Linux Audio

Check our new training course

Loading...
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 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_inode.h"
  16#include "xfs_dir2.h"
  17#include "xfs_ialloc.h"
  18#include "xfs_alloc.h"
  19#include "xfs_rtalloc.h"
  20#include "xfs_bmap.h"
  21#include "xfs_trans.h"
  22#include "xfs_trans_priv.h"
  23#include "xfs_log.h"
  24#include "xfs_log_priv.h"
  25#include "xfs_error.h"
  26#include "xfs_quota.h"
  27#include "xfs_fsops.h"
 
  28#include "xfs_icache.h"
  29#include "xfs_sysfs.h"
  30#include "xfs_rmap_btree.h"
  31#include "xfs_refcount_btree.h"
  32#include "xfs_reflink.h"
  33#include "xfs_extent_busy.h"
  34#include "xfs_health.h"
  35#include "xfs_trace.h"
  36#include "xfs_ag.h"
  37#include "scrub/stats.h"
  38
  39static DEFINE_MUTEX(xfs_uuid_table_mutex);
  40static int xfs_uuid_table_size;
  41static uuid_t *xfs_uuid_table;
  42
  43void
  44xfs_uuid_table_free(void)
  45{
  46	if (xfs_uuid_table_size == 0)
  47		return;
  48	kfree(xfs_uuid_table);
  49	xfs_uuid_table = NULL;
  50	xfs_uuid_table_size = 0;
  51}
  52
  53/*
  54 * See if the UUID is unique among mounted XFS filesystems.
  55 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  56 */
  57STATIC int
  58xfs_uuid_mount(
  59	struct xfs_mount	*mp)
  60{
  61	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  62	int			hole, i;
  63
  64	/* Publish UUID in struct super_block */
  65	super_set_uuid(mp->m_super, uuid->b, sizeof(*uuid));
  66
  67	if (xfs_has_nouuid(mp))
  68		return 0;
  69
  70	if (uuid_is_null(uuid)) {
  71		xfs_warn(mp, "Filesystem has null UUID - can't mount");
  72		return -EINVAL;
  73	}
  74
  75	mutex_lock(&xfs_uuid_table_mutex);
  76	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  77		if (uuid_is_null(&xfs_uuid_table[i])) {
  78			hole = i;
  79			continue;
  80		}
  81		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  82			goto out_duplicate;
  83	}
  84
  85	if (hole < 0) {
  86		xfs_uuid_table = krealloc(xfs_uuid_table,
  87			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
  88			GFP_KERNEL | __GFP_NOFAIL);
  89		hole = xfs_uuid_table_size++;
  90	}
  91	xfs_uuid_table[hole] = *uuid;
  92	mutex_unlock(&xfs_uuid_table_mutex);
  93
  94	return 0;
  95
  96 out_duplicate:
  97	mutex_unlock(&xfs_uuid_table_mutex);
  98	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
  99	return -EINVAL;
 100}
 101
 102STATIC void
 103xfs_uuid_unmount(
 104	struct xfs_mount	*mp)
 105{
 106	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 107	int			i;
 108
 109	if (xfs_has_nouuid(mp))
 110		return;
 111
 112	mutex_lock(&xfs_uuid_table_mutex);
 113	for (i = 0; i < xfs_uuid_table_size; i++) {
 114		if (uuid_is_null(&xfs_uuid_table[i]))
 115			continue;
 116		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 117			continue;
 118		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 119		break;
 120	}
 121	ASSERT(i < xfs_uuid_table_size);
 122	mutex_unlock(&xfs_uuid_table_mutex);
 123}
 124
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 125/*
 126 * Check size of device based on the (data/realtime) block count.
 127 * Note: this check is used by the growfs code as well as mount.
 128 */
 129int
 130xfs_sb_validate_fsb_count(
 131	xfs_sb_t	*sbp,
 132	uint64_t	nblocks)
 133{
 134	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 135	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 136
 137	/* Limited by ULONG_MAX of page cache index */
 138	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 139		return -EFBIG;
 140	return 0;
 141}
 142
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 143/*
 144 * xfs_readsb
 145 *
 146 * Does the initial read of the superblock.
 147 */
 148int
 149xfs_readsb(
 150	struct xfs_mount *mp,
 151	int		flags)
 152{
 153	unsigned int	sector_size;
 154	struct xfs_buf	*bp;
 155	struct xfs_sb	*sbp = &mp->m_sb;
 156	int		error;
 157	int		loud = !(flags & XFS_MFSI_QUIET);
 158	const struct xfs_buf_ops *buf_ops;
 159
 160	ASSERT(mp->m_sb_bp == NULL);
 161	ASSERT(mp->m_ddev_targp != NULL);
 162
 163	/*
 164	 * For the initial read, we must guess at the sector
 165	 * size based on the block device.  It's enough to
 166	 * get the sb_sectsize out of the superblock and
 167	 * then reread with the proper length.
 168	 * We don't verify it yet, because it may not be complete.
 169	 */
 170	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 171	buf_ops = NULL;
 172
 173	/*
 174	 * Allocate a (locked) buffer to hold the superblock. This will be kept
 175	 * around at all times to optimize access to the superblock. Therefore,
 176	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
 177	 * elevated.
 178	 */
 179reread:
 180	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 181				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
 182				      buf_ops);
 183	if (error) {
 184		if (loud)
 185			xfs_warn(mp, "SB validate failed with error %d.", error);
 186		/* bad CRC means corrupted metadata */
 187		if (error == -EFSBADCRC)
 188			error = -EFSCORRUPTED;
 189		return error;
 190	}
 191
 192	/*
 193	 * Initialize the mount structure from the superblock.
 194	 */
 195	xfs_sb_from_disk(sbp, bp->b_addr);
 196
 197	/*
 198	 * If we haven't validated the superblock, do so now before we try
 199	 * to check the sector size and reread the superblock appropriately.
 200	 */
 201	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 202		if (loud)
 203			xfs_warn(mp, "Invalid superblock magic number");
 204		error = -EINVAL;
 205		goto release_buf;
 206	}
 207
 208	/*
 209	 * We must be able to do sector-sized and sector-aligned IO.
 210	 */
 211	if (sector_size > sbp->sb_sectsize) {
 212		if (loud)
 213			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 214				sector_size, sbp->sb_sectsize);
 215		error = -ENOSYS;
 216		goto release_buf;
 217	}
 218
 219	if (buf_ops == NULL) {
 220		/*
 221		 * Re-read the superblock so the buffer is correctly sized,
 222		 * and properly verified.
 223		 */
 224		xfs_buf_relse(bp);
 225		sector_size = sbp->sb_sectsize;
 226		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 227		goto reread;
 228	}
 229
 230	mp->m_features |= xfs_sb_version_to_features(sbp);
 231	xfs_reinit_percpu_counters(mp);
 232
 233	/* no need to be quiet anymore, so reset the buf ops */
 234	bp->b_ops = &xfs_sb_buf_ops;
 235
 236	mp->m_sb_bp = bp;
 237	xfs_buf_unlock(bp);
 238	return 0;
 239
 240release_buf:
 241	xfs_buf_relse(bp);
 242	return error;
 243}
 244
 245/*
 246 * If the sunit/swidth change would move the precomputed root inode value, we
 247 * must reject the ondisk change because repair will stumble over that.
 248 * However, we allow the mount to proceed because we never rejected this
 249 * combination before.  Returns true to update the sb, false otherwise.
 250 */
 251static inline int
 252xfs_check_new_dalign(
 253	struct xfs_mount	*mp,
 254	int			new_dalign,
 255	bool			*update_sb)
 256{
 257	struct xfs_sb		*sbp = &mp->m_sb;
 258	xfs_ino_t		calc_ino;
 259
 260	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
 261	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 262
 263	if (sbp->sb_rootino == calc_ino) {
 264		*update_sb = true;
 265		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 266	}
 267
 268	xfs_warn(mp,
 269"Cannot change stripe alignment; would require moving root inode.");
 270
 271	/*
 272	 * XXX: Next time we add a new incompat feature, this should start
 273	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
 274	 * that we're ignoring the administrator's instructions.
 275	 */
 276	xfs_warn(mp, "Skipping superblock stripe alignment update.");
 277	*update_sb = false;
 278	return 0;
 279}
 280
 281/*
 282 * If we were provided with new sunit/swidth values as mount options, make sure
 283 * that they pass basic alignment and superblock feature checks, and convert
 284 * them into the same units (FSB) that everything else expects.  This step
 285 * /must/ be done before computing the inode geometry.
 286 */
 287STATIC int
 288xfs_validate_new_dalign(
 289	struct xfs_mount	*mp)
 290{
 291	if (mp->m_dalign == 0)
 292		return 0;
 293
 294	/*
 295	 * If stripe unit and stripe width are not multiples
 296	 * of the fs blocksize turn off alignment.
 297	 */
 298	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 299	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 300		xfs_warn(mp,
 301	"alignment check failed: sunit/swidth vs. blocksize(%d)",
 302			mp->m_sb.sb_blocksize);
 303		return -EINVAL;
 304	}
 305
 306	/*
 307	 * Convert the stripe unit and width to FSBs.
 308	 */
 309	mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 310	if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
 311		xfs_warn(mp,
 312	"alignment check failed: sunit/swidth vs. agsize(%d)",
 313			mp->m_sb.sb_agblocks);
 314		return -EINVAL;
 315	}
 316
 317	if (!mp->m_dalign) {
 318		xfs_warn(mp,
 319	"alignment check failed: sunit(%d) less than bsize(%d)",
 320			mp->m_dalign, mp->m_sb.sb_blocksize);
 321		return -EINVAL;
 322	}
 323
 324	mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 325
 326	if (!xfs_has_dalign(mp)) {
 327		xfs_warn(mp,
 328"cannot change alignment: superblock does not support data alignment");
 329		return -EINVAL;
 
 
 
 
 
 
 330	}
 331
 332	return 0;
 333}
 334
 335/* Update alignment values based on mount options and sb values. */
 336STATIC int
 337xfs_update_alignment(
 338	struct xfs_mount	*mp)
 
 
 
 
 339{
 340	struct xfs_sb		*sbp = &mp->m_sb;
 341
 342	if (mp->m_dalign) {
 343		bool		update_sb;
 344		int		error;
 345
 346		if (sbp->sb_unit == mp->m_dalign &&
 347		    sbp->sb_width == mp->m_swidth)
 348			return 0;
 349
 350		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
 351		if (error || !update_sb)
 352			return error;
 353
 354		sbp->sb_unit = mp->m_dalign;
 355		sbp->sb_width = mp->m_swidth;
 356		mp->m_update_sb = true;
 357	} else if (!xfs_has_noalign(mp) && xfs_has_dalign(mp)) {
 358		mp->m_dalign = sbp->sb_unit;
 359		mp->m_swidth = sbp->sb_width;
 
 
 
 
 
 360	}
 361
 362	return 0;
 
 
 
 
 
 
 
 
 
 
 
 363}
 364
 365/*
 366 * precalculate the low space thresholds for dynamic speculative preallocation.
 367 */
 368void
 369xfs_set_low_space_thresholds(
 370	struct xfs_mount	*mp)
 371{
 372	uint64_t		dblocks = mp->m_sb.sb_dblocks;
 373	uint64_t		rtexts = mp->m_sb.sb_rextents;
 374	int			i;
 375
 376	do_div(dblocks, 100);
 377	do_div(rtexts, 100);
 378
 379	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 380		mp->m_low_space[i] = dblocks * (i + 1);
 381		mp->m_low_rtexts[i] = rtexts * (i + 1);
 
 
 382	}
 383}
 384
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 385/*
 386 * Check that the data (and log if separate) is an ok size.
 387 */
 388STATIC int
 389xfs_check_sizes(
 390	struct xfs_mount *mp)
 391{
 392	struct xfs_buf	*bp;
 393	xfs_daddr_t	d;
 394	int		error;
 395
 396	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 397	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 398		xfs_warn(mp, "filesystem size mismatch detected");
 399		return -EFBIG;
 400	}
 401	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 402					d - XFS_FSS_TO_BB(mp, 1),
 403					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 404	if (error) {
 405		xfs_warn(mp, "last sector read failed");
 406		return error;
 407	}
 408	xfs_buf_relse(bp);
 409
 410	if (mp->m_logdev_targp == mp->m_ddev_targp)
 411		return 0;
 412
 413	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 414	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 415		xfs_warn(mp, "log size mismatch detected");
 416		return -EFBIG;
 417	}
 418	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 419					d - XFS_FSB_TO_BB(mp, 1),
 420					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
 421	if (error) {
 422		xfs_warn(mp, "log device read failed");
 423		return error;
 424	}
 425	xfs_buf_relse(bp);
 426	return 0;
 427}
 428
 429/*
 430 * Clear the quotaflags in memory and in the superblock.
 431 */
 432int
 433xfs_mount_reset_sbqflags(
 434	struct xfs_mount	*mp)
 435{
 436	mp->m_qflags = 0;
 437
 438	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 439	if (mp->m_sb.sb_qflags == 0)
 440		return 0;
 441	spin_lock(&mp->m_sb_lock);
 442	mp->m_sb.sb_qflags = 0;
 443	spin_unlock(&mp->m_sb_lock);
 444
 445	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 446		return 0;
 447
 448	return xfs_sync_sb(mp, false);
 449}
 450
 451uint64_t
 452xfs_default_resblks(xfs_mount_t *mp)
 453{
 454	uint64_t resblks;
 455
 456	/*
 457	 * We default to 5% or 8192 fsbs of space reserved, whichever is
 458	 * smaller.  This is intended to cover concurrent allocation
 459	 * transactions when we initially hit enospc. These each require a 4
 460	 * block reservation. Hence by default we cover roughly 2000 concurrent
 461	 * allocation reservations.
 462	 */
 463	resblks = mp->m_sb.sb_dblocks;
 464	do_div(resblks, 20);
 465	resblks = min_t(uint64_t, resblks, 8192);
 466	return resblks;
 467}
 468
 469/* Ensure the summary counts are correct. */
 470STATIC int
 471xfs_check_summary_counts(
 472	struct xfs_mount	*mp)
 473{
 474	int			error = 0;
 475
 476	/*
 477	 * The AG0 superblock verifier rejects in-progress filesystems,
 478	 * so we should never see the flag set this far into mounting.
 479	 */
 480	if (mp->m_sb.sb_inprogress) {
 481		xfs_err(mp, "sb_inprogress set after log recovery??");
 482		WARN_ON(1);
 483		return -EFSCORRUPTED;
 484	}
 485
 486	/*
 487	 * Now the log is mounted, we know if it was an unclean shutdown or
 488	 * not. If it was, with the first phase of recovery has completed, we
 489	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 490	 * but they are recovered transactionally in the second recovery phase
 491	 * later.
 492	 *
 493	 * If the log was clean when we mounted, we can check the summary
 494	 * counters.  If any of them are obviously incorrect, we can recompute
 495	 * them from the AGF headers in the next step.
 496	 */
 497	if (xfs_is_clean(mp) &&
 498	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
 499	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
 500	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
 501		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
 502
 503	/*
 504	 * We can safely re-initialise incore superblock counters from the
 505	 * per-ag data. These may not be correct if the filesystem was not
 506	 * cleanly unmounted, so we waited for recovery to finish before doing
 507	 * this.
 508	 *
 509	 * If the filesystem was cleanly unmounted or the previous check did
 510	 * not flag anything weird, then we can trust the values in the
 511	 * superblock to be correct and we don't need to do anything here.
 512	 * Otherwise, recalculate the summary counters.
 513	 */
 514	if ((xfs_has_lazysbcount(mp) && !xfs_is_clean(mp)) ||
 515	    xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) {
 516		error = xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
 517		if (error)
 518			return error;
 519	}
 520
 521	/*
 522	 * Older kernels misused sb_frextents to reflect both incore
 523	 * reservations made by running transactions and the actual count of
 524	 * free rt extents in the ondisk metadata.  Transactions committed
 525	 * during runtime can therefore contain a superblock update that
 526	 * undercounts the number of free rt extents tracked in the rt bitmap.
 527	 * A clean unmount record will have the correct frextents value since
 528	 * there can be no other transactions running at that point.
 529	 *
 530	 * If we're mounting the rt volume after recovering the log, recompute
 531	 * frextents from the rtbitmap file to fix the inconsistency.
 532	 */
 533	if (xfs_has_realtime(mp) && !xfs_is_clean(mp)) {
 534		error = xfs_rtalloc_reinit_frextents(mp);
 535		if (error)
 536			return error;
 537	}
 538
 539	return 0;
 540}
 541
 542static void
 543xfs_unmount_check(
 544	struct xfs_mount	*mp)
 545{
 546	if (xfs_is_shutdown(mp))
 547		return;
 548
 549	if (percpu_counter_sum(&mp->m_ifree) >
 550			percpu_counter_sum(&mp->m_icount)) {
 551		xfs_alert(mp, "ifree/icount mismatch at unmount");
 552		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
 553	}
 554}
 555
 556/*
 557 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
 558 * internal inode structures can be sitting in the CIL and AIL at this point,
 559 * so we need to unpin them, write them back and/or reclaim them before unmount
 560 * can proceed.  In other words, callers are required to have inactivated all
 561 * inodes.
 562 *
 563 * An inode cluster that has been freed can have its buffer still pinned in
 564 * memory because the transaction is still sitting in a iclog. The stale inodes
 565 * on that buffer will be pinned to the buffer until the transaction hits the
 566 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
 567 * may never see the pinned buffer, so nothing will push out the iclog and
 568 * unpin the buffer.
 569 *
 570 * Hence we need to force the log to unpin everything first. However, log
 571 * forces don't wait for the discards they issue to complete, so we have to
 572 * explicitly wait for them to complete here as well.
 573 *
 574 * Then we can tell the world we are unmounting so that error handling knows
 575 * that the filesystem is going away and we should error out anything that we
 576 * have been retrying in the background.  This will prevent never-ending
 577 * retries in AIL pushing from hanging the unmount.
 578 *
 579 * Finally, we can push the AIL to clean all the remaining dirty objects, then
 580 * reclaim the remaining inodes that are still in memory at this point in time.
 581 */
 582static void
 583xfs_unmount_flush_inodes(
 584	struct xfs_mount	*mp)
 585{
 586	xfs_log_force(mp, XFS_LOG_SYNC);
 587	xfs_extent_busy_wait_all(mp);
 588	flush_workqueue(xfs_discard_wq);
 589
 590	set_bit(XFS_OPSTATE_UNMOUNTING, &mp->m_opstate);
 591
 592	xfs_ail_push_all_sync(mp->m_ail);
 593	xfs_inodegc_stop(mp);
 594	cancel_delayed_work_sync(&mp->m_reclaim_work);
 595	xfs_reclaim_inodes(mp);
 596	xfs_health_unmount(mp);
 597}
 598
 599static void
 600xfs_mount_setup_inode_geom(
 601	struct xfs_mount	*mp)
 602{
 603	struct xfs_ino_geometry *igeo = M_IGEO(mp);
 604
 605	igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
 606	ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
 607
 608	xfs_ialloc_setup_geometry(mp);
 609}
 610
 611/* Compute maximum possible height for per-AG btree types for this fs. */
 612static inline void
 613xfs_agbtree_compute_maxlevels(
 614	struct xfs_mount	*mp)
 615{
 616	unsigned int		levels;
 617
 618	levels = max(mp->m_alloc_maxlevels, M_IGEO(mp)->inobt_maxlevels);
 619	levels = max(levels, mp->m_rmap_maxlevels);
 620	mp->m_agbtree_maxlevels = max(levels, mp->m_refc_maxlevels);
 621}
 622
 623/*
 624 * This function does the following on an initial mount of a file system:
 625 *	- reads the superblock from disk and init the mount struct
 626 *	- if we're a 32-bit kernel, do a size check on the superblock
 627 *		so we don't mount terabyte filesystems
 628 *	- init mount struct realtime fields
 629 *	- allocate inode hash table for fs
 630 *	- init directory manager
 631 *	- perform recovery and init the log manager
 632 */
 633int
 634xfs_mountfs(
 635	struct xfs_mount	*mp)
 636{
 637	struct xfs_sb		*sbp = &(mp->m_sb);
 638	struct xfs_inode	*rip;
 639	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
 640	uint			quotamount = 0;
 641	uint			quotaflags = 0;
 642	int			error = 0;
 643
 644	xfs_sb_mount_common(mp, sbp);
 645
 646	/*
 647	 * Check for a mismatched features2 values.  Older kernels read & wrote
 648	 * into the wrong sb offset for sb_features2 on some platforms due to
 649	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 650	 * which made older superblock reading/writing routines swap it as a
 651	 * 64-bit value.
 652	 *
 653	 * For backwards compatibility, we make both slots equal.
 654	 *
 655	 * If we detect a mismatched field, we OR the set bits into the existing
 656	 * features2 field in case it has already been modified; we don't want
 657	 * to lose any features.  We then update the bad location with the ORed
 658	 * value so that older kernels will see any features2 flags. The
 659	 * superblock writeback code ensures the new sb_features2 is copied to
 660	 * sb_bad_features2 before it is logged or written to disk.
 661	 */
 662	if (xfs_sb_has_mismatched_features2(sbp)) {
 663		xfs_warn(mp, "correcting sb_features alignment problem");
 664		sbp->sb_features2 |= sbp->sb_bad_features2;
 665		mp->m_update_sb = true;
 
 
 
 
 
 
 
 
 666	}
 667
 
 
 
 
 
 
 
 
 
 668
 669	/* always use v2 inodes by default now */
 670	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 671		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 672		mp->m_features |= XFS_FEAT_NLINK;
 673		mp->m_update_sb = true;
 674	}
 675
 676	/*
 677	 * If we were given new sunit/swidth options, do some basic validation
 678	 * checks and convert the incore dalign and swidth values to the
 679	 * same units (FSB) that everything else uses.  This /must/ happen
 680	 * before computing the inode geometry.
 681	 */
 682	error = xfs_validate_new_dalign(mp);
 683	if (error)
 684		goto out;
 685
 686	xfs_alloc_compute_maxlevels(mp);
 687	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 688	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 689	xfs_mount_setup_inode_geom(mp);
 690	xfs_rmapbt_compute_maxlevels(mp);
 691	xfs_refcountbt_compute_maxlevels(mp);
 692
 693	xfs_agbtree_compute_maxlevels(mp);
 694
 695	/*
 696	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
 697	 * is NOT aligned turn off m_dalign since allocator alignment is within
 698	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
 699	 * we must compute the free space and rmap btree geometry before doing
 700	 * this.
 701	 */
 702	error = xfs_update_alignment(mp);
 703	if (error)
 704		goto out;
 705
 706	/* enable fail_at_unmount as default */
 707	mp->m_fail_unmount = true;
 708
 709	super_set_sysfs_name_id(mp->m_super);
 710
 711	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
 712			       NULL, mp->m_super->s_id);
 713	if (error)
 714		goto out;
 715
 716	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
 717			       &mp->m_kobj, "stats");
 718	if (error)
 719		goto out_remove_sysfs;
 720
 721	xchk_stats_register(mp->m_scrub_stats, mp->m_debugfs);
 722
 723	error = xfs_error_sysfs_init(mp);
 724	if (error)
 725		goto out_remove_scrub_stats;
 726
 727	error = xfs_errortag_init(mp);
 728	if (error)
 729		goto out_remove_error_sysfs;
 730
 731	error = xfs_uuid_mount(mp);
 732	if (error)
 733		goto out_remove_errortag;
 734
 735	/*
 736	 * Update the preferred write size based on the information from the
 737	 * on-disk superblock.
 738	 */
 739	mp->m_allocsize_log =
 740		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
 741	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
 742
 743	/* set the low space thresholds for dynamic preallocation */
 744	xfs_set_low_space_thresholds(mp);
 745
 746	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 747	 * If enabled, sparse inode chunk alignment is expected to match the
 748	 * cluster size. Full inode chunk alignment must match the chunk size,
 749	 * but that is checked on sb read verification...
 750	 */
 751	if (xfs_has_sparseinodes(mp) &&
 752	    mp->m_sb.sb_spino_align !=
 753			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
 754		xfs_warn(mp,
 755	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 756			 mp->m_sb.sb_spino_align,
 757			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
 758		error = -EINVAL;
 759		goto out_remove_uuid;
 760	}
 761
 762	/*
 
 
 
 
 
 763	 * Check that the data (and log if separate) is an ok size.
 764	 */
 765	error = xfs_check_sizes(mp);
 766	if (error)
 767		goto out_remove_uuid;
 768
 769	/*
 770	 * Initialize realtime fields in the mount structure
 771	 */
 772	error = xfs_rtmount_init(mp);
 773	if (error) {
 774		xfs_warn(mp, "RT mount failed");
 775		goto out_remove_uuid;
 776	}
 777
 778	/*
 779	 *  Copies the low order bits of the timestamp and the randomly
 780	 *  set "sequence" number out of a UUID.
 781	 */
 782	mp->m_fixedfsid[0] =
 783		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
 784		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
 785	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
 786
 787	error = xfs_da_mount(mp);
 788	if (error) {
 789		xfs_warn(mp, "Failed dir/attr init: %d", error);
 790		goto out_remove_uuid;
 791	}
 792
 793	/*
 794	 * Initialize the precomputed transaction reservations values.
 795	 */
 796	xfs_trans_init(mp);
 797
 798	/*
 799	 * Allocate and initialize the per-ag data.
 800	 */
 801	error = xfs_initialize_perag(mp, sbp->sb_agcount, mp->m_sb.sb_dblocks,
 802			&mp->m_maxagi);
 
 803	if (error) {
 804		xfs_warn(mp, "Failed per-ag init: %d", error);
 805		goto out_free_dir;
 806	}
 807
 808	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
 809		xfs_warn(mp, "no log defined");
 
 810		error = -EFSCORRUPTED;
 811		goto out_free_perag;
 812	}
 813
 814	error = xfs_inodegc_register_shrinker(mp);
 815	if (error)
 816		goto out_fail_wait;
 817
 818	/*
 819	 * Log's mount-time initialization. The first part of recovery can place
 820	 * some items on the AIL, to be handled when recovery is finished or
 821	 * cancelled.
 822	 */
 823	error = xfs_log_mount(mp, mp->m_logdev_targp,
 824			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 825			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 826	if (error) {
 827		xfs_warn(mp, "log mount failed");
 828		goto out_inodegc_shrinker;
 829	}
 830
 831	/* Enable background inode inactivation workers. */
 832	xfs_inodegc_start(mp);
 833	xfs_blockgc_start(mp);
 834
 835	/*
 836	 * Now that we've recovered any pending superblock feature bit
 837	 * additions, we can finish setting up the attr2 behaviour for the
 838	 * mount. The noattr2 option overrides the superblock flag, so only
 839	 * check the superblock feature flag if the mount option is not set.
 
 
 
 
 
 
 
 
 
 
 
 
 
 840	 */
 841	if (xfs_has_noattr2(mp)) {
 842		mp->m_features &= ~XFS_FEAT_ATTR2;
 843	} else if (!xfs_has_attr2(mp) &&
 844		   (mp->m_sb.sb_features2 & XFS_SB_VERSION2_ATTR2BIT)) {
 845		mp->m_features |= XFS_FEAT_ATTR2;
 
 846	}
 847
 848	/*
 849	 * Get and sanity-check the root inode.
 850	 * Save the pointer to it in the mount structure.
 851	 */
 852	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
 853			 XFS_ILOCK_EXCL, &rip);
 854	if (error) {
 855		xfs_warn(mp,
 856			"Failed to read root inode 0x%llx, error %d",
 857			sbp->sb_rootino, -error);
 858		goto out_log_dealloc;
 859	}
 860
 861	ASSERT(rip != NULL);
 862
 863	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
 864		xfs_warn(mp, "corrupted root inode %llu: not a directory",
 865			(unsigned long long)rip->i_ino);
 866		xfs_iunlock(rip, XFS_ILOCK_EXCL);
 
 
 867		error = -EFSCORRUPTED;
 868		goto out_rele_rip;
 869	}
 870	mp->m_rootip = rip;	/* save it */
 871
 872	xfs_iunlock(rip, XFS_ILOCK_EXCL);
 873
 874	/*
 875	 * Initialize realtime inode pointers in the mount structure
 876	 */
 877	error = xfs_rtmount_inodes(mp);
 878	if (error) {
 879		/*
 880		 * Free up the root inode.
 881		 */
 882		xfs_warn(mp, "failed to read RT inodes");
 883		goto out_rele_rip;
 884	}
 885
 886	/* Make sure the summary counts are ok. */
 887	error = xfs_check_summary_counts(mp);
 888	if (error)
 889		goto out_rtunmount;
 890
 891	/*
 892	 * If this is a read-only mount defer the superblock updates until
 893	 * the next remount into writeable mode.  Otherwise we would never
 894	 * perform the update e.g. for the root filesystem.
 895	 */
 896	if (mp->m_update_sb && !xfs_is_readonly(mp)) {
 897		error = xfs_sync_sb(mp, false);
 898		if (error) {
 899			xfs_warn(mp, "failed to write sb changes");
 900			goto out_rtunmount;
 901		}
 902	}
 903
 904	/*
 905	 * Initialise the XFS quota management subsystem for this mount
 906	 */
 907	if (XFS_IS_QUOTA_ON(mp)) {
 908		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
 909		if (error)
 910			goto out_rtunmount;
 911	} else {
 
 
 912		/*
 913		 * If a file system had quotas running earlier, but decided to
 914		 * mount without -o uquota/pquota/gquota options, revoke the
 915		 * quotachecked license.
 916		 */
 917		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
 918			xfs_notice(mp, "resetting quota flags");
 919			error = xfs_mount_reset_sbqflags(mp);
 920			if (error)
 921				goto out_rtunmount;
 922		}
 923	}
 924
 925	/*
 
 
 
 
 
 
 
 
 
 926	 * Finish recovering the file system.  This part needed to be delayed
 927	 * until after the root and real-time bitmap inodes were consistently
 928	 * read in.  Temporarily create per-AG space reservations for metadata
 929	 * btree shape changes because space freeing transactions (for inode
 930	 * inactivation) require the per-AG reservation in lieu of reserving
 931	 * blocks.
 932	 */
 933	error = xfs_fs_reserve_ag_blocks(mp);
 934	if (error && error == -ENOSPC)
 935		xfs_warn(mp,
 936	"ENOSPC reserving per-AG metadata pool, log recovery may fail.");
 937	error = xfs_log_mount_finish(mp);
 938	xfs_fs_unreserve_ag_blocks(mp);
 939	if (error) {
 940		xfs_warn(mp, "log mount finish failed");
 941		goto out_rtunmount;
 942	}
 943
 944	/*
 945	 * Now the log is fully replayed, we can transition to full read-only
 946	 * mode for read-only mounts. This will sync all the metadata and clean
 947	 * the log so that the recovery we just performed does not have to be
 948	 * replayed again on the next mount.
 949	 *
 950	 * We use the same quiesce mechanism as the rw->ro remount, as they are
 951	 * semantically identical operations.
 952	 */
 953	if (xfs_is_readonly(mp) && !xfs_has_norecovery(mp))
 954		xfs_log_clean(mp);
 
 
 955
 956	/*
 957	 * Complete the quota initialisation, post-log-replay component.
 958	 */
 959	if (quotamount) {
 960		ASSERT(mp->m_qflags == 0);
 961		mp->m_qflags = quotaflags;
 962
 963		xfs_qm_mount_quotas(mp);
 964	}
 965
 966	/*
 967	 * Now we are mounted, reserve a small amount of unused space for
 968	 * privileged transactions. This is needed so that transaction
 969	 * space required for critical operations can dip into this pool
 970	 * when at ENOSPC. This is needed for operations like create with
 971	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
 972	 * are not allowed to use this reserved space.
 973	 *
 974	 * This may drive us straight to ENOSPC on mount, but that implies
 975	 * we were already there on the last unmount. Warn if this occurs.
 976	 */
 977	if (!xfs_is_readonly(mp)) {
 978		error = xfs_reserve_blocks(mp, xfs_default_resblks(mp));
 
 979		if (error)
 980			xfs_warn(mp,
 981	"Unable to allocate reserve blocks. Continuing without reserve pool.");
 982
 
 
 
 
 
 
 
 
 
 983		/* Reserve AG blocks for future btree expansion. */
 984		error = xfs_fs_reserve_ag_blocks(mp);
 985		if (error && error != -ENOSPC)
 986			goto out_agresv;
 987	}
 988
 989	return 0;
 990
 991 out_agresv:
 992	xfs_fs_unreserve_ag_blocks(mp);
 
 993	xfs_qm_unmount_quotas(mp);
 994 out_rtunmount:
 
 995	xfs_rtunmount_inodes(mp);
 996 out_rele_rip:
 997	xfs_irele(rip);
 998	/* Clean out dquots that might be in memory after quotacheck. */
 999	xfs_qm_unmount(mp);
1000
1001	/*
1002	 * Inactivate all inodes that might still be in memory after a log
1003	 * intent recovery failure so that reclaim can free them.  Metadata
1004	 * inodes and the root directory shouldn't need inactivation, but the
1005	 * mount failed for some reason, so pull down all the state and flee.
1006	 */
1007	xfs_inodegc_flush(mp);
1008
1009	/*
1010	 * Flush all inode reclamation work and flush the log.
1011	 * We have to do this /after/ rtunmount and qm_unmount because those
1012	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1013	 *
1014	 * This is slightly different from the unmountfs call sequence
1015	 * because we could be tearing down a partially set up mount.  In
1016	 * particular, if log_mount_finish fails we bail out without calling
1017	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1018	 * quota inodes.
1019	 */
1020	xfs_unmount_flush_inodes(mp);
1021 out_log_dealloc:
 
1022	xfs_log_mount_cancel(mp);
1023 out_inodegc_shrinker:
1024	shrinker_free(mp->m_inodegc_shrinker);
1025 out_fail_wait:
1026	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1027		xfs_buftarg_drain(mp->m_logdev_targp);
1028	xfs_buftarg_drain(mp->m_ddev_targp);
1029 out_free_perag:
1030	xfs_free_perag(mp);
1031 out_free_dir:
1032	xfs_da_unmount(mp);
1033 out_remove_uuid:
1034	xfs_uuid_unmount(mp);
1035 out_remove_errortag:
1036	xfs_errortag_del(mp);
1037 out_remove_error_sysfs:
1038	xfs_error_sysfs_del(mp);
1039 out_remove_scrub_stats:
1040	xchk_stats_unregister(mp->m_scrub_stats);
1041	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1042 out_remove_sysfs:
1043	xfs_sysfs_del(&mp->m_kobj);
1044 out:
1045	return error;
1046}
1047
1048/*
1049 * This flushes out the inodes,dquots and the superblock, unmounts the
1050 * log and makes sure that incore structures are freed.
1051 */
1052void
1053xfs_unmountfs(
1054	struct xfs_mount	*mp)
1055{
 
1056	int			error;
1057
1058	/*
1059	 * Perform all on-disk metadata updates required to inactivate inodes
1060	 * that the VFS evicted earlier in the unmount process.  Freeing inodes
1061	 * and discarding CoW fork preallocations can cause shape changes to
1062	 * the free inode and refcount btrees, respectively, so we must finish
1063	 * this before we discard the metadata space reservations.  Metadata
1064	 * inodes and the root directory do not require inactivation.
1065	 */
1066	xfs_inodegc_flush(mp);
1067
1068	xfs_blockgc_stop(mp);
1069	xfs_fs_unreserve_ag_blocks(mp);
1070	xfs_qm_unmount_quotas(mp);
1071	xfs_rtunmount_inodes(mp);
1072	xfs_irele(mp->m_rootip);
1073
1074	xfs_unmount_flush_inodes(mp);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1075
1076	xfs_qm_unmount(mp);
1077
1078	/*
1079	 * Unreserve any blocks we have so that when we unmount we don't account
1080	 * the reserved free space as used. This is really only necessary for
1081	 * lazy superblock counting because it trusts the incore superblock
1082	 * counters to be absolutely correct on clean unmount.
1083	 *
1084	 * We don't bother correcting this elsewhere for lazy superblock
1085	 * counting because on mount of an unclean filesystem we reconstruct the
1086	 * correct counter value and this is irrelevant.
1087	 *
1088	 * For non-lazy counter filesystems, this doesn't matter at all because
1089	 * we only every apply deltas to the superblock and hence the incore
1090	 * value does not matter....
1091	 */
1092	error = xfs_reserve_blocks(mp, 0);
 
1093	if (error)
1094		xfs_warn(mp, "Unable to free reserved block pool. "
1095				"Freespace may not be correct on next mount.");
1096	xfs_unmount_check(mp);
 
 
 
 
 
1097
1098	xfs_log_unmount(mp);
1099	xfs_da_unmount(mp);
1100	xfs_uuid_unmount(mp);
1101
1102#if defined(DEBUG)
1103	xfs_errortag_clearall(mp);
1104#endif
1105	shrinker_free(mp->m_inodegc_shrinker);
1106	xfs_free_perag(mp);
1107
1108	xfs_errortag_del(mp);
1109	xfs_error_sysfs_del(mp);
1110	xchk_stats_unregister(mp->m_scrub_stats);
1111	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1112	xfs_sysfs_del(&mp->m_kobj);
1113}
1114
1115/*
1116 * Determine whether modifications can proceed. The caller specifies the minimum
1117 * freeze level for which modifications should not be allowed. This allows
1118 * certain operations to proceed while the freeze sequence is in progress, if
1119 * necessary.
1120 */
1121bool
1122xfs_fs_writable(
1123	struct xfs_mount	*mp,
1124	int			level)
1125{
1126	ASSERT(level > SB_UNFROZEN);
1127	if ((mp->m_super->s_writers.frozen >= level) ||
1128	    xfs_is_shutdown(mp) || xfs_is_readonly(mp))
1129		return false;
1130
1131	return true;
1132}
1133
1134/* Adjust m_fdblocks or m_frextents. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1135int
1136xfs_mod_freecounter(
1137	struct xfs_mount	*mp,
1138	struct percpu_counter	*counter,
1139	int64_t			delta,
1140	bool			rsvd)
1141{
1142	int64_t			lcounter;
1143	long long		res_used;
1144	uint64_t		set_aside = 0;
1145	s32			batch;
1146	bool			has_resv_pool;
1147
1148	ASSERT(counter == &mp->m_fdblocks || counter == &mp->m_frextents);
1149	has_resv_pool = (counter == &mp->m_fdblocks);
1150	if (rsvd)
1151		ASSERT(has_resv_pool);
1152
1153	if (delta > 0) {
1154		/*
1155		 * If the reserve pool is depleted, put blocks back into it
1156		 * first. Most of the time the pool is full.
1157		 */
1158		if (likely(!has_resv_pool ||
1159			   mp->m_resblks == mp->m_resblks_avail)) {
1160			percpu_counter_add(counter, delta);
1161			return 0;
1162		}
1163
1164		spin_lock(&mp->m_sb_lock);
1165		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1166
1167		if (res_used > delta) {
1168			mp->m_resblks_avail += delta;
1169		} else {
1170			delta -= res_used;
1171			mp->m_resblks_avail = mp->m_resblks;
1172			percpu_counter_add(counter, delta);
1173		}
1174		spin_unlock(&mp->m_sb_lock);
1175		return 0;
1176	}
1177
1178	/*
1179	 * Taking blocks away, need to be more accurate the closer we
1180	 * are to zero.
1181	 *
1182	 * If the counter has a value of less than 2 * max batch size,
1183	 * then make everything serialise as we are real close to
1184	 * ENOSPC.
1185	 */
1186	if (__percpu_counter_compare(counter, 2 * XFS_FDBLOCKS_BATCH,
1187				     XFS_FDBLOCKS_BATCH) < 0)
1188		batch = 1;
1189	else
1190		batch = XFS_FDBLOCKS_BATCH;
1191
1192	/*
1193	 * Set aside allocbt blocks because these blocks are tracked as free
1194	 * space but not available for allocation. Technically this means that a
1195	 * single reservation cannot consume all remaining free space, but the
1196	 * ratio of allocbt blocks to usable free blocks should be rather small.
1197	 * The tradeoff without this is that filesystems that maintain high
1198	 * perag block reservations can over reserve physical block availability
1199	 * and fail physical allocation, which leads to much more serious
1200	 * problems (i.e. transaction abort, pagecache discards, etc.) than
1201	 * slightly premature -ENOSPC.
1202	 */
1203	if (has_resv_pool)
1204		set_aside = xfs_fdblocks_unavailable(mp);
1205	percpu_counter_add_batch(counter, delta, batch);
1206	if (__percpu_counter_compare(counter, set_aside,
1207				     XFS_FDBLOCKS_BATCH) >= 0) {
1208		/* we had space! */
1209		return 0;
1210	}
1211
1212	/*
1213	 * lock up the sb for dipping into reserves before releasing the space
1214	 * that took us to ENOSPC.
1215	 */
1216	spin_lock(&mp->m_sb_lock);
1217	percpu_counter_add(counter, -delta);
1218	if (!has_resv_pool || !rsvd)
1219		goto fdblocks_enospc;
1220
1221	lcounter = (long long)mp->m_resblks_avail + delta;
1222	if (lcounter >= 0) {
1223		mp->m_resblks_avail = lcounter;
1224		spin_unlock(&mp->m_sb_lock);
1225		return 0;
1226	}
1227	xfs_warn_once(mp,
1228"Reserve blocks depleted! Consider increasing reserve pool size.");
1229
 
1230fdblocks_enospc:
1231	spin_unlock(&mp->m_sb_lock);
1232	return -ENOSPC;
1233}
1234
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1235/*
1236 * Used to free the superblock along various error paths.
1237 */
1238void
1239xfs_freesb(
1240	struct xfs_mount	*mp)
1241{
1242	struct xfs_buf		*bp = mp->m_sb_bp;
1243
1244	xfs_buf_lock(bp);
1245	mp->m_sb_bp = NULL;
1246	xfs_buf_relse(bp);
1247}
1248
1249/*
1250 * If the underlying (data/log/rt) device is readonly, there are some
1251 * operations that cannot proceed.
1252 */
1253int
1254xfs_dev_is_read_only(
1255	struct xfs_mount	*mp,
1256	char			*message)
1257{
1258	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1259	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1260	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1261		xfs_notice(mp, "%s required on read-only device.", message);
1262		xfs_notice(mp, "write access unavailable, cannot proceed.");
1263		return -EROFS;
1264	}
1265	return 0;
1266}
1267
1268/* Force the summary counters to be recalculated at next mount. */
1269void
1270xfs_force_summary_recalc(
1271	struct xfs_mount	*mp)
1272{
1273	if (!xfs_has_lazysbcount(mp))
1274		return;
1275
1276	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1277}
1278
1279/*
1280 * Enable a log incompat feature flag in the primary superblock.  The caller
1281 * cannot have any other transactions in progress.
1282 */
1283int
1284xfs_add_incompat_log_feature(
1285	struct xfs_mount	*mp,
1286	uint32_t		feature)
1287{
1288	struct xfs_dsb		*dsb;
1289	int			error;
1290
1291	ASSERT(hweight32(feature) == 1);
1292	ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1293
1294	/*
1295	 * Force the log to disk and kick the background AIL thread to reduce
1296	 * the chances that the bwrite will stall waiting for the AIL to unpin
1297	 * the primary superblock buffer.  This isn't a data integrity
1298	 * operation, so we don't need a synchronous push.
1299	 */
1300	error = xfs_log_force(mp, XFS_LOG_SYNC);
1301	if (error)
1302		return error;
1303	xfs_ail_push_all(mp->m_ail);
1304
1305	/*
1306	 * Lock the primary superblock buffer to serialize all callers that
1307	 * are trying to set feature bits.
1308	 */
1309	xfs_buf_lock(mp->m_sb_bp);
1310	xfs_buf_hold(mp->m_sb_bp);
1311
1312	if (xfs_is_shutdown(mp)) {
1313		error = -EIO;
1314		goto rele;
1315	}
1316
1317	if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1318		goto rele;
1319
1320	/*
1321	 * Write the primary superblock to disk immediately, because we need
1322	 * the log_incompat bit to be set in the primary super now to protect
1323	 * the log items that we're going to commit later.
1324	 */
1325	dsb = mp->m_sb_bp->b_addr;
1326	xfs_sb_to_disk(dsb, &mp->m_sb);
1327	dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1328	error = xfs_bwrite(mp->m_sb_bp);
1329	if (error)
1330		goto shutdown;
1331
1332	/*
1333	 * Add the feature bits to the incore superblock before we unlock the
1334	 * buffer.
1335	 */
1336	xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1337	xfs_buf_relse(mp->m_sb_bp);
1338
1339	/* Log the superblock to disk. */
1340	return xfs_sync_sb(mp, false);
1341shutdown:
1342	xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1343rele:
1344	xfs_buf_relse(mp->m_sb_bp);
1345	return error;
1346}
1347
1348/*
1349 * Clear all the log incompat flags from the superblock.
1350 *
1351 * The caller cannot be in a transaction, must ensure that the log does not
1352 * contain any log items protected by any log incompat bit, and must ensure
1353 * that there are no other threads that depend on the state of the log incompat
1354 * feature flags in the primary super.
1355 *
1356 * Returns true if the superblock is dirty.
1357 */
1358bool
1359xfs_clear_incompat_log_features(
1360	struct xfs_mount	*mp)
1361{
1362	bool			ret = false;
1363
1364	if (!xfs_has_crc(mp) ||
1365	    !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1366				XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1367	    xfs_is_shutdown(mp))
1368		return false;
1369
1370	/*
1371	 * Update the incore superblock.  We synchronize on the primary super
1372	 * buffer lock to be consistent with the add function, though at least
1373	 * in theory this shouldn't be necessary.
1374	 */
1375	xfs_buf_lock(mp->m_sb_bp);
1376	xfs_buf_hold(mp->m_sb_bp);
1377
1378	if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1379				XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1380		xfs_sb_remove_incompat_log_features(&mp->m_sb);
1381		ret = true;
1382	}
1383
1384	xfs_buf_relse(mp->m_sb_bp);
1385	return ret;
1386}
1387
1388/*
1389 * Update the in-core delayed block counter.
1390 *
1391 * We prefer to update the counter without having to take a spinlock for every
1392 * counter update (i.e. batching).  Each change to delayed allocation
1393 * reservations can change can easily exceed the default percpu counter
1394 * batching, so we use a larger batch factor here.
1395 *
1396 * Note that we don't currently have any callers requiring fast summation
1397 * (e.g. percpu_counter_read) so we can use a big batch value here.
1398 */
1399#define XFS_DELALLOC_BATCH	(4096)
1400void
1401xfs_mod_delalloc(
1402	struct xfs_mount	*mp,
1403	int64_t			delta)
1404{
1405	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1406			XFS_DELALLOC_BATCH);
1407}
v4.10.11
 
   1/*
   2 * Copyright (c) 2000-2005 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_sb.h"
  26#include "xfs_mount.h"
  27#include "xfs_defer.h"
  28#include "xfs_da_format.h"
  29#include "xfs_da_btree.h"
  30#include "xfs_inode.h"
  31#include "xfs_dir2.h"
  32#include "xfs_ialloc.h"
  33#include "xfs_alloc.h"
  34#include "xfs_rtalloc.h"
  35#include "xfs_bmap.h"
  36#include "xfs_trans.h"
  37#include "xfs_trans_priv.h"
  38#include "xfs_log.h"
 
  39#include "xfs_error.h"
  40#include "xfs_quota.h"
  41#include "xfs_fsops.h"
  42#include "xfs_trace.h"
  43#include "xfs_icache.h"
  44#include "xfs_sysfs.h"
  45#include "xfs_rmap_btree.h"
  46#include "xfs_refcount_btree.h"
  47#include "xfs_reflink.h"
  48
 
 
 
 
  49
  50static DEFINE_MUTEX(xfs_uuid_table_mutex);
  51static int xfs_uuid_table_size;
  52static uuid_t *xfs_uuid_table;
  53
  54void
  55xfs_uuid_table_free(void)
  56{
  57	if (xfs_uuid_table_size == 0)
  58		return;
  59	kmem_free(xfs_uuid_table);
  60	xfs_uuid_table = NULL;
  61	xfs_uuid_table_size = 0;
  62}
  63
  64/*
  65 * See if the UUID is unique among mounted XFS filesystems.
  66 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
  67 */
  68STATIC int
  69xfs_uuid_mount(
  70	struct xfs_mount	*mp)
  71{
  72	uuid_t			*uuid = &mp->m_sb.sb_uuid;
  73	int			hole, i;
  74
  75	if (mp->m_flags & XFS_MOUNT_NOUUID)
 
 
 
  76		return 0;
  77
  78	if (uuid_is_nil(uuid)) {
  79		xfs_warn(mp, "Filesystem has nil UUID - can't mount");
  80		return -EINVAL;
  81	}
  82
  83	mutex_lock(&xfs_uuid_table_mutex);
  84	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
  85		if (uuid_is_nil(&xfs_uuid_table[i])) {
  86			hole = i;
  87			continue;
  88		}
  89		if (uuid_equal(uuid, &xfs_uuid_table[i]))
  90			goto out_duplicate;
  91	}
  92
  93	if (hole < 0) {
  94		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
  95			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
  96			KM_SLEEP);
  97		hole = xfs_uuid_table_size++;
  98	}
  99	xfs_uuid_table[hole] = *uuid;
 100	mutex_unlock(&xfs_uuid_table_mutex);
 101
 102	return 0;
 103
 104 out_duplicate:
 105	mutex_unlock(&xfs_uuid_table_mutex);
 106	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
 107	return -EINVAL;
 108}
 109
 110STATIC void
 111xfs_uuid_unmount(
 112	struct xfs_mount	*mp)
 113{
 114	uuid_t			*uuid = &mp->m_sb.sb_uuid;
 115	int			i;
 116
 117	if (mp->m_flags & XFS_MOUNT_NOUUID)
 118		return;
 119
 120	mutex_lock(&xfs_uuid_table_mutex);
 121	for (i = 0; i < xfs_uuid_table_size; i++) {
 122		if (uuid_is_nil(&xfs_uuid_table[i]))
 123			continue;
 124		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
 125			continue;
 126		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
 127		break;
 128	}
 129	ASSERT(i < xfs_uuid_table_size);
 130	mutex_unlock(&xfs_uuid_table_mutex);
 131}
 132
 133
 134STATIC void
 135__xfs_free_perag(
 136	struct rcu_head	*head)
 137{
 138	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 139
 140	ASSERT(atomic_read(&pag->pag_ref) == 0);
 141	kmem_free(pag);
 142}
 143
 144/*
 145 * Free up the per-ag resources associated with the mount structure.
 146 */
 147STATIC void
 148xfs_free_perag(
 149	xfs_mount_t	*mp)
 150{
 151	xfs_agnumber_t	agno;
 152	struct xfs_perag *pag;
 153
 154	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
 155		spin_lock(&mp->m_perag_lock);
 156		pag = radix_tree_delete(&mp->m_perag_tree, agno);
 157		spin_unlock(&mp->m_perag_lock);
 158		ASSERT(pag);
 159		ASSERT(atomic_read(&pag->pag_ref) == 0);
 160		xfs_buf_hash_destroy(pag);
 161		call_rcu(&pag->rcu_head, __xfs_free_perag);
 162	}
 163}
 164
 165/*
 166 * Check size of device based on the (data/realtime) block count.
 167 * Note: this check is used by the growfs code as well as mount.
 168 */
 169int
 170xfs_sb_validate_fsb_count(
 171	xfs_sb_t	*sbp,
 172	__uint64_t	nblocks)
 173{
 174	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
 175	ASSERT(sbp->sb_blocklog >= BBSHIFT);
 176
 177	/* Limited by ULONG_MAX of page cache index */
 178	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
 179		return -EFBIG;
 180	return 0;
 181}
 182
 183int
 184xfs_initialize_perag(
 185	xfs_mount_t	*mp,
 186	xfs_agnumber_t	agcount,
 187	xfs_agnumber_t	*maxagi)
 188{
 189	xfs_agnumber_t	index;
 190	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
 191	xfs_perag_t	*pag;
 192	int		error = -ENOMEM;
 193
 194	/*
 195	 * Walk the current per-ag tree so we don't try to initialise AGs
 196	 * that already exist (growfs case). Allocate and insert all the
 197	 * AGs we don't find ready for initialisation.
 198	 */
 199	for (index = 0; index < agcount; index++) {
 200		pag = xfs_perag_get(mp, index);
 201		if (pag) {
 202			xfs_perag_put(pag);
 203			continue;
 204		}
 205
 206		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
 207		if (!pag)
 208			goto out_unwind_new_pags;
 209		pag->pag_agno = index;
 210		pag->pag_mount = mp;
 211		spin_lock_init(&pag->pag_ici_lock);
 212		mutex_init(&pag->pag_ici_reclaim_lock);
 213		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 214		if (xfs_buf_hash_init(pag))
 215			goto out_free_pag;
 216
 217		if (radix_tree_preload(GFP_NOFS))
 218			goto out_hash_destroy;
 219
 220		spin_lock(&mp->m_perag_lock);
 221		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
 222			BUG();
 223			spin_unlock(&mp->m_perag_lock);
 224			radix_tree_preload_end();
 225			error = -EEXIST;
 226			goto out_hash_destroy;
 227		}
 228		spin_unlock(&mp->m_perag_lock);
 229		radix_tree_preload_end();
 230		/* first new pag is fully initialized */
 231		if (first_initialised == NULLAGNUMBER)
 232			first_initialised = index;
 233	}
 234
 235	index = xfs_set_inode_alloc(mp, agcount);
 236
 237	if (maxagi)
 238		*maxagi = index;
 239
 240	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
 241	return 0;
 242
 243out_hash_destroy:
 244	xfs_buf_hash_destroy(pag);
 245out_free_pag:
 246	kmem_free(pag);
 247out_unwind_new_pags:
 248	/* unwind any prior newly initialized pags */
 249	for (index = first_initialised; index < agcount; index++) {
 250		pag = radix_tree_delete(&mp->m_perag_tree, index);
 251		if (!pag)
 252			break;
 253		xfs_buf_hash_destroy(pag);
 254		kmem_free(pag);
 255	}
 256	return error;
 257}
 258
 259/*
 260 * xfs_readsb
 261 *
 262 * Does the initial read of the superblock.
 263 */
 264int
 265xfs_readsb(
 266	struct xfs_mount *mp,
 267	int		flags)
 268{
 269	unsigned int	sector_size;
 270	struct xfs_buf	*bp;
 271	struct xfs_sb	*sbp = &mp->m_sb;
 272	int		error;
 273	int		loud = !(flags & XFS_MFSI_QUIET);
 274	const struct xfs_buf_ops *buf_ops;
 275
 276	ASSERT(mp->m_sb_bp == NULL);
 277	ASSERT(mp->m_ddev_targp != NULL);
 278
 279	/*
 280	 * For the initial read, we must guess at the sector
 281	 * size based on the block device.  It's enough to
 282	 * get the sb_sectsize out of the superblock and
 283	 * then reread with the proper length.
 284	 * We don't verify it yet, because it may not be complete.
 285	 */
 286	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
 287	buf_ops = NULL;
 288
 289	/*
 290	 * Allocate a (locked) buffer to hold the superblock. This will be kept
 291	 * around at all times to optimize access to the superblock. Therefore,
 292	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
 293	 * elevated.
 294	 */
 295reread:
 296	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
 297				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
 298				      buf_ops);
 299	if (error) {
 300		if (loud)
 301			xfs_warn(mp, "SB validate failed with error %d.", error);
 302		/* bad CRC means corrupted metadata */
 303		if (error == -EFSBADCRC)
 304			error = -EFSCORRUPTED;
 305		return error;
 306	}
 307
 308	/*
 309	 * Initialize the mount structure from the superblock.
 310	 */
 311	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
 312
 313	/*
 314	 * If we haven't validated the superblock, do so now before we try
 315	 * to check the sector size and reread the superblock appropriately.
 316	 */
 317	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
 318		if (loud)
 319			xfs_warn(mp, "Invalid superblock magic number");
 320		error = -EINVAL;
 321		goto release_buf;
 322	}
 323
 324	/*
 325	 * We must be able to do sector-sized and sector-aligned IO.
 326	 */
 327	if (sector_size > sbp->sb_sectsize) {
 328		if (loud)
 329			xfs_warn(mp, "device supports %u byte sectors (not %u)",
 330				sector_size, sbp->sb_sectsize);
 331		error = -ENOSYS;
 332		goto release_buf;
 333	}
 334
 335	if (buf_ops == NULL) {
 336		/*
 337		 * Re-read the superblock so the buffer is correctly sized,
 338		 * and properly verified.
 339		 */
 340		xfs_buf_relse(bp);
 341		sector_size = sbp->sb_sectsize;
 342		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
 343		goto reread;
 344	}
 345
 
 346	xfs_reinit_percpu_counters(mp);
 347
 348	/* no need to be quiet anymore, so reset the buf ops */
 349	bp->b_ops = &xfs_sb_buf_ops;
 350
 351	mp->m_sb_bp = bp;
 352	xfs_buf_unlock(bp);
 353	return 0;
 354
 355release_buf:
 356	xfs_buf_relse(bp);
 357	return error;
 358}
 359
 360/*
 361 * Update alignment values based on mount options and sb values
 
 
 
 362 */
 363STATIC int
 364xfs_update_alignment(xfs_mount_t *mp)
 
 
 
 365{
 366	xfs_sb_t	*sbp = &(mp->m_sb);
 
 367
 368	if (mp->m_dalign) {
 369		/*
 370		 * If stripe unit and stripe width are not multiples
 371		 * of the fs blocksize turn off alignment.
 372		 */
 373		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
 374		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
 375			xfs_warn(mp,
 376		"alignment check failed: sunit/swidth vs. blocksize(%d)",
 377				sbp->sb_blocksize);
 378			return -EINVAL;
 379		} else {
 380			/*
 381			 * Convert the stripe unit and width to FSBs.
 382			 */
 383			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
 384			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
 385				xfs_warn(mp,
 386			"alignment check failed: sunit/swidth vs. agsize(%d)",
 387					 sbp->sb_agblocks);
 388				return -EINVAL;
 389			} else if (mp->m_dalign) {
 390				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
 391			} else {
 392				xfs_warn(mp,
 393			"alignment check failed: sunit(%d) less than bsize(%d)",
 394					 mp->m_dalign, sbp->sb_blocksize);
 395				return -EINVAL;
 396			}
 397		}
 398
 399		/*
 400		 * Update superblock with new values
 401		 * and log changes
 402		 */
 403		if (xfs_sb_version_hasdalign(sbp)) {
 404			if (sbp->sb_unit != mp->m_dalign) {
 405				sbp->sb_unit = mp->m_dalign;
 406				mp->m_update_sb = true;
 407			}
 408			if (sbp->sb_width != mp->m_swidth) {
 409				sbp->sb_width = mp->m_swidth;
 410				mp->m_update_sb = true;
 411			}
 412		} else {
 413			xfs_warn(mp,
 414	"cannot change alignment: superblock does not support data alignment");
 415			return -EINVAL;
 416		}
 417	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
 418		    xfs_sb_version_hasdalign(&mp->m_sb)) {
 419			mp->m_dalign = sbp->sb_unit;
 420			mp->m_swidth = sbp->sb_width;
 421	}
 422
 
 
 
 
 
 
 
 
 
 
 423	return 0;
 424}
 425
 426/*
 427 * Set the maximum inode count for this filesystem
 
 
 
 428 */
 429STATIC void
 430xfs_set_maxicount(xfs_mount_t *mp)
 
 431{
 432	xfs_sb_t	*sbp = &(mp->m_sb);
 433	__uint64_t	icount;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 434
 435	if (sbp->sb_imax_pct) {
 436		/*
 437		 * Make sure the maximum inode count is a multiple
 438		 * of the units we allocate inodes in.
 439		 */
 440		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 441		do_div(icount, 100);
 442		do_div(icount, mp->m_ialloc_blks);
 443		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
 444				   sbp->sb_inopblog;
 445	} else {
 446		mp->m_maxicount = 0;
 447	}
 
 
 448}
 449
 450/*
 451 * Set the default minimum read and write sizes unless
 452 * already specified in a mount option.
 453 * We use smaller I/O sizes when the file system
 454 * is being used for NFS service (wsync mount option).
 455 */
 456STATIC void
 457xfs_set_rw_sizes(xfs_mount_t *mp)
 458{
 459	xfs_sb_t	*sbp = &(mp->m_sb);
 460	int		readio_log, writeio_log;
 
 
 
 
 
 
 
 
 
 
 
 461
 462	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
 463		if (mp->m_flags & XFS_MOUNT_WSYNC) {
 464			readio_log = XFS_WSYNC_READIO_LOG;
 465			writeio_log = XFS_WSYNC_WRITEIO_LOG;
 466		} else {
 467			readio_log = XFS_READIO_LOG_LARGE;
 468			writeio_log = XFS_WRITEIO_LOG_LARGE;
 469		}
 470	} else {
 471		readio_log = mp->m_readio_log;
 472		writeio_log = mp->m_writeio_log;
 473	}
 474
 475	if (sbp->sb_blocklog > readio_log) {
 476		mp->m_readio_log = sbp->sb_blocklog;
 477	} else {
 478		mp->m_readio_log = readio_log;
 479	}
 480	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
 481	if (sbp->sb_blocklog > writeio_log) {
 482		mp->m_writeio_log = sbp->sb_blocklog;
 483	} else {
 484		mp->m_writeio_log = writeio_log;
 485	}
 486	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 487}
 488
 489/*
 490 * precalculate the low space thresholds for dynamic speculative preallocation.
 491 */
 492void
 493xfs_set_low_space_thresholds(
 494	struct xfs_mount	*mp)
 495{
 496	int i;
 
 
 
 
 
 497
 498	for (i = 0; i < XFS_LOWSP_MAX; i++) {
 499		__uint64_t space = mp->m_sb.sb_dblocks;
 500
 501		do_div(space, 100);
 502		mp->m_low_space[i] = space * (i + 1);
 503	}
 504}
 505
 506
 507/*
 508 * Set whether we're using inode alignment.
 509 */
 510STATIC void
 511xfs_set_inoalignment(xfs_mount_t *mp)
 512{
 513	if (xfs_sb_version_hasalign(&mp->m_sb) &&
 514		mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
 515		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
 516	else
 517		mp->m_inoalign_mask = 0;
 518	/*
 519	 * If we are using stripe alignment, check whether
 520	 * the stripe unit is a multiple of the inode alignment
 521	 */
 522	if (mp->m_dalign && mp->m_inoalign_mask &&
 523	    !(mp->m_dalign & mp->m_inoalign_mask))
 524		mp->m_sinoalign = mp->m_dalign;
 525	else
 526		mp->m_sinoalign = 0;
 527}
 528
 529/*
 530 * Check that the data (and log if separate) is an ok size.
 531 */
 532STATIC int
 533xfs_check_sizes(
 534	struct xfs_mount *mp)
 535{
 536	struct xfs_buf	*bp;
 537	xfs_daddr_t	d;
 538	int		error;
 539
 540	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
 541	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
 542		xfs_warn(mp, "filesystem size mismatch detected");
 543		return -EFBIG;
 544	}
 545	error = xfs_buf_read_uncached(mp->m_ddev_targp,
 546					d - XFS_FSS_TO_BB(mp, 1),
 547					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 548	if (error) {
 549		xfs_warn(mp, "last sector read failed");
 550		return error;
 551	}
 552	xfs_buf_relse(bp);
 553
 554	if (mp->m_logdev_targp == mp->m_ddev_targp)
 555		return 0;
 556
 557	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
 558	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
 559		xfs_warn(mp, "log size mismatch detected");
 560		return -EFBIG;
 561	}
 562	error = xfs_buf_read_uncached(mp->m_logdev_targp,
 563					d - XFS_FSB_TO_BB(mp, 1),
 564					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
 565	if (error) {
 566		xfs_warn(mp, "log device read failed");
 567		return error;
 568	}
 569	xfs_buf_relse(bp);
 570	return 0;
 571}
 572
 573/*
 574 * Clear the quotaflags in memory and in the superblock.
 575 */
 576int
 577xfs_mount_reset_sbqflags(
 578	struct xfs_mount	*mp)
 579{
 580	mp->m_qflags = 0;
 581
 582	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
 583	if (mp->m_sb.sb_qflags == 0)
 584		return 0;
 585	spin_lock(&mp->m_sb_lock);
 586	mp->m_sb.sb_qflags = 0;
 587	spin_unlock(&mp->m_sb_lock);
 588
 589	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
 590		return 0;
 591
 592	return xfs_sync_sb(mp, false);
 593}
 594
 595__uint64_t
 596xfs_default_resblks(xfs_mount_t *mp)
 597{
 598	__uint64_t resblks;
 599
 600	/*
 601	 * We default to 5% or 8192 fsbs of space reserved, whichever is
 602	 * smaller.  This is intended to cover concurrent allocation
 603	 * transactions when we initially hit enospc. These each require a 4
 604	 * block reservation. Hence by default we cover roughly 2000 concurrent
 605	 * allocation reservations.
 606	 */
 607	resblks = mp->m_sb.sb_dblocks;
 608	do_div(resblks, 20);
 609	resblks = min_t(__uint64_t, resblks, 8192);
 610	return resblks;
 611}
 612
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 613/*
 614 * This function does the following on an initial mount of a file system:
 615 *	- reads the superblock from disk and init the mount struct
 616 *	- if we're a 32-bit kernel, do a size check on the superblock
 617 *		so we don't mount terabyte filesystems
 618 *	- init mount struct realtime fields
 619 *	- allocate inode hash table for fs
 620 *	- init directory manager
 621 *	- perform recovery and init the log manager
 622 */
 623int
 624xfs_mountfs(
 625	struct xfs_mount	*mp)
 626{
 627	struct xfs_sb		*sbp = &(mp->m_sb);
 628	struct xfs_inode	*rip;
 629	__uint64_t		resblks;
 630	uint			quotamount = 0;
 631	uint			quotaflags = 0;
 632	int			error = 0;
 633
 634	xfs_sb_mount_common(mp, sbp);
 635
 636	/*
 637	 * Check for a mismatched features2 values.  Older kernels read & wrote
 638	 * into the wrong sb offset for sb_features2 on some platforms due to
 639	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
 640	 * which made older superblock reading/writing routines swap it as a
 641	 * 64-bit value.
 642	 *
 643	 * For backwards compatibility, we make both slots equal.
 644	 *
 645	 * If we detect a mismatched field, we OR the set bits into the existing
 646	 * features2 field in case it has already been modified; we don't want
 647	 * to lose any features.  We then update the bad location with the ORed
 648	 * value so that older kernels will see any features2 flags. The
 649	 * superblock writeback code ensures the new sb_features2 is copied to
 650	 * sb_bad_features2 before it is logged or written to disk.
 651	 */
 652	if (xfs_sb_has_mismatched_features2(sbp)) {
 653		xfs_warn(mp, "correcting sb_features alignment problem");
 654		sbp->sb_features2 |= sbp->sb_bad_features2;
 655		mp->m_update_sb = true;
 656
 657		/*
 658		 * Re-check for ATTR2 in case it was found in bad_features2
 659		 * slot.
 660		 */
 661		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 662		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
 663			mp->m_flags |= XFS_MOUNT_ATTR2;
 664	}
 665
 666	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
 667	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
 668		xfs_sb_version_removeattr2(&mp->m_sb);
 669		mp->m_update_sb = true;
 670
 671		/* update sb_versionnum for the clearing of the morebits */
 672		if (!sbp->sb_features2)
 673			mp->m_update_sb = true;
 674	}
 675
 676	/* always use v2 inodes by default now */
 677	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
 678		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
 
 679		mp->m_update_sb = true;
 680	}
 681
 682	/*
 683	 * Check if sb_agblocks is aligned at stripe boundary
 684	 * If sb_agblocks is NOT aligned turn off m_dalign since
 685	 * allocator alignment is within an ag, therefore ag has
 686	 * to be aligned at stripe boundary.
 687	 */
 688	error = xfs_update_alignment(mp);
 689	if (error)
 690		goto out;
 691
 692	xfs_alloc_compute_maxlevels(mp);
 693	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
 694	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
 695	xfs_ialloc_compute_maxlevels(mp);
 696	xfs_rmapbt_compute_maxlevels(mp);
 697	xfs_refcountbt_compute_maxlevels(mp);
 698
 699	xfs_set_maxicount(mp);
 
 
 
 
 
 
 
 
 
 
 
 700
 701	/* enable fail_at_unmount as default */
 702	mp->m_fail_unmount = 1;
 703
 704	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
 
 
 
 705	if (error)
 706		goto out;
 707
 708	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
 709			       &mp->m_kobj, "stats");
 710	if (error)
 711		goto out_remove_sysfs;
 712
 
 
 713	error = xfs_error_sysfs_init(mp);
 714	if (error)
 715		goto out_del_stats;
 716
 
 
 
 717
 718	error = xfs_uuid_mount(mp);
 719	if (error)
 720		goto out_remove_error_sysfs;
 721
 722	/*
 723	 * Set the minimum read and write sizes
 
 724	 */
 725	xfs_set_rw_sizes(mp);
 
 
 726
 727	/* set the low space thresholds for dynamic preallocation */
 728	xfs_set_low_space_thresholds(mp);
 729
 730	/*
 731	 * Set the inode cluster size.
 732	 * This may still be overridden by the file system
 733	 * block size if it is larger than the chosen cluster size.
 734	 *
 735	 * For v5 filesystems, scale the cluster size with the inode size to
 736	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
 737	 * has set the inode alignment value appropriately for larger cluster
 738	 * sizes.
 739	 */
 740	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
 741	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 742		int	new_size = mp->m_inode_cluster_size;
 743
 744		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
 745		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
 746			mp->m_inode_cluster_size = new_size;
 747	}
 748
 749	/*
 750	 * If enabled, sparse inode chunk alignment is expected to match the
 751	 * cluster size. Full inode chunk alignment must match the chunk size,
 752	 * but that is checked on sb read verification...
 753	 */
 754	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
 755	    mp->m_sb.sb_spino_align !=
 756			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
 757		xfs_warn(mp,
 758	"Sparse inode block alignment (%u) must match cluster size (%llu).",
 759			 mp->m_sb.sb_spino_align,
 760			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
 761		error = -EINVAL;
 762		goto out_remove_uuid;
 763	}
 764
 765	/*
 766	 * Set inode alignment fields
 767	 */
 768	xfs_set_inoalignment(mp);
 769
 770	/*
 771	 * Check that the data (and log if separate) is an ok size.
 772	 */
 773	error = xfs_check_sizes(mp);
 774	if (error)
 775		goto out_remove_uuid;
 776
 777	/*
 778	 * Initialize realtime fields in the mount structure
 779	 */
 780	error = xfs_rtmount_init(mp);
 781	if (error) {
 782		xfs_warn(mp, "RT mount failed");
 783		goto out_remove_uuid;
 784	}
 785
 786	/*
 787	 *  Copies the low order bits of the timestamp and the randomly
 788	 *  set "sequence" number out of a UUID.
 789	 */
 790	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
 791
 792	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
 
 793
 794	error = xfs_da_mount(mp);
 795	if (error) {
 796		xfs_warn(mp, "Failed dir/attr init: %d", error);
 797		goto out_remove_uuid;
 798	}
 799
 800	/*
 801	 * Initialize the precomputed transaction reservations values.
 802	 */
 803	xfs_trans_init(mp);
 804
 805	/*
 806	 * Allocate and initialize the per-ag data.
 807	 */
 808	spin_lock_init(&mp->m_perag_lock);
 809	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
 810	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
 811	if (error) {
 812		xfs_warn(mp, "Failed per-ag init: %d", error);
 813		goto out_free_dir;
 814	}
 815
 816	if (!sbp->sb_logblocks) {
 817		xfs_warn(mp, "no log defined");
 818		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
 819		error = -EFSCORRUPTED;
 820		goto out_free_perag;
 821	}
 822
 
 
 
 
 823	/*
 824	 * Log's mount-time initialization. The first part of recovery can place
 825	 * some items on the AIL, to be handled when recovery is finished or
 826	 * cancelled.
 827	 */
 828	error = xfs_log_mount(mp, mp->m_logdev_targp,
 829			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
 830			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
 831	if (error) {
 832		xfs_warn(mp, "log mount failed");
 833		goto out_fail_wait;
 834	}
 835
 
 
 
 
 836	/*
 837	 * Now the log is mounted, we know if it was an unclean shutdown or
 838	 * not. If it was, with the first phase of recovery has completed, we
 839	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
 840	 * but they are recovered transactionally in the second recovery phase
 841	 * later.
 842	 *
 843	 * Hence we can safely re-initialise incore superblock counters from
 844	 * the per-ag data. These may not be correct if the filesystem was not
 845	 * cleanly unmounted, so we need to wait for recovery to finish before
 846	 * doing this.
 847	 *
 848	 * If the filesystem was cleanly unmounted, then we can trust the
 849	 * values in the superblock to be correct and we don't need to do
 850	 * anything here.
 851	 *
 852	 * If we are currently making the filesystem, the initialisation will
 853	 * fail as the perag data is in an undefined state.
 854	 */
 855	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
 856	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
 857	     !mp->m_sb.sb_inprogress) {
 858		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
 859		if (error)
 860			goto out_log_dealloc;
 861	}
 862
 863	/*
 864	 * Get and sanity-check the root inode.
 865	 * Save the pointer to it in the mount structure.
 866	 */
 867	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
 
 868	if (error) {
 869		xfs_warn(mp, "failed to read root inode");
 
 
 870		goto out_log_dealloc;
 871	}
 872
 873	ASSERT(rip != NULL);
 874
 875	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
 876		xfs_warn(mp, "corrupted root inode %llu: not a directory",
 877			(unsigned long long)rip->i_ino);
 878		xfs_iunlock(rip, XFS_ILOCK_EXCL);
 879		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
 880				 mp);
 881		error = -EFSCORRUPTED;
 882		goto out_rele_rip;
 883	}
 884	mp->m_rootip = rip;	/* save it */
 885
 886	xfs_iunlock(rip, XFS_ILOCK_EXCL);
 887
 888	/*
 889	 * Initialize realtime inode pointers in the mount structure
 890	 */
 891	error = xfs_rtmount_inodes(mp);
 892	if (error) {
 893		/*
 894		 * Free up the root inode.
 895		 */
 896		xfs_warn(mp, "failed to read RT inodes");
 897		goto out_rele_rip;
 898	}
 899
 
 
 
 
 
 900	/*
 901	 * If this is a read-only mount defer the superblock updates until
 902	 * the next remount into writeable mode.  Otherwise we would never
 903	 * perform the update e.g. for the root filesystem.
 904	 */
 905	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
 906		error = xfs_sync_sb(mp, false);
 907		if (error) {
 908			xfs_warn(mp, "failed to write sb changes");
 909			goto out_rtunmount;
 910		}
 911	}
 912
 913	/*
 914	 * Initialise the XFS quota management subsystem for this mount
 915	 */
 916	if (XFS_IS_QUOTA_RUNNING(mp)) {
 917		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
 918		if (error)
 919			goto out_rtunmount;
 920	} else {
 921		ASSERT(!XFS_IS_QUOTA_ON(mp));
 922
 923		/*
 924		 * If a file system had quotas running earlier, but decided to
 925		 * mount without -o uquota/pquota/gquota options, revoke the
 926		 * quotachecked license.
 927		 */
 928		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
 929			xfs_notice(mp, "resetting quota flags");
 930			error = xfs_mount_reset_sbqflags(mp);
 931			if (error)
 932				goto out_rtunmount;
 933		}
 934	}
 935
 936	/*
 937	 * During the second phase of log recovery, we need iget and
 938	 * iput to behave like they do for an active filesystem.
 939	 * xfs_fs_drop_inode needs to be able to prevent the deletion
 940	 * of inodes before we're done replaying log items on those
 941	 * inodes.
 942	 */
 943	mp->m_super->s_flags |= MS_ACTIVE;
 944
 945	/*
 946	 * Finish recovering the file system.  This part needed to be delayed
 947	 * until after the root and real-time bitmap inodes were consistently
 948	 * read in.
 
 
 
 949	 */
 
 
 
 
 950	error = xfs_log_mount_finish(mp);
 
 951	if (error) {
 952		xfs_warn(mp, "log mount finish failed");
 953		goto out_rtunmount;
 954	}
 955
 956	/*
 957	 * Now the log is fully replayed, we can transition to full read-only
 958	 * mode for read-only mounts. This will sync all the metadata and clean
 959	 * the log so that the recovery we just performed does not have to be
 960	 * replayed again on the next mount.
 961	 *
 962	 * We use the same quiesce mechanism as the rw->ro remount, as they are
 963	 * semantically identical operations.
 964	 */
 965	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
 966							XFS_MOUNT_RDONLY) {
 967		xfs_quiesce_attr(mp);
 968	}
 969
 970	/*
 971	 * Complete the quota initialisation, post-log-replay component.
 972	 */
 973	if (quotamount) {
 974		ASSERT(mp->m_qflags == 0);
 975		mp->m_qflags = quotaflags;
 976
 977		xfs_qm_mount_quotas(mp);
 978	}
 979
 980	/*
 981	 * Now we are mounted, reserve a small amount of unused space for
 982	 * privileged transactions. This is needed so that transaction
 983	 * space required for critical operations can dip into this pool
 984	 * when at ENOSPC. This is needed for operations like create with
 985	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
 986	 * are not allowed to use this reserved space.
 987	 *
 988	 * This may drive us straight to ENOSPC on mount, but that implies
 989	 * we were already there on the last unmount. Warn if this occurs.
 990	 */
 991	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
 992		resblks = xfs_default_resblks(mp);
 993		error = xfs_reserve_blocks(mp, &resblks, NULL);
 994		if (error)
 995			xfs_warn(mp,
 996	"Unable to allocate reserve blocks. Continuing without reserve pool.");
 997
 998		/* Recover any CoW blocks that never got remapped. */
 999		error = xfs_reflink_recover_cow(mp);
1000		if (error) {
1001			xfs_err(mp,
1002	"Error %d recovering leftover CoW allocations.", error);
1003			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1004			goto out_quota;
1005		}
1006
1007		/* Reserve AG blocks for future btree expansion. */
1008		error = xfs_fs_reserve_ag_blocks(mp);
1009		if (error && error != -ENOSPC)
1010			goto out_agresv;
1011	}
1012
1013	return 0;
1014
1015 out_agresv:
1016	xfs_fs_unreserve_ag_blocks(mp);
1017 out_quota:
1018	xfs_qm_unmount_quotas(mp);
1019 out_rtunmount:
1020	mp->m_super->s_flags &= ~MS_ACTIVE;
1021	xfs_rtunmount_inodes(mp);
1022 out_rele_rip:
1023	IRELE(rip);
1024	cancel_delayed_work_sync(&mp->m_reclaim_work);
1025	xfs_reclaim_inodes(mp, SYNC_WAIT);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1026 out_log_dealloc:
1027	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1028	xfs_log_mount_cancel(mp);
 
 
1029 out_fail_wait:
1030	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1031		xfs_wait_buftarg(mp->m_logdev_targp);
1032	xfs_wait_buftarg(mp->m_ddev_targp);
1033 out_free_perag:
1034	xfs_free_perag(mp);
1035 out_free_dir:
1036	xfs_da_unmount(mp);
1037 out_remove_uuid:
1038	xfs_uuid_unmount(mp);
 
 
1039 out_remove_error_sysfs:
1040	xfs_error_sysfs_del(mp);
1041 out_del_stats:
 
1042	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1043 out_remove_sysfs:
1044	xfs_sysfs_del(&mp->m_kobj);
1045 out:
1046	return error;
1047}
1048
1049/*
1050 * This flushes out the inodes,dquots and the superblock, unmounts the
1051 * log and makes sure that incore structures are freed.
1052 */
1053void
1054xfs_unmountfs(
1055	struct xfs_mount	*mp)
1056{
1057	__uint64_t		resblks;
1058	int			error;
1059
1060	cancel_delayed_work_sync(&mp->m_eofblocks_work);
1061	cancel_delayed_work_sync(&mp->m_cowblocks_work);
 
 
 
 
 
 
 
1062
 
1063	xfs_fs_unreserve_ag_blocks(mp);
1064	xfs_qm_unmount_quotas(mp);
1065	xfs_rtunmount_inodes(mp);
1066	IRELE(mp->m_rootip);
1067
1068	/*
1069	 * We can potentially deadlock here if we have an inode cluster
1070	 * that has been freed has its buffer still pinned in memory because
1071	 * the transaction is still sitting in a iclog. The stale inodes
1072	 * on that buffer will have their flush locks held until the
1073	 * transaction hits the disk and the callbacks run. the inode
1074	 * flush takes the flush lock unconditionally and with nothing to
1075	 * push out the iclog we will never get that unlocked. hence we
1076	 * need to force the log first.
1077	 */
1078	xfs_log_force(mp, XFS_LOG_SYNC);
1079
1080	/*
1081	 * We now need to tell the world we are unmounting. This will allow
1082	 * us to detect that the filesystem is going away and we should error
1083	 * out anything that we have been retrying in the background. This will
1084	 * prevent neverending retries in AIL pushing from hanging the unmount.
1085	 */
1086	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1087
1088	/*
1089	 * Flush all pending changes from the AIL.
1090	 */
1091	xfs_ail_push_all_sync(mp->m_ail);
1092
1093	/*
1094	 * And reclaim all inodes.  At this point there should be no dirty
1095	 * inodes and none should be pinned or locked, but use synchronous
1096	 * reclaim just to be sure. We can stop background inode reclaim
1097	 * here as well if it is still running.
1098	 */
1099	cancel_delayed_work_sync(&mp->m_reclaim_work);
1100	xfs_reclaim_inodes(mp, SYNC_WAIT);
1101
1102	xfs_qm_unmount(mp);
1103
1104	/*
1105	 * Unreserve any blocks we have so that when we unmount we don't account
1106	 * the reserved free space as used. This is really only necessary for
1107	 * lazy superblock counting because it trusts the incore superblock
1108	 * counters to be absolutely correct on clean unmount.
1109	 *
1110	 * We don't bother correcting this elsewhere for lazy superblock
1111	 * counting because on mount of an unclean filesystem we reconstruct the
1112	 * correct counter value and this is irrelevant.
1113	 *
1114	 * For non-lazy counter filesystems, this doesn't matter at all because
1115	 * we only every apply deltas to the superblock and hence the incore
1116	 * value does not matter....
1117	 */
1118	resblks = 0;
1119	error = xfs_reserve_blocks(mp, &resblks, NULL);
1120	if (error)
1121		xfs_warn(mp, "Unable to free reserved block pool. "
1122				"Freespace may not be correct on next mount.");
1123
1124	error = xfs_log_sbcount(mp);
1125	if (error)
1126		xfs_warn(mp, "Unable to update superblock counters. "
1127				"Freespace may not be correct on next mount.");
1128
1129
1130	xfs_log_unmount(mp);
1131	xfs_da_unmount(mp);
1132	xfs_uuid_unmount(mp);
1133
1134#if defined(DEBUG)
1135	xfs_errortag_clearall(mp, 0);
1136#endif
 
1137	xfs_free_perag(mp);
1138
 
1139	xfs_error_sysfs_del(mp);
 
1140	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1141	xfs_sysfs_del(&mp->m_kobj);
1142}
1143
1144/*
1145 * Determine whether modifications can proceed. The caller specifies the minimum
1146 * freeze level for which modifications should not be allowed. This allows
1147 * certain operations to proceed while the freeze sequence is in progress, if
1148 * necessary.
1149 */
1150bool
1151xfs_fs_writable(
1152	struct xfs_mount	*mp,
1153	int			level)
1154{
1155	ASSERT(level > SB_UNFROZEN);
1156	if ((mp->m_super->s_writers.frozen >= level) ||
1157	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1158		return false;
1159
1160	return true;
1161}
1162
1163/*
1164 * xfs_log_sbcount
1165 *
1166 * Sync the superblock counters to disk.
1167 *
1168 * Note this code can be called during the process of freezing, so we use the
1169 * transaction allocator that does not block when the transaction subsystem is
1170 * in its frozen state.
1171 */
1172int
1173xfs_log_sbcount(xfs_mount_t *mp)
1174{
1175	/* allow this to proceed during the freeze sequence... */
1176	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1177		return 0;
1178
1179	/*
1180	 * we don't need to do this if we are updating the superblock
1181	 * counters on every modification.
1182	 */
1183	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1184		return 0;
1185
1186	return xfs_sync_sb(mp, true);
1187}
1188
1189/*
1190 * Deltas for the inode count are +/-64, hence we use a large batch size
1191 * of 128 so we don't need to take the counter lock on every update.
1192 */
1193#define XFS_ICOUNT_BATCH	128
1194int
1195xfs_mod_icount(
1196	struct xfs_mount	*mp,
1197	int64_t			delta)
1198{
1199	__percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1200	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1201		ASSERT(0);
1202		percpu_counter_add(&mp->m_icount, -delta);
1203		return -EINVAL;
1204	}
1205	return 0;
1206}
1207
1208int
1209xfs_mod_ifree(
1210	struct xfs_mount	*mp,
1211	int64_t			delta)
1212{
1213	percpu_counter_add(&mp->m_ifree, delta);
1214	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1215		ASSERT(0);
1216		percpu_counter_add(&mp->m_ifree, -delta);
1217		return -EINVAL;
1218	}
1219	return 0;
1220}
1221
1222/*
1223 * Deltas for the block count can vary from 1 to very large, but lock contention
1224 * only occurs on frequent small block count updates such as in the delayed
1225 * allocation path for buffered writes (page a time updates). Hence we set
1226 * a large batch count (1024) to minimise global counter updates except when
1227 * we get near to ENOSPC and we have to be very accurate with our updates.
1228 */
1229#define XFS_FDBLOCKS_BATCH	1024
1230int
1231xfs_mod_fdblocks(
1232	struct xfs_mount	*mp,
 
1233	int64_t			delta,
1234	bool			rsvd)
1235{
1236	int64_t			lcounter;
1237	long long		res_used;
 
1238	s32			batch;
 
 
 
 
 
 
1239
1240	if (delta > 0) {
1241		/*
1242		 * If the reserve pool is depleted, put blocks back into it
1243		 * first. Most of the time the pool is full.
1244		 */
1245		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1246			percpu_counter_add(&mp->m_fdblocks, delta);
 
1247			return 0;
1248		}
1249
1250		spin_lock(&mp->m_sb_lock);
1251		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1252
1253		if (res_used > delta) {
1254			mp->m_resblks_avail += delta;
1255		} else {
1256			delta -= res_used;
1257			mp->m_resblks_avail = mp->m_resblks;
1258			percpu_counter_add(&mp->m_fdblocks, delta);
1259		}
1260		spin_unlock(&mp->m_sb_lock);
1261		return 0;
1262	}
1263
1264	/*
1265	 * Taking blocks away, need to be more accurate the closer we
1266	 * are to zero.
1267	 *
1268	 * If the counter has a value of less than 2 * max batch size,
1269	 * then make everything serialise as we are real close to
1270	 * ENOSPC.
1271	 */
1272	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1273				     XFS_FDBLOCKS_BATCH) < 0)
1274		batch = 1;
1275	else
1276		batch = XFS_FDBLOCKS_BATCH;
1277
1278	__percpu_counter_add(&mp->m_fdblocks, delta, batch);
1279	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
 
 
 
 
 
 
 
 
 
 
 
 
 
1280				     XFS_FDBLOCKS_BATCH) >= 0) {
1281		/* we had space! */
1282		return 0;
1283	}
1284
1285	/*
1286	 * lock up the sb for dipping into reserves before releasing the space
1287	 * that took us to ENOSPC.
1288	 */
1289	spin_lock(&mp->m_sb_lock);
1290	percpu_counter_add(&mp->m_fdblocks, -delta);
1291	if (!rsvd)
1292		goto fdblocks_enospc;
1293
1294	lcounter = (long long)mp->m_resblks_avail + delta;
1295	if (lcounter >= 0) {
1296		mp->m_resblks_avail = lcounter;
1297		spin_unlock(&mp->m_sb_lock);
1298		return 0;
1299	}
1300	printk_once(KERN_WARNING
1301		"Filesystem \"%s\": reserve blocks depleted! "
1302		"Consider increasing reserve pool size.",
1303		mp->m_fsname);
1304fdblocks_enospc:
1305	spin_unlock(&mp->m_sb_lock);
1306	return -ENOSPC;
1307}
1308
1309int
1310xfs_mod_frextents(
1311	struct xfs_mount	*mp,
1312	int64_t			delta)
1313{
1314	int64_t			lcounter;
1315	int			ret = 0;
1316
1317	spin_lock(&mp->m_sb_lock);
1318	lcounter = mp->m_sb.sb_frextents + delta;
1319	if (lcounter < 0)
1320		ret = -ENOSPC;
1321	else
1322		mp->m_sb.sb_frextents = lcounter;
1323	spin_unlock(&mp->m_sb_lock);
1324	return ret;
1325}
1326
1327/*
1328 * xfs_getsb() is called to obtain the buffer for the superblock.
1329 * The buffer is returned locked and read in from disk.
1330 * The buffer should be released with a call to xfs_brelse().
1331 *
1332 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1333 * the superblock buffer if it can be locked without sleeping.
1334 * If it can't then we'll return NULL.
1335 */
1336struct xfs_buf *
1337xfs_getsb(
1338	struct xfs_mount	*mp,
1339	int			flags)
1340{
1341	struct xfs_buf		*bp = mp->m_sb_bp;
1342
1343	if (!xfs_buf_trylock(bp)) {
1344		if (flags & XBF_TRYLOCK)
1345			return NULL;
1346		xfs_buf_lock(bp);
1347	}
1348
1349	xfs_buf_hold(bp);
1350	ASSERT(bp->b_flags & XBF_DONE);
1351	return bp;
1352}
1353
1354/*
1355 * Used to free the superblock along various error paths.
1356 */
1357void
1358xfs_freesb(
1359	struct xfs_mount	*mp)
1360{
1361	struct xfs_buf		*bp = mp->m_sb_bp;
1362
1363	xfs_buf_lock(bp);
1364	mp->m_sb_bp = NULL;
1365	xfs_buf_relse(bp);
1366}
1367
1368/*
1369 * If the underlying (data/log/rt) device is readonly, there are some
1370 * operations that cannot proceed.
1371 */
1372int
1373xfs_dev_is_read_only(
1374	struct xfs_mount	*mp,
1375	char			*message)
1376{
1377	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1378	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1379	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1380		xfs_notice(mp, "%s required on read-only device.", message);
1381		xfs_notice(mp, "write access unavailable, cannot proceed.");
1382		return -EROFS;
1383	}
1384	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1385}