1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6
   7#include "xfs.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_sb.h"
  13#include "xfs_mount.h"
  14#include "xfs_inode.h"
  15#include "xfs_btree.h"
  16#include "xfs_bmap.h"
  17#include "xfs_alloc.h"
  18#include "xfs_fsops.h"
  19#include "xfs_trans.h"
  20#include "xfs_buf_item.h"
  21#include "xfs_log.h"
  22#include "xfs_log_priv.h"
  23#include "xfs_dir2.h"
  24#include "xfs_extfree_item.h"
  25#include "xfs_mru_cache.h"
  26#include "xfs_inode_item.h"
  27#include "xfs_icache.h"
  28#include "xfs_trace.h"
  29#include "xfs_icreate_item.h"
  30#include "xfs_filestream.h"
  31#include "xfs_quota.h"
  32#include "xfs_sysfs.h"
  33#include "xfs_ondisk.h"
  34#include "xfs_rmap_item.h"
  35#include "xfs_refcount_item.h"
  36#include "xfs_bmap_item.h"
  37#include "xfs_reflink.h"
  38#include "xfs_pwork.h"
  39#include "xfs_ag.h"
  40#include "xfs_defer.h"
  41#include "xfs_attr_item.h"
  42#include "xfs_xattr.h"
  43#include "xfs_iunlink_item.h"
  44#include "xfs_dahash_test.h"
  45#include "xfs_rtbitmap.h"
  46#include "scrub/stats.h"
  47
  48#include <linux/magic.h>
  49#include <linux/fs_context.h>
  50#include <linux/fs_parser.h>
  51
  52static const struct super_operations xfs_super_operations;
 
  53
  54static struct dentry *xfs_debugfs;	/* top-level xfs debugfs dir */
  55static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
  56#ifdef DEBUG
  57static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
  58#endif
  59
  60enum xfs_dax_mode {
  61	XFS_DAX_INODE = 0,
  62	XFS_DAX_ALWAYS = 1,
  63	XFS_DAX_NEVER = 2,
  64};
  65
  66static void
  67xfs_mount_set_dax_mode(
  68	struct xfs_mount	*mp,
  69	enum xfs_dax_mode	mode)
  70{
  71	switch (mode) {
  72	case XFS_DAX_INODE:
  73		mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
  74		break;
  75	case XFS_DAX_ALWAYS:
  76		mp->m_features |= XFS_FEAT_DAX_ALWAYS;
  77		mp->m_features &= ~XFS_FEAT_DAX_NEVER;
  78		break;
  79	case XFS_DAX_NEVER:
  80		mp->m_features |= XFS_FEAT_DAX_NEVER;
  81		mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
  82		break;
  83	}
  84}
  85
  86static const struct constant_table dax_param_enums[] = {
  87	{"inode",	XFS_DAX_INODE },
  88	{"always",	XFS_DAX_ALWAYS },
  89	{"never",	XFS_DAX_NEVER },
  90	{}
  91};
  92
  93/*
  94 * Table driven mount option parser.
  95 */
  96enum {
  97	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
  98	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
  99	Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
 100	Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
 101	Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
 102	Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
 103	Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
 104	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
 105	Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
 106};
 107
 108static const struct fs_parameter_spec xfs_fs_parameters[] = {
 109	fsparam_u32("logbufs",		Opt_logbufs),
 110	fsparam_string("logbsize",	Opt_logbsize),
 111	fsparam_string("logdev",	Opt_logdev),
 112	fsparam_string("rtdev",		Opt_rtdev),
 113	fsparam_flag("wsync",		Opt_wsync),
 114	fsparam_flag("noalign",		Opt_noalign),
 115	fsparam_flag("swalloc",		Opt_swalloc),
 116	fsparam_u32("sunit",		Opt_sunit),
 117	fsparam_u32("swidth",		Opt_swidth),
 118	fsparam_flag("nouuid",		Opt_nouuid),
 119	fsparam_flag("grpid",		Opt_grpid),
 120	fsparam_flag("nogrpid",		Opt_nogrpid),
 121	fsparam_flag("bsdgroups",	Opt_bsdgroups),
 122	fsparam_flag("sysvgroups",	Opt_sysvgroups),
 123	fsparam_string("allocsize",	Opt_allocsize),
 124	fsparam_flag("norecovery",	Opt_norecovery),
 125	fsparam_flag("inode64",		Opt_inode64),
 126	fsparam_flag("inode32",		Opt_inode32),
 127	fsparam_flag("ikeep",		Opt_ikeep),
 128	fsparam_flag("noikeep",		Opt_noikeep),
 129	fsparam_flag("largeio",		Opt_largeio),
 130	fsparam_flag("nolargeio",	Opt_nolargeio),
 131	fsparam_flag("attr2",		Opt_attr2),
 132	fsparam_flag("noattr2",		Opt_noattr2),
 133	fsparam_flag("filestreams",	Opt_filestreams),
 134	fsparam_flag("quota",		Opt_quota),
 135	fsparam_flag("noquota",		Opt_noquota),
 136	fsparam_flag("usrquota",	Opt_usrquota),
 137	fsparam_flag("grpquota",	Opt_grpquota),
 138	fsparam_flag("prjquota",	Opt_prjquota),
 139	fsparam_flag("uquota",		Opt_uquota),
 140	fsparam_flag("gquota",		Opt_gquota),
 141	fsparam_flag("pquota",		Opt_pquota),
 142	fsparam_flag("uqnoenforce",	Opt_uqnoenforce),
 143	fsparam_flag("gqnoenforce",	Opt_gqnoenforce),
 144	fsparam_flag("pqnoenforce",	Opt_pqnoenforce),
 145	fsparam_flag("qnoenforce",	Opt_qnoenforce),
 146	fsparam_flag("discard",		Opt_discard),
 147	fsparam_flag("nodiscard",	Opt_nodiscard),
 148	fsparam_flag("dax",		Opt_dax),
 149	fsparam_enum("dax",		Opt_dax_enum, dax_param_enums),
 150	{}
 
 
 151};
 152
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 153struct proc_xfs_info {
 154	uint64_t	flag;
 155	char		*str;
 156};
 157
 158static int
 159xfs_fs_show_options(
 160	struct seq_file		*m,
 161	struct dentry		*root)
 162{
 163	static struct proc_xfs_info xfs_info_set[] = {
 164		/* the few simple ones we can get from the mount struct */
 165		{ XFS_FEAT_IKEEP,		",ikeep" },
 166		{ XFS_FEAT_WSYNC,		",wsync" },
 167		{ XFS_FEAT_NOALIGN,		",noalign" },
 168		{ XFS_FEAT_SWALLOC,		",swalloc" },
 169		{ XFS_FEAT_NOUUID,		",nouuid" },
 170		{ XFS_FEAT_NORECOVERY,		",norecovery" },
 171		{ XFS_FEAT_ATTR2,		",attr2" },
 172		{ XFS_FEAT_FILESTREAMS,		",filestreams" },
 173		{ XFS_FEAT_GRPID,		",grpid" },
 174		{ XFS_FEAT_DISCARD,		",discard" },
 175		{ XFS_FEAT_LARGE_IOSIZE,	",largeio" },
 176		{ XFS_FEAT_DAX_ALWAYS,		",dax=always" },
 177		{ XFS_FEAT_DAX_NEVER,		",dax=never" },
 
 
 
 
 
 178		{ 0, NULL }
 179	};
 180	struct xfs_mount	*mp = XFS_M(root->d_sb);
 181	struct proc_xfs_info	*xfs_infop;
 182
 183	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
 184		if (mp->m_features & xfs_infop->flag)
 
 
 
 
 185			seq_puts(m, xfs_infop->str);
 186	}
 187
 188	seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
 189
 190	if (xfs_has_allocsize(mp))
 191		seq_printf(m, ",allocsize=%dk",
 192			   (1 << mp->m_allocsize_log) >> 10);
 193
 194	if (mp->m_logbufs > 0)
 195		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
 196	if (mp->m_logbsize > 0)
 197		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
 198
 199	if (mp->m_logname)
 200		seq_show_option(m, "logdev", mp->m_logname);
 201	if (mp->m_rtname)
 202		seq_show_option(m, "rtdev", mp->m_rtname);
 203
 204	if (mp->m_dalign > 0)
 205		seq_printf(m, ",sunit=%d",
 206				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
 207	if (mp->m_swidth > 0)
 208		seq_printf(m, ",swidth=%d",
 209				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
 210
 211	if (mp->m_qflags & XFS_UQUOTA_ENFD)
 212		seq_puts(m, ",usrquota");
 213	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
 214		seq_puts(m, ",uqnoenforce");
 215
 216	if (mp->m_qflags & XFS_PQUOTA_ENFD)
 217		seq_puts(m, ",prjquota");
 218	else if (mp->m_qflags & XFS_PQUOTA_ACCT)
 219		seq_puts(m, ",pqnoenforce");
 220
 221	if (mp->m_qflags & XFS_GQUOTA_ENFD)
 222		seq_puts(m, ",grpquota");
 223	else if (mp->m_qflags & XFS_GQUOTA_ACCT)
 224		seq_puts(m, ",gqnoenforce");
 
 
 
 225
 226	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
 227		seq_puts(m, ",noquota");
 228
 229	return 0;
 230}
 231
 232static bool
 233xfs_set_inode_alloc_perag(
 234	struct xfs_perag	*pag,
 235	xfs_ino_t		ino,
 236	xfs_agnumber_t		max_metadata)
 237{
 238	if (!xfs_is_inode32(pag->pag_mount)) {
 239		set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 240		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 241		return false;
 242	}
 243
 244	if (ino > XFS_MAXINUMBER_32) {
 245		clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 246		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 247		return false;
 248	}
 249
 250	set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
 251	if (pag->pag_agno < max_metadata)
 252		set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 253	else
 254		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
 255	return true;
 256}
 257
 258/*
 259 * Set parameters for inode allocation heuristics, taking into account
 260 * filesystem size and inode32/inode64 mount options; i.e. specifically
 261 * whether or not XFS_FEAT_SMALL_INUMS is set.
 262 *
 263 * Inode allocation patterns are altered only if inode32 is requested
 264 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
 265 * If altered, XFS_OPSTATE_INODE32 is set as well.
 266 *
 267 * An agcount independent of that in the mount structure is provided
 268 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
 269 * to the potentially higher ag count.
 270 *
 271 * Returns the maximum AG index which may contain inodes.
 272 */
 273xfs_agnumber_t
 274xfs_set_inode_alloc(
 275	struct xfs_mount *mp,
 276	xfs_agnumber_t	agcount)
 277{
 278	xfs_agnumber_t	index;
 279	xfs_agnumber_t	maxagi = 0;
 280	xfs_sb_t	*sbp = &mp->m_sb;
 281	xfs_agnumber_t	max_metadata;
 282	xfs_agino_t	agino;
 283	xfs_ino_t	ino;
 284
 285	/*
 286	 * Calculate how much should be reserved for inodes to meet
 287	 * the max inode percentage.  Used only for inode32.
 288	 */
 289	if (M_IGEO(mp)->maxicount) {
 290		uint64_t	icount;
 291
 292		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
 293		do_div(icount, 100);
 294		icount += sbp->sb_agblocks - 1;
 295		do_div(icount, sbp->sb_agblocks);
 296		max_metadata = icount;
 297	} else {
 298		max_metadata = agcount;
 299	}
 300
 301	/* Get the last possible inode in the filesystem */
 302	agino =	XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
 303	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
 304
 305	/*
 306	 * If user asked for no more than 32-bit inodes, and the fs is
 307	 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
 308	 * the allocator to accommodate the request.
 309	 */
 310	if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
 311		set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 312	else
 313		clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
 314
 315	for (index = 0; index < agcount; index++) {
 316		struct xfs_perag	*pag;
 317
 318		ino = XFS_AGINO_TO_INO(mp, index, agino);
 319
 320		pag = xfs_perag_get(mp, index);
 321		if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
 322			maxagi++;
 323		xfs_perag_put(pag);
 324	}
 325
 326	return xfs_is_inode32(mp) ? maxagi : agcount;
 327}
 328
 329static int
 330xfs_setup_dax_always(
 331	struct xfs_mount	*mp)
 332{
 333	if (!mp->m_ddev_targp->bt_daxdev &&
 334	    (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
 335		xfs_alert(mp,
 336			"DAX unsupported by block device. Turning off DAX.");
 337		goto disable_dax;
 338	}
 339
 340	if (mp->m_super->s_blocksize != PAGE_SIZE) {
 341		xfs_alert(mp,
 342			"DAX not supported for blocksize. Turning off DAX.");
 343		goto disable_dax;
 344	}
 
 
 
 
 
 
 
 
 
 
 
 345
 346	if (xfs_has_reflink(mp) &&
 347	    bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
 348		xfs_alert(mp,
 349			"DAX and reflink cannot work with multi-partitions!");
 350		return -EINVAL;
 351	}
 352
 353	return 0;
 354
 355disable_dax:
 356	xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
 357	return 0;
 358}
 359
 360STATIC int
 361xfs_blkdev_get(
 362	xfs_mount_t		*mp,
 363	const char		*name,
 364	struct bdev_handle	**handlep)
 365{
 366	int			error = 0;
 367
 368	*handlep = bdev_open_by_path(name,
 369		BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES,
 370		mp->m_super, &fs_holder_ops);
 371	if (IS_ERR(*handlep)) {
 372		error = PTR_ERR(*handlep);
 373		*handlep = NULL;
 374		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
 375	}
 376
 377	return error;
 378}
 379
 380STATIC void
 381xfs_shutdown_devices(
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 382	struct xfs_mount	*mp)
 383{
 384	/*
 385	 * Udev is triggered whenever anyone closes a block device or unmounts
 386	 * a file systemm on a block device.
 387	 * The default udev rules invoke blkid to read the fs super and create
 388	 * symlinks to the bdev under /dev/disk.  For this, it uses buffered
 389	 * reads through the page cache.
 390	 *
 391	 * xfs_db also uses buffered reads to examine metadata.  There is no
 392	 * coordination between xfs_db and udev, which means that they can run
 393	 * concurrently.  Note there is no coordination between the kernel and
 394	 * blkid either.
 395	 *
 396	 * On a system with 64k pages, the page cache can cache the superblock
 397	 * and the root inode (and hence the root directory) with the same 64k
 398	 * page.  If udev spawns blkid after the mkfs and the system is busy
 399	 * enough that it is still running when xfs_db starts up, they'll both
 400	 * read from the same page in the pagecache.
 401	 *
 402	 * The unmount writes updated inode metadata to disk directly.  The XFS
 403	 * buffer cache does not use the bdev pagecache, so it needs to
 404	 * invalidate that pagecache on unmount.  If the above scenario occurs,
 405	 * the pagecache no longer reflects what's on disk, xfs_db reads the
 406	 * stale metadata, and fails to find /a.  Most of the time this succeeds
 407	 * because closing a bdev invalidates the page cache, but when processes
 408	 * race, everyone loses.
 409	 */
 410	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 411		blkdev_issue_flush(mp->m_logdev_targp->bt_bdev);
 412		invalidate_bdev(mp->m_logdev_targp->bt_bdev);
 
 
 
 
 413	}
 414	if (mp->m_rtdev_targp) {
 415		blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
 416		invalidate_bdev(mp->m_rtdev_targp->bt_bdev);
 
 
 
 
 417	}
 418	blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
 419	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
 420}
 421
 422/*
 423 * The file system configurations are:
 424 *	(1) device (partition) with data and internal log
 425 *	(2) logical volume with data and log subvolumes.
 426 *	(3) logical volume with data, log, and realtime subvolumes.
 427 *
 428 * We only have to handle opening the log and realtime volumes here if
 429 * they are present.  The data subvolume has already been opened by
 430 * get_sb_bdev() and is stored in sb->s_bdev.
 431 */
 432STATIC int
 433xfs_open_devices(
 434	struct xfs_mount	*mp)
 435{
 436	struct super_block	*sb = mp->m_super;
 437	struct block_device	*ddev = sb->s_bdev;
 438	struct bdev_handle	*logdev_handle = NULL, *rtdev_handle = NULL;
 
 439	int			error;
 440
 441	/*
 442	 * Open real time and log devices - order is important.
 443	 */
 444	if (mp->m_logname) {
 445		error = xfs_blkdev_get(mp, mp->m_logname, &logdev_handle);
 446		if (error)
 447			return error;
 
 448	}
 449
 450	if (mp->m_rtname) {
 451		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev_handle);
 452		if (error)
 453			goto out_close_logdev;
 454
 455		if (rtdev_handle->bdev == ddev ||
 456		    (logdev_handle &&
 457		     rtdev_handle->bdev == logdev_handle->bdev)) {
 458			xfs_warn(mp,
 459	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
 460			error = -EINVAL;
 461			goto out_close_rtdev;
 462		}
 
 463	}
 464
 465	/*
 466	 * Setup xfs_mount buffer target pointers
 467	 */
 468	error = -ENOMEM;
 469	mp->m_ddev_targp = xfs_alloc_buftarg(mp, sb->s_bdev_handle);
 470	if (!mp->m_ddev_targp)
 471		goto out_close_rtdev;
 472
 473	if (rtdev_handle) {
 474		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev_handle);
 475		if (!mp->m_rtdev_targp)
 476			goto out_free_ddev_targ;
 477	}
 478
 479	if (logdev_handle && logdev_handle->bdev != ddev) {
 480		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev_handle);
 481		if (!mp->m_logdev_targp)
 482			goto out_free_rtdev_targ;
 483	} else {
 484		mp->m_logdev_targp = mp->m_ddev_targp;
 485		/* Handle won't be used, drop it */
 486		if (logdev_handle)
 487			bdev_release(logdev_handle);
 488	}
 489
 490	return 0;
 491
 492 out_free_rtdev_targ:
 493	if (mp->m_rtdev_targp)
 494		xfs_free_buftarg(mp->m_rtdev_targp);
 495 out_free_ddev_targ:
 496	xfs_free_buftarg(mp->m_ddev_targp);
 497 out_close_rtdev:
 498	 if (rtdev_handle)
 499		bdev_release(rtdev_handle);
 500 out_close_logdev:
 501	if (logdev_handle)
 502		bdev_release(logdev_handle);
 
 
 
 
 503	return error;
 504}
 505
 506/*
 507 * Setup xfs_mount buffer target pointers based on superblock
 508 */
 509STATIC int
 510xfs_setup_devices(
 511	struct xfs_mount	*mp)
 512{
 513	int			error;
 514
 515	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
 516	if (error)
 517		return error;
 518
 519	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
 520		unsigned int	log_sector_size = BBSIZE;
 521
 522		if (xfs_has_sector(mp))
 523			log_sector_size = mp->m_sb.sb_logsectsize;
 524		error = xfs_setsize_buftarg(mp->m_logdev_targp,
 525					    log_sector_size);
 526		if (error)
 527			return error;
 528	}
 529	if (mp->m_rtdev_targp) {
 530		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
 531					    mp->m_sb.sb_sectsize);
 532		if (error)
 533			return error;
 534	}
 535
 536	return 0;
 537}
 538
 539STATIC int
 540xfs_init_mount_workqueues(
 541	struct xfs_mount	*mp)
 542{
 543	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
 544			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 545			1, mp->m_super->s_id);
 546	if (!mp->m_buf_workqueue)
 547		goto out;
 548
 549	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
 550			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 551			0, mp->m_super->s_id);
 552	if (!mp->m_unwritten_workqueue)
 553		goto out_destroy_buf;
 554
 
 
 
 
 
 
 555	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
 556			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 557			0, mp->m_super->s_id);
 558	if (!mp->m_reclaim_workqueue)
 559		goto out_destroy_unwritten;
 560
 561	mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
 562			XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 563			0, mp->m_super->s_id);
 564	if (!mp->m_blockgc_wq)
 565		goto out_destroy_reclaim;
 566
 567	mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
 568			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 569			1, mp->m_super->s_id);
 570	if (!mp->m_inodegc_wq)
 571		goto out_destroy_blockgc;
 572
 573	mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
 574			XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
 575	if (!mp->m_sync_workqueue)
 576		goto out_destroy_inodegc;
 577
 578	return 0;
 579
 580out_destroy_inodegc:
 581	destroy_workqueue(mp->m_inodegc_wq);
 582out_destroy_blockgc:
 583	destroy_workqueue(mp->m_blockgc_wq);
 584out_destroy_reclaim:
 585	destroy_workqueue(mp->m_reclaim_workqueue);
 
 
 586out_destroy_unwritten:
 587	destroy_workqueue(mp->m_unwritten_workqueue);
 588out_destroy_buf:
 589	destroy_workqueue(mp->m_buf_workqueue);
 590out:
 591	return -ENOMEM;
 592}
 593
 594STATIC void
 595xfs_destroy_mount_workqueues(
 596	struct xfs_mount	*mp)
 597{
 598	destroy_workqueue(mp->m_sync_workqueue);
 599	destroy_workqueue(mp->m_blockgc_wq);
 600	destroy_workqueue(mp->m_inodegc_wq);
 601	destroy_workqueue(mp->m_reclaim_workqueue);
 
 602	destroy_workqueue(mp->m_unwritten_workqueue);
 603	destroy_workqueue(mp->m_buf_workqueue);
 604}
 605
 606static void
 607xfs_flush_inodes_worker(
 608	struct work_struct	*work)
 609{
 610	struct xfs_mount	*mp = container_of(work, struct xfs_mount,
 611						   m_flush_inodes_work);
 612	struct super_block	*sb = mp->m_super;
 613
 614	if (down_read_trylock(&sb->s_umount)) {
 615		sync_inodes_sb(sb);
 616		up_read(&sb->s_umount);
 617	}
 618}
 619
 620/*
 621 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
 622 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
 623 * for IO to complete so that we effectively throttle multiple callers to the
 624 * rate at which IO is completing.
 625 */
 626void
 627xfs_flush_inodes(
 628	struct xfs_mount	*mp)
 629{
 630	/*
 631	 * If flush_work() returns true then that means we waited for a flush
 632	 * which was already in progress.  Don't bother running another scan.
 633	 */
 634	if (flush_work(&mp->m_flush_inodes_work))
 635		return;
 636
 637	queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
 638	flush_work(&mp->m_flush_inodes_work);
 
 
 639}
 640
 641/* Catch misguided souls that try to use this interface on XFS */
 642STATIC struct inode *
 643xfs_fs_alloc_inode(
 644	struct super_block	*sb)
 645{
 646	BUG();
 647	return NULL;
 648}
 649
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 650/*
 651 * Now that the generic code is guaranteed not to be accessing
 652 * the linux inode, we can inactivate and reclaim the inode.
 653 */
 654STATIC void
 655xfs_fs_destroy_inode(
 656	struct inode		*inode)
 657{
 658	struct xfs_inode	*ip = XFS_I(inode);
 659
 660	trace_xfs_destroy_inode(ip);
 661
 662	ASSERT(!rwsem_is_locked(&inode->i_rwsem));
 663	XFS_STATS_INC(ip->i_mount, vn_rele);
 664	XFS_STATS_INC(ip->i_mount, vn_remove);
 665	xfs_inode_mark_reclaimable(ip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 666}
 667
 668static void
 669xfs_fs_dirty_inode(
 670	struct inode			*inode,
 671	int				flags)
 672{
 673	struct xfs_inode		*ip = XFS_I(inode);
 674	struct xfs_mount		*mp = ip->i_mount;
 675	struct xfs_trans		*tp;
 676
 677	if (!(inode->i_sb->s_flags & SB_LAZYTIME))
 678		return;
 679
 680	/*
 681	 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
 682	 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
 683	 * in flags possibly together with I_DIRTY_SYNC.
 684	 */
 685	if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
 686		return;
 687
 688	if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
 689		return;
 690	xfs_ilock(ip, XFS_ILOCK_EXCL);
 691	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 692	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 693	xfs_trans_commit(tp);
 694}
 695
 696/*
 697 * Slab object creation initialisation for the XFS inode.
 698 * This covers only the idempotent fields in the XFS inode;
 699 * all other fields need to be initialised on allocation
 700 * from the slab. This avoids the need to repeatedly initialise
 701 * fields in the xfs inode that left in the initialise state
 702 * when freeing the inode.
 703 */
 704STATIC void
 705xfs_fs_inode_init_once(
 706	void			*inode)
 707{
 708	struct xfs_inode	*ip = inode;
 709
 710	memset(ip, 0, sizeof(struct xfs_inode));
 711
 712	/* vfs inode */
 713	inode_init_once(VFS_I(ip));
 714
 715	/* xfs inode */
 716	atomic_set(&ip->i_pincount, 0);
 717	spin_lock_init(&ip->i_flags_lock);
 718
 
 
 719	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
 720		     "xfsino", ip->i_ino);
 721}
 722
 723/*
 724 * We do an unlocked check for XFS_IDONTCACHE here because we are already
 725 * serialised against cache hits here via the inode->i_lock and igrab() in
 726 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
 727 * racing with us, and it avoids needing to grab a spinlock here for every inode
 728 * we drop the final reference on.
 729 */
 730STATIC int
 731xfs_fs_drop_inode(
 732	struct inode		*inode)
 733{
 734	struct xfs_inode	*ip = XFS_I(inode);
 735
 736	/*
 737	 * If this unlinked inode is in the middle of recovery, don't
 738	 * drop the inode just yet; log recovery will take care of
 739	 * that.  See the comment for this inode flag.
 740	 */
 741	if (ip->i_flags & XFS_IRECOVERY) {
 742		ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
 743		return 0;
 744	}
 745
 746	return generic_drop_inode(inode);
 747}
 748
 749static void
 750xfs_mount_free(
 751	struct xfs_mount	*mp)
 752{
 753	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
 754		xfs_free_buftarg(mp->m_logdev_targp);
 755	if (mp->m_rtdev_targp)
 756		xfs_free_buftarg(mp->m_rtdev_targp);
 757	if (mp->m_ddev_targp)
 758		xfs_free_buftarg(mp->m_ddev_targp);
 759
 760	debugfs_remove(mp->m_debugfs);
 761	kfree(mp->m_rtname);
 762	kfree(mp->m_logname);
 763	kmem_free(mp);
 764}
 765
 766STATIC int
 767xfs_fs_sync_fs(
 768	struct super_block	*sb,
 769	int			wait)
 770{
 771	struct xfs_mount	*mp = XFS_M(sb);
 772	int			error;
 773
 774	trace_xfs_fs_sync_fs(mp, __return_address);
 775
 776	/*
 777	 * Doing anything during the async pass would be counterproductive.
 778	 */
 779	if (!wait)
 780		return 0;
 781
 782	error = xfs_log_force(mp, XFS_LOG_SYNC);
 783	if (error)
 784		return error;
 785
 786	if (laptop_mode) {
 787		/*
 788		 * The disk must be active because we're syncing.
 789		 * We schedule log work now (now that the disk is
 790		 * active) instead of later (when it might not be).
 791		 */
 792		flush_delayed_work(&mp->m_log->l_work);
 793	}
 794
 795	/*
 796	 * If we are called with page faults frozen out, it means we are about
 797	 * to freeze the transaction subsystem. Take the opportunity to shut
 798	 * down inodegc because once SB_FREEZE_FS is set it's too late to
 799	 * prevent inactivation races with freeze. The fs doesn't get called
 800	 * again by the freezing process until after SB_FREEZE_FS has been set,
 801	 * so it's now or never.  Same logic applies to speculative allocation
 802	 * garbage collection.
 803	 *
 804	 * We don't care if this is a normal syncfs call that does this or
 805	 * freeze that does this - we can run this multiple times without issue
 806	 * and we won't race with a restart because a restart can only occur
 807	 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
 808	 */
 809	if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
 810		xfs_inodegc_stop(mp);
 811		xfs_blockgc_stop(mp);
 812	}
 813
 814	return 0;
 815}
 816
 817STATIC int
 818xfs_fs_statfs(
 819	struct dentry		*dentry,
 820	struct kstatfs		*statp)
 821{
 822	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
 823	xfs_sb_t		*sbp = &mp->m_sb;
 824	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
 825	uint64_t		fakeinos, id;
 826	uint64_t		icount;
 827	uint64_t		ifree;
 828	uint64_t		fdblocks;
 829	xfs_extlen_t		lsize;
 830	int64_t			ffree;
 831
 832	/*
 833	 * Expedite background inodegc but don't wait. We do not want to block
 834	 * here waiting hours for a billion extent file to be truncated.
 835	 */
 836	xfs_inodegc_push(mp);
 837
 838	statp->f_type = XFS_SUPER_MAGIC;
 839	statp->f_namelen = MAXNAMELEN - 1;
 840
 841	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
 842	statp->f_fsid = u64_to_fsid(id);
 
 843
 844	icount = percpu_counter_sum(&mp->m_icount);
 845	ifree = percpu_counter_sum(&mp->m_ifree);
 846	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
 847
 848	spin_lock(&mp->m_sb_lock);
 849	statp->f_bsize = sbp->sb_blocksize;
 850	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
 851	statp->f_blocks = sbp->sb_dblocks - lsize;
 852	spin_unlock(&mp->m_sb_lock);
 853
 854	/* make sure statp->f_bfree does not underflow */
 855	statp->f_bfree = max_t(int64_t, 0,
 856				fdblocks - xfs_fdblocks_unavailable(mp));
 857	statp->f_bavail = statp->f_bfree;
 858
 859	fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
 860	statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
 861	if (M_IGEO(mp)->maxicount)
 862		statp->f_files = min_t(typeof(statp->f_files),
 863					statp->f_files,
 864					M_IGEO(mp)->maxicount);
 865
 866	/* If sb_icount overshot maxicount, report actual allocation */
 867	statp->f_files = max_t(typeof(statp->f_files),
 868					statp->f_files,
 869					sbp->sb_icount);
 870
 871	/* make sure statp->f_ffree does not underflow */
 872	ffree = statp->f_files - (icount - ifree);
 873	statp->f_ffree = max_t(int64_t, ffree, 0);
 874
 875
 876	if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
 877	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
 878			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
 879		xfs_qm_statvfs(ip, statp);
 880
 881	if (XFS_IS_REALTIME_MOUNT(mp) &&
 882	    (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
 883		s64	freertx;
 884
 885		statp->f_blocks = sbp->sb_rblocks;
 886		freertx = percpu_counter_sum_positive(&mp->m_frextents);
 887		statp->f_bavail = statp->f_bfree = xfs_rtx_to_rtb(mp, freertx);
 888	}
 889
 890	return 0;
 891}
 892
 893STATIC void
 894xfs_save_resvblks(struct xfs_mount *mp)
 895{
 
 
 896	mp->m_resblks_save = mp->m_resblks;
 897	xfs_reserve_blocks(mp, 0);
 898}
 899
 900STATIC void
 901xfs_restore_resvblks(struct xfs_mount *mp)
 902{
 903	uint64_t resblks;
 904
 905	if (mp->m_resblks_save) {
 906		resblks = mp->m_resblks_save;
 907		mp->m_resblks_save = 0;
 908	} else
 909		resblks = xfs_default_resblks(mp);
 910
 911	xfs_reserve_blocks(mp, resblks);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 912}
 913
 914/*
 915 * Second stage of a freeze. The data is already frozen so we only
 916 * need to take care of the metadata. Once that's done sync the superblock
 917 * to the log to dirty it in case of a crash while frozen. This ensures that we
 918 * will recover the unlinked inode lists on the next mount.
 919 */
 920STATIC int
 921xfs_fs_freeze(
 922	struct super_block	*sb)
 923{
 924	struct xfs_mount	*mp = XFS_M(sb);
 925	unsigned int		flags;
 926	int			ret;
 927
 928	/*
 929	 * The filesystem is now frozen far enough that memory reclaim
 930	 * cannot safely operate on the filesystem. Hence we need to
 931	 * set a GFP_NOFS context here to avoid recursion deadlocks.
 932	 */
 933	flags = memalloc_nofs_save();
 934	xfs_save_resvblks(mp);
 935	ret = xfs_log_quiesce(mp);
 936	memalloc_nofs_restore(flags);
 937
 938	/*
 939	 * For read-write filesystems, we need to restart the inodegc on error
 940	 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
 941	 * going to be run to restart it now.  We are at SB_FREEZE_FS level
 942	 * here, so we can restart safely without racing with a stop in
 943	 * xfs_fs_sync_fs().
 944	 */
 945	if (ret && !xfs_is_readonly(mp)) {
 946		xfs_blockgc_start(mp);
 947		xfs_inodegc_start(mp);
 948	}
 949
 950	return ret;
 951}
 952
 953STATIC int
 954xfs_fs_unfreeze(
 955	struct super_block	*sb)
 956{
 957	struct xfs_mount	*mp = XFS_M(sb);
 958
 959	xfs_restore_resvblks(mp);
 960	xfs_log_work_queue(mp);
 
 
 
 961
 962	/*
 963	 * Don't reactivate the inodegc worker on a readonly filesystem because
 964	 * inodes are sent directly to reclaim.  Don't reactivate the blockgc
 965	 * worker because there are no speculative preallocations on a readonly
 966	 * filesystem.
 967	 */
 968	if (!xfs_is_readonly(mp)) {
 969		xfs_blockgc_start(mp);
 970		xfs_inodegc_start(mp);
 971	}
 972
 973	return 0;
 974}
 975
 976/*
 977 * This function fills in xfs_mount_t fields based on mount args.
 978 * Note: the superblock _has_ now been read in.
 979 */
 980STATIC int
 981xfs_finish_flags(
 982	struct xfs_mount	*mp)
 983{
 
 
 984	/* Fail a mount where the logbuf is smaller than the log stripe */
 985	if (xfs_has_logv2(mp)) {
 986		if (mp->m_logbsize <= 0 &&
 987		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
 988			mp->m_logbsize = mp->m_sb.sb_logsunit;
 989		} else if (mp->m_logbsize > 0 &&
 990			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
 991			xfs_warn(mp,
 992		"logbuf size must be greater than or equal to log stripe size");
 993			return -EINVAL;
 994		}
 995	} else {
 996		/* Fail a mount if the logbuf is larger than 32K */
 997		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
 998			xfs_warn(mp,
 999		"logbuf size for version 1 logs must be 16K or 32K");
1000			return -EINVAL;
1001		}
1002	}
1003
1004	/*
1005	 * V5 filesystems always use attr2 format for attributes.
1006	 */
1007	if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
 
1008		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1009			     "attr2 is always enabled for V5 filesystems.");
1010		return -EINVAL;
1011	}
1012
1013	/*
 
 
 
 
 
 
 
 
1014	 * prohibit r/w mounts of read-only filesystems
1015	 */
1016	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1017		xfs_warn(mp,
1018			"cannot mount a read-only filesystem as read-write");
1019		return -EROFS;
1020	}
1021
1022	if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1023	    (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1024	    !xfs_has_pquotino(mp)) {
1025		xfs_warn(mp,
1026		  "Super block does not support project and group quota together");
1027		return -EINVAL;
1028	}
1029
1030	return 0;
1031}
1032
1033static int
1034xfs_init_percpu_counters(
1035	struct xfs_mount	*mp)
1036{
1037	int		error;
1038
1039	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1040	if (error)
1041		return -ENOMEM;
1042
1043	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1044	if (error)
1045		goto free_icount;
1046
1047	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1048	if (error)
1049		goto free_ifree;
1050
1051	error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1052	if (error)
1053		goto free_fdblocks;
1054
1055	error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1056	if (error)
1057		goto free_delalloc;
1058
1059	return 0;
1060
1061free_delalloc:
1062	percpu_counter_destroy(&mp->m_delalloc_blks);
1063free_fdblocks:
1064	percpu_counter_destroy(&mp->m_fdblocks);
1065free_ifree:
1066	percpu_counter_destroy(&mp->m_ifree);
1067free_icount:
1068	percpu_counter_destroy(&mp->m_icount);
1069	return -ENOMEM;
1070}
1071
1072void
1073xfs_reinit_percpu_counters(
1074	struct xfs_mount	*mp)
1075{
1076	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1077	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1078	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1079	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1080}
1081
1082static void
1083xfs_destroy_percpu_counters(
1084	struct xfs_mount	*mp)
1085{
1086	percpu_counter_destroy(&mp->m_icount);
1087	percpu_counter_destroy(&mp->m_ifree);
1088	percpu_counter_destroy(&mp->m_fdblocks);
1089	ASSERT(xfs_is_shutdown(mp) ||
1090	       percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1091	percpu_counter_destroy(&mp->m_delalloc_blks);
1092	percpu_counter_destroy(&mp->m_frextents);
1093}
1094
1095static int
1096xfs_inodegc_init_percpu(
1097	struct xfs_mount	*mp)
1098{
1099	struct xfs_inodegc	*gc;
1100	int			cpu;
1101
1102	mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1103	if (!mp->m_inodegc)
1104		return -ENOMEM;
1105
1106	for_each_possible_cpu(cpu) {
1107		gc = per_cpu_ptr(mp->m_inodegc, cpu);
1108		gc->cpu = cpu;
1109		gc->mp = mp;
1110		init_llist_head(&gc->list);
1111		gc->items = 0;
1112		gc->error = 0;
1113		INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1114	}
1115	return 0;
1116}
1117
1118static void
1119xfs_inodegc_free_percpu(
1120	struct xfs_mount	*mp)
1121{
1122	if (!mp->m_inodegc)
1123		return;
1124	free_percpu(mp->m_inodegc);
1125}
1126
1127static void
1128xfs_fs_put_super(
1129	struct super_block	*sb)
1130{
1131	struct xfs_mount	*mp = XFS_M(sb);
1132
1133	xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1134	xfs_filestream_unmount(mp);
1135	xfs_unmountfs(mp);
1136
1137	xfs_freesb(mp);
1138	xchk_mount_stats_free(mp);
1139	free_percpu(mp->m_stats.xs_stats);
1140	xfs_inodegc_free_percpu(mp);
1141	xfs_destroy_percpu_counters(mp);
1142	xfs_destroy_mount_workqueues(mp);
1143	xfs_shutdown_devices(mp);
1144}
1145
1146static long
1147xfs_fs_nr_cached_objects(
1148	struct super_block	*sb,
1149	struct shrink_control	*sc)
1150{
1151	/* Paranoia: catch incorrect calls during mount setup or teardown */
1152	if (WARN_ON_ONCE(!sb->s_fs_info))
1153		return 0;
1154	return xfs_reclaim_inodes_count(XFS_M(sb));
1155}
1156
1157static long
1158xfs_fs_free_cached_objects(
1159	struct super_block	*sb,
1160	struct shrink_control	*sc)
1161{
1162	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1163}
1164
1165static void
1166xfs_fs_shutdown(
1167	struct super_block	*sb)
1168{
1169	xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED);
1170}
1171
1172static const struct super_operations xfs_super_operations = {
1173	.alloc_inode		= xfs_fs_alloc_inode,
1174	.destroy_inode		= xfs_fs_destroy_inode,
1175	.dirty_inode		= xfs_fs_dirty_inode,
1176	.drop_inode		= xfs_fs_drop_inode,
1177	.put_super		= xfs_fs_put_super,
1178	.sync_fs		= xfs_fs_sync_fs,
1179	.freeze_fs		= xfs_fs_freeze,
1180	.unfreeze_fs		= xfs_fs_unfreeze,
1181	.statfs			= xfs_fs_statfs,
1182	.show_options		= xfs_fs_show_options,
1183	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1184	.free_cached_objects	= xfs_fs_free_cached_objects,
1185	.shutdown		= xfs_fs_shutdown,
1186};
1187
1188static int
1189suffix_kstrtoint(
1190	const char	*s,
1191	unsigned int	base,
1192	int		*res)
1193{
1194	int		last, shift_left_factor = 0, _res;
1195	char		*value;
1196	int		ret = 0;
1197
1198	value = kstrdup(s, GFP_KERNEL);
1199	if (!value)
1200		return -ENOMEM;
1201
1202	last = strlen(value) - 1;
1203	if (value[last] == 'K' || value[last] == 'k') {
1204		shift_left_factor = 10;
1205		value[last] = '\0';
1206	}
1207	if (value[last] == 'M' || value[last] == 'm') {
1208		shift_left_factor = 20;
1209		value[last] = '\0';
1210	}
1211	if (value[last] == 'G' || value[last] == 'g') {
1212		shift_left_factor = 30;
1213		value[last] = '\0';
1214	}
1215
1216	if (kstrtoint(value, base, &_res))
1217		ret = -EINVAL;
1218	kfree(value);
1219	*res = _res << shift_left_factor;
1220	return ret;
1221}
1222
1223static inline void
1224xfs_fs_warn_deprecated(
1225	struct fs_context	*fc,
1226	struct fs_parameter	*param,
1227	uint64_t		flag,
1228	bool			value)
1229{
1230	/* Don't print the warning if reconfiguring and current mount point
1231	 * already had the flag set
1232	 */
1233	if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1234            !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1235		return;
1236	xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1237}
1238
1239/*
1240 * Set mount state from a mount option.
1241 *
1242 * NOTE: mp->m_super is NULL here!
1243 */
1244static int
1245xfs_fs_parse_param(
1246	struct fs_context	*fc,
1247	struct fs_parameter	*param)
1248{
1249	struct xfs_mount	*parsing_mp = fc->s_fs_info;
1250	struct fs_parse_result	result;
1251	int			size = 0;
1252	int			opt;
1253
1254	opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1255	if (opt < 0)
1256		return opt;
1257
1258	switch (opt) {
1259	case Opt_logbufs:
1260		parsing_mp->m_logbufs = result.uint_32;
1261		return 0;
1262	case Opt_logbsize:
1263		if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1264			return -EINVAL;
1265		return 0;
1266	case Opt_logdev:
1267		kfree(parsing_mp->m_logname);
1268		parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1269		if (!parsing_mp->m_logname)
1270			return -ENOMEM;
1271		return 0;
1272	case Opt_rtdev:
1273		kfree(parsing_mp->m_rtname);
1274		parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1275		if (!parsing_mp->m_rtname)
1276			return -ENOMEM;
1277		return 0;
1278	case Opt_allocsize:
1279		if (suffix_kstrtoint(param->string, 10, &size))
1280			return -EINVAL;
1281		parsing_mp->m_allocsize_log = ffs(size) - 1;
1282		parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1283		return 0;
1284	case Opt_grpid:
1285	case Opt_bsdgroups:
1286		parsing_mp->m_features |= XFS_FEAT_GRPID;
1287		return 0;
1288	case Opt_nogrpid:
1289	case Opt_sysvgroups:
1290		parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1291		return 0;
1292	case Opt_wsync:
1293		parsing_mp->m_features |= XFS_FEAT_WSYNC;
1294		return 0;
1295	case Opt_norecovery:
1296		parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1297		return 0;
1298	case Opt_noalign:
1299		parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1300		return 0;
1301	case Opt_swalloc:
1302		parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1303		return 0;
1304	case Opt_sunit:
1305		parsing_mp->m_dalign = result.uint_32;
1306		return 0;
1307	case Opt_swidth:
1308		parsing_mp->m_swidth = result.uint_32;
1309		return 0;
1310	case Opt_inode32:
1311		parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1312		return 0;
1313	case Opt_inode64:
1314		parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1315		return 0;
1316	case Opt_nouuid:
1317		parsing_mp->m_features |= XFS_FEAT_NOUUID;
1318		return 0;
1319	case Opt_largeio:
1320		parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1321		return 0;
1322	case Opt_nolargeio:
1323		parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1324		return 0;
1325	case Opt_filestreams:
1326		parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1327		return 0;
1328	case Opt_noquota:
1329		parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1330		parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1331		return 0;
1332	case Opt_quota:
1333	case Opt_uquota:
1334	case Opt_usrquota:
1335		parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1336		return 0;
1337	case Opt_qnoenforce:
1338	case Opt_uqnoenforce:
1339		parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1340		parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1341		return 0;
1342	case Opt_pquota:
1343	case Opt_prjquota:
1344		parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1345		return 0;
1346	case Opt_pqnoenforce:
1347		parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1348		parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1349		return 0;
1350	case Opt_gquota:
1351	case Opt_grpquota:
1352		parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1353		return 0;
1354	case Opt_gqnoenforce:
1355		parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1356		parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1357		return 0;
1358	case Opt_discard:
1359		parsing_mp->m_features |= XFS_FEAT_DISCARD;
1360		return 0;
1361	case Opt_nodiscard:
1362		parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1363		return 0;
1364#ifdef CONFIG_FS_DAX
1365	case Opt_dax:
1366		xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1367		return 0;
1368	case Opt_dax_enum:
1369		xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1370		return 0;
1371#endif
1372	/* Following mount options will be removed in September 2025 */
1373	case Opt_ikeep:
1374		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1375		parsing_mp->m_features |= XFS_FEAT_IKEEP;
1376		return 0;
1377	case Opt_noikeep:
1378		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1379		parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1380		return 0;
1381	case Opt_attr2:
1382		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1383		parsing_mp->m_features |= XFS_FEAT_ATTR2;
1384		return 0;
1385	case Opt_noattr2:
1386		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1387		parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1388		return 0;
1389	default:
1390		xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1391		return -EINVAL;
1392	}
1393
1394	return 0;
1395}
1396
1397static int
1398xfs_fs_validate_params(
1399	struct xfs_mount	*mp)
1400{
1401	/* No recovery flag requires a read-only mount */
1402	if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1403		xfs_warn(mp, "no-recovery mounts must be read-only.");
1404		return -EINVAL;
1405	}
1406
 
 
 
 
 
 
 
 
 
 
 
1407	/*
1408	 * We have not read the superblock at this point, so only the attr2
1409	 * mount option can set the attr2 feature by this stage.
 
 
1410	 */
1411	if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1412		xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1413		return -EINVAL;
1414	}
1415
1416
1417	if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1418		xfs_warn(mp,
1419	"sunit and swidth options incompatible with the noalign option");
1420		return -EINVAL;
1421	}
1422
1423	if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1424		xfs_warn(mp, "quota support not available in this kernel.");
1425		return -EINVAL;
1426	}
1427
1428	if ((mp->m_dalign && !mp->m_swidth) ||
1429	    (!mp->m_dalign && mp->m_swidth)) {
1430		xfs_warn(mp, "sunit and swidth must be specified together");
1431		return -EINVAL;
1432	}
1433
1434	if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1435		xfs_warn(mp,
1436	"stripe width (%d) must be a multiple of the stripe unit (%d)",
1437			mp->m_swidth, mp->m_dalign);
1438		return -EINVAL;
1439	}
1440
1441	if (mp->m_logbufs != -1 &&
1442	    mp->m_logbufs != 0 &&
1443	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1444	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1445		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1446			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1447		return -EINVAL;
1448	}
1449
1450	if (mp->m_logbsize != -1 &&
1451	    mp->m_logbsize !=  0 &&
1452	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1453	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1454	     !is_power_of_2(mp->m_logbsize))) {
1455		xfs_warn(mp,
1456			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1457			mp->m_logbsize);
1458		return -EINVAL;
1459	}
1460
1461	if (xfs_has_allocsize(mp) &&
1462	    (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1463	     mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1464		xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1465			mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1466		return -EINVAL;
1467	}
1468
1469	return 0;
1470}
1471
1472struct dentry *
1473xfs_debugfs_mkdir(
1474	const char	*name,
1475	struct dentry	*parent)
1476{
1477	struct dentry	*child;
1478
1479	/* Apparently we're expected to ignore error returns?? */
1480	child = debugfs_create_dir(name, parent);
1481	if (IS_ERR(child))
1482		return NULL;
1483
1484	return child;
1485}
1486
1487static int
1488xfs_fs_fill_super(
1489	struct super_block	*sb,
1490	struct fs_context	*fc)
 
1491{
1492	struct xfs_mount	*mp = sb->s_fs_info;
1493	struct inode		*root;
1494	int			flags = 0, error;
1495
1496	mp->m_super = sb;
1497
1498	/*
1499	 * Copy VFS mount flags from the context now that all parameter parsing
1500	 * is guaranteed to have been completed by either the old mount API or
1501	 * the newer fsopen/fsconfig API.
1502	 */
1503	if (fc->sb_flags & SB_RDONLY)
1504		set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1505	if (fc->sb_flags & SB_DIRSYNC)
1506		mp->m_features |= XFS_FEAT_DIRSYNC;
1507	if (fc->sb_flags & SB_SYNCHRONOUS)
1508		mp->m_features |= XFS_FEAT_WSYNC;
1509
1510	error = xfs_fs_validate_params(mp);
1511	if (error)
1512		return error;
1513
1514	sb_min_blocksize(sb, BBSIZE);
1515	sb->s_xattr = xfs_xattr_handlers;
1516	sb->s_export_op = &xfs_export_operations;
1517#ifdef CONFIG_XFS_QUOTA
1518	sb->s_qcop = &xfs_quotactl_operations;
1519	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1520#endif
1521	sb->s_op = &xfs_super_operations;
1522
1523	/*
1524	 * Delay mount work if the debug hook is set. This is debug
1525	 * instrumention to coordinate simulation of xfs mount failures with
1526	 * VFS superblock operations
1527	 */
1528	if (xfs_globals.mount_delay) {
1529		xfs_notice(mp, "Delaying mount for %d seconds.",
1530			xfs_globals.mount_delay);
1531		msleep(xfs_globals.mount_delay * 1000);
1532	}
1533
1534	if (fc->sb_flags & SB_SILENT)
1535		flags |= XFS_MFSI_QUIET;
1536
1537	error = xfs_open_devices(mp);
1538	if (error)
1539		return error;
1540
1541	if (xfs_debugfs) {
1542		mp->m_debugfs = xfs_debugfs_mkdir(mp->m_super->s_id,
1543						  xfs_debugfs);
1544	} else {
1545		mp->m_debugfs = NULL;
1546	}
1547
1548	error = xfs_init_mount_workqueues(mp);
1549	if (error)
1550		goto out_shutdown_devices;
1551
1552	error = xfs_init_percpu_counters(mp);
1553	if (error)
1554		goto out_destroy_workqueues;
1555
1556	error = xfs_inodegc_init_percpu(mp);
1557	if (error)
1558		goto out_destroy_counters;
1559
1560	/* Allocate stats memory before we do operations that might use it */
1561	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1562	if (!mp->m_stats.xs_stats) {
1563		error = -ENOMEM;
1564		goto out_destroy_inodegc;
1565	}
1566
1567	error = xchk_mount_stats_alloc(mp);
1568	if (error)
1569		goto out_free_stats;
1570
1571	error = xfs_readsb(mp, flags);
1572	if (error)
1573		goto out_free_scrub_stats;
1574
1575	error = xfs_finish_flags(mp);
1576	if (error)
1577		goto out_free_sb;
1578
1579	error = xfs_setup_devices(mp);
1580	if (error)
1581		goto out_free_sb;
1582
1583	/* V4 support is undergoing deprecation. */
1584	if (!xfs_has_crc(mp)) {
1585#ifdef CONFIG_XFS_SUPPORT_V4
1586		xfs_warn_once(mp,
1587	"Deprecated V4 format (crc=0) will not be supported after September 2030.");
1588#else
1589		xfs_warn(mp,
1590	"Deprecated V4 format (crc=0) not supported by kernel.");
1591		error = -EINVAL;
1592		goto out_free_sb;
1593#endif
1594	}
1595
1596	/* ASCII case insensitivity is undergoing deprecation. */
1597	if (xfs_has_asciici(mp)) {
1598#ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1599		xfs_warn_once(mp,
1600	"Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1601#else
1602		xfs_warn(mp,
1603	"Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1604		error = -EINVAL;
1605		goto out_free_sb;
1606#endif
1607	}
1608
1609	/* Filesystem claims it needs repair, so refuse the mount. */
1610	if (xfs_has_needsrepair(mp)) {
1611		xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1612		error = -EFSCORRUPTED;
1613		goto out_free_sb;
1614	}
1615
1616	/*
1617	 * Don't touch the filesystem if a user tool thinks it owns the primary
1618	 * superblock.  mkfs doesn't clear the flag from secondary supers, so
1619	 * we don't check them at all.
1620	 */
1621	if (mp->m_sb.sb_inprogress) {
1622		xfs_warn(mp, "Offline file system operation in progress!");
1623		error = -EFSCORRUPTED;
1624		goto out_free_sb;
1625	}
1626
1627	/*
1628	 * Until this is fixed only page-sized or smaller data blocks work.
1629	 */
1630	if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1631		xfs_warn(mp,
1632		"File system with blocksize %d bytes. "
1633		"Only pagesize (%ld) or less will currently work.",
1634				mp->m_sb.sb_blocksize, PAGE_SIZE);
1635		error = -ENOSYS;
1636		goto out_free_sb;
1637	}
1638
1639	/* Ensure this filesystem fits in the page cache limits */
1640	if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1641	    xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1642		xfs_warn(mp,
1643		"file system too large to be mounted on this system.");
1644		error = -EFBIG;
1645		goto out_free_sb;
1646	}
1647
1648	/*
1649	 * XFS block mappings use 54 bits to store the logical block offset.
1650	 * This should suffice to handle the maximum file size that the VFS
1651	 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1652	 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1653	 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1654	 * to check this assertion.
1655	 *
1656	 * Avoid integer overflow by comparing the maximum bmbt offset to the
1657	 * maximum pagecache offset in units of fs blocks.
1658	 */
1659	if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1660		xfs_warn(mp,
1661"MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1662			 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1663			 XFS_MAX_FILEOFF);
1664		error = -EINVAL;
1665		goto out_free_sb;
1666	}
1667
1668	error = xfs_filestream_mount(mp);
1669	if (error)
1670		goto out_free_sb;
1671
1672	/*
1673	 * we must configure the block size in the superblock before we run the
1674	 * full mount process as the mount process can lookup and cache inodes.
1675	 */
1676	sb->s_magic = XFS_SUPER_MAGIC;
1677	sb->s_blocksize = mp->m_sb.sb_blocksize;
1678	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1679	sb->s_maxbytes = MAX_LFS_FILESIZE;
1680	sb->s_max_links = XFS_MAXLINK;
1681	sb->s_time_gran = 1;
1682	if (xfs_has_bigtime(mp)) {
1683		sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1684		sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1685	} else {
1686		sb->s_time_min = XFS_LEGACY_TIME_MIN;
1687		sb->s_time_max = XFS_LEGACY_TIME_MAX;
1688	}
1689	trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1690	sb->s_iflags |= SB_I_CGROUPWB;
1691
1692	set_posix_acl_flag(sb);
1693
1694	/* version 5 superblocks support inode version counters. */
1695	if (xfs_has_crc(mp))
1696		sb->s_flags |= SB_I_VERSION;
1697
1698	if (xfs_has_dax_always(mp)) {
1699		error = xfs_setup_dax_always(mp);
1700		if (error)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1701			goto out_filestream_unmount;
 
1702	}
1703
1704	if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1705		xfs_warn(mp,
1706	"mounting with \"discard\" option, but the device does not support discard");
1707		mp->m_features &= ~XFS_FEAT_DISCARD;
 
 
 
 
1708	}
1709
1710	if (xfs_has_reflink(mp)) {
1711		if (mp->m_sb.sb_rblocks) {
1712			xfs_alert(mp,
1713	"reflink not compatible with realtime device!");
1714			error = -EINVAL;
1715			goto out_filestream_unmount;
1716		}
1717
1718		if (xfs_globals.always_cow) {
1719			xfs_info(mp, "using DEBUG-only always_cow mode.");
1720			mp->m_always_cow = true;
1721		}
1722	}
1723
1724	if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1725		xfs_alert(mp,
1726	"reverse mapping btree not compatible with realtime device!");
1727		error = -EINVAL;
1728		goto out_filestream_unmount;
1729	}
1730
1731	error = xfs_mountfs(mp);
1732	if (error)
1733		goto out_filestream_unmount;
1734
1735	root = igrab(VFS_I(mp->m_rootip));
1736	if (!root) {
1737		error = -ENOENT;
1738		goto out_unmount;
1739	}
1740	sb->s_root = d_make_root(root);
1741	if (!sb->s_root) {
1742		error = -ENOMEM;
1743		goto out_unmount;
1744	}
1745
1746	return 0;
1747
1748 out_filestream_unmount:
1749	xfs_filestream_unmount(mp);
1750 out_free_sb:
1751	xfs_freesb(mp);
1752 out_free_scrub_stats:
1753	xchk_mount_stats_free(mp);
1754 out_free_stats:
1755	free_percpu(mp->m_stats.xs_stats);
1756 out_destroy_inodegc:
1757	xfs_inodegc_free_percpu(mp);
1758 out_destroy_counters:
1759	xfs_destroy_percpu_counters(mp);
1760 out_destroy_workqueues:
1761	xfs_destroy_mount_workqueues(mp);
1762 out_shutdown_devices:
1763	xfs_shutdown_devices(mp);
 
 
 
 
 
1764	return error;
1765
1766 out_unmount:
1767	xfs_filestream_unmount(mp);
1768	xfs_unmountfs(mp);
1769	goto out_free_sb;
1770}
1771
1772static int
1773xfs_fs_get_tree(
1774	struct fs_context	*fc)
1775{
1776	return get_tree_bdev(fc, xfs_fs_fill_super);
1777}
1778
1779static int
1780xfs_remount_rw(
1781	struct xfs_mount	*mp)
1782{
1783	struct xfs_sb		*sbp = &mp->m_sb;
1784	int error;
1785
1786	if (xfs_has_norecovery(mp)) {
1787		xfs_warn(mp,
1788			"ro->rw transition prohibited on norecovery mount");
1789		return -EINVAL;
1790	}
1791
1792	if (xfs_sb_is_v5(sbp) &&
1793	    xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1794		xfs_warn(mp,
1795	"ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1796			(sbp->sb_features_ro_compat &
1797				XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1798		return -EINVAL;
1799	}
1800
1801	clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1802
1803	/*
1804	 * If this is the first remount to writeable state we might have some
1805	 * superblock changes to update.
1806	 */
1807	if (mp->m_update_sb) {
1808		error = xfs_sync_sb(mp, false);
1809		if (error) {
1810			xfs_warn(mp, "failed to write sb changes");
1811			return error;
1812		}
1813		mp->m_update_sb = false;
1814	}
1815
1816	/*
1817	 * Fill out the reserve pool if it is empty. Use the stashed value if
1818	 * it is non-zero, otherwise go with the default.
1819	 */
1820	xfs_restore_resvblks(mp);
1821	xfs_log_work_queue(mp);
1822	xfs_blockgc_start(mp);
1823
1824	/* Create the per-AG metadata reservation pool .*/
1825	error = xfs_fs_reserve_ag_blocks(mp);
1826	if (error && error != -ENOSPC)
1827		return error;
1828
1829	/* Re-enable the background inode inactivation worker. */
1830	xfs_inodegc_start(mp);
 
 
 
1831
1832	return 0;
 
 
1833}
1834
1835static int
1836xfs_remount_ro(
1837	struct xfs_mount	*mp)
 
 
 
1838{
1839	struct xfs_icwalk	icw = {
1840		.icw_flags	= XFS_ICWALK_FLAG_SYNC,
1841	};
1842	int			error;
1843
1844	/* Flush all the dirty data to disk. */
1845	error = sync_filesystem(mp->m_super);
1846	if (error)
1847		return error;
1848
1849	/*
1850	 * Cancel background eofb scanning so it cannot race with the final
1851	 * log force+buftarg wait and deadlock the remount.
1852	 */
1853	xfs_blockgc_stop(mp);
1854
1855	/*
1856	 * Clear out all remaining COW staging extents and speculative post-EOF
1857	 * preallocations so that we don't leave inodes requiring inactivation
1858	 * cleanups during reclaim on a read-only mount.  We must process every
1859	 * cached inode, so this requires a synchronous cache scan.
1860	 */
1861	error = xfs_blockgc_free_space(mp, &icw);
1862	if (error) {
1863		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1864		return error;
1865	}
1866
1867	/*
1868	 * Stop the inodegc background worker.  xfs_fs_reconfigure already
1869	 * flushed all pending inodegc work when it sync'd the filesystem.
1870	 * The VFS holds s_umount, so we know that inodes cannot enter
1871	 * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1872	 * we send inodes straight to reclaim, so no inodes will be queued.
1873	 */
1874	xfs_inodegc_stop(mp);
1875
1876	/* Free the per-AG metadata reservation pool. */
1877	error = xfs_fs_unreserve_ag_blocks(mp);
1878	if (error) {
1879		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1880		return error;
1881	}
1882
1883	/*
1884	 * Before we sync the metadata, we need to free up the reserve block
1885	 * pool so that the used block count in the superblock on disk is
1886	 * correct at the end of the remount. Stash the current* reserve pool
1887	 * size so that if we get remounted rw, we can return it to the same
1888	 * size.
1889	 */
1890	xfs_save_resvblks(mp);
1891
1892	xfs_log_clean(mp);
1893	set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1894
1895	return 0;
1896}
1897
1898/*
1899 * Logically we would return an error here to prevent users from believing
1900 * they might have changed mount options using remount which can't be changed.
1901 *
1902 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1903 * arguments in some cases so we can't blindly reject options, but have to
1904 * check for each specified option if it actually differs from the currently
1905 * set option and only reject it if that's the case.
1906 *
1907 * Until that is implemented we return success for every remount request, and
1908 * silently ignore all options that we can't actually change.
1909 */
1910static int
1911xfs_fs_reconfigure(
1912	struct fs_context *fc)
1913{
1914	struct xfs_mount	*mp = XFS_M(fc->root->d_sb);
1915	struct xfs_mount        *new_mp = fc->s_fs_info;
1916	int			flags = fc->sb_flags;
1917	int			error;
1918
1919	/* version 5 superblocks always support version counters. */
1920	if (xfs_has_crc(mp))
1921		fc->sb_flags |= SB_I_VERSION;
1922
1923	error = xfs_fs_validate_params(new_mp);
1924	if (error)
1925		return error;
1926
1927	/* inode32 -> inode64 */
1928	if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1929		mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1930		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1931	}
1932
1933	/* inode64 -> inode32 */
1934	if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1935		mp->m_features |= XFS_FEAT_SMALL_INUMS;
1936		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1937	}
1938
1939	/* ro -> rw */
1940	if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1941		error = xfs_remount_rw(mp);
1942		if (error)
1943			return error;
1944	}
1945
1946	/* rw -> ro */
1947	if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1948		error = xfs_remount_ro(mp);
1949		if (error)
1950			return error;
1951	}
1952
1953	return 0;
1954}
1955
1956static void
1957xfs_fs_free(
1958	struct fs_context	*fc)
 
1959{
1960	struct xfs_mount	*mp = fc->s_fs_info;
1961
1962	/*
1963	 * mp is stored in the fs_context when it is initialized.
1964	 * mp is transferred to the superblock on a successful mount,
1965	 * but if an error occurs before the transfer we have to free
1966	 * it here.
1967	 */
1968	if (mp)
1969		xfs_mount_free(mp);
1970}
1971
1972static const struct fs_context_operations xfs_context_ops = {
1973	.parse_param = xfs_fs_parse_param,
1974	.get_tree    = xfs_fs_get_tree,
1975	.reconfigure = xfs_fs_reconfigure,
1976	.free        = xfs_fs_free,
 
 
 
 
 
 
 
 
 
1977};
1978
1979/*
1980 * WARNING: do not initialise any parameters in this function that depend on
1981 * mount option parsing having already been performed as this can be called from
1982 * fsopen() before any parameters have been set.
1983 */
1984static int xfs_init_fs_context(
1985	struct fs_context	*fc)
1986{
1987	struct xfs_mount	*mp;
1988
1989	mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1990	if (!mp)
1991		return -ENOMEM;
1992
1993	spin_lock_init(&mp->m_sb_lock);
1994	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1995	spin_lock_init(&mp->m_perag_lock);
1996	mutex_init(&mp->m_growlock);
1997	INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1998	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1999	mp->m_kobj.kobject.kset = xfs_kset;
2000	/*
2001	 * We don't create the finobt per-ag space reservation until after log
2002	 * recovery, so we must set this to true so that an ifree transaction
2003	 * started during log recovery will not depend on space reservations
2004	 * for finobt expansion.
2005	 */
2006	mp->m_finobt_nores = true;
2007
2008	/*
2009	 * These can be overridden by the mount option parsing.
2010	 */
2011	mp->m_logbufs = -1;
2012	mp->m_logbsize = -1;
2013	mp->m_allocsize_log = 16; /* 64k */
2014
2015	fc->s_fs_info = mp;
2016	fc->ops = &xfs_context_ops;
2017
2018	return 0;
2019}
2020
2021static void
2022xfs_kill_sb(
2023	struct super_block		*sb)
2024{
2025	kill_block_super(sb);
2026	xfs_mount_free(XFS_M(sb));
2027}
2028
2029static struct file_system_type xfs_fs_type = {
2030	.owner			= THIS_MODULE,
2031	.name			= "xfs",
2032	.init_fs_context	= xfs_init_fs_context,
2033	.parameters		= xfs_fs_parameters,
2034	.kill_sb		= xfs_kill_sb,
2035	.fs_flags		= FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
2036};
2037MODULE_ALIAS_FS("xfs");
2038
2039STATIC int __init
2040xfs_init_caches(void)
2041{
2042	int		error;
2043
2044	xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2045					 SLAB_HWCACHE_ALIGN |
2046					 SLAB_RECLAIM_ACCOUNT |
2047					 SLAB_MEM_SPREAD,
2048					 NULL);
2049	if (!xfs_buf_cache)
2050		goto out;
2051
2052	xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
2053						sizeof(struct xlog_ticket),
2054						0, 0, NULL);
2055	if (!xfs_log_ticket_cache)
2056		goto out_destroy_buf_cache;
2057
2058	error = xfs_btree_init_cur_caches();
2059	if (error)
2060		goto out_destroy_log_ticket_cache;
2061
2062	error = xfs_defer_init_item_caches();
2063	if (error)
2064		goto out_destroy_btree_cur_cache;
2065
2066	xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2067					      sizeof(struct xfs_da_state),
2068					      0, 0, NULL);
2069	if (!xfs_da_state_cache)
2070		goto out_destroy_defer_item_cache;
2071
2072	xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2073					   sizeof(struct xfs_ifork),
2074					   0, 0, NULL);
2075	if (!xfs_ifork_cache)
2076		goto out_destroy_da_state_cache;
2077
2078	xfs_trans_cache = kmem_cache_create("xfs_trans",
2079					   sizeof(struct xfs_trans),
2080					   0, 0, NULL);
2081	if (!xfs_trans_cache)
2082		goto out_destroy_ifork_cache;
2083
2084
2085	/*
2086	 * The size of the cache-allocated buf log item is the maximum
2087	 * size possible under XFS.  This wastes a little bit of memory,
2088	 * but it is much faster.
2089	 */
2090	xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2091					      sizeof(struct xfs_buf_log_item),
2092					      0, 0, NULL);
2093	if (!xfs_buf_item_cache)
2094		goto out_destroy_trans_cache;
2095
2096	xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2097			xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2098			0, 0, NULL);
2099	if (!xfs_efd_cache)
2100		goto out_destroy_buf_item_cache;
2101
2102	xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2103			xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2104			0, 0, NULL);
2105	if (!xfs_efi_cache)
2106		goto out_destroy_efd_cache;
2107
2108	xfs_inode_cache = kmem_cache_create("xfs_inode",
2109					   sizeof(struct xfs_inode), 0,
2110					   (SLAB_HWCACHE_ALIGN |
2111					    SLAB_RECLAIM_ACCOUNT |
2112					    SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2113					   xfs_fs_inode_init_once);
2114	if (!xfs_inode_cache)
2115		goto out_destroy_efi_cache;
2116
2117	xfs_ili_cache = kmem_cache_create("xfs_ili",
2118					 sizeof(struct xfs_inode_log_item), 0,
2119					 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2120					 NULL);
2121	if (!xfs_ili_cache)
2122		goto out_destroy_inode_cache;
2123
2124	xfs_icreate_cache = kmem_cache_create("xfs_icr",
2125					     sizeof(struct xfs_icreate_item),
2126					     0, 0, NULL);
2127	if (!xfs_icreate_cache)
2128		goto out_destroy_ili_cache;
2129
2130	xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2131					 sizeof(struct xfs_rud_log_item),
2132					 0, 0, NULL);
2133	if (!xfs_rud_cache)
2134		goto out_destroy_icreate_cache;
2135
2136	xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2137			xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2138			0, 0, NULL);
2139	if (!xfs_rui_cache)
2140		goto out_destroy_rud_cache;
2141
2142	xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2143					 sizeof(struct xfs_cud_log_item),
2144					 0, 0, NULL);
2145	if (!xfs_cud_cache)
2146		goto out_destroy_rui_cache;
2147
2148	xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2149			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2150			0, 0, NULL);
2151	if (!xfs_cui_cache)
2152		goto out_destroy_cud_cache;
2153
2154	xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2155					 sizeof(struct xfs_bud_log_item),
2156					 0, 0, NULL);
2157	if (!xfs_bud_cache)
2158		goto out_destroy_cui_cache;
2159
2160	xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2161			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2162			0, 0, NULL);
2163	if (!xfs_bui_cache)
2164		goto out_destroy_bud_cache;
2165
2166	xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2167					    sizeof(struct xfs_attrd_log_item),
2168					    0, 0, NULL);
2169	if (!xfs_attrd_cache)
2170		goto out_destroy_bui_cache;
2171
2172	xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2173					    sizeof(struct xfs_attri_log_item),
2174					    0, 0, NULL);
2175	if (!xfs_attri_cache)
2176		goto out_destroy_attrd_cache;
2177
2178	xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2179					     sizeof(struct xfs_iunlink_item),
2180					     0, 0, NULL);
2181	if (!xfs_iunlink_cache)
2182		goto out_destroy_attri_cache;
2183
2184	return 0;
2185
2186 out_destroy_attri_cache:
2187	kmem_cache_destroy(xfs_attri_cache);
2188 out_destroy_attrd_cache:
2189	kmem_cache_destroy(xfs_attrd_cache);
2190 out_destroy_bui_cache:
2191	kmem_cache_destroy(xfs_bui_cache);
2192 out_destroy_bud_cache:
2193	kmem_cache_destroy(xfs_bud_cache);
2194 out_destroy_cui_cache:
2195	kmem_cache_destroy(xfs_cui_cache);
2196 out_destroy_cud_cache:
2197	kmem_cache_destroy(xfs_cud_cache);
2198 out_destroy_rui_cache:
2199	kmem_cache_destroy(xfs_rui_cache);
2200 out_destroy_rud_cache:
2201	kmem_cache_destroy(xfs_rud_cache);
2202 out_destroy_icreate_cache:
2203	kmem_cache_destroy(xfs_icreate_cache);
2204 out_destroy_ili_cache:
2205	kmem_cache_destroy(xfs_ili_cache);
2206 out_destroy_inode_cache:
2207	kmem_cache_destroy(xfs_inode_cache);
2208 out_destroy_efi_cache:
2209	kmem_cache_destroy(xfs_efi_cache);
2210 out_destroy_efd_cache:
2211	kmem_cache_destroy(xfs_efd_cache);
2212 out_destroy_buf_item_cache:
2213	kmem_cache_destroy(xfs_buf_item_cache);
2214 out_destroy_trans_cache:
2215	kmem_cache_destroy(xfs_trans_cache);
2216 out_destroy_ifork_cache:
2217	kmem_cache_destroy(xfs_ifork_cache);
2218 out_destroy_da_state_cache:
2219	kmem_cache_destroy(xfs_da_state_cache);
2220 out_destroy_defer_item_cache:
2221	xfs_defer_destroy_item_caches();
2222 out_destroy_btree_cur_cache:
2223	xfs_btree_destroy_cur_caches();
2224 out_destroy_log_ticket_cache:
2225	kmem_cache_destroy(xfs_log_ticket_cache);
2226 out_destroy_buf_cache:
2227	kmem_cache_destroy(xfs_buf_cache);
2228 out:
2229	return -ENOMEM;
2230}
2231
2232STATIC void
2233xfs_destroy_caches(void)
2234{
2235	/*
2236	 * Make sure all delayed rcu free are flushed before we
2237	 * destroy caches.
2238	 */
2239	rcu_barrier();
2240	kmem_cache_destroy(xfs_iunlink_cache);
2241	kmem_cache_destroy(xfs_attri_cache);
2242	kmem_cache_destroy(xfs_attrd_cache);
2243	kmem_cache_destroy(xfs_bui_cache);
2244	kmem_cache_destroy(xfs_bud_cache);
2245	kmem_cache_destroy(xfs_cui_cache);
2246	kmem_cache_destroy(xfs_cud_cache);
2247	kmem_cache_destroy(xfs_rui_cache);
2248	kmem_cache_destroy(xfs_rud_cache);
2249	kmem_cache_destroy(xfs_icreate_cache);
2250	kmem_cache_destroy(xfs_ili_cache);
2251	kmem_cache_destroy(xfs_inode_cache);
2252	kmem_cache_destroy(xfs_efi_cache);
2253	kmem_cache_destroy(xfs_efd_cache);
2254	kmem_cache_destroy(xfs_buf_item_cache);
2255	kmem_cache_destroy(xfs_trans_cache);
2256	kmem_cache_destroy(xfs_ifork_cache);
2257	kmem_cache_destroy(xfs_da_state_cache);
2258	xfs_defer_destroy_item_caches();
2259	xfs_btree_destroy_cur_caches();
2260	kmem_cache_destroy(xfs_log_ticket_cache);
2261	kmem_cache_destroy(xfs_buf_cache);
2262}
2263
2264STATIC int __init
2265xfs_init_workqueues(void)
2266{
2267	/*
2268	 * The allocation workqueue can be used in memory reclaim situations
2269	 * (writepage path), and parallelism is only limited by the number of
2270	 * AGs in all the filesystems mounted. Hence use the default large
2271	 * max_active value for this workqueue.
2272	 */
2273	xfs_alloc_wq = alloc_workqueue("xfsalloc",
2274			XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2275	if (!xfs_alloc_wq)
2276		return -ENOMEM;
2277
2278	xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2279			0);
2280	if (!xfs_discard_wq)
2281		goto out_free_alloc_wq;
2282
2283	return 0;
2284out_free_alloc_wq:
2285	destroy_workqueue(xfs_alloc_wq);
2286	return -ENOMEM;
2287}
2288
2289STATIC void
2290xfs_destroy_workqueues(void)
2291{
2292	destroy_workqueue(xfs_discard_wq);
2293	destroy_workqueue(xfs_alloc_wq);
2294}
2295
2296STATIC int __init
2297init_xfs_fs(void)
2298{
2299	int			error;
2300
2301	xfs_check_ondisk_structs();
2302
2303	error = xfs_dahash_test();
2304	if (error)
2305		return error;
2306
2307	printk(KERN_INFO XFS_VERSION_STRING " with "
2308			 XFS_BUILD_OPTIONS " enabled\n");
2309
2310	xfs_dir_startup();
2311
2312	error = xfs_init_caches();
2313	if (error)
2314		goto out;
2315
2316	error = xfs_init_workqueues();
2317	if (error)
2318		goto out_destroy_caches;
2319
2320	error = xfs_mru_cache_init();
2321	if (error)
2322		goto out_destroy_wq;
2323
2324	error = xfs_init_procfs();
2325	if (error)
2326		goto out_mru_cache_uninit;
2327
 
 
 
 
2328	error = xfs_sysctl_register();
2329	if (error)
2330		goto out_cleanup_procfs;
2331
2332	xfs_debugfs = xfs_debugfs_mkdir("xfs", NULL);
2333
2334	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2335	if (!xfs_kset) {
2336		error = -ENOMEM;
2337		goto out_debugfs_unregister;
2338	}
2339
2340	xfsstats.xs_kobj.kobject.kset = xfs_kset;
2341
2342	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2343	if (!xfsstats.xs_stats) {
2344		error = -ENOMEM;
2345		goto out_kset_unregister;
2346	}
2347
2348	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2349			       "stats");
2350	if (error)
2351		goto out_free_stats;
2352
2353	error = xchk_global_stats_setup(xfs_debugfs);
2354	if (error)
2355		goto out_remove_stats_kobj;
2356
2357#ifdef DEBUG
2358	xfs_dbg_kobj.kobject.kset = xfs_kset;
2359	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2360	if (error)
2361		goto out_remove_scrub_stats;
2362#endif
2363
2364	error = xfs_qm_init();
2365	if (error)
2366		goto out_remove_dbg_kobj;
2367
2368	error = register_filesystem(&xfs_fs_type);
2369	if (error)
2370		goto out_qm_exit;
2371	return 0;
2372
2373 out_qm_exit:
2374	xfs_qm_exit();
2375 out_remove_dbg_kobj:
2376#ifdef DEBUG
2377	xfs_sysfs_del(&xfs_dbg_kobj);
2378 out_remove_scrub_stats:
2379#endif
2380	xchk_global_stats_teardown();
2381 out_remove_stats_kobj:
 
2382	xfs_sysfs_del(&xfsstats.xs_kobj);
2383 out_free_stats:
2384	free_percpu(xfsstats.xs_stats);
2385 out_kset_unregister:
2386	kset_unregister(xfs_kset);
2387 out_debugfs_unregister:
2388	debugfs_remove(xfs_debugfs);
2389	xfs_sysctl_unregister();
2390 out_cleanup_procfs:
2391	xfs_cleanup_procfs();
 
 
2392 out_mru_cache_uninit:
2393	xfs_mru_cache_uninit();
2394 out_destroy_wq:
2395	xfs_destroy_workqueues();
2396 out_destroy_caches:
2397	xfs_destroy_caches();
2398 out:
2399	return error;
2400}
2401
2402STATIC void __exit
2403exit_xfs_fs(void)
2404{
2405	xfs_qm_exit();
2406	unregister_filesystem(&xfs_fs_type);
2407#ifdef DEBUG
2408	xfs_sysfs_del(&xfs_dbg_kobj);
2409#endif
2410	xchk_global_stats_teardown();
2411	xfs_sysfs_del(&xfsstats.xs_kobj);
2412	free_percpu(xfsstats.xs_stats);
2413	kset_unregister(xfs_kset);
2414	debugfs_remove(xfs_debugfs);
2415	xfs_sysctl_unregister();
2416	xfs_cleanup_procfs();
 
2417	xfs_mru_cache_uninit();
2418	xfs_destroy_workqueues();
2419	xfs_destroy_caches();
2420	xfs_uuid_table_free();
2421}
2422
2423module_init(init_xfs_fs);
2424module_exit(exit_xfs_fs);
2425
2426MODULE_AUTHOR("Silicon Graphics, Inc.");
2427MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2428MODULE_LICENSE("GPL");