be
2612		 * replayed in the wrong order!
2613		 *
2614		 * The recovery function can free the log item, so we must not
2615		 * access dfp->dfp_intent after it returns.  It must dispose of
2616		 * @dfp if it returns 0.
2617		 */
2618		error = xfs_defer_finish_recovery(log->l_mp, dfp,
2619				&capture_list);
2620		if (error)
2621			break;
 
2622	}
2623	if (error)
2624		goto err;
2625
2626	error = xlog_finish_defer_ops(log->l_mp, &capture_list);
2627	if (error)
2628		goto err;
2629
2630	return 0;
2631err:
2632	xlog_abort_defer_ops(log->l_mp, &capture_list);
2633	return error;
2634}
2635
2636/*
2637 * A cancel occurs when the mount has failed and we're bailing out.  Release all
2638 * pending log intent items that we haven't started recovery on so they don't
2639 * pin the AIL.
2640 */
2641STATIC void
2642xlog_recover_cancel_intents(
2643	struct xlog			*log)
2644{
2645	struct xfs_defer_pending	*dfp, *n;
2646
2647	list_for_each_entry_safe(dfp, n, &log->r_dfops, dfp_list) {
2648		ASSERT(xlog_item_is_intent(dfp->dfp_intent));
2649
2650		xfs_defer_cancel_recovery(log->l_mp, dfp);
2651	}
2652}
2653
2654/*
2655 * Transfer ownership of the recovered pending work to the recovery transaction
2656 * and try to finish the work.  If there is more work to be done, the dfp will
2657 * remain attached to the transaction.  If not, the dfp is freed.
2658 */
2659int
2660xlog_recover_finish_intent(
2661	struct xfs_trans		*tp,
2662	struct xfs_defer_pending	*dfp)
2663{
2664	int				error;
2665
2666	list_move(&dfp->dfp_list, &tp->t_dfops);
2667	error = xfs_defer_finish_one(tp, dfp);
2668	if (error == -EAGAIN)
2669		return 0;
2670	return error;
2671}
2672
2673/*
2674 * This routine performs a transaction to null out a bad inode pointer
2675 * in an agi unlinked inode hash bucket.
2676 */
2677STATIC void
2678xlog_recover_clear_agi_bucket(
2679	struct xfs_perag	*pag,
2680	int			bucket)
2681{
2682	struct xfs_mount	*mp = pag_mount(pag);
2683	struct xfs_trans	*tp;
2684	struct xfs_agi		*agi;
2685	struct xfs_buf		*agibp;
2686	int			offset;
2687	int			error;
2688
2689	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_clearagi, 0, 0, 0, &tp);
 
2690	if (error)
2691		goto out_error;
2692
2693	error = xfs_read_agi(pag, tp, 0, &agibp);
2694	if (error)
2695		goto out_abort;
2696
2697	agi = agibp->b_addr;
2698	agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
2699	offset = offsetof(xfs_agi_t, agi_unlinked) +
2700		 (sizeof(xfs_agino_t) * bucket);
2701	xfs_trans_log_buf(tp, agibp, offset,
2702			  (offset + sizeof(xfs_agino_t) - 1));
2703
2704	error = xfs_trans_commit(tp);
2705	if (error)
2706		goto out_error;
2707	return;
2708
2709out_abort:
2710	xfs_trans_cancel(tp);
2711out_error:
2712	xfs_warn(mp, "%s: failed to clear agi %d. Continuing.", __func__,
2713			pag_agno(pag));
2714	return;
2715}
2716
2717static int
2718xlog_recover_iunlink_bucket(
2719	struct xfs_perag	*pag,
2720	struct xfs_agi		*agi,
2721	int			bucket)
2722{
2723	struct xfs_mount	*mp = pag_mount(pag);
2724	struct xfs_inode	*prev_ip = NULL;
2725	struct xfs_inode	*ip;
2726	xfs_agino_t		prev_agino, agino;
2727	int			error = 0;
 
2728
2729	agino = be32_to_cpu(agi->agi_unlinked[bucket]);
2730	while (agino != NULLAGINO) {
2731		error = xfs_iget(mp, NULL, xfs_agino_to_ino(pag, agino), 0, 0,
2732				&ip);
2733		if (error)
2734			break;
2735
2736		ASSERT(VFS_I(ip)->i_nlink == 0);
2737		ASSERT(VFS_I(ip)->i_mode != 0);
2738		xfs_iflags_clear(ip, XFS_IRECOVERY);
2739		agino = ip->i_next_unlinked;
 
 
2740
2741		if (prev_ip) {
2742			ip->i_prev_unlinked = prev_agino;
2743			xfs_irele(prev_ip);
2744
2745			/*
2746			 * Ensure the inode is removed from the unlinked list
2747			 * before we continue so that it won't race with
2748			 * building the in-memory list here. This could be
2749			 * serialised with the agibp lock, but that just
2750			 * serialises via lockstepping and it's much simpler
2751			 * just to flush the inodegc queue and wait for it to
2752			 * complete.
2753			 */
2754			error = xfs_inodegc_flush(mp);
2755			if (error)
2756				break;
2757		}
2758
2759		prev_agino = agino;
2760		prev_ip = ip;
2761	}
 
 
2762
2763	if (prev_ip) {
2764		int	error2;
2765
2766		ip->i_prev_unlinked = prev_agino;
2767		xfs_irele(prev_ip);
2768
2769		error2 = xfs_inodegc_flush(mp);
2770		if (error2 && !error)
2771			return error2;
2772	}
2773	return error;
 
 
 
 
 
2774}
2775
2776/*
2777 * Recover AGI unlinked lists
2778 *
2779 * This is called during recovery to process any inodes which we unlinked but
2780 * not freed when the system crashed.  These inodes will be on the lists in the
2781 * AGI blocks. What we do here is scan all the AGIs and fully truncate and free
2782 * any inodes found on the lists. Each inode is removed from the lists when it
2783 * has been fully truncated and is freed. The freeing of the inode and its
2784 * removal from the list must be atomic.
2785 *
2786 * If everything we touch in the agi processing loop is already in memory, this
2787 * loop can hold the cpu for a long time. It runs without lock contention,
2788 * memory allocation contention, the need wait for IO, etc, and so will run
2789 * until we either run out of inodes to process, run low on memory or we run out
2790 * of log space.
2791 *
2792 * This behaviour is bad for latency on single CPU and non-preemptible kernels,
2793 * and can prevent other filesystem work (such as CIL pushes) from running. This
2794 * can lead to deadlocks if the recovery process runs out of log reservation
2795 * space. Hence we need to yield the CPU when there is other kernel work
2796 * scheduled on this CPU to ensure other scheduled work can run without undue
2797 * latency.
2798 */
2799static void
2800xlog_recover_iunlink_ag(
2801	struct xfs_perag	*pag)
2802{
2803	struct xfs_agi		*agi;
2804	struct xfs_buf		*agibp;
2805	int			bucket;
2806	int			error;
 
 
 
 
2807
2808	error = xfs_read_agi(pag, NULL, 0, &agibp);
2809	if (error) {
2810		/*
2811		 * AGI is b0rked. Don't process it.
2812		 *
2813		 * We should probably mark the filesystem as corrupt after we've
2814		 * recovered all the ag's we can....
2815		 */
2816		return;
2817	}
2818
2819	/*
2820	 * Unlock the buffer so that it can be acquired in the normal course of
2821	 * the transaction to truncate and free each inode.  Because we are not
2822	 * racing with anyone else here for the AGI buffer, we don't even need
2823	 * to hold it locked to read the initial unlinked bucket entries out of
2824	 * the buffer. We keep buffer reference though, so that it stays pinned
2825	 * in memory while we need the buffer.
2826	 */
2827	agi = agibp->b_addr;
2828	xfs_buf_unlock(agibp);
2829
2830	for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++) {
2831		error = xlog_recover_iunlink_bucket(pag, agi, bucket);
 
 
 
2832		if (error) {
2833			/*
2834			 * Bucket is unrecoverable, so only a repair scan can
2835			 * free the remaining unlinked inodes. Just empty the
2836			 * bucket and remaining inodes on it unreferenced and
2837			 * unfreeable.
2838			 */
2839			xlog_recover_clear_agi_bucket(pag, bucket);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2840		}
 
2841	}
2842
2843	xfs_buf_rele(agibp);
2844}
2845
2846static void
2847xlog_recover_process_iunlinks(
2848	struct xlog	*log)
 
 
 
 
 
 
 
 
 
 
 
 
2849{
2850	struct xfs_perag	*pag = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2851
2852	while ((pag = xfs_perag_next(log->l_mp, pag)))
2853		xlog_recover_iunlink_ag(pag);
2854}
2855
2856STATIC void
2857xlog_unpack_data(
2858	struct xlog_rec_header	*rhead,
2859	char			*dp,
2860	struct xlog		*log)
2861{
2862	int			i, j, k;
 
 
 
 
 
2863
2864	for (i = 0; i < BTOBB(be32_to_cpu(rhead->h_len)) &&
2865		  i < (XLOG_HEADER_CYCLE_SIZE / BBSIZE); i++) {
2866		*(__be32 *)dp = *(__be32 *)&rhead->h_cycle_data[i];
2867		dp += BBSIZE;
2868	}
2869
2870	if (xfs_has_logv2(log->l_mp)) {
2871		xlog_in_core_2_t *xhdr = (xlog_in_core_2_t *)rhead;
2872		for ( ; i < BTOBB(be32_to_cpu(rhead->h_len)); i++) {
2873			j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
2874			k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
2875			*(__be32 *)dp = xhdr[j].hic_xheader.xh_cycle_data[k];
2876			dp += BBSIZE;
2877		}
2878	}
2879}
2880
2881/*
2882 * CRC check, unpack and process a log record.
2883 */
2884STATIC int
2885xlog_recover_process(
2886	struct xlog		*log,
2887	struct hlist_head	rhash[],
2888	struct xlog_rec_header	*rhead,
2889	char			*dp,
2890	int			pass,
2891	struct list_head	*buffer_list)
2892{
2893	__le32			old_crc = rhead->h_crc;
2894	__le32			crc;
2895
2896	crc = xlog_cksum(log, rhead, dp, be32_to_cpu(rhead->h_len));
2897
2898	/*
2899	 * Nothing else to do if this is a CRC verification pass. Just return
2900	 * if this a record with a non-zero crc. Unfortunately, mkfs always
2901	 * sets old_crc to 0 so we must consider this valid even on v5 supers.
2902	 * Otherwise, return EFSBADCRC on failure so the callers up the stack
2903	 * know precisely what failed.
2904	 */
2905	if (pass == XLOG_RECOVER_CRCPASS) {
2906		if (old_crc && crc != old_crc)
2907			return -EFSBADCRC;
2908		return 0;
2909	}
2910
2911	/*
2912	 * We're in the normal recovery path. Issue a warning if and only if the
2913	 * CRC in the header is non-zero. This is an advisory warning and the
2914	 * zero CRC check prevents warnings from being emitted when upgrading
2915	 * the kernel from one that does not add CRCs by default.
2916	 */
2917	if (crc != old_crc) {
2918		if (old_crc || xfs_has_crc(log->l_mp)) {
2919			xfs_alert(log->l_mp,
2920		"log record CRC mismatch: found 0x%x, expected 0x%x.",
2921					le32_to_cpu(old_crc),
2922					le32_to_cpu(crc));
2923			xfs_hex_dump(dp, 32);
2924		}
2925
2926		/*
2927		 * If the filesystem is CRC enabled, this mismatch becomes a
2928		 * fatal log corruption failure.
2929		 */
2930		if (xfs_has_crc(log->l_mp)) {
2931			XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
2932			return -EFSCORRUPTED;
2933		}
2934	}
2935
2936	xlog_unpack_data(rhead, dp, log);
2937
2938	return xlog_recover_process_data(log, rhash, rhead, dp, pass,
2939					 buffer_list);
2940}
2941
2942STATIC int
2943xlog_valid_rec_header(
2944	struct xlog		*log,
2945	struct xlog_rec_header	*rhead,
2946	xfs_daddr_t		blkno,
2947	int			bufsize)
2948{
2949	int			hlen;
2950
2951	if (XFS_IS_CORRUPT(log->l_mp,
2952			   rhead->h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)))
2953		return -EFSCORRUPTED;
2954	if (XFS_IS_CORRUPT(log->l_mp,
2955			   (!rhead->h_version ||
2956			   (be32_to_cpu(rhead->h_version) &
2957			    (~XLOG_VERSION_OKBITS))))) {
 
2958		xfs_warn(log->l_mp, "%s: unrecognised log version (%d).",
2959			__func__, be32_to_cpu(rhead->h_version));
2960		return -EFSCORRUPTED;
2961	}
2962
2963	/*
2964	 * LR body must have data (or it wouldn't have been written)
2965	 * and h_len must not be greater than LR buffer size.
2966	 */
2967	hlen = be32_to_cpu(rhead->h_len);
2968	if (XFS_IS_CORRUPT(log->l_mp, hlen <= 0 || hlen > bufsize))
2969		return -EFSCORRUPTED;
2970
2971	if (XFS_IS_CORRUPT(log->l_mp,
2972			   blkno > log->l_logBBsize || blkno > INT_MAX))
2973		return -EFSCORRUPTED;
 
 
 
 
2974	return 0;
2975}
2976
2977/*
2978 * Read the log from tail to head and process the log records found.
2979 * Handle the two cases where the tail and head are in the same cycle
2980 * and where the active portion of the log wraps around the end of
2981 * the physical log separately.  The pass parameter is passed through
2982 * to the routines called to process the data and is not looked at
2983 * here.
2984 */
2985STATIC int
2986xlog_do_recovery_pass(
2987	struct xlog		*log,
2988	xfs_daddr_t		head_blk,
2989	xfs_daddr_t		tail_blk,
2990	int			pass,
2991	xfs_daddr_t		*first_bad)	/* out: first bad log rec */
2992{
2993	xlog_rec_header_t	*rhead;
2994	xfs_daddr_t		blk_no, rblk_no;
2995	xfs_daddr_t		rhead_blk;
2996	char			*offset;
2997	char			*hbp, *dbp;
2998	int			error = 0, h_size, h_len;
2999	int			error2 = 0;
3000	int			bblks, split_bblks;
3001	int			hblks = 1, split_hblks, wrapped_hblks;
3002	int			i;
3003	struct hlist_head	rhash[XLOG_RHASH_SIZE];
3004	LIST_HEAD		(buffer_list);
3005
3006	ASSERT(head_blk != tail_blk);
3007	blk_no = rhead_blk = tail_blk;
3008
3009	for (i = 0; i < XLOG_RHASH_SIZE; i++)
3010		INIT_HLIST_HEAD(&rhash[i]);
3011
3012	hbp = xlog_alloc_buffer(log, hblks);
3013	if (!hbp)
3014		return -ENOMEM;
3015
3016	/*
3017	 * Read the header of the tail block and get the iclog buffer size from
3018	 * h_size.  Use this to tell how many sectors make up the log header.
3019	 */
3020	if (xfs_has_logv2(log->l_mp)) {
3021		/*
3022		 * When using variable length iclogs, read first sector of
3023		 * iclog header and extract the header size from it.  Get a
3024		 * new hbp that is the correct size.
3025		 */
 
 
 
 
3026		error = xlog_bread(log, tail_blk, 1, hbp, &offset);
3027		if (error)
3028			goto bread_err1;
3029
3030		rhead = (xlog_rec_header_t *)offset;
3031
3032		/*
3033		 * xfsprogs has a bug where record length is based on lsunit but
3034		 * h_size (iclog size) is hardcoded to 32k. Now that we
3035		 * unconditionally CRC verify the unmount record, this means the
3036		 * log buffer can be too small for the record and cause an
3037		 * overrun.
3038		 *
3039		 * Detect this condition here. Use lsunit for the buffer size as
3040		 * long as this looks like the mkfs case. Otherwise, return an
3041		 * error to avoid a buffer overrun.
3042		 */
3043		h_size = be32_to_cpu(rhead->h_size);
3044		h_len = be32_to_cpu(rhead->h_len);
3045		if (h_len > h_size && h_len <= log->l_mp->m_logbsize &&
3046		    rhead->h_num_logops == cpu_to_be32(1)) {
3047			xfs_warn(log->l_mp,
3048		"invalid iclog size (%d bytes), using lsunit (%d bytes)",
3049				 h_size, log->l_mp->m_logbsize);
3050			h_size = log->l_mp->m_logbsize;
3051		}
3052
3053		error = xlog_valid_rec_header(log, rhead, tail_blk, h_size);
3054		if (error)
3055			goto bread_err1;
3056
3057		/*
3058		 * This open codes xlog_logrec_hblks so that we can reuse the
3059		 * fixed up h_size value calculated above.  Without that we'd
3060		 * still allocate the buffer based on the incorrect on-disk
3061		 * size.
3062		 */
3063		if (h_size > XLOG_HEADER_CYCLE_SIZE &&
3064		    (rhead->h_version & cpu_to_be32(XLOG_VERSION_2))) {
3065			hblks = DIV_ROUND_UP(h_size, XLOG_HEADER_CYCLE_SIZE);
3066			if (hblks > 1) {
3067				kvfree(hbp);
3068				hbp = xlog_alloc_buffer(log, hblks);
3069				if (!hbp)
3070					return -ENOMEM;
3071			}
3072		}
3073	} else {
3074		ASSERT(log->l_sectBBsize == 1);
 
 
3075		h_size = XLOG_BIG_RECORD_BSIZE;
3076	}
3077
3078	dbp = xlog_alloc_buffer(log, BTOBB(h_size));
 
 
3079	if (!dbp) {
3080		kvfree(hbp);
3081		return -ENOMEM;
3082	}
3083
3084	memset(rhash, 0, sizeof(rhash));
3085	if (tail_blk > head_blk) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3086		/*
3087		 * Perform recovery around the end of the physical log.
3088		 * When the head is not on the same cycle number as the tail,
3089		 * we can't do a sequential recovery.
3090		 */
 
3091		while (blk_no < log->l_logBBsize) {
3092			/*
3093			 * Check for header wrapping around physical end-of-log
3094			 */
3095			offset = hbp;
3096			split_hblks = 0;
3097			wrapped_hblks = 0;
3098			if (blk_no + hblks <= log->l_logBBsize) {
3099				/* Read header in one read */
3100				error = xlog_bread(log, blk_no, hblks, hbp,
3101						   &offset);
3102				if (error)
3103					goto bread_err2;
3104			} else {
3105				/* This LR is split across physical log end */
3106				if (blk_no != log->l_logBBsize) {
3107					/* some data before physical log end */
3108					ASSERT(blk_no <= INT_MAX);
3109					split_hblks = log->l_logBBsize - (int)blk_no;
3110					ASSERT(split_hblks > 0);
3111					error = xlog_bread(log, blk_no,
3112							   split_hblks, hbp,
3113							   &offset);
3114					if (error)
3115						goto bread_err2;
3116				}
3117
3118				/*
3119				 * Note: this black magic still works with
3120				 * large sector sizes (non-512) only because:
3121				 * - we increased the buffer size originally
3122				 *   by 1 sector giving us enough extra space
3123				 *   for the second read;
3124				 * - the log start is guaranteed to be sector
3125				 *   aligned;
3126				 * - we read the log end (LR header start)
3127				 *   _first_, then the log start (LR header end)
3128				 *   - order is important.
3129				 */
3130				wrapped_hblks = hblks - split_hblks;
3131				error = xlog_bread_noalign(log, 0,
3132						wrapped_hblks,
3133						offset + BBTOB(split_hblks));
3134				if (error)
3135					goto bread_err2;
3136			}
3137			rhead = (xlog_rec_header_t *)offset;
3138			error = xlog_valid_rec_header(log, rhead,
3139					split_hblks ? blk_no : 0, h_size);
3140			if (error)
3141				goto bread_err2;
3142
3143			bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3144			blk_no += hblks;
3145
3146			/*
3147			 * Read the log record data in multiple reads if it
3148			 * wraps around the end of the log. Note that if the
3149			 * header already wrapped, blk_no could point past the
3150			 * end of the log. The record data is contiguous in
3151			 * that case.
3152			 */
3153			if (blk_no + bblks <= log->l_logBBsize ||
3154			    blk_no >= log->l_logBBsize) {
3155				rblk_no = xlog_wrap_logbno(log, blk_no);
3156				error = xlog_bread(log, rblk_no, bblks, dbp,
3157						   &offset);
3158				if (error)
3159					goto bread_err2;
3160			} else {
3161				/* This log record is split across the
3162				 * physical end of log */
3163				offset = dbp;
3164				split_bblks = 0;
3165				if (blk_no != log->l_logBBsize) {
3166					/* some data is before the physical
3167					 * end of log */
3168					ASSERT(!wrapped_hblks);
3169					ASSERT(blk_no <= INT_MAX);
3170					split_bblks =
3171						log->l_logBBsize - (int)blk_no;
3172					ASSERT(split_bblks > 0);
3173					error = xlog_bread(log, blk_no,
3174							split_bblks, dbp,
3175							&offset);
3176					if (error)
3177						goto bread_err2;
3178				}
3179
3180				/*
3181				 * Note: this black magic still works with
3182				 * large sector sizes (non-512) only because:
3183				 * - we increased the buffer size originally
3184				 *   by 1 sector giving us enough extra space
3185				 *   for the second read;
3186				 * - the log start is guaranteed to be sector
3187				 *   aligned;
3188				 * - we read the log end (LR header start)
3189				 *   _first_, then the log start (LR header end)
3190				 *   - order is important.
3191				 */
3192				error = xlog_bread_noalign(log, 0,
3193						bblks - split_bblks,
3194						offset + BBTOB(split_bblks));
3195				if (error)
3196					goto bread_err2;
3197			}
3198
3199			error = xlog_recover_process(log, rhash, rhead, offset,
3200						     pass, &buffer_list);
3201			if (error)
3202				goto bread_err2;
3203
 
 
 
 
3204			blk_no += bblks;
3205			rhead_blk = blk_no;
3206		}
3207
3208		ASSERT(blk_no >= log->l_logBBsize);
3209		blk_no -= log->l_logBBsize;
3210		rhead_blk = blk_no;
3211	}
3212
3213	/* read first part of physical log */
3214	while (blk_no < head_blk) {
3215		error = xlog_bread(log, blk_no, hblks, hbp, &offset);
3216		if (error)
3217			goto bread_err2;
3218
3219		rhead = (xlog_rec_header_t *)offset;
3220		error = xlog_valid_rec_header(log, rhead, blk_no, h_size);
3221		if (error)
3222			goto bread_err2;
3223
3224		/* blocks in data section */
3225		bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
3226		error = xlog_bread(log, blk_no+hblks, bblks, dbp,
3227				   &offset);
3228		if (error)
3229			goto bread_err2;
3230
3231		error = xlog_recover_process(log, rhash, rhead, offset, pass,
3232					     &buffer_list);
3233		if (error)
3234			goto bread_err2;
3235
3236		blk_no += bblks + hblks;
3237		rhead_blk = blk_no;
 
 
 
 
3238	}
3239
3240 bread_err2:
3241	kvfree(dbp);
3242 bread_err1:
3243	kvfree(hbp);
3244
3245	/*
3246	 * Submit buffers that have been dirtied by the last record recovered.
3247	 */
3248	if (!list_empty(&buffer_list)) {
3249		if (error) {
3250			/*
3251			 * If there has been an item recovery error then we
3252			 * cannot allow partial checkpoint writeback to
3253			 * occur.  We might have multiple checkpoints with the
3254			 * same start LSN in this buffer list, and partial
3255			 * writeback of a checkpoint in this situation can
3256			 * prevent future recovery of all the changes in the
3257			 * checkpoints at this start LSN.
3258			 *
3259			 * Note: Shutting down the filesystem will result in the
3260			 * delwri submission marking all the buffers stale,
3261			 * completing them and cleaning up _XBF_LOGRECOVERY
3262			 * state without doing any IO.
3263			 */
3264			xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
3265		}
3266		error2 = xfs_buf_delwri_submit(&buffer_list);
3267	}
3268
3269	if (error && first_bad)
3270		*first_bad = rhead_blk;
3271
3272	/*
3273	 * Transactions are freed at commit time but transactions without commit
3274	 * records on disk are never committed. Free any that may be left in the
3275	 * hash table.
3276	 */
3277	for (i = 0; i < XLOG_RHASH_SIZE; i++) {
3278		struct hlist_node	*tmp;
3279		struct xlog_recover	*trans;
3280
3281		hlist_for_each_entry_safe(trans, tmp, &rhash[i], r_list)
3282			xlog_recover_free_trans(trans);
3283	}
3284
3285	return error ? error : error2;
3286}
3287
3288/*
3289 * Do the recovery of the log.  We actually do this in two phases.
3290 * The two passes are necessary in order to implement the function
3291 * of cancelling a record written into the log.  The first pass
3292 * determines those things which have been cancelled, and the
3293 * second pass replays log items normally except for those which
3294 * have been cancelled.  The handling of the replay and cancellations
3295 * takes place in the log item type specific routines.
3296 *
3297 * The table of items which have cancel records in the log is allocated
3298 * and freed at this level, since only here do we know when all of
3299 * the log recovery has been completed.
3300 */
3301STATIC int
3302xlog_do_log_recovery(
3303	struct xlog	*log,
3304	xfs_daddr_t	head_blk,
3305	xfs_daddr_t	tail_blk)
3306{
3307	int		error;
3308
3309	ASSERT(head_blk != tail_blk);
3310
3311	/*
3312	 * First do a pass to find all of the cancelled buf log items.
3313	 * Store them in the buf_cancel_table for use in the second pass.
3314	 */
3315	error = xlog_alloc_buf_cancel_table(log);
3316	if (error)
3317		return error;
 
 
3318
3319	error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3320				      XLOG_RECOVER_PASS1, NULL);
3321	if (error != 0)
3322		goto out_cancel;
3323
 
 
3324	/*
3325	 * Then do a second pass to actually recover the items in the log.
3326	 * When it is complete free the table of buf cancel items.
3327	 */
3328	error = xlog_do_recovery_pass(log, head_blk, tail_blk,
3329				      XLOG_RECOVER_PASS2, NULL);
3330	if (!error)
3331		xlog_check_buf_cancel_table(log);
3332out_cancel:
3333	xlog_free_buf_cancel_table(log);
 
 
 
 
 
 
 
 
3334	return error;
3335}
3336
3337/*
3338 * Do the actual recovery
3339 */
3340STATIC int
3341xlog_do_recover(
3342	struct xlog		*log,
3343	xfs_daddr_t		head_blk,
3344	xfs_daddr_t		tail_blk)
3345{
3346	struct xfs_mount	*mp = log->l_mp;
3347	struct xfs_buf		*bp = mp->m_sb_bp;
3348	struct xfs_sb		*sbp = &mp->m_sb;
3349	int			error;
3350
3351	trace_xfs_log_recover(log, head_blk, tail_blk);
3352
3353	/*
3354	 * First replay the images in the log.
3355	 */
3356	error = xlog_do_log_recovery(log, head_blk, tail_blk);
3357	if (error)
3358		return error;
3359
3360	if (xlog_is_shutdown(log))
3361		return -EIO;
 
 
 
 
3362
3363	/*
3364	 * We now update the tail_lsn since much of the recovery has completed
3365	 * and there may be space available to use.  If there were no extent or
3366	 * iunlinks, we can free up the entire log.  This was set in
3367	 * xlog_find_tail to be the lsn of the last known good LR on disk.  If
3368	 * there are extent frees or iunlinks they will have some entries in the
3369	 * AIL; so we look at the AIL to determine how to set the tail_lsn.
3370	 */
3371	xfs_ail_assign_tail_lsn(log->l_ailp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3372
3373	/*
3374	 * Now that we've finished replaying all buffer and inode updates,
3375	 * re-read the superblock and reverify it.
3376	 */
3377	xfs_buf_lock(bp);
3378	xfs_buf_hold(bp);
3379	error = _xfs_buf_read(bp, XBF_READ);
3380	if (error) {
3381		if (!xlog_is_shutdown(log)) {
3382			xfs_buf_ioerror_alert(bp, __this_address);
3383			ASSERT(0);
3384		}
3385		xfs_buf_relse(bp);
3386		return error;
3387	}
3388
3389	/* Convert superblock from on-disk format */
3390	xfs_sb_from_disk(sbp, bp->b_addr);
 
 
 
3391	xfs_buf_relse(bp);
3392
3393	/* re-initialise in-core superblock and geometry structures */
3394	mp->m_features |= xfs_sb_version_to_features(sbp);
3395	xfs_reinit_percpu_counters(mp);
 
3396
3397	/* Normal transactions can now occur */
3398	clear_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);
3399	return 0;
3400}
3401
3402/*
3403 * Perform recovery and re-initialize some log variables in xlog_find_tail.
3404 *
3405 * Return error or zero.
3406 */
3407int
3408xlog_recover(
3409	struct xlog	*log)
3410{
3411	xfs_daddr_t	head_blk, tail_blk;
3412	int		error;
3413
3414	/* find the tail of the log */
3415	error = xlog_find_tail(log, &head_blk, &tail_blk);
3416	if (error)
3417		return error;
3418
3419	/*
3420	 * The superblock was read before the log was available and thus the LSN
3421	 * could not be verified. Check the superblock LSN against the current
3422	 * LSN now that it's known.
3423	 */
3424	if (xfs_has_crc(log->l_mp) &&
3425	    !xfs_log_check_lsn(log->l_mp, log->l_mp->m_sb.sb_lsn))
3426		return -EINVAL;
3427
3428	if (tail_blk != head_blk) {
3429		/* There used to be a comment here:
3430		 *
3431		 * disallow recovery on read-only mounts.  note -- mount
3432		 * checks for ENOSPC and turns it into an intelligent
3433		 * error message.
3434		 * ...but this is no longer true.  Now, unless you specify
3435		 * NORECOVERY (in which case this function would never be
3436		 * called), we just go ahead and recover.  We do this all
3437		 * under the vfs layer, so we can get away with it unless
3438		 * the device itself is read-only, in which case we fail.
3439		 */
3440		if ((error = xfs_dev_is_read_only(log->l_mp, "recovery"))) {
3441			return error;
3442		}
3443
3444		/*
3445		 * Version 5 superblock log feature mask validation. We know the
3446		 * log is dirty so check if there are any unknown log features
3447		 * in what we need to recover. If there are unknown features
3448		 * (e.g. unsupported transactions, then simply reject the
3449		 * attempt at recovery before touching anything.
3450		 */
3451		if (xfs_sb_is_v5(&log->l_mp->m_sb) &&
3452		    xfs_sb_has_incompat_log_feature(&log->l_mp->m_sb,
3453					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)) {
3454			xfs_warn(log->l_mp,
3455"Superblock has unknown incompatible log features (0x%x) enabled.",
 
 
3456				(log->l_mp->m_sb.sb_features_log_incompat &
3457					XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
3458			xfs_warn(log->l_mp,
3459"The log can not be fully and/or safely recovered by this kernel.");
3460			xfs_warn(log->l_mp,
3461"Please recover the log on a kernel that supports the unknown features.");
3462			return -EINVAL;
3463		}
3464
3465		/*
3466		 * Delay log recovery if the debug hook is set. This is debug
3467		 * instrumentation to coordinate simulation of I/O failures with
3468		 * log recovery.
3469		 */
3470		if (xfs_globals.log_recovery_delay) {
3471			xfs_notice(log->l_mp,
3472				"Delaying log recovery for %d seconds.",
3473				xfs_globals.log_recovery_delay);
3474			msleep(xfs_globals.log_recovery_delay * 1000);
3475		}
3476
3477		xfs_notice(log->l_mp, "Starting recovery (logdev: %s)",
3478				log->l_mp->m_logname ? log->l_mp->m_logname
3479						     : "internal");
3480
3481		error = xlog_do_recover(log, head_blk, tail_blk);
3482		set_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);
3483	}
3484	return error;
3485}
3486
3487/*
3488 * In the first part of recovery we replay inodes and buffers and build up the
3489 * list of intents which need to be processed. Here we process the intents and
3490 * clean up the on disk unlinked inode lists. This is separated from the first
3491 * part of recovery so that the root and real-time bitmap inodes can be read in
3492 * from disk in between the two stages.  This is necessary so that we can free
3493 * space in the real-time portion of the file system.
3494 *
3495 * We run this whole process under GFP_NOFS allocation context. We do a
3496 * combination of non-transactional and transactional work, yet we really don't
3497 * want to recurse into the filesystem from direct reclaim during any of this
3498 * processing. This allows all the recovery code run here not to care about the
3499 * memory allocation context it is running in.
3500 */
3501int
3502xlog_recover_finish(
3503	struct xlog	*log)
3504{
3505	unsigned int	nofs_flags = memalloc_nofs_save();
3506	int		error;
3507
3508	error = xlog_recover_process_intents(log);
3509	if (error) {
 
 
 
 
 
 
 
 
 
 
3510		/*
3511		 * Cancel all the unprocessed intent items now so that we don't
3512		 * leave them pinned in the AIL.  This can cause the AIL to
3513		 * livelock on the pinned item if anyone tries to push the AIL
3514		 * (inode reclaim does this) before we get around to
3515		 * xfs_log_mount_cancel.
3516		 */
3517		xlog_recover_cancel_intents(log);
3518		xfs_alert(log->l_mp, "Failed to recover intents");
3519		xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
3520		goto out_error;
3521	}
3522
3523	/*
3524	 * Sync the log to get all the intents out of the AIL.  This isn't
3525	 * absolutely necessary, but it helps in case the unlink transactions
3526	 * would have problems pushing the intents out of the way.
3527	 */
3528	xfs_log_force(log->l_mp, XFS_LOG_SYNC);
3529
3530	xlog_recover_process_iunlinks(log);
3531
3532	/*
3533	 * Recover any CoW staging blocks that are still referenced by the
3534	 * ondisk refcount metadata.  During mount there cannot be any live
3535	 * staging extents as we have not permitted any user modifications.
3536	 * Therefore, it is safe to free them all right now, even on a
3537	 * read-only mount.
3538	 */
3539	error = xfs_reflink_recover_cow(log->l_mp);
3540	if (error) {
3541		xfs_alert(log->l_mp,
3542	"Failed to recover leftover CoW staging extents, err %d.",
3543				error);
3544		/*
3545		 * If we get an error here, make sure the log is shut down
3546		 * but return zero so that any log items committed since the
3547		 * end of intents processing can be pushed through the CIL
3548		 * and AIL.
3549		 */
3550		xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);
3551		error = 0;
3552		goto out_error;
3553	}
 
 
3554
3555out_error:
3556	memalloc_nofs_restore(nofs_flags);
3557	return error;
3558}
3559
 
 
 
 
 
3560void
3561xlog_recover_cancel(
3562	struct xlog	*log)
3563{
3564	if (xlog_recovery_needed(log))
3565		xlog_recover_cancel_intents(log);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3566}
3567