xref: /linux/fs/xfs/xfs_super.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 "xfs_exchmaps_item.h"
47 #include "xfs_parent.h"
48 #include "xfs_rtalloc.h"
49 #include "scrub/stats.h"
50 #include "scrub/rcbag_btree.h"
51 
52 #include <linux/magic.h>
53 #include <linux/fs_context.h>
54 #include <linux/fs_parser.h>
55 
56 static const struct super_operations xfs_super_operations;
57 
58 static struct dentry *xfs_debugfs;	/* top-level xfs debugfs dir */
59 static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
60 #ifdef DEBUG
61 static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
62 #endif
63 
64 enum xfs_dax_mode {
65 	XFS_DAX_INODE = 0,
66 	XFS_DAX_ALWAYS = 1,
67 	XFS_DAX_NEVER = 2,
68 };
69 
70 /* Were quota mount options provided?  Must use the upper 16 bits of qflags. */
71 #define XFS_QFLAGS_MNTOPTS	(1U << 31)
72 
73 static void
74 xfs_mount_set_dax_mode(
75 	struct xfs_mount	*mp,
76 	enum xfs_dax_mode	mode)
77 {
78 	switch (mode) {
79 	case XFS_DAX_INODE:
80 		mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
81 		break;
82 	case XFS_DAX_ALWAYS:
83 		mp->m_features |= XFS_FEAT_DAX_ALWAYS;
84 		mp->m_features &= ~XFS_FEAT_DAX_NEVER;
85 		break;
86 	case XFS_DAX_NEVER:
87 		mp->m_features |= XFS_FEAT_DAX_NEVER;
88 		mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
89 		break;
90 	}
91 }
92 
93 static const struct constant_table dax_param_enums[] = {
94 	{"inode",	XFS_DAX_INODE },
95 	{"always",	XFS_DAX_ALWAYS },
96 	{"never",	XFS_DAX_NEVER },
97 	{}
98 };
99 
100 /*
101  * Table driven mount option parser.
102  */
103 enum {
104 	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
105 	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
106 	Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
107 	Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
108 	Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
109 	Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
110 	Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
111 	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
112 	Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
113 };
114 
115 static const struct fs_parameter_spec xfs_fs_parameters[] = {
116 	fsparam_u32("logbufs",		Opt_logbufs),
117 	fsparam_string("logbsize",	Opt_logbsize),
118 	fsparam_string("logdev",	Opt_logdev),
119 	fsparam_string("rtdev",		Opt_rtdev),
120 	fsparam_flag("wsync",		Opt_wsync),
121 	fsparam_flag("noalign",		Opt_noalign),
122 	fsparam_flag("swalloc",		Opt_swalloc),
123 	fsparam_u32("sunit",		Opt_sunit),
124 	fsparam_u32("swidth",		Opt_swidth),
125 	fsparam_flag("nouuid",		Opt_nouuid),
126 	fsparam_flag("grpid",		Opt_grpid),
127 	fsparam_flag("nogrpid",		Opt_nogrpid),
128 	fsparam_flag("bsdgroups",	Opt_bsdgroups),
129 	fsparam_flag("sysvgroups",	Opt_sysvgroups),
130 	fsparam_string("allocsize",	Opt_allocsize),
131 	fsparam_flag("norecovery",	Opt_norecovery),
132 	fsparam_flag("inode64",		Opt_inode64),
133 	fsparam_flag("inode32",		Opt_inode32),
134 	fsparam_flag("ikeep",		Opt_ikeep),
135 	fsparam_flag("noikeep",		Opt_noikeep),
136 	fsparam_flag("largeio",		Opt_largeio),
137 	fsparam_flag("nolargeio",	Opt_nolargeio),
138 	fsparam_flag("attr2",		Opt_attr2),
139 	fsparam_flag("noattr2",		Opt_noattr2),
140 	fsparam_flag("filestreams",	Opt_filestreams),
141 	fsparam_flag("quota",		Opt_quota),
142 	fsparam_flag("noquota",		Opt_noquota),
143 	fsparam_flag("usrquota",	Opt_usrquota),
144 	fsparam_flag("grpquota",	Opt_grpquota),
145 	fsparam_flag("prjquota",	Opt_prjquota),
146 	fsparam_flag("uquota",		Opt_uquota),
147 	fsparam_flag("gquota",		Opt_gquota),
148 	fsparam_flag("pquota",		Opt_pquota),
149 	fsparam_flag("uqnoenforce",	Opt_uqnoenforce),
150 	fsparam_flag("gqnoenforce",	Opt_gqnoenforce),
151 	fsparam_flag("pqnoenforce",	Opt_pqnoenforce),
152 	fsparam_flag("qnoenforce",	Opt_qnoenforce),
153 	fsparam_flag("discard",		Opt_discard),
154 	fsparam_flag("nodiscard",	Opt_nodiscard),
155 	fsparam_flag("dax",		Opt_dax),
156 	fsparam_enum("dax",		Opt_dax_enum, dax_param_enums),
157 	{}
158 };
159 
160 struct proc_xfs_info {
161 	uint64_t	flag;
162 	char		*str;
163 };
164 
165 static int
166 xfs_fs_show_options(
167 	struct seq_file		*m,
168 	struct dentry		*root)
169 {
170 	static struct proc_xfs_info xfs_info_set[] = {
171 		/* the few simple ones we can get from the mount struct */
172 		{ XFS_FEAT_IKEEP,		",ikeep" },
173 		{ XFS_FEAT_WSYNC,		",wsync" },
174 		{ XFS_FEAT_NOALIGN,		",noalign" },
175 		{ XFS_FEAT_SWALLOC,		",swalloc" },
176 		{ XFS_FEAT_NOUUID,		",nouuid" },
177 		{ XFS_FEAT_NORECOVERY,		",norecovery" },
178 		{ XFS_FEAT_ATTR2,		",attr2" },
179 		{ XFS_FEAT_FILESTREAMS,		",filestreams" },
180 		{ XFS_FEAT_GRPID,		",grpid" },
181 		{ XFS_FEAT_DISCARD,		",discard" },
182 		{ XFS_FEAT_LARGE_IOSIZE,	",largeio" },
183 		{ XFS_FEAT_DAX_ALWAYS,		",dax=always" },
184 		{ XFS_FEAT_DAX_NEVER,		",dax=never" },
185 		{ 0, NULL }
186 	};
187 	struct xfs_mount	*mp = XFS_M(root->d_sb);
188 	struct proc_xfs_info	*xfs_infop;
189 
190 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
191 		if (mp->m_features & xfs_infop->flag)
192 			seq_puts(m, xfs_infop->str);
193 	}
194 
195 	seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
196 
197 	if (xfs_has_allocsize(mp))
198 		seq_printf(m, ",allocsize=%dk",
199 			   (1 << mp->m_allocsize_log) >> 10);
200 
201 	if (mp->m_logbufs > 0)
202 		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
203 	if (mp->m_logbsize > 0)
204 		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
205 
206 	if (mp->m_logname)
207 		seq_show_option(m, "logdev", mp->m_logname);
208 	if (mp->m_rtname)
209 		seq_show_option(m, "rtdev", mp->m_rtname);
210 
211 	if (mp->m_dalign > 0)
212 		seq_printf(m, ",sunit=%d",
213 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
214 	if (mp->m_swidth > 0)
215 		seq_printf(m, ",swidth=%d",
216 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
217 
218 	if (mp->m_qflags & XFS_UQUOTA_ENFD)
219 		seq_puts(m, ",usrquota");
220 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
221 		seq_puts(m, ",uqnoenforce");
222 
223 	if (mp->m_qflags & XFS_PQUOTA_ENFD)
224 		seq_puts(m, ",prjquota");
225 	else if (mp->m_qflags & XFS_PQUOTA_ACCT)
226 		seq_puts(m, ",pqnoenforce");
227 
228 	if (mp->m_qflags & XFS_GQUOTA_ENFD)
229 		seq_puts(m, ",grpquota");
230 	else if (mp->m_qflags & XFS_GQUOTA_ACCT)
231 		seq_puts(m, ",gqnoenforce");
232 
233 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
234 		seq_puts(m, ",noquota");
235 
236 	return 0;
237 }
238 
239 static bool
240 xfs_set_inode_alloc_perag(
241 	struct xfs_perag	*pag,
242 	xfs_ino_t		ino,
243 	xfs_agnumber_t		max_metadata)
244 {
245 	if (!xfs_is_inode32(pag_mount(pag))) {
246 		set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
247 		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
248 		return false;
249 	}
250 
251 	if (ino > XFS_MAXINUMBER_32) {
252 		clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
253 		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
254 		return false;
255 	}
256 
257 	set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
258 	if (pag_agno(pag) < max_metadata)
259 		set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
260 	else
261 		clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
262 	return true;
263 }
264 
265 /*
266  * Set parameters for inode allocation heuristics, taking into account
267  * filesystem size and inode32/inode64 mount options; i.e. specifically
268  * whether or not XFS_FEAT_SMALL_INUMS is set.
269  *
270  * Inode allocation patterns are altered only if inode32 is requested
271  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
272  * If altered, XFS_OPSTATE_INODE32 is set as well.
273  *
274  * An agcount independent of that in the mount structure is provided
275  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
276  * to the potentially higher ag count.
277  *
278  * Returns the maximum AG index which may contain inodes.
279  */
280 xfs_agnumber_t
281 xfs_set_inode_alloc(
282 	struct xfs_mount *mp,
283 	xfs_agnumber_t	agcount)
284 {
285 	xfs_agnumber_t	index;
286 	xfs_agnumber_t	maxagi = 0;
287 	xfs_sb_t	*sbp = &mp->m_sb;
288 	xfs_agnumber_t	max_metadata;
289 	xfs_agino_t	agino;
290 	xfs_ino_t	ino;
291 
292 	/*
293 	 * Calculate how much should be reserved for inodes to meet
294 	 * the max inode percentage.  Used only for inode32.
295 	 */
296 	if (M_IGEO(mp)->maxicount) {
297 		uint64_t	icount;
298 
299 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
300 		do_div(icount, 100);
301 		icount += sbp->sb_agblocks - 1;
302 		do_div(icount, sbp->sb_agblocks);
303 		max_metadata = icount;
304 	} else {
305 		max_metadata = agcount;
306 	}
307 
308 	/* Get the last possible inode in the filesystem */
309 	agino =	XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
310 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
311 
312 	/*
313 	 * If user asked for no more than 32-bit inodes, and the fs is
314 	 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
315 	 * the allocator to accommodate the request.
316 	 */
317 	if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
318 		xfs_set_inode32(mp);
319 	else
320 		xfs_clear_inode32(mp);
321 
322 	for (index = 0; index < agcount; index++) {
323 		struct xfs_perag	*pag;
324 
325 		ino = XFS_AGINO_TO_INO(mp, index, agino);
326 
327 		pag = xfs_perag_get(mp, index);
328 		if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
329 			maxagi++;
330 		xfs_perag_put(pag);
331 	}
332 
333 	return xfs_is_inode32(mp) ? maxagi : agcount;
334 }
335 
336 static int
337 xfs_setup_dax_always(
338 	struct xfs_mount	*mp)
339 {
340 	if (!mp->m_ddev_targp->bt_daxdev &&
341 	    (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
342 		xfs_alert(mp,
343 			"DAX unsupported by block device. Turning off DAX.");
344 		goto disable_dax;
345 	}
346 
347 	if (mp->m_super->s_blocksize != PAGE_SIZE) {
348 		xfs_alert(mp,
349 			"DAX not supported for blocksize. Turning off DAX.");
350 		goto disable_dax;
351 	}
352 
353 	if (xfs_has_reflink(mp) &&
354 	    bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
355 		xfs_alert(mp,
356 			"DAX and reflink cannot work with multi-partitions!");
357 		return -EINVAL;
358 	}
359 
360 	return 0;
361 
362 disable_dax:
363 	xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
364 	return 0;
365 }
366 
367 STATIC int
368 xfs_blkdev_get(
369 	xfs_mount_t		*mp,
370 	const char		*name,
371 	struct file		**bdev_filep)
372 {
373 	int			error = 0;
374 
375 	*bdev_filep = bdev_file_open_by_path(name,
376 		BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES,
377 		mp->m_super, &fs_holder_ops);
378 	if (IS_ERR(*bdev_filep)) {
379 		error = PTR_ERR(*bdev_filep);
380 		*bdev_filep = NULL;
381 		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
382 	}
383 
384 	return error;
385 }
386 
387 STATIC void
388 xfs_shutdown_devices(
389 	struct xfs_mount	*mp)
390 {
391 	/*
392 	 * Udev is triggered whenever anyone closes a block device or unmounts
393 	 * a file systemm on a block device.
394 	 * The default udev rules invoke blkid to read the fs super and create
395 	 * symlinks to the bdev under /dev/disk.  For this, it uses buffered
396 	 * reads through the page cache.
397 	 *
398 	 * xfs_db also uses buffered reads to examine metadata.  There is no
399 	 * coordination between xfs_db and udev, which means that they can run
400 	 * concurrently.  Note there is no coordination between the kernel and
401 	 * blkid either.
402 	 *
403 	 * On a system with 64k pages, the page cache can cache the superblock
404 	 * and the root inode (and hence the root directory) with the same 64k
405 	 * page.  If udev spawns blkid after the mkfs and the system is busy
406 	 * enough that it is still running when xfs_db starts up, they'll both
407 	 * read from the same page in the pagecache.
408 	 *
409 	 * The unmount writes updated inode metadata to disk directly.  The XFS
410 	 * buffer cache does not use the bdev pagecache, so it needs to
411 	 * invalidate that pagecache on unmount.  If the above scenario occurs,
412 	 * the pagecache no longer reflects what's on disk, xfs_db reads the
413 	 * stale metadata, and fails to find /a.  Most of the time this succeeds
414 	 * because closing a bdev invalidates the page cache, but when processes
415 	 * race, everyone loses.
416 	 */
417 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
418 		blkdev_issue_flush(mp->m_logdev_targp->bt_bdev);
419 		invalidate_bdev(mp->m_logdev_targp->bt_bdev);
420 	}
421 	if (mp->m_rtdev_targp) {
422 		blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
423 		invalidate_bdev(mp->m_rtdev_targp->bt_bdev);
424 	}
425 	blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
426 	invalidate_bdev(mp->m_ddev_targp->bt_bdev);
427 }
428 
429 /*
430  * The file system configurations are:
431  *	(1) device (partition) with data and internal log
432  *	(2) logical volume with data and log subvolumes.
433  *	(3) logical volume with data, log, and realtime subvolumes.
434  *
435  * We only have to handle opening the log and realtime volumes here if
436  * they are present.  The data subvolume has already been opened by
437  * get_sb_bdev() and is stored in sb->s_bdev.
438  */
439 STATIC int
440 xfs_open_devices(
441 	struct xfs_mount	*mp)
442 {
443 	struct super_block	*sb = mp->m_super;
444 	struct block_device	*ddev = sb->s_bdev;
445 	struct file		*logdev_file = NULL, *rtdev_file = NULL;
446 	int			error;
447 
448 	/*
449 	 * Open real time and log devices - order is important.
450 	 */
451 	if (mp->m_logname) {
452 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev_file);
453 		if (error)
454 			return error;
455 	}
456 
457 	if (mp->m_rtname) {
458 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev_file);
459 		if (error)
460 			goto out_close_logdev;
461 
462 		if (file_bdev(rtdev_file) == ddev ||
463 		    (logdev_file &&
464 		     file_bdev(rtdev_file) == file_bdev(logdev_file))) {
465 			xfs_warn(mp,
466 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
467 			error = -EINVAL;
468 			goto out_close_rtdev;
469 		}
470 	}
471 
472 	/*
473 	 * Setup xfs_mount buffer target pointers
474 	 */
475 	error = -ENOMEM;
476 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, sb->s_bdev_file);
477 	if (!mp->m_ddev_targp)
478 		goto out_close_rtdev;
479 
480 	if (rtdev_file) {
481 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev_file);
482 		if (!mp->m_rtdev_targp)
483 			goto out_free_ddev_targ;
484 	}
485 
486 	if (logdev_file && file_bdev(logdev_file) != ddev) {
487 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev_file);
488 		if (!mp->m_logdev_targp)
489 			goto out_free_rtdev_targ;
490 	} else {
491 		mp->m_logdev_targp = mp->m_ddev_targp;
492 		/* Handle won't be used, drop it */
493 		if (logdev_file)
494 			bdev_fput(logdev_file);
495 	}
496 
497 	return 0;
498 
499  out_free_rtdev_targ:
500 	if (mp->m_rtdev_targp)
501 		xfs_free_buftarg(mp->m_rtdev_targp);
502  out_free_ddev_targ:
503 	xfs_free_buftarg(mp->m_ddev_targp);
504  out_close_rtdev:
505 	 if (rtdev_file)
506 		bdev_fput(rtdev_file);
507  out_close_logdev:
508 	if (logdev_file)
509 		bdev_fput(logdev_file);
510 	return error;
511 }
512 
513 /*
514  * Setup xfs_mount buffer target pointers based on superblock
515  */
516 STATIC int
517 xfs_setup_devices(
518 	struct xfs_mount	*mp)
519 {
520 	int			error;
521 
522 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
523 	if (error)
524 		return error;
525 
526 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
527 		unsigned int	log_sector_size = BBSIZE;
528 
529 		if (xfs_has_sector(mp))
530 			log_sector_size = mp->m_sb.sb_logsectsize;
531 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
532 					    log_sector_size);
533 		if (error)
534 			return error;
535 	}
536 	if (mp->m_rtdev_targp) {
537 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
538 					    mp->m_sb.sb_sectsize);
539 		if (error)
540 			return error;
541 	}
542 
543 	return 0;
544 }
545 
546 STATIC int
547 xfs_init_mount_workqueues(
548 	struct xfs_mount	*mp)
549 {
550 	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
551 			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
552 			1, mp->m_super->s_id);
553 	if (!mp->m_buf_workqueue)
554 		goto out;
555 
556 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
557 			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
558 			0, mp->m_super->s_id);
559 	if (!mp->m_unwritten_workqueue)
560 		goto out_destroy_buf;
561 
562 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
563 			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
564 			0, mp->m_super->s_id);
565 	if (!mp->m_reclaim_workqueue)
566 		goto out_destroy_unwritten;
567 
568 	mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
569 			XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
570 			0, mp->m_super->s_id);
571 	if (!mp->m_blockgc_wq)
572 		goto out_destroy_reclaim;
573 
574 	mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
575 			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
576 			1, mp->m_super->s_id);
577 	if (!mp->m_inodegc_wq)
578 		goto out_destroy_blockgc;
579 
580 	mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
581 			XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
582 	if (!mp->m_sync_workqueue)
583 		goto out_destroy_inodegc;
584 
585 	return 0;
586 
587 out_destroy_inodegc:
588 	destroy_workqueue(mp->m_inodegc_wq);
589 out_destroy_blockgc:
590 	destroy_workqueue(mp->m_blockgc_wq);
591 out_destroy_reclaim:
592 	destroy_workqueue(mp->m_reclaim_workqueue);
593 out_destroy_unwritten:
594 	destroy_workqueue(mp->m_unwritten_workqueue);
595 out_destroy_buf:
596 	destroy_workqueue(mp->m_buf_workqueue);
597 out:
598 	return -ENOMEM;
599 }
600 
601 STATIC void
602 xfs_destroy_mount_workqueues(
603 	struct xfs_mount	*mp)
604 {
605 	destroy_workqueue(mp->m_sync_workqueue);
606 	destroy_workqueue(mp->m_blockgc_wq);
607 	destroy_workqueue(mp->m_inodegc_wq);
608 	destroy_workqueue(mp->m_reclaim_workqueue);
609 	destroy_workqueue(mp->m_unwritten_workqueue);
610 	destroy_workqueue(mp->m_buf_workqueue);
611 }
612 
613 static void
614 xfs_flush_inodes_worker(
615 	struct work_struct	*work)
616 {
617 	struct xfs_mount	*mp = container_of(work, struct xfs_mount,
618 						   m_flush_inodes_work);
619 	struct super_block	*sb = mp->m_super;
620 
621 	if (down_read_trylock(&sb->s_umount)) {
622 		sync_inodes_sb(sb);
623 		up_read(&sb->s_umount);
624 	}
625 }
626 
627 /*
628  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
629  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
630  * for IO to complete so that we effectively throttle multiple callers to the
631  * rate at which IO is completing.
632  */
633 void
634 xfs_flush_inodes(
635 	struct xfs_mount	*mp)
636 {
637 	/*
638 	 * If flush_work() returns true then that means we waited for a flush
639 	 * which was already in progress.  Don't bother running another scan.
640 	 */
641 	if (flush_work(&mp->m_flush_inodes_work))
642 		return;
643 
644 	queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
645 	flush_work(&mp->m_flush_inodes_work);
646 }
647 
648 /* Catch misguided souls that try to use this interface on XFS */
649 STATIC struct inode *
650 xfs_fs_alloc_inode(
651 	struct super_block	*sb)
652 {
653 	BUG();
654 	return NULL;
655 }
656 
657 /*
658  * Now that the generic code is guaranteed not to be accessing
659  * the linux inode, we can inactivate and reclaim the inode.
660  */
661 STATIC void
662 xfs_fs_destroy_inode(
663 	struct inode		*inode)
664 {
665 	struct xfs_inode	*ip = XFS_I(inode);
666 
667 	trace_xfs_destroy_inode(ip);
668 
669 	ASSERT(!rwsem_is_locked(&inode->i_rwsem));
670 	XFS_STATS_INC(ip->i_mount, vn_rele);
671 	XFS_STATS_INC(ip->i_mount, vn_remove);
672 	xfs_inode_mark_reclaimable(ip);
673 }
674 
675 static void
676 xfs_fs_dirty_inode(
677 	struct inode			*inode,
678 	int				flags)
679 {
680 	struct xfs_inode		*ip = XFS_I(inode);
681 	struct xfs_mount		*mp = ip->i_mount;
682 	struct xfs_trans		*tp;
683 
684 	if (!(inode->i_sb->s_flags & SB_LAZYTIME))
685 		return;
686 
687 	/*
688 	 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
689 	 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
690 	 * in flags possibly together with I_DIRTY_SYNC.
691 	 */
692 	if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
693 		return;
694 
695 	if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
696 		return;
697 	xfs_ilock(ip, XFS_ILOCK_EXCL);
698 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
699 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
700 	xfs_trans_commit(tp);
701 }
702 
703 /*
704  * Slab object creation initialisation for the XFS inode.
705  * This covers only the idempotent fields in the XFS inode;
706  * all other fields need to be initialised on allocation
707  * from the slab. This avoids the need to repeatedly initialise
708  * fields in the xfs inode that left in the initialise state
709  * when freeing the inode.
710  */
711 STATIC void
712 xfs_fs_inode_init_once(
713 	void			*inode)
714 {
715 	struct xfs_inode	*ip = inode;
716 
717 	memset(ip, 0, sizeof(struct xfs_inode));
718 
719 	/* vfs inode */
720 	inode_init_once(VFS_I(ip));
721 
722 	/* xfs inode */
723 	atomic_set(&ip->i_pincount, 0);
724 	spin_lock_init(&ip->i_flags_lock);
725 	init_rwsem(&ip->i_lock);
726 }
727 
728 /*
729  * We do an unlocked check for XFS_IDONTCACHE here because we are already
730  * serialised against cache hits here via the inode->i_lock and igrab() in
731  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
732  * racing with us, and it avoids needing to grab a spinlock here for every inode
733  * we drop the final reference on.
734  */
735 STATIC int
736 xfs_fs_drop_inode(
737 	struct inode		*inode)
738 {
739 	struct xfs_inode	*ip = XFS_I(inode);
740 
741 	/*
742 	 * If this unlinked inode is in the middle of recovery, don't
743 	 * drop the inode just yet; log recovery will take care of
744 	 * that.  See the comment for this inode flag.
745 	 */
746 	if (ip->i_flags & XFS_IRECOVERY) {
747 		ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
748 		return 0;
749 	}
750 
751 	return generic_drop_inode(inode);
752 }
753 
754 static void
755 xfs_mount_free(
756 	struct xfs_mount	*mp)
757 {
758 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
759 		xfs_free_buftarg(mp->m_logdev_targp);
760 	if (mp->m_rtdev_targp)
761 		xfs_free_buftarg(mp->m_rtdev_targp);
762 	if (mp->m_ddev_targp)
763 		xfs_free_buftarg(mp->m_ddev_targp);
764 
765 	debugfs_remove(mp->m_debugfs);
766 	kfree(mp->m_rtname);
767 	kfree(mp->m_logname);
768 	kfree(mp);
769 }
770 
771 STATIC int
772 xfs_fs_sync_fs(
773 	struct super_block	*sb,
774 	int			wait)
775 {
776 	struct xfs_mount	*mp = XFS_M(sb);
777 	int			error;
778 
779 	trace_xfs_fs_sync_fs(mp, __return_address);
780 
781 	/*
782 	 * Doing anything during the async pass would be counterproductive.
783 	 */
784 	if (!wait)
785 		return 0;
786 
787 	error = xfs_log_force(mp, XFS_LOG_SYNC);
788 	if (error)
789 		return error;
790 
791 	if (laptop_mode) {
792 		/*
793 		 * The disk must be active because we're syncing.
794 		 * We schedule log work now (now that the disk is
795 		 * active) instead of later (when it might not be).
796 		 */
797 		flush_delayed_work(&mp->m_log->l_work);
798 	}
799 
800 	/*
801 	 * If we are called with page faults frozen out, it means we are about
802 	 * to freeze the transaction subsystem. Take the opportunity to shut
803 	 * down inodegc because once SB_FREEZE_FS is set it's too late to
804 	 * prevent inactivation races with freeze. The fs doesn't get called
805 	 * again by the freezing process until after SB_FREEZE_FS has been set,
806 	 * so it's now or never.  Same logic applies to speculative allocation
807 	 * garbage collection.
808 	 *
809 	 * We don't care if this is a normal syncfs call that does this or
810 	 * freeze that does this - we can run this multiple times without issue
811 	 * and we won't race with a restart because a restart can only occur
812 	 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
813 	 */
814 	if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
815 		xfs_inodegc_stop(mp);
816 		xfs_blockgc_stop(mp);
817 	}
818 
819 	return 0;
820 }
821 
822 STATIC int
823 xfs_fs_statfs(
824 	struct dentry		*dentry,
825 	struct kstatfs		*statp)
826 {
827 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
828 	xfs_sb_t		*sbp = &mp->m_sb;
829 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
830 	uint64_t		fakeinos, id;
831 	uint64_t		icount;
832 	uint64_t		ifree;
833 	uint64_t		fdblocks;
834 	xfs_extlen_t		lsize;
835 	int64_t			ffree;
836 
837 	/*
838 	 * Expedite background inodegc but don't wait. We do not want to block
839 	 * here waiting hours for a billion extent file to be truncated.
840 	 */
841 	xfs_inodegc_push(mp);
842 
843 	statp->f_type = XFS_SUPER_MAGIC;
844 	statp->f_namelen = MAXNAMELEN - 1;
845 
846 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
847 	statp->f_fsid = u64_to_fsid(id);
848 
849 	icount = percpu_counter_sum(&mp->m_icount);
850 	ifree = percpu_counter_sum(&mp->m_ifree);
851 	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
852 
853 	spin_lock(&mp->m_sb_lock);
854 	statp->f_bsize = sbp->sb_blocksize;
855 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
856 	statp->f_blocks = sbp->sb_dblocks - lsize;
857 	spin_unlock(&mp->m_sb_lock);
858 
859 	/* make sure statp->f_bfree does not underflow */
860 	statp->f_bfree = max_t(int64_t, 0,
861 				fdblocks - xfs_fdblocks_unavailable(mp));
862 	statp->f_bavail = statp->f_bfree;
863 
864 	fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
865 	statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
866 	if (M_IGEO(mp)->maxicount)
867 		statp->f_files = min_t(typeof(statp->f_files),
868 					statp->f_files,
869 					M_IGEO(mp)->maxicount);
870 
871 	/* If sb_icount overshot maxicount, report actual allocation */
872 	statp->f_files = max_t(typeof(statp->f_files),
873 					statp->f_files,
874 					sbp->sb_icount);
875 
876 	/* make sure statp->f_ffree does not underflow */
877 	ffree = statp->f_files - (icount - ifree);
878 	statp->f_ffree = max_t(int64_t, ffree, 0);
879 
880 	if (XFS_IS_REALTIME_MOUNT(mp) &&
881 	    (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
882 		s64	freertx;
883 
884 		statp->f_blocks = sbp->sb_rblocks;
885 		freertx = percpu_counter_sum_positive(&mp->m_frextents);
886 		statp->f_bavail = statp->f_bfree =
887 			xfs_rtbxlen_to_blen(mp, freertx);
888 	}
889 
890 	if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
891 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
892 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
893 		xfs_qm_statvfs(ip, statp);
894 
895 	return 0;
896 }
897 
898 STATIC void
899 xfs_save_resvblks(struct xfs_mount *mp)
900 {
901 	mp->m_resblks_save = mp->m_resblks;
902 	xfs_reserve_blocks(mp, 0);
903 }
904 
905 STATIC void
906 xfs_restore_resvblks(struct xfs_mount *mp)
907 {
908 	uint64_t resblks;
909 
910 	if (mp->m_resblks_save) {
911 		resblks = mp->m_resblks_save;
912 		mp->m_resblks_save = 0;
913 	} else
914 		resblks = xfs_default_resblks(mp);
915 
916 	xfs_reserve_blocks(mp, resblks);
917 }
918 
919 /*
920  * Second stage of a freeze. The data is already frozen so we only
921  * need to take care of the metadata. Once that's done sync the superblock
922  * to the log to dirty it in case of a crash while frozen. This ensures that we
923  * will recover the unlinked inode lists on the next mount.
924  */
925 STATIC int
926 xfs_fs_freeze(
927 	struct super_block	*sb)
928 {
929 	struct xfs_mount	*mp = XFS_M(sb);
930 	unsigned int		flags;
931 	int			ret;
932 
933 	/*
934 	 * The filesystem is now frozen far enough that memory reclaim
935 	 * cannot safely operate on the filesystem. Hence we need to
936 	 * set a GFP_NOFS context here to avoid recursion deadlocks.
937 	 */
938 	flags = memalloc_nofs_save();
939 	xfs_save_resvblks(mp);
940 	ret = xfs_log_quiesce(mp);
941 	memalloc_nofs_restore(flags);
942 
943 	/*
944 	 * For read-write filesystems, we need to restart the inodegc on error
945 	 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
946 	 * going to be run to restart it now.  We are at SB_FREEZE_FS level
947 	 * here, so we can restart safely without racing with a stop in
948 	 * xfs_fs_sync_fs().
949 	 */
950 	if (ret && !xfs_is_readonly(mp)) {
951 		xfs_blockgc_start(mp);
952 		xfs_inodegc_start(mp);
953 	}
954 
955 	return ret;
956 }
957 
958 STATIC int
959 xfs_fs_unfreeze(
960 	struct super_block	*sb)
961 {
962 	struct xfs_mount	*mp = XFS_M(sb);
963 
964 	xfs_restore_resvblks(mp);
965 	xfs_log_work_queue(mp);
966 
967 	/*
968 	 * Don't reactivate the inodegc worker on a readonly filesystem because
969 	 * inodes are sent directly to reclaim.  Don't reactivate the blockgc
970 	 * worker because there are no speculative preallocations on a readonly
971 	 * filesystem.
972 	 */
973 	if (!xfs_is_readonly(mp)) {
974 		xfs_blockgc_start(mp);
975 		xfs_inodegc_start(mp);
976 	}
977 
978 	return 0;
979 }
980 
981 /*
982  * This function fills in xfs_mount_t fields based on mount args.
983  * Note: the superblock _has_ now been read in.
984  */
985 STATIC int
986 xfs_finish_flags(
987 	struct xfs_mount	*mp)
988 {
989 	/* Fail a mount where the logbuf is smaller than the log stripe */
990 	if (xfs_has_logv2(mp)) {
991 		if (mp->m_logbsize <= 0 &&
992 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
993 			mp->m_logbsize = mp->m_sb.sb_logsunit;
994 		} else if (mp->m_logbsize > 0 &&
995 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
996 			xfs_warn(mp,
997 		"logbuf size must be greater than or equal to log stripe size");
998 			return -EINVAL;
999 		}
1000 	} else {
1001 		/* Fail a mount if the logbuf is larger than 32K */
1002 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1003 			xfs_warn(mp,
1004 		"logbuf size for version 1 logs must be 16K or 32K");
1005 			return -EINVAL;
1006 		}
1007 	}
1008 
1009 	/*
1010 	 * V5 filesystems always use attr2 format for attributes.
1011 	 */
1012 	if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
1013 		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1014 			     "attr2 is always enabled for V5 filesystems.");
1015 		return -EINVAL;
1016 	}
1017 
1018 	/*
1019 	 * prohibit r/w mounts of read-only filesystems
1020 	 */
1021 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1022 		xfs_warn(mp,
1023 			"cannot mount a read-only filesystem as read-write");
1024 		return -EROFS;
1025 	}
1026 
1027 	if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1028 	    (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1029 	    !xfs_has_pquotino(mp)) {
1030 		xfs_warn(mp,
1031 		  "Super block does not support project and group quota together");
1032 		return -EINVAL;
1033 	}
1034 
1035 	return 0;
1036 }
1037 
1038 static int
1039 xfs_init_percpu_counters(
1040 	struct xfs_mount	*mp)
1041 {
1042 	int		error;
1043 
1044 	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1045 	if (error)
1046 		return -ENOMEM;
1047 
1048 	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1049 	if (error)
1050 		goto free_icount;
1051 
1052 	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1053 	if (error)
1054 		goto free_ifree;
1055 
1056 	error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1057 	if (error)
1058 		goto free_fdblocks;
1059 
1060 	error = percpu_counter_init(&mp->m_delalloc_rtextents, 0, GFP_KERNEL);
1061 	if (error)
1062 		goto free_delalloc;
1063 
1064 	error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1065 	if (error)
1066 		goto free_delalloc_rt;
1067 
1068 	return 0;
1069 
1070 free_delalloc_rt:
1071 	percpu_counter_destroy(&mp->m_delalloc_rtextents);
1072 free_delalloc:
1073 	percpu_counter_destroy(&mp->m_delalloc_blks);
1074 free_fdblocks:
1075 	percpu_counter_destroy(&mp->m_fdblocks);
1076 free_ifree:
1077 	percpu_counter_destroy(&mp->m_ifree);
1078 free_icount:
1079 	percpu_counter_destroy(&mp->m_icount);
1080 	return -ENOMEM;
1081 }
1082 
1083 void
1084 xfs_reinit_percpu_counters(
1085 	struct xfs_mount	*mp)
1086 {
1087 	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1088 	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1089 	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1090 	percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1091 }
1092 
1093 static void
1094 xfs_destroy_percpu_counters(
1095 	struct xfs_mount	*mp)
1096 {
1097 	percpu_counter_destroy(&mp->m_icount);
1098 	percpu_counter_destroy(&mp->m_ifree);
1099 	percpu_counter_destroy(&mp->m_fdblocks);
1100 	ASSERT(xfs_is_shutdown(mp) ||
1101 	       percpu_counter_sum(&mp->m_delalloc_rtextents) == 0);
1102 	percpu_counter_destroy(&mp->m_delalloc_rtextents);
1103 	ASSERT(xfs_is_shutdown(mp) ||
1104 	       percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1105 	percpu_counter_destroy(&mp->m_delalloc_blks);
1106 	percpu_counter_destroy(&mp->m_frextents);
1107 }
1108 
1109 static int
1110 xfs_inodegc_init_percpu(
1111 	struct xfs_mount	*mp)
1112 {
1113 	struct xfs_inodegc	*gc;
1114 	int			cpu;
1115 
1116 	mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1117 	if (!mp->m_inodegc)
1118 		return -ENOMEM;
1119 
1120 	for_each_possible_cpu(cpu) {
1121 		gc = per_cpu_ptr(mp->m_inodegc, cpu);
1122 		gc->cpu = cpu;
1123 		gc->mp = mp;
1124 		init_llist_head(&gc->list);
1125 		gc->items = 0;
1126 		gc->error = 0;
1127 		INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1128 	}
1129 	return 0;
1130 }
1131 
1132 static void
1133 xfs_inodegc_free_percpu(
1134 	struct xfs_mount	*mp)
1135 {
1136 	if (!mp->m_inodegc)
1137 		return;
1138 	free_percpu(mp->m_inodegc);
1139 }
1140 
1141 static void
1142 xfs_fs_put_super(
1143 	struct super_block	*sb)
1144 {
1145 	struct xfs_mount	*mp = XFS_M(sb);
1146 
1147 	xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1148 	xfs_filestream_unmount(mp);
1149 	xfs_unmountfs(mp);
1150 
1151 	xfs_rtmount_freesb(mp);
1152 	xfs_freesb(mp);
1153 	xchk_mount_stats_free(mp);
1154 	free_percpu(mp->m_stats.xs_stats);
1155 	xfs_inodegc_free_percpu(mp);
1156 	xfs_destroy_percpu_counters(mp);
1157 	xfs_destroy_mount_workqueues(mp);
1158 	xfs_shutdown_devices(mp);
1159 }
1160 
1161 static long
1162 xfs_fs_nr_cached_objects(
1163 	struct super_block	*sb,
1164 	struct shrink_control	*sc)
1165 {
1166 	/* Paranoia: catch incorrect calls during mount setup or teardown */
1167 	if (WARN_ON_ONCE(!sb->s_fs_info))
1168 		return 0;
1169 	return xfs_reclaim_inodes_count(XFS_M(sb));
1170 }
1171 
1172 static long
1173 xfs_fs_free_cached_objects(
1174 	struct super_block	*sb,
1175 	struct shrink_control	*sc)
1176 {
1177 	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1178 }
1179 
1180 static void
1181 xfs_fs_shutdown(
1182 	struct super_block	*sb)
1183 {
1184 	xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED);
1185 }
1186 
1187 static const struct super_operations xfs_super_operations = {
1188 	.alloc_inode		= xfs_fs_alloc_inode,
1189 	.destroy_inode		= xfs_fs_destroy_inode,
1190 	.dirty_inode		= xfs_fs_dirty_inode,
1191 	.drop_inode		= xfs_fs_drop_inode,
1192 	.put_super		= xfs_fs_put_super,
1193 	.sync_fs		= xfs_fs_sync_fs,
1194 	.freeze_fs		= xfs_fs_freeze,
1195 	.unfreeze_fs		= xfs_fs_unfreeze,
1196 	.statfs			= xfs_fs_statfs,
1197 	.show_options		= xfs_fs_show_options,
1198 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1199 	.free_cached_objects	= xfs_fs_free_cached_objects,
1200 	.shutdown		= xfs_fs_shutdown,
1201 };
1202 
1203 static int
1204 suffix_kstrtoint(
1205 	const char	*s,
1206 	unsigned int	base,
1207 	int		*res)
1208 {
1209 	int		last, shift_left_factor = 0, _res;
1210 	char		*value;
1211 	int		ret = 0;
1212 
1213 	value = kstrdup(s, GFP_KERNEL);
1214 	if (!value)
1215 		return -ENOMEM;
1216 
1217 	last = strlen(value) - 1;
1218 	if (value[last] == 'K' || value[last] == 'k') {
1219 		shift_left_factor = 10;
1220 		value[last] = '\0';
1221 	}
1222 	if (value[last] == 'M' || value[last] == 'm') {
1223 		shift_left_factor = 20;
1224 		value[last] = '\0';
1225 	}
1226 	if (value[last] == 'G' || value[last] == 'g') {
1227 		shift_left_factor = 30;
1228 		value[last] = '\0';
1229 	}
1230 
1231 	if (kstrtoint(value, base, &_res))
1232 		ret = -EINVAL;
1233 	kfree(value);
1234 	*res = _res << shift_left_factor;
1235 	return ret;
1236 }
1237 
1238 static inline void
1239 xfs_fs_warn_deprecated(
1240 	struct fs_context	*fc,
1241 	struct fs_parameter	*param,
1242 	uint64_t		flag,
1243 	bool			value)
1244 {
1245 	/* Don't print the warning if reconfiguring and current mount point
1246 	 * already had the flag set
1247 	 */
1248 	if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1249             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1250 		return;
1251 	xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1252 }
1253 
1254 /*
1255  * Set mount state from a mount option.
1256  *
1257  * NOTE: mp->m_super is NULL here!
1258  */
1259 static int
1260 xfs_fs_parse_param(
1261 	struct fs_context	*fc,
1262 	struct fs_parameter	*param)
1263 {
1264 	struct xfs_mount	*parsing_mp = fc->s_fs_info;
1265 	struct fs_parse_result	result;
1266 	int			size = 0;
1267 	int			opt;
1268 
1269 	BUILD_BUG_ON(XFS_QFLAGS_MNTOPTS & XFS_MOUNT_QUOTA_ALL);
1270 
1271 	opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1272 	if (opt < 0)
1273 		return opt;
1274 
1275 	switch (opt) {
1276 	case Opt_logbufs:
1277 		parsing_mp->m_logbufs = result.uint_32;
1278 		return 0;
1279 	case Opt_logbsize:
1280 		if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1281 			return -EINVAL;
1282 		return 0;
1283 	case Opt_logdev:
1284 		kfree(parsing_mp->m_logname);
1285 		parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1286 		if (!parsing_mp->m_logname)
1287 			return -ENOMEM;
1288 		return 0;
1289 	case Opt_rtdev:
1290 		kfree(parsing_mp->m_rtname);
1291 		parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1292 		if (!parsing_mp->m_rtname)
1293 			return -ENOMEM;
1294 		return 0;
1295 	case Opt_allocsize:
1296 		if (suffix_kstrtoint(param->string, 10, &size))
1297 			return -EINVAL;
1298 		parsing_mp->m_allocsize_log = ffs(size) - 1;
1299 		parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1300 		return 0;
1301 	case Opt_grpid:
1302 	case Opt_bsdgroups:
1303 		parsing_mp->m_features |= XFS_FEAT_GRPID;
1304 		return 0;
1305 	case Opt_nogrpid:
1306 	case Opt_sysvgroups:
1307 		parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1308 		return 0;
1309 	case Opt_wsync:
1310 		parsing_mp->m_features |= XFS_FEAT_WSYNC;
1311 		return 0;
1312 	case Opt_norecovery:
1313 		parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1314 		return 0;
1315 	case Opt_noalign:
1316 		parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1317 		return 0;
1318 	case Opt_swalloc:
1319 		parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1320 		return 0;
1321 	case Opt_sunit:
1322 		parsing_mp->m_dalign = result.uint_32;
1323 		return 0;
1324 	case Opt_swidth:
1325 		parsing_mp->m_swidth = result.uint_32;
1326 		return 0;
1327 	case Opt_inode32:
1328 		parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1329 		return 0;
1330 	case Opt_inode64:
1331 		parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1332 		return 0;
1333 	case Opt_nouuid:
1334 		parsing_mp->m_features |= XFS_FEAT_NOUUID;
1335 		return 0;
1336 	case Opt_largeio:
1337 		parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1338 		return 0;
1339 	case Opt_nolargeio:
1340 		parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1341 		return 0;
1342 	case Opt_filestreams:
1343 		parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1344 		return 0;
1345 	case Opt_noquota:
1346 		parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1347 		parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1348 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1349 		return 0;
1350 	case Opt_quota:
1351 	case Opt_uquota:
1352 	case Opt_usrquota:
1353 		parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1354 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1355 		return 0;
1356 	case Opt_qnoenforce:
1357 	case Opt_uqnoenforce:
1358 		parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1359 		parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1360 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1361 		return 0;
1362 	case Opt_pquota:
1363 	case Opt_prjquota:
1364 		parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1365 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1366 		return 0;
1367 	case Opt_pqnoenforce:
1368 		parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1369 		parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1370 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1371 		return 0;
1372 	case Opt_gquota:
1373 	case Opt_grpquota:
1374 		parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1375 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1376 		return 0;
1377 	case Opt_gqnoenforce:
1378 		parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1379 		parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1380 		parsing_mp->m_qflags |= XFS_QFLAGS_MNTOPTS;
1381 		return 0;
1382 	case Opt_discard:
1383 		parsing_mp->m_features |= XFS_FEAT_DISCARD;
1384 		return 0;
1385 	case Opt_nodiscard:
1386 		parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1387 		return 0;
1388 #ifdef CONFIG_FS_DAX
1389 	case Opt_dax:
1390 		xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1391 		return 0;
1392 	case Opt_dax_enum:
1393 		xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1394 		return 0;
1395 #endif
1396 	/* Following mount options will be removed in September 2025 */
1397 	case Opt_ikeep:
1398 		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1399 		parsing_mp->m_features |= XFS_FEAT_IKEEP;
1400 		return 0;
1401 	case Opt_noikeep:
1402 		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1403 		parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1404 		return 0;
1405 	case Opt_attr2:
1406 		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1407 		parsing_mp->m_features |= XFS_FEAT_ATTR2;
1408 		return 0;
1409 	case Opt_noattr2:
1410 		xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1411 		parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1412 		return 0;
1413 	default:
1414 		xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1415 		return -EINVAL;
1416 	}
1417 
1418 	return 0;
1419 }
1420 
1421 static int
1422 xfs_fs_validate_params(
1423 	struct xfs_mount	*mp)
1424 {
1425 	/* No recovery flag requires a read-only mount */
1426 	if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1427 		xfs_warn(mp, "no-recovery mounts must be read-only.");
1428 		return -EINVAL;
1429 	}
1430 
1431 	/*
1432 	 * We have not read the superblock at this point, so only the attr2
1433 	 * mount option can set the attr2 feature by this stage.
1434 	 */
1435 	if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1436 		xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1437 		return -EINVAL;
1438 	}
1439 
1440 
1441 	if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1442 		xfs_warn(mp,
1443 	"sunit and swidth options incompatible with the noalign option");
1444 		return -EINVAL;
1445 	}
1446 
1447 	if (!IS_ENABLED(CONFIG_XFS_QUOTA) &&
1448 	    (mp->m_qflags & ~XFS_QFLAGS_MNTOPTS)) {
1449 		xfs_warn(mp, "quota support not available in this kernel.");
1450 		return -EINVAL;
1451 	}
1452 
1453 	if ((mp->m_dalign && !mp->m_swidth) ||
1454 	    (!mp->m_dalign && mp->m_swidth)) {
1455 		xfs_warn(mp, "sunit and swidth must be specified together");
1456 		return -EINVAL;
1457 	}
1458 
1459 	if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1460 		xfs_warn(mp,
1461 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
1462 			mp->m_swidth, mp->m_dalign);
1463 		return -EINVAL;
1464 	}
1465 
1466 	if (mp->m_logbufs != -1 &&
1467 	    mp->m_logbufs != 0 &&
1468 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1469 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1470 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1471 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1472 		return -EINVAL;
1473 	}
1474 
1475 	if (mp->m_logbsize != -1 &&
1476 	    mp->m_logbsize !=  0 &&
1477 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1478 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1479 	     !is_power_of_2(mp->m_logbsize))) {
1480 		xfs_warn(mp,
1481 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1482 			mp->m_logbsize);
1483 		return -EINVAL;
1484 	}
1485 
1486 	if (xfs_has_allocsize(mp) &&
1487 	    (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1488 	     mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1489 		xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1490 			mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1491 		return -EINVAL;
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 struct dentry *
1498 xfs_debugfs_mkdir(
1499 	const char	*name,
1500 	struct dentry	*parent)
1501 {
1502 	struct dentry	*child;
1503 
1504 	/* Apparently we're expected to ignore error returns?? */
1505 	child = debugfs_create_dir(name, parent);
1506 	if (IS_ERR(child))
1507 		return NULL;
1508 
1509 	return child;
1510 }
1511 
1512 static int
1513 xfs_fs_fill_super(
1514 	struct super_block	*sb,
1515 	struct fs_context	*fc)
1516 {
1517 	struct xfs_mount	*mp = sb->s_fs_info;
1518 	struct inode		*root;
1519 	int			flags = 0, error;
1520 
1521 	mp->m_super = sb;
1522 
1523 	/*
1524 	 * Copy VFS mount flags from the context now that all parameter parsing
1525 	 * is guaranteed to have been completed by either the old mount API or
1526 	 * the newer fsopen/fsconfig API.
1527 	 */
1528 	if (fc->sb_flags & SB_RDONLY)
1529 		xfs_set_readonly(mp);
1530 	if (fc->sb_flags & SB_DIRSYNC)
1531 		mp->m_features |= XFS_FEAT_DIRSYNC;
1532 	if (fc->sb_flags & SB_SYNCHRONOUS)
1533 		mp->m_features |= XFS_FEAT_WSYNC;
1534 
1535 	error = xfs_fs_validate_params(mp);
1536 	if (error)
1537 		return error;
1538 
1539 	sb_min_blocksize(sb, BBSIZE);
1540 	sb->s_xattr = xfs_xattr_handlers;
1541 	sb->s_export_op = &xfs_export_operations;
1542 #ifdef CONFIG_XFS_QUOTA
1543 	sb->s_qcop = &xfs_quotactl_operations;
1544 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1545 #endif
1546 	sb->s_op = &xfs_super_operations;
1547 
1548 	/*
1549 	 * Delay mount work if the debug hook is set. This is debug
1550 	 * instrumention to coordinate simulation of xfs mount failures with
1551 	 * VFS superblock operations
1552 	 */
1553 	if (xfs_globals.mount_delay) {
1554 		xfs_notice(mp, "Delaying mount for %d seconds.",
1555 			xfs_globals.mount_delay);
1556 		msleep(xfs_globals.mount_delay * 1000);
1557 	}
1558 
1559 	if (fc->sb_flags & SB_SILENT)
1560 		flags |= XFS_MFSI_QUIET;
1561 
1562 	error = xfs_open_devices(mp);
1563 	if (error)
1564 		return error;
1565 
1566 	if (xfs_debugfs) {
1567 		mp->m_debugfs = xfs_debugfs_mkdir(mp->m_super->s_id,
1568 						  xfs_debugfs);
1569 	} else {
1570 		mp->m_debugfs = NULL;
1571 	}
1572 
1573 	error = xfs_init_mount_workqueues(mp);
1574 	if (error)
1575 		goto out_shutdown_devices;
1576 
1577 	error = xfs_init_percpu_counters(mp);
1578 	if (error)
1579 		goto out_destroy_workqueues;
1580 
1581 	error = xfs_inodegc_init_percpu(mp);
1582 	if (error)
1583 		goto out_destroy_counters;
1584 
1585 	/* Allocate stats memory before we do operations that might use it */
1586 	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1587 	if (!mp->m_stats.xs_stats) {
1588 		error = -ENOMEM;
1589 		goto out_destroy_inodegc;
1590 	}
1591 
1592 	error = xchk_mount_stats_alloc(mp);
1593 	if (error)
1594 		goto out_free_stats;
1595 
1596 	error = xfs_readsb(mp, flags);
1597 	if (error)
1598 		goto out_free_scrub_stats;
1599 
1600 	error = xfs_finish_flags(mp);
1601 	if (error)
1602 		goto out_free_sb;
1603 
1604 	error = xfs_setup_devices(mp);
1605 	if (error)
1606 		goto out_free_sb;
1607 
1608 	/*
1609 	 * V4 support is undergoing deprecation.
1610 	 *
1611 	 * Note: this has to use an open coded m_features check as xfs_has_crc
1612 	 * always returns false for !CONFIG_XFS_SUPPORT_V4.
1613 	 */
1614 	if (!(mp->m_features & XFS_FEAT_CRC)) {
1615 		if (!IS_ENABLED(CONFIG_XFS_SUPPORT_V4)) {
1616 			xfs_warn(mp,
1617 	"Deprecated V4 format (crc=0) not supported by kernel.");
1618 			error = -EINVAL;
1619 			goto out_free_sb;
1620 		}
1621 		xfs_warn_once(mp,
1622 	"Deprecated V4 format (crc=0) will not be supported after September 2030.");
1623 	}
1624 
1625 	/* ASCII case insensitivity is undergoing deprecation. */
1626 	if (xfs_has_asciici(mp)) {
1627 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1628 		xfs_warn_once(mp,
1629 	"Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1630 #else
1631 		xfs_warn(mp,
1632 	"Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1633 		error = -EINVAL;
1634 		goto out_free_sb;
1635 #endif
1636 	}
1637 
1638 	/* Filesystem claims it needs repair, so refuse the mount. */
1639 	if (xfs_has_needsrepair(mp)) {
1640 		xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1641 		error = -EFSCORRUPTED;
1642 		goto out_free_sb;
1643 	}
1644 
1645 	/*
1646 	 * Don't touch the filesystem if a user tool thinks it owns the primary
1647 	 * superblock.  mkfs doesn't clear the flag from secondary supers, so
1648 	 * we don't check them at all.
1649 	 */
1650 	if (mp->m_sb.sb_inprogress) {
1651 		xfs_warn(mp, "Offline file system operation in progress!");
1652 		error = -EFSCORRUPTED;
1653 		goto out_free_sb;
1654 	}
1655 
1656 	if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1657 		size_t max_folio_size = mapping_max_folio_size_supported();
1658 
1659 		if (!xfs_has_crc(mp)) {
1660 			xfs_warn(mp,
1661 "V4 Filesystem with blocksize %d bytes. Only pagesize (%ld) or less is supported.",
1662 				mp->m_sb.sb_blocksize, PAGE_SIZE);
1663 			error = -ENOSYS;
1664 			goto out_free_sb;
1665 		}
1666 
1667 		if (mp->m_sb.sb_blocksize > max_folio_size) {
1668 			xfs_warn(mp,
1669 "block size (%u bytes) not supported; Only block size (%zu) or less is supported",
1670 				mp->m_sb.sb_blocksize, max_folio_size);
1671 			error = -ENOSYS;
1672 			goto out_free_sb;
1673 		}
1674 
1675 		xfs_warn_experimental(mp, XFS_EXPERIMENTAL_LBS);
1676 	}
1677 
1678 	/* Ensure this filesystem fits in the page cache limits */
1679 	if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1680 	    xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1681 		xfs_warn(mp,
1682 		"file system too large to be mounted on this system.");
1683 		error = -EFBIG;
1684 		goto out_free_sb;
1685 	}
1686 
1687 	/*
1688 	 * XFS block mappings use 54 bits to store the logical block offset.
1689 	 * This should suffice to handle the maximum file size that the VFS
1690 	 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1691 	 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1692 	 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1693 	 * to check this assertion.
1694 	 *
1695 	 * Avoid integer overflow by comparing the maximum bmbt offset to the
1696 	 * maximum pagecache offset in units of fs blocks.
1697 	 */
1698 	if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1699 		xfs_warn(mp,
1700 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1701 			 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1702 			 XFS_MAX_FILEOFF);
1703 		error = -EINVAL;
1704 		goto out_free_sb;
1705 	}
1706 
1707 	error = xfs_rtmount_readsb(mp);
1708 	if (error)
1709 		goto out_free_sb;
1710 
1711 	error = xfs_filestream_mount(mp);
1712 	if (error)
1713 		goto out_free_rtsb;
1714 
1715 	/*
1716 	 * we must configure the block size in the superblock before we run the
1717 	 * full mount process as the mount process can lookup and cache inodes.
1718 	 */
1719 	sb->s_magic = XFS_SUPER_MAGIC;
1720 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1721 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1722 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1723 	sb->s_max_links = XFS_MAXLINK;
1724 	sb->s_time_gran = 1;
1725 	if (xfs_has_bigtime(mp)) {
1726 		sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1727 		sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1728 	} else {
1729 		sb->s_time_min = XFS_LEGACY_TIME_MIN;
1730 		sb->s_time_max = XFS_LEGACY_TIME_MAX;
1731 	}
1732 	trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1733 	sb->s_iflags |= SB_I_CGROUPWB;
1734 
1735 	set_posix_acl_flag(sb);
1736 
1737 	/* version 5 superblocks support inode version counters. */
1738 	if (xfs_has_crc(mp))
1739 		sb->s_flags |= SB_I_VERSION;
1740 
1741 	if (xfs_has_dax_always(mp)) {
1742 		error = xfs_setup_dax_always(mp);
1743 		if (error)
1744 			goto out_filestream_unmount;
1745 	}
1746 
1747 	if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1748 		xfs_warn(mp,
1749 	"mounting with \"discard\" option, but the device does not support discard");
1750 		mp->m_features &= ~XFS_FEAT_DISCARD;
1751 	}
1752 
1753 	if (xfs_has_metadir(mp))
1754 		xfs_warn_experimental(mp, XFS_EXPERIMENTAL_METADIR);
1755 
1756 	if (xfs_has_reflink(mp)) {
1757 		if (mp->m_sb.sb_rblocks) {
1758 			xfs_alert(mp,
1759 	"reflink not compatible with realtime device!");
1760 			error = -EINVAL;
1761 			goto out_filestream_unmount;
1762 		}
1763 
1764 		if (xfs_globals.always_cow) {
1765 			xfs_info(mp, "using DEBUG-only always_cow mode.");
1766 			mp->m_always_cow = true;
1767 		}
1768 	}
1769 
1770 	if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1771 		xfs_alert(mp,
1772 	"reverse mapping btree not compatible with realtime device!");
1773 		error = -EINVAL;
1774 		goto out_filestream_unmount;
1775 	}
1776 
1777 	if (xfs_has_exchange_range(mp))
1778 		xfs_warn_experimental(mp, XFS_EXPERIMENTAL_EXCHRANGE);
1779 
1780 	if (xfs_has_parent(mp))
1781 		xfs_warn_experimental(mp, XFS_EXPERIMENTAL_PPTR);
1782 
1783 	/*
1784 	 * If no quota mount options were provided, maybe we'll try to pick
1785 	 * up the quota accounting and enforcement flags from the ondisk sb.
1786 	 */
1787 	if (!(mp->m_qflags & XFS_QFLAGS_MNTOPTS))
1788 		xfs_set_resuming_quotaon(mp);
1789 	mp->m_qflags &= ~XFS_QFLAGS_MNTOPTS;
1790 
1791 	error = xfs_mountfs(mp);
1792 	if (error)
1793 		goto out_filestream_unmount;
1794 
1795 	root = igrab(VFS_I(mp->m_rootip));
1796 	if (!root) {
1797 		error = -ENOENT;
1798 		goto out_unmount;
1799 	}
1800 	sb->s_root = d_make_root(root);
1801 	if (!sb->s_root) {
1802 		error = -ENOMEM;
1803 		goto out_unmount;
1804 	}
1805 
1806 	return 0;
1807 
1808  out_filestream_unmount:
1809 	xfs_filestream_unmount(mp);
1810  out_free_rtsb:
1811 	xfs_rtmount_freesb(mp);
1812  out_free_sb:
1813 	xfs_freesb(mp);
1814  out_free_scrub_stats:
1815 	xchk_mount_stats_free(mp);
1816  out_free_stats:
1817 	free_percpu(mp->m_stats.xs_stats);
1818  out_destroy_inodegc:
1819 	xfs_inodegc_free_percpu(mp);
1820  out_destroy_counters:
1821 	xfs_destroy_percpu_counters(mp);
1822  out_destroy_workqueues:
1823 	xfs_destroy_mount_workqueues(mp);
1824  out_shutdown_devices:
1825 	xfs_shutdown_devices(mp);
1826 	return error;
1827 
1828  out_unmount:
1829 	xfs_filestream_unmount(mp);
1830 	xfs_unmountfs(mp);
1831 	goto out_free_rtsb;
1832 }
1833 
1834 static int
1835 xfs_fs_get_tree(
1836 	struct fs_context	*fc)
1837 {
1838 	return get_tree_bdev(fc, xfs_fs_fill_super);
1839 }
1840 
1841 static int
1842 xfs_remount_rw(
1843 	struct xfs_mount	*mp)
1844 {
1845 	struct xfs_sb		*sbp = &mp->m_sb;
1846 	int error;
1847 
1848 	if (xfs_has_norecovery(mp)) {
1849 		xfs_warn(mp,
1850 			"ro->rw transition prohibited on norecovery mount");
1851 		return -EINVAL;
1852 	}
1853 
1854 	if (xfs_sb_is_v5(sbp) &&
1855 	    xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1856 		xfs_warn(mp,
1857 	"ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1858 			(sbp->sb_features_ro_compat &
1859 				XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1860 		return -EINVAL;
1861 	}
1862 
1863 	xfs_clear_readonly(mp);
1864 
1865 	/*
1866 	 * If this is the first remount to writeable state we might have some
1867 	 * superblock changes to update.
1868 	 */
1869 	if (mp->m_update_sb) {
1870 		error = xfs_sync_sb(mp, false);
1871 		if (error) {
1872 			xfs_warn(mp, "failed to write sb changes");
1873 			return error;
1874 		}
1875 		mp->m_update_sb = false;
1876 	}
1877 
1878 	/*
1879 	 * Fill out the reserve pool if it is empty. Use the stashed value if
1880 	 * it is non-zero, otherwise go with the default.
1881 	 */
1882 	xfs_restore_resvblks(mp);
1883 	xfs_log_work_queue(mp);
1884 	xfs_blockgc_start(mp);
1885 
1886 	/* Create the per-AG metadata reservation pool .*/
1887 	error = xfs_fs_reserve_ag_blocks(mp);
1888 	if (error && error != -ENOSPC)
1889 		return error;
1890 
1891 	/* Re-enable the background inode inactivation worker. */
1892 	xfs_inodegc_start(mp);
1893 
1894 	return 0;
1895 }
1896 
1897 static int
1898 xfs_remount_ro(
1899 	struct xfs_mount	*mp)
1900 {
1901 	struct xfs_icwalk	icw = {
1902 		.icw_flags	= XFS_ICWALK_FLAG_SYNC,
1903 	};
1904 	int			error;
1905 
1906 	/* Flush all the dirty data to disk. */
1907 	error = sync_filesystem(mp->m_super);
1908 	if (error)
1909 		return error;
1910 
1911 	/*
1912 	 * Cancel background eofb scanning so it cannot race with the final
1913 	 * log force+buftarg wait and deadlock the remount.
1914 	 */
1915 	xfs_blockgc_stop(mp);
1916 
1917 	/*
1918 	 * Clear out all remaining COW staging extents and speculative post-EOF
1919 	 * preallocations so that we don't leave inodes requiring inactivation
1920 	 * cleanups during reclaim on a read-only mount.  We must process every
1921 	 * cached inode, so this requires a synchronous cache scan.
1922 	 */
1923 	error = xfs_blockgc_free_space(mp, &icw);
1924 	if (error) {
1925 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1926 		return error;
1927 	}
1928 
1929 	/*
1930 	 * Stop the inodegc background worker.  xfs_fs_reconfigure already
1931 	 * flushed all pending inodegc work when it sync'd the filesystem.
1932 	 * The VFS holds s_umount, so we know that inodes cannot enter
1933 	 * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1934 	 * we send inodes straight to reclaim, so no inodes will be queued.
1935 	 */
1936 	xfs_inodegc_stop(mp);
1937 
1938 	/* Free the per-AG metadata reservation pool. */
1939 	xfs_fs_unreserve_ag_blocks(mp);
1940 
1941 	/*
1942 	 * Before we sync the metadata, we need to free up the reserve block
1943 	 * pool so that the used block count in the superblock on disk is
1944 	 * correct at the end of the remount. Stash the current* reserve pool
1945 	 * size so that if we get remounted rw, we can return it to the same
1946 	 * size.
1947 	 */
1948 	xfs_save_resvblks(mp);
1949 
1950 	xfs_log_clean(mp);
1951 	xfs_set_readonly(mp);
1952 
1953 	return 0;
1954 }
1955 
1956 /*
1957  * Logically we would return an error here to prevent users from believing
1958  * they might have changed mount options using remount which can't be changed.
1959  *
1960  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1961  * arguments in some cases so we can't blindly reject options, but have to
1962  * check for each specified option if it actually differs from the currently
1963  * set option and only reject it if that's the case.
1964  *
1965  * Until that is implemented we return success for every remount request, and
1966  * silently ignore all options that we can't actually change.
1967  */
1968 static int
1969 xfs_fs_reconfigure(
1970 	struct fs_context *fc)
1971 {
1972 	struct xfs_mount	*mp = XFS_M(fc->root->d_sb);
1973 	struct xfs_mount        *new_mp = fc->s_fs_info;
1974 	int			flags = fc->sb_flags;
1975 	int			error;
1976 
1977 	new_mp->m_qflags &= ~XFS_QFLAGS_MNTOPTS;
1978 
1979 	/* version 5 superblocks always support version counters. */
1980 	if (xfs_has_crc(mp))
1981 		fc->sb_flags |= SB_I_VERSION;
1982 
1983 	error = xfs_fs_validate_params(new_mp);
1984 	if (error)
1985 		return error;
1986 
1987 	/* inode32 -> inode64 */
1988 	if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1989 		mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1990 		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1991 	}
1992 
1993 	/* inode64 -> inode32 */
1994 	if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1995 		mp->m_features |= XFS_FEAT_SMALL_INUMS;
1996 		mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1997 	}
1998 
1999 	/* ro -> rw */
2000 	if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
2001 		error = xfs_remount_rw(mp);
2002 		if (error)
2003 			return error;
2004 	}
2005 
2006 	/* rw -> ro */
2007 	if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
2008 		error = xfs_remount_ro(mp);
2009 		if (error)
2010 			return error;
2011 	}
2012 
2013 	return 0;
2014 }
2015 
2016 static void
2017 xfs_fs_free(
2018 	struct fs_context	*fc)
2019 {
2020 	struct xfs_mount	*mp = fc->s_fs_info;
2021 
2022 	/*
2023 	 * mp is stored in the fs_context when it is initialized.
2024 	 * mp is transferred to the superblock on a successful mount,
2025 	 * but if an error occurs before the transfer we have to free
2026 	 * it here.
2027 	 */
2028 	if (mp)
2029 		xfs_mount_free(mp);
2030 }
2031 
2032 static const struct fs_context_operations xfs_context_ops = {
2033 	.parse_param = xfs_fs_parse_param,
2034 	.get_tree    = xfs_fs_get_tree,
2035 	.reconfigure = xfs_fs_reconfigure,
2036 	.free        = xfs_fs_free,
2037 };
2038 
2039 /*
2040  * WARNING: do not initialise any parameters in this function that depend on
2041  * mount option parsing having already been performed as this can be called from
2042  * fsopen() before any parameters have been set.
2043  */
2044 static int
2045 xfs_init_fs_context(
2046 	struct fs_context	*fc)
2047 {
2048 	struct xfs_mount	*mp;
2049 	int			i;
2050 
2051 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL | __GFP_NOFAIL);
2052 	if (!mp)
2053 		return -ENOMEM;
2054 
2055 	spin_lock_init(&mp->m_sb_lock);
2056 	for (i = 0; i < XG_TYPE_MAX; i++)
2057 		xa_init(&mp->m_groups[i].xa);
2058 	mutex_init(&mp->m_growlock);
2059 	INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
2060 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
2061 	mp->m_kobj.kobject.kset = xfs_kset;
2062 	/*
2063 	 * We don't create the finobt per-ag space reservation until after log
2064 	 * recovery, so we must set this to true so that an ifree transaction
2065 	 * started during log recovery will not depend on space reservations
2066 	 * for finobt expansion.
2067 	 */
2068 	mp->m_finobt_nores = true;
2069 
2070 	/*
2071 	 * These can be overridden by the mount option parsing.
2072 	 */
2073 	mp->m_logbufs = -1;
2074 	mp->m_logbsize = -1;
2075 	mp->m_allocsize_log = 16; /* 64k */
2076 
2077 	xfs_hooks_init(&mp->m_dir_update_hooks);
2078 
2079 	fc->s_fs_info = mp;
2080 	fc->ops = &xfs_context_ops;
2081 
2082 	return 0;
2083 }
2084 
2085 static void
2086 xfs_kill_sb(
2087 	struct super_block		*sb)
2088 {
2089 	kill_block_super(sb);
2090 	xfs_mount_free(XFS_M(sb));
2091 }
2092 
2093 static struct file_system_type xfs_fs_type = {
2094 	.owner			= THIS_MODULE,
2095 	.name			= "xfs",
2096 	.init_fs_context	= xfs_init_fs_context,
2097 	.parameters		= xfs_fs_parameters,
2098 	.kill_sb		= xfs_kill_sb,
2099 	.fs_flags		= FS_REQUIRES_DEV | FS_ALLOW_IDMAP | FS_MGTIME,
2100 };
2101 MODULE_ALIAS_FS("xfs");
2102 
2103 STATIC int __init
2104 xfs_init_caches(void)
2105 {
2106 	int		error;
2107 
2108 	xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2109 					 SLAB_HWCACHE_ALIGN |
2110 					 SLAB_RECLAIM_ACCOUNT,
2111 					 NULL);
2112 	if (!xfs_buf_cache)
2113 		goto out;
2114 
2115 	xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
2116 						sizeof(struct xlog_ticket),
2117 						0, 0, NULL);
2118 	if (!xfs_log_ticket_cache)
2119 		goto out_destroy_buf_cache;
2120 
2121 	error = xfs_btree_init_cur_caches();
2122 	if (error)
2123 		goto out_destroy_log_ticket_cache;
2124 
2125 	error = rcbagbt_init_cur_cache();
2126 	if (error)
2127 		goto out_destroy_btree_cur_cache;
2128 
2129 	error = xfs_defer_init_item_caches();
2130 	if (error)
2131 		goto out_destroy_rcbagbt_cur_cache;
2132 
2133 	xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2134 					      sizeof(struct xfs_da_state),
2135 					      0, 0, NULL);
2136 	if (!xfs_da_state_cache)
2137 		goto out_destroy_defer_item_cache;
2138 
2139 	xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2140 					   sizeof(struct xfs_ifork),
2141 					   0, 0, NULL);
2142 	if (!xfs_ifork_cache)
2143 		goto out_destroy_da_state_cache;
2144 
2145 	xfs_trans_cache = kmem_cache_create("xfs_trans",
2146 					   sizeof(struct xfs_trans),
2147 					   0, 0, NULL);
2148 	if (!xfs_trans_cache)
2149 		goto out_destroy_ifork_cache;
2150 
2151 
2152 	/*
2153 	 * The size of the cache-allocated buf log item is the maximum
2154 	 * size possible under XFS.  This wastes a little bit of memory,
2155 	 * but it is much faster.
2156 	 */
2157 	xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2158 					      sizeof(struct xfs_buf_log_item),
2159 					      0, 0, NULL);
2160 	if (!xfs_buf_item_cache)
2161 		goto out_destroy_trans_cache;
2162 
2163 	xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2164 			xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2165 			0, 0, NULL);
2166 	if (!xfs_efd_cache)
2167 		goto out_destroy_buf_item_cache;
2168 
2169 	xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2170 			xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2171 			0, 0, NULL);
2172 	if (!xfs_efi_cache)
2173 		goto out_destroy_efd_cache;
2174 
2175 	xfs_inode_cache = kmem_cache_create("xfs_inode",
2176 					   sizeof(struct xfs_inode), 0,
2177 					   (SLAB_HWCACHE_ALIGN |
2178 					    SLAB_RECLAIM_ACCOUNT |
2179 					    SLAB_ACCOUNT),
2180 					   xfs_fs_inode_init_once);
2181 	if (!xfs_inode_cache)
2182 		goto out_destroy_efi_cache;
2183 
2184 	xfs_ili_cache = kmem_cache_create("xfs_ili",
2185 					 sizeof(struct xfs_inode_log_item), 0,
2186 					 SLAB_RECLAIM_ACCOUNT,
2187 					 NULL);
2188 	if (!xfs_ili_cache)
2189 		goto out_destroy_inode_cache;
2190 
2191 	xfs_icreate_cache = kmem_cache_create("xfs_icr",
2192 					     sizeof(struct xfs_icreate_item),
2193 					     0, 0, NULL);
2194 	if (!xfs_icreate_cache)
2195 		goto out_destroy_ili_cache;
2196 
2197 	xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2198 					 sizeof(struct xfs_rud_log_item),
2199 					 0, 0, NULL);
2200 	if (!xfs_rud_cache)
2201 		goto out_destroy_icreate_cache;
2202 
2203 	xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2204 			xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2205 			0, 0, NULL);
2206 	if (!xfs_rui_cache)
2207 		goto out_destroy_rud_cache;
2208 
2209 	xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2210 					 sizeof(struct xfs_cud_log_item),
2211 					 0, 0, NULL);
2212 	if (!xfs_cud_cache)
2213 		goto out_destroy_rui_cache;
2214 
2215 	xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2216 			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2217 			0, 0, NULL);
2218 	if (!xfs_cui_cache)
2219 		goto out_destroy_cud_cache;
2220 
2221 	xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2222 					 sizeof(struct xfs_bud_log_item),
2223 					 0, 0, NULL);
2224 	if (!xfs_bud_cache)
2225 		goto out_destroy_cui_cache;
2226 
2227 	xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2228 			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2229 			0, 0, NULL);
2230 	if (!xfs_bui_cache)
2231 		goto out_destroy_bud_cache;
2232 
2233 	xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2234 					    sizeof(struct xfs_attrd_log_item),
2235 					    0, 0, NULL);
2236 	if (!xfs_attrd_cache)
2237 		goto out_destroy_bui_cache;
2238 
2239 	xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2240 					    sizeof(struct xfs_attri_log_item),
2241 					    0, 0, NULL);
2242 	if (!xfs_attri_cache)
2243 		goto out_destroy_attrd_cache;
2244 
2245 	xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2246 					     sizeof(struct xfs_iunlink_item),
2247 					     0, 0, NULL);
2248 	if (!xfs_iunlink_cache)
2249 		goto out_destroy_attri_cache;
2250 
2251 	xfs_xmd_cache = kmem_cache_create("xfs_xmd_item",
2252 					 sizeof(struct xfs_xmd_log_item),
2253 					 0, 0, NULL);
2254 	if (!xfs_xmd_cache)
2255 		goto out_destroy_iul_cache;
2256 
2257 	xfs_xmi_cache = kmem_cache_create("xfs_xmi_item",
2258 					 sizeof(struct xfs_xmi_log_item),
2259 					 0, 0, NULL);
2260 	if (!xfs_xmi_cache)
2261 		goto out_destroy_xmd_cache;
2262 
2263 	xfs_parent_args_cache = kmem_cache_create("xfs_parent_args",
2264 					     sizeof(struct xfs_parent_args),
2265 					     0, 0, NULL);
2266 	if (!xfs_parent_args_cache)
2267 		goto out_destroy_xmi_cache;
2268 
2269 	return 0;
2270 
2271  out_destroy_xmi_cache:
2272 	kmem_cache_destroy(xfs_xmi_cache);
2273  out_destroy_xmd_cache:
2274 	kmem_cache_destroy(xfs_xmd_cache);
2275  out_destroy_iul_cache:
2276 	kmem_cache_destroy(xfs_iunlink_cache);
2277  out_destroy_attri_cache:
2278 	kmem_cache_destroy(xfs_attri_cache);
2279  out_destroy_attrd_cache:
2280 	kmem_cache_destroy(xfs_attrd_cache);
2281  out_destroy_bui_cache:
2282 	kmem_cache_destroy(xfs_bui_cache);
2283  out_destroy_bud_cache:
2284 	kmem_cache_destroy(xfs_bud_cache);
2285  out_destroy_cui_cache:
2286 	kmem_cache_destroy(xfs_cui_cache);
2287  out_destroy_cud_cache:
2288 	kmem_cache_destroy(xfs_cud_cache);
2289  out_destroy_rui_cache:
2290 	kmem_cache_destroy(xfs_rui_cache);
2291  out_destroy_rud_cache:
2292 	kmem_cache_destroy(xfs_rud_cache);
2293  out_destroy_icreate_cache:
2294 	kmem_cache_destroy(xfs_icreate_cache);
2295  out_destroy_ili_cache:
2296 	kmem_cache_destroy(xfs_ili_cache);
2297  out_destroy_inode_cache:
2298 	kmem_cache_destroy(xfs_inode_cache);
2299  out_destroy_efi_cache:
2300 	kmem_cache_destroy(xfs_efi_cache);
2301  out_destroy_efd_cache:
2302 	kmem_cache_destroy(xfs_efd_cache);
2303  out_destroy_buf_item_cache:
2304 	kmem_cache_destroy(xfs_buf_item_cache);
2305  out_destroy_trans_cache:
2306 	kmem_cache_destroy(xfs_trans_cache);
2307  out_destroy_ifork_cache:
2308 	kmem_cache_destroy(xfs_ifork_cache);
2309  out_destroy_da_state_cache:
2310 	kmem_cache_destroy(xfs_da_state_cache);
2311  out_destroy_defer_item_cache:
2312 	xfs_defer_destroy_item_caches();
2313  out_destroy_rcbagbt_cur_cache:
2314 	rcbagbt_destroy_cur_cache();
2315  out_destroy_btree_cur_cache:
2316 	xfs_btree_destroy_cur_caches();
2317  out_destroy_log_ticket_cache:
2318 	kmem_cache_destroy(xfs_log_ticket_cache);
2319  out_destroy_buf_cache:
2320 	kmem_cache_destroy(xfs_buf_cache);
2321  out:
2322 	return -ENOMEM;
2323 }
2324 
2325 STATIC void
2326 xfs_destroy_caches(void)
2327 {
2328 	/*
2329 	 * Make sure all delayed rcu free are flushed before we
2330 	 * destroy caches.
2331 	 */
2332 	rcu_barrier();
2333 	kmem_cache_destroy(xfs_parent_args_cache);
2334 	kmem_cache_destroy(xfs_xmd_cache);
2335 	kmem_cache_destroy(xfs_xmi_cache);
2336 	kmem_cache_destroy(xfs_iunlink_cache);
2337 	kmem_cache_destroy(xfs_attri_cache);
2338 	kmem_cache_destroy(xfs_attrd_cache);
2339 	kmem_cache_destroy(xfs_bui_cache);
2340 	kmem_cache_destroy(xfs_bud_cache);
2341 	kmem_cache_destroy(xfs_cui_cache);
2342 	kmem_cache_destroy(xfs_cud_cache);
2343 	kmem_cache_destroy(xfs_rui_cache);
2344 	kmem_cache_destroy(xfs_rud_cache);
2345 	kmem_cache_destroy(xfs_icreate_cache);
2346 	kmem_cache_destroy(xfs_ili_cache);
2347 	kmem_cache_destroy(xfs_inode_cache);
2348 	kmem_cache_destroy(xfs_efi_cache);
2349 	kmem_cache_destroy(xfs_efd_cache);
2350 	kmem_cache_destroy(xfs_buf_item_cache);
2351 	kmem_cache_destroy(xfs_trans_cache);
2352 	kmem_cache_destroy(xfs_ifork_cache);
2353 	kmem_cache_destroy(xfs_da_state_cache);
2354 	xfs_defer_destroy_item_caches();
2355 	rcbagbt_destroy_cur_cache();
2356 	xfs_btree_destroy_cur_caches();
2357 	kmem_cache_destroy(xfs_log_ticket_cache);
2358 	kmem_cache_destroy(xfs_buf_cache);
2359 }
2360 
2361 STATIC int __init
2362 xfs_init_workqueues(void)
2363 {
2364 	/*
2365 	 * The allocation workqueue can be used in memory reclaim situations
2366 	 * (writepage path), and parallelism is only limited by the number of
2367 	 * AGs in all the filesystems mounted. Hence use the default large
2368 	 * max_active value for this workqueue.
2369 	 */
2370 	xfs_alloc_wq = alloc_workqueue("xfsalloc",
2371 			XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2372 	if (!xfs_alloc_wq)
2373 		return -ENOMEM;
2374 
2375 	xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2376 			0);
2377 	if (!xfs_discard_wq)
2378 		goto out_free_alloc_wq;
2379 
2380 	return 0;
2381 out_free_alloc_wq:
2382 	destroy_workqueue(xfs_alloc_wq);
2383 	return -ENOMEM;
2384 }
2385 
2386 STATIC void
2387 xfs_destroy_workqueues(void)
2388 {
2389 	destroy_workqueue(xfs_discard_wq);
2390 	destroy_workqueue(xfs_alloc_wq);
2391 }
2392 
2393 STATIC int __init
2394 init_xfs_fs(void)
2395 {
2396 	int			error;
2397 
2398 	xfs_check_ondisk_structs();
2399 
2400 	error = xfs_dahash_test();
2401 	if (error)
2402 		return error;
2403 
2404 	printk(KERN_INFO XFS_VERSION_STRING " with "
2405 			 XFS_BUILD_OPTIONS " enabled\n");
2406 
2407 	xfs_dir_startup();
2408 
2409 	error = xfs_init_caches();
2410 	if (error)
2411 		goto out;
2412 
2413 	error = xfs_init_workqueues();
2414 	if (error)
2415 		goto out_destroy_caches;
2416 
2417 	error = xfs_mru_cache_init();
2418 	if (error)
2419 		goto out_destroy_wq;
2420 
2421 	error = xfs_init_procfs();
2422 	if (error)
2423 		goto out_mru_cache_uninit;
2424 
2425 	error = xfs_sysctl_register();
2426 	if (error)
2427 		goto out_cleanup_procfs;
2428 
2429 	xfs_debugfs = xfs_debugfs_mkdir("xfs", NULL);
2430 
2431 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2432 	if (!xfs_kset) {
2433 		error = -ENOMEM;
2434 		goto out_debugfs_unregister;
2435 	}
2436 
2437 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
2438 
2439 	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2440 	if (!xfsstats.xs_stats) {
2441 		error = -ENOMEM;
2442 		goto out_kset_unregister;
2443 	}
2444 
2445 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2446 			       "stats");
2447 	if (error)
2448 		goto out_free_stats;
2449 
2450 	error = xchk_global_stats_setup(xfs_debugfs);
2451 	if (error)
2452 		goto out_remove_stats_kobj;
2453 
2454 #ifdef DEBUG
2455 	xfs_dbg_kobj.kobject.kset = xfs_kset;
2456 	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2457 	if (error)
2458 		goto out_remove_scrub_stats;
2459 #endif
2460 
2461 	error = xfs_qm_init();
2462 	if (error)
2463 		goto out_remove_dbg_kobj;
2464 
2465 	error = register_filesystem(&xfs_fs_type);
2466 	if (error)
2467 		goto out_qm_exit;
2468 	return 0;
2469 
2470  out_qm_exit:
2471 	xfs_qm_exit();
2472  out_remove_dbg_kobj:
2473 #ifdef DEBUG
2474 	xfs_sysfs_del(&xfs_dbg_kobj);
2475  out_remove_scrub_stats:
2476 #endif
2477 	xchk_global_stats_teardown();
2478  out_remove_stats_kobj:
2479 	xfs_sysfs_del(&xfsstats.xs_kobj);
2480  out_free_stats:
2481 	free_percpu(xfsstats.xs_stats);
2482  out_kset_unregister:
2483 	kset_unregister(xfs_kset);
2484  out_debugfs_unregister:
2485 	debugfs_remove(xfs_debugfs);
2486 	xfs_sysctl_unregister();
2487  out_cleanup_procfs:
2488 	xfs_cleanup_procfs();
2489  out_mru_cache_uninit:
2490 	xfs_mru_cache_uninit();
2491  out_destroy_wq:
2492 	xfs_destroy_workqueues();
2493  out_destroy_caches:
2494 	xfs_destroy_caches();
2495  out:
2496 	return error;
2497 }
2498 
2499 STATIC void __exit
2500 exit_xfs_fs(void)
2501 {
2502 	xfs_qm_exit();
2503 	unregister_filesystem(&xfs_fs_type);
2504 #ifdef DEBUG
2505 	xfs_sysfs_del(&xfs_dbg_kobj);
2506 #endif
2507 	xchk_global_stats_teardown();
2508 	xfs_sysfs_del(&xfsstats.xs_kobj);
2509 	free_percpu(xfsstats.xs_stats);
2510 	kset_unregister(xfs_kset);
2511 	debugfs_remove(xfs_debugfs);
2512 	xfs_sysctl_unregister();
2513 	xfs_cleanup_procfs();
2514 	xfs_mru_cache_uninit();
2515 	xfs_destroy_workqueues();
2516 	xfs_destroy_caches();
2517 	xfs_uuid_table_free();
2518 }
2519 
2520 module_init(init_xfs_fs);
2521 module_exit(exit_xfs_fs);
2522 
2523 MODULE_AUTHOR("Silicon Graphics, Inc.");
2524 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2525 MODULE_LICENSE("GPL");
2526