xref: /linux/fs/xfs/xfs_super.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 
19 #include "xfs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
32 #include "xfs_fsops.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_log.h"
36 #include "xfs_log_priv.h"
37 #include "xfs_da_btree.h"
38 #include "xfs_dir2.h"
39 #include "xfs_extfree_item.h"
40 #include "xfs_mru_cache.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_icache.h"
43 #include "xfs_trace.h"
44 #include "xfs_icreate_item.h"
45 #include "xfs_filestream.h"
46 #include "xfs_quota.h"
47 #include "xfs_sysfs.h"
48 #include "xfs_ondisk.h"
49 #include "xfs_rmap_item.h"
50 #include "xfs_refcount_item.h"
51 #include "xfs_bmap_item.h"
52 #include "xfs_reflink.h"
53 
54 #include <linux/namei.h>
55 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mount.h>
58 #include <linux/mempool.h>
59 #include <linux/writeback.h>
60 #include <linux/kthread.h>
61 #include <linux/freezer.h>
62 #include <linux/parser.h>
63 
64 static const struct super_operations xfs_super_operations;
65 struct bio_set *xfs_ioend_bioset;
66 
67 static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
68 #ifdef DEBUG
69 static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
70 #endif
71 
72 /*
73  * Table driven mount option parser.
74  */
75 enum {
76 	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
77 	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
78 	Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
79 	Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
80 	Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
81 	Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
82 	Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
83 	Opt_uquota, Opt_gquota, Opt_pquota,
84 	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
85 	Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
86 };
87 
88 static const match_table_t tokens = {
89 	{Opt_logbufs,	"logbufs=%u"},	/* number of XFS log buffers */
90 	{Opt_logbsize,	"logbsize=%s"},	/* size of XFS log buffers */
91 	{Opt_logdev,	"logdev=%s"},	/* log device */
92 	{Opt_rtdev,	"rtdev=%s"},	/* realtime I/O device */
93 	{Opt_biosize,	"biosize=%u"},	/* log2 of preferred buffered io size */
94 	{Opt_wsync,	"wsync"},	/* safe-mode nfs compatible mount */
95 	{Opt_noalign,	"noalign"},	/* turn off stripe alignment */
96 	{Opt_swalloc,	"swalloc"},	/* turn on stripe width allocation */
97 	{Opt_sunit,	"sunit=%u"},	/* data volume stripe unit */
98 	{Opt_swidth,	"swidth=%u"},	/* data volume stripe width */
99 	{Opt_nouuid,	"nouuid"},	/* ignore filesystem UUID */
100 	{Opt_mtpt,	"mtpt"},	/* filesystem mount point */
101 	{Opt_grpid,	"grpid"},	/* group-ID from parent directory */
102 	{Opt_nogrpid,	"nogrpid"},	/* group-ID from current process */
103 	{Opt_bsdgroups,	"bsdgroups"},	/* group-ID from parent directory */
104 	{Opt_sysvgroups,"sysvgroups"},	/* group-ID from current process */
105 	{Opt_allocsize,	"allocsize=%s"},/* preferred allocation size */
106 	{Opt_norecovery,"norecovery"},	/* don't run XFS recovery */
107 	{Opt_inode64,	"inode64"},	/* inodes can be allocated anywhere */
108 	{Opt_inode32,   "inode32"},	/* inode allocation limited to
109 					 * XFS_MAXINUMBER_32 */
110 	{Opt_ikeep,	"ikeep"},	/* do not free empty inode clusters */
111 	{Opt_noikeep,	"noikeep"},	/* free empty inode clusters */
112 	{Opt_largeio,	"largeio"},	/* report large I/O sizes in stat() */
113 	{Opt_nolargeio,	"nolargeio"},	/* do not report large I/O sizes
114 					 * in stat(). */
115 	{Opt_attr2,	"attr2"},	/* do use attr2 attribute format */
116 	{Opt_noattr2,	"noattr2"},	/* do not use attr2 attribute format */
117 	{Opt_filestreams,"filestreams"},/* use filestreams allocator */
118 	{Opt_quota,	"quota"},	/* disk quotas (user) */
119 	{Opt_noquota,	"noquota"},	/* no quotas */
120 	{Opt_usrquota,	"usrquota"},	/* user quota enabled */
121 	{Opt_grpquota,	"grpquota"},	/* group quota enabled */
122 	{Opt_prjquota,	"prjquota"},	/* project quota enabled */
123 	{Opt_uquota,	"uquota"},	/* user quota (IRIX variant) */
124 	{Opt_gquota,	"gquota"},	/* group quota (IRIX variant) */
125 	{Opt_pquota,	"pquota"},	/* project quota (IRIX variant) */
126 	{Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
127 	{Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
128 	{Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
129 	{Opt_qnoenforce, "qnoenforce"},	/* same as uqnoenforce */
130 	{Opt_discard,	"discard"},	/* Discard unused blocks */
131 	{Opt_nodiscard,	"nodiscard"},	/* Do not discard unused blocks */
132 
133 	{Opt_dax,	"dax"},		/* Enable direct access to bdev pages */
134 
135 	/* Deprecated mount options scheduled for removal */
136 	{Opt_barrier,	"barrier"},	/* use writer barriers for log write and
137 					 * unwritten extent conversion */
138 	{Opt_nobarrier,	"nobarrier"},	/* .. disable */
139 
140 	{Opt_err,	NULL},
141 };
142 
143 
144 STATIC int
145 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
146 {
147 	int	last, shift_left_factor = 0, _res;
148 	char	*value;
149 	int	ret = 0;
150 
151 	value = match_strdup(s);
152 	if (!value)
153 		return -ENOMEM;
154 
155 	last = strlen(value) - 1;
156 	if (value[last] == 'K' || value[last] == 'k') {
157 		shift_left_factor = 10;
158 		value[last] = '\0';
159 	}
160 	if (value[last] == 'M' || value[last] == 'm') {
161 		shift_left_factor = 20;
162 		value[last] = '\0';
163 	}
164 	if (value[last] == 'G' || value[last] == 'g') {
165 		shift_left_factor = 30;
166 		value[last] = '\0';
167 	}
168 
169 	if (kstrtoint(value, base, &_res))
170 		ret = -EINVAL;
171 	kfree(value);
172 	*res = _res << shift_left_factor;
173 	return ret;
174 }
175 
176 /*
177  * This function fills in xfs_mount_t fields based on mount args.
178  * Note: the superblock has _not_ yet been read in.
179  *
180  * Note that this function leaks the various device name allocations on
181  * failure.  The caller takes care of them.
182  *
183  * *sb is const because this is also used to test options on the remount
184  * path, and we don't want this to have any side effects at remount time.
185  * Today this function does not change *sb, but just to future-proof...
186  */
187 STATIC int
188 xfs_parseargs(
189 	struct xfs_mount	*mp,
190 	char			*options)
191 {
192 	const struct super_block *sb = mp->m_super;
193 	char			*p;
194 	substring_t		args[MAX_OPT_ARGS];
195 	int			dsunit = 0;
196 	int			dswidth = 0;
197 	int			iosize = 0;
198 	__uint8_t		iosizelog = 0;
199 
200 	/*
201 	 * set up the mount name first so all the errors will refer to the
202 	 * correct device.
203 	 */
204 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
205 	if (!mp->m_fsname)
206 		return -ENOMEM;
207 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
208 
209 	/*
210 	 * Copy binary VFS mount flags we are interested in.
211 	 */
212 	if (sb->s_flags & MS_RDONLY)
213 		mp->m_flags |= XFS_MOUNT_RDONLY;
214 	if (sb->s_flags & MS_DIRSYNC)
215 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
216 	if (sb->s_flags & MS_SYNCHRONOUS)
217 		mp->m_flags |= XFS_MOUNT_WSYNC;
218 
219 	/*
220 	 * Set some default flags that could be cleared by the mount option
221 	 * parsing.
222 	 */
223 	mp->m_flags |= XFS_MOUNT_BARRIER;
224 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
225 
226 	/*
227 	 * These can be overridden by the mount option parsing.
228 	 */
229 	mp->m_logbufs = -1;
230 	mp->m_logbsize = -1;
231 
232 	if (!options)
233 		goto done;
234 
235 	while ((p = strsep(&options, ",")) != NULL) {
236 		int		token;
237 
238 		if (!*p)
239 			continue;
240 
241 		token = match_token(p, tokens, args);
242 		switch (token) {
243 		case Opt_logbufs:
244 			if (match_int(args, &mp->m_logbufs))
245 				return -EINVAL;
246 			break;
247 		case Opt_logbsize:
248 			if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
249 				return -EINVAL;
250 			break;
251 		case Opt_logdev:
252 			mp->m_logname = match_strdup(args);
253 			if (!mp->m_logname)
254 				return -ENOMEM;
255 			break;
256 		case Opt_mtpt:
257 			xfs_warn(mp, "%s option not allowed on this system", p);
258 			return -EINVAL;
259 		case Opt_rtdev:
260 			mp->m_rtname = match_strdup(args);
261 			if (!mp->m_rtname)
262 				return -ENOMEM;
263 			break;
264 		case Opt_allocsize:
265 		case Opt_biosize:
266 			if (suffix_kstrtoint(args, 10, &iosize))
267 				return -EINVAL;
268 			iosizelog = ffs(iosize) - 1;
269 			break;
270 		case Opt_grpid:
271 		case Opt_bsdgroups:
272 			mp->m_flags |= XFS_MOUNT_GRPID;
273 			break;
274 		case Opt_nogrpid:
275 		case Opt_sysvgroups:
276 			mp->m_flags &= ~XFS_MOUNT_GRPID;
277 			break;
278 		case Opt_wsync:
279 			mp->m_flags |= XFS_MOUNT_WSYNC;
280 			break;
281 		case Opt_norecovery:
282 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
283 			break;
284 		case Opt_noalign:
285 			mp->m_flags |= XFS_MOUNT_NOALIGN;
286 			break;
287 		case Opt_swalloc:
288 			mp->m_flags |= XFS_MOUNT_SWALLOC;
289 			break;
290 		case Opt_sunit:
291 			if (match_int(args, &dsunit))
292 				return -EINVAL;
293 			break;
294 		case Opt_swidth:
295 			if (match_int(args, &dswidth))
296 				return -EINVAL;
297 			break;
298 		case Opt_inode32:
299 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
300 			break;
301 		case Opt_inode64:
302 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
303 			break;
304 		case Opt_nouuid:
305 			mp->m_flags |= XFS_MOUNT_NOUUID;
306 			break;
307 		case Opt_ikeep:
308 			mp->m_flags |= XFS_MOUNT_IKEEP;
309 			break;
310 		case Opt_noikeep:
311 			mp->m_flags &= ~XFS_MOUNT_IKEEP;
312 			break;
313 		case Opt_largeio:
314 			mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
315 			break;
316 		case Opt_nolargeio:
317 			mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
318 			break;
319 		case Opt_attr2:
320 			mp->m_flags |= XFS_MOUNT_ATTR2;
321 			break;
322 		case Opt_noattr2:
323 			mp->m_flags &= ~XFS_MOUNT_ATTR2;
324 			mp->m_flags |= XFS_MOUNT_NOATTR2;
325 			break;
326 		case Opt_filestreams:
327 			mp->m_flags |= XFS_MOUNT_FILESTREAMS;
328 			break;
329 		case Opt_noquota:
330 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
331 			mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
332 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
333 			break;
334 		case Opt_quota:
335 		case Opt_uquota:
336 		case Opt_usrquota:
337 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
338 					 XFS_UQUOTA_ENFD);
339 			break;
340 		case Opt_qnoenforce:
341 		case Opt_uqnoenforce:
342 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
343 			mp->m_qflags &= ~XFS_UQUOTA_ENFD;
344 			break;
345 		case Opt_pquota:
346 		case Opt_prjquota:
347 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
348 					 XFS_PQUOTA_ENFD);
349 			break;
350 		case Opt_pqnoenforce:
351 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
352 			mp->m_qflags &= ~XFS_PQUOTA_ENFD;
353 			break;
354 		case Opt_gquota:
355 		case Opt_grpquota:
356 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
357 					 XFS_GQUOTA_ENFD);
358 			break;
359 		case Opt_gqnoenforce:
360 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
361 			mp->m_qflags &= ~XFS_GQUOTA_ENFD;
362 			break;
363 		case Opt_discard:
364 			mp->m_flags |= XFS_MOUNT_DISCARD;
365 			break;
366 		case Opt_nodiscard:
367 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
368 			break;
369 #ifdef CONFIG_FS_DAX
370 		case Opt_dax:
371 			mp->m_flags |= XFS_MOUNT_DAX;
372 			break;
373 #endif
374 		case Opt_barrier:
375 			xfs_warn(mp, "%s option is deprecated, ignoring.", p);
376 			mp->m_flags |= XFS_MOUNT_BARRIER;
377 			break;
378 		case Opt_nobarrier:
379 			xfs_warn(mp, "%s option is deprecated, ignoring.", p);
380 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
381 			break;
382 		default:
383 			xfs_warn(mp, "unknown mount option [%s].", p);
384 			return -EINVAL;
385 		}
386 	}
387 
388 	/*
389 	 * no recovery flag requires a read-only mount
390 	 */
391 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
392 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
393 		xfs_warn(mp, "no-recovery mounts must be read-only.");
394 		return -EINVAL;
395 	}
396 
397 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
398 		xfs_warn(mp,
399 	"sunit and swidth options incompatible with the noalign option");
400 		return -EINVAL;
401 	}
402 
403 #ifndef CONFIG_XFS_QUOTA
404 	if (XFS_IS_QUOTA_RUNNING(mp)) {
405 		xfs_warn(mp, "quota support not available in this kernel.");
406 		return -EINVAL;
407 	}
408 #endif
409 
410 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
411 		xfs_warn(mp, "sunit and swidth must be specified together");
412 		return -EINVAL;
413 	}
414 
415 	if (dsunit && (dswidth % dsunit != 0)) {
416 		xfs_warn(mp,
417 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
418 			dswidth, dsunit);
419 		return -EINVAL;
420 	}
421 
422 done:
423 	if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
424 		/*
425 		 * At this point the superblock has not been read
426 		 * in, therefore we do not know the block size.
427 		 * Before the mount call ends we will convert
428 		 * these to FSBs.
429 		 */
430 		mp->m_dalign = dsunit;
431 		mp->m_swidth = dswidth;
432 	}
433 
434 	if (mp->m_logbufs != -1 &&
435 	    mp->m_logbufs != 0 &&
436 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
437 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
438 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
439 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
440 		return -EINVAL;
441 	}
442 	if (mp->m_logbsize != -1 &&
443 	    mp->m_logbsize !=  0 &&
444 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
445 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
446 	     !is_power_of_2(mp->m_logbsize))) {
447 		xfs_warn(mp,
448 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
449 			mp->m_logbsize);
450 		return -EINVAL;
451 	}
452 
453 	if (iosizelog) {
454 		if (iosizelog > XFS_MAX_IO_LOG ||
455 		    iosizelog < XFS_MIN_IO_LOG) {
456 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
457 				iosizelog, XFS_MIN_IO_LOG,
458 				XFS_MAX_IO_LOG);
459 			return -EINVAL;
460 		}
461 
462 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
463 		mp->m_readio_log = iosizelog;
464 		mp->m_writeio_log = iosizelog;
465 	}
466 
467 	return 0;
468 }
469 
470 struct proc_xfs_info {
471 	uint64_t	flag;
472 	char		*str;
473 };
474 
475 STATIC int
476 xfs_showargs(
477 	struct xfs_mount	*mp,
478 	struct seq_file		*m)
479 {
480 	static struct proc_xfs_info xfs_info_set[] = {
481 		/* the few simple ones we can get from the mount struct */
482 		{ XFS_MOUNT_IKEEP,		",ikeep" },
483 		{ XFS_MOUNT_WSYNC,		",wsync" },
484 		{ XFS_MOUNT_NOALIGN,		",noalign" },
485 		{ XFS_MOUNT_SWALLOC,		",swalloc" },
486 		{ XFS_MOUNT_NOUUID,		",nouuid" },
487 		{ XFS_MOUNT_NORECOVERY,		",norecovery" },
488 		{ XFS_MOUNT_ATTR2,		",attr2" },
489 		{ XFS_MOUNT_FILESTREAMS,	",filestreams" },
490 		{ XFS_MOUNT_GRPID,		",grpid" },
491 		{ XFS_MOUNT_DISCARD,		",discard" },
492 		{ XFS_MOUNT_SMALL_INUMS,	",inode32" },
493 		{ XFS_MOUNT_DAX,		",dax" },
494 		{ 0, NULL }
495 	};
496 	static struct proc_xfs_info xfs_info_unset[] = {
497 		/* the few simple ones we can get from the mount struct */
498 		{ XFS_MOUNT_COMPAT_IOSIZE,	",largeio" },
499 		{ XFS_MOUNT_BARRIER,		",nobarrier" },
500 		{ XFS_MOUNT_SMALL_INUMS,	",inode64" },
501 		{ 0, NULL }
502 	};
503 	struct proc_xfs_info	*xfs_infop;
504 
505 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
506 		if (mp->m_flags & xfs_infop->flag)
507 			seq_puts(m, xfs_infop->str);
508 	}
509 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
510 		if (!(mp->m_flags & xfs_infop->flag))
511 			seq_puts(m, xfs_infop->str);
512 	}
513 
514 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
515 		seq_printf(m, ",allocsize=%dk",
516 				(int)(1 << mp->m_writeio_log) >> 10);
517 
518 	if (mp->m_logbufs > 0)
519 		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
520 	if (mp->m_logbsize > 0)
521 		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
522 
523 	if (mp->m_logname)
524 		seq_show_option(m, "logdev", mp->m_logname);
525 	if (mp->m_rtname)
526 		seq_show_option(m, "rtdev", mp->m_rtname);
527 
528 	if (mp->m_dalign > 0)
529 		seq_printf(m, ",sunit=%d",
530 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
531 	if (mp->m_swidth > 0)
532 		seq_printf(m, ",swidth=%d",
533 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
534 
535 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
536 		seq_puts(m, ",usrquota");
537 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
538 		seq_puts(m, ",uqnoenforce");
539 
540 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
541 		if (mp->m_qflags & XFS_PQUOTA_ENFD)
542 			seq_puts(m, ",prjquota");
543 		else
544 			seq_puts(m, ",pqnoenforce");
545 	}
546 	if (mp->m_qflags & XFS_GQUOTA_ACCT) {
547 		if (mp->m_qflags & XFS_GQUOTA_ENFD)
548 			seq_puts(m, ",grpquota");
549 		else
550 			seq_puts(m, ",gqnoenforce");
551 	}
552 
553 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
554 		seq_puts(m, ",noquota");
555 
556 	return 0;
557 }
558 static __uint64_t
559 xfs_max_file_offset(
560 	unsigned int		blockshift)
561 {
562 	unsigned int		pagefactor = 1;
563 	unsigned int		bitshift = BITS_PER_LONG - 1;
564 
565 	/* Figure out maximum filesize, on Linux this can depend on
566 	 * the filesystem blocksize (on 32 bit platforms).
567 	 * __block_write_begin does this in an [unsigned] long...
568 	 *      page->index << (PAGE_SHIFT - bbits)
569 	 * So, for page sized blocks (4K on 32 bit platforms),
570 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
571 	 *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
572 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
573 	 * Note1: get_block_t takes a long (implicit cast from above)
574 	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
575 	 * can optionally convert the [unsigned] long from above into
576 	 * an [unsigned] long long.
577 	 */
578 
579 #if BITS_PER_LONG == 32
580 # if defined(CONFIG_LBDAF)
581 	ASSERT(sizeof(sector_t) == 8);
582 	pagefactor = PAGE_SIZE;
583 	bitshift = BITS_PER_LONG;
584 # else
585 	pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
586 # endif
587 #endif
588 
589 	return (((__uint64_t)pagefactor) << bitshift) - 1;
590 }
591 
592 /*
593  * Set parameters for inode allocation heuristics, taking into account
594  * filesystem size and inode32/inode64 mount options; i.e. specifically
595  * whether or not XFS_MOUNT_SMALL_INUMS is set.
596  *
597  * Inode allocation patterns are altered only if inode32 is requested
598  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
599  * If altered, XFS_MOUNT_32BITINODES is set as well.
600  *
601  * An agcount independent of that in the mount structure is provided
602  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
603  * to the potentially higher ag count.
604  *
605  * Returns the maximum AG index which may contain inodes.
606  */
607 xfs_agnumber_t
608 xfs_set_inode_alloc(
609 	struct xfs_mount *mp,
610 	xfs_agnumber_t	agcount)
611 {
612 	xfs_agnumber_t	index;
613 	xfs_agnumber_t	maxagi = 0;
614 	xfs_sb_t	*sbp = &mp->m_sb;
615 	xfs_agnumber_t	max_metadata;
616 	xfs_agino_t	agino;
617 	xfs_ino_t	ino;
618 
619 	/*
620 	 * Calculate how much should be reserved for inodes to meet
621 	 * the max inode percentage.  Used only for inode32.
622 	 */
623 	if (mp->m_maxicount) {
624 		__uint64_t	icount;
625 
626 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
627 		do_div(icount, 100);
628 		icount += sbp->sb_agblocks - 1;
629 		do_div(icount, sbp->sb_agblocks);
630 		max_metadata = icount;
631 	} else {
632 		max_metadata = agcount;
633 	}
634 
635 	/* Get the last possible inode in the filesystem */
636 	agino =	XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
637 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
638 
639 	/*
640 	 * If user asked for no more than 32-bit inodes, and the fs is
641 	 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
642 	 * the allocator to accommodate the request.
643 	 */
644 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
645 		mp->m_flags |= XFS_MOUNT_32BITINODES;
646 	else
647 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
648 
649 	for (index = 0; index < agcount; index++) {
650 		struct xfs_perag	*pag;
651 
652 		ino = XFS_AGINO_TO_INO(mp, index, agino);
653 
654 		pag = xfs_perag_get(mp, index);
655 
656 		if (mp->m_flags & XFS_MOUNT_32BITINODES) {
657 			if (ino > XFS_MAXINUMBER_32) {
658 				pag->pagi_inodeok = 0;
659 				pag->pagf_metadata = 0;
660 			} else {
661 				pag->pagi_inodeok = 1;
662 				maxagi++;
663 				if (index < max_metadata)
664 					pag->pagf_metadata = 1;
665 				else
666 					pag->pagf_metadata = 0;
667 			}
668 		} else {
669 			pag->pagi_inodeok = 1;
670 			pag->pagf_metadata = 0;
671 		}
672 
673 		xfs_perag_put(pag);
674 	}
675 
676 	return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
677 }
678 
679 STATIC int
680 xfs_blkdev_get(
681 	xfs_mount_t		*mp,
682 	const char		*name,
683 	struct block_device	**bdevp)
684 {
685 	int			error = 0;
686 
687 	*bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
688 				    mp);
689 	if (IS_ERR(*bdevp)) {
690 		error = PTR_ERR(*bdevp);
691 		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
692 	}
693 
694 	return error;
695 }
696 
697 STATIC void
698 xfs_blkdev_put(
699 	struct block_device	*bdev)
700 {
701 	if (bdev)
702 		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
703 }
704 
705 void
706 xfs_blkdev_issue_flush(
707 	xfs_buftarg_t		*buftarg)
708 {
709 	blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
710 }
711 
712 STATIC void
713 xfs_close_devices(
714 	struct xfs_mount	*mp)
715 {
716 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
717 		struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
718 		xfs_free_buftarg(mp, mp->m_logdev_targp);
719 		xfs_blkdev_put(logdev);
720 	}
721 	if (mp->m_rtdev_targp) {
722 		struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
723 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
724 		xfs_blkdev_put(rtdev);
725 	}
726 	xfs_free_buftarg(mp, mp->m_ddev_targp);
727 }
728 
729 /*
730  * The file system configurations are:
731  *	(1) device (partition) with data and internal log
732  *	(2) logical volume with data and log subvolumes.
733  *	(3) logical volume with data, log, and realtime subvolumes.
734  *
735  * We only have to handle opening the log and realtime volumes here if
736  * they are present.  The data subvolume has already been opened by
737  * get_sb_bdev() and is stored in sb->s_bdev.
738  */
739 STATIC int
740 xfs_open_devices(
741 	struct xfs_mount	*mp)
742 {
743 	struct block_device	*ddev = mp->m_super->s_bdev;
744 	struct block_device	*logdev = NULL, *rtdev = NULL;
745 	int			error;
746 
747 	/*
748 	 * Open real time and log devices - order is important.
749 	 */
750 	if (mp->m_logname) {
751 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
752 		if (error)
753 			goto out;
754 	}
755 
756 	if (mp->m_rtname) {
757 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
758 		if (error)
759 			goto out_close_logdev;
760 
761 		if (rtdev == ddev || rtdev == logdev) {
762 			xfs_warn(mp,
763 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
764 			error = -EINVAL;
765 			goto out_close_rtdev;
766 		}
767 	}
768 
769 	/*
770 	 * Setup xfs_mount buffer target pointers
771 	 */
772 	error = -ENOMEM;
773 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
774 	if (!mp->m_ddev_targp)
775 		goto out_close_rtdev;
776 
777 	if (rtdev) {
778 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
779 		if (!mp->m_rtdev_targp)
780 			goto out_free_ddev_targ;
781 	}
782 
783 	if (logdev && logdev != ddev) {
784 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
785 		if (!mp->m_logdev_targp)
786 			goto out_free_rtdev_targ;
787 	} else {
788 		mp->m_logdev_targp = mp->m_ddev_targp;
789 	}
790 
791 	return 0;
792 
793  out_free_rtdev_targ:
794 	if (mp->m_rtdev_targp)
795 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
796  out_free_ddev_targ:
797 	xfs_free_buftarg(mp, mp->m_ddev_targp);
798  out_close_rtdev:
799 	xfs_blkdev_put(rtdev);
800  out_close_logdev:
801 	if (logdev && logdev != ddev)
802 		xfs_blkdev_put(logdev);
803  out:
804 	return error;
805 }
806 
807 /*
808  * Setup xfs_mount buffer target pointers based on superblock
809  */
810 STATIC int
811 xfs_setup_devices(
812 	struct xfs_mount	*mp)
813 {
814 	int			error;
815 
816 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
817 	if (error)
818 		return error;
819 
820 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
821 		unsigned int	log_sector_size = BBSIZE;
822 
823 		if (xfs_sb_version_hassector(&mp->m_sb))
824 			log_sector_size = mp->m_sb.sb_logsectsize;
825 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
826 					    log_sector_size);
827 		if (error)
828 			return error;
829 	}
830 	if (mp->m_rtdev_targp) {
831 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
832 					    mp->m_sb.sb_sectsize);
833 		if (error)
834 			return error;
835 	}
836 
837 	return 0;
838 }
839 
840 STATIC int
841 xfs_init_mount_workqueues(
842 	struct xfs_mount	*mp)
843 {
844 	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
845 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
846 	if (!mp->m_buf_workqueue)
847 		goto out;
848 
849 	mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
850 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
851 	if (!mp->m_data_workqueue)
852 		goto out_destroy_buf;
853 
854 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
855 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
856 	if (!mp->m_unwritten_workqueue)
857 		goto out_destroy_data_iodone_queue;
858 
859 	mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
860 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
861 	if (!mp->m_cil_workqueue)
862 		goto out_destroy_unwritten;
863 
864 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
865 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
866 	if (!mp->m_reclaim_workqueue)
867 		goto out_destroy_cil;
868 
869 	mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
870 			WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
871 			mp->m_fsname);
872 	if (!mp->m_log_workqueue)
873 		goto out_destroy_reclaim;
874 
875 	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
876 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
877 	if (!mp->m_eofblocks_workqueue)
878 		goto out_destroy_log;
879 
880 	return 0;
881 
882 out_destroy_log:
883 	destroy_workqueue(mp->m_log_workqueue);
884 out_destroy_reclaim:
885 	destroy_workqueue(mp->m_reclaim_workqueue);
886 out_destroy_cil:
887 	destroy_workqueue(mp->m_cil_workqueue);
888 out_destroy_unwritten:
889 	destroy_workqueue(mp->m_unwritten_workqueue);
890 out_destroy_data_iodone_queue:
891 	destroy_workqueue(mp->m_data_workqueue);
892 out_destroy_buf:
893 	destroy_workqueue(mp->m_buf_workqueue);
894 out:
895 	return -ENOMEM;
896 }
897 
898 STATIC void
899 xfs_destroy_mount_workqueues(
900 	struct xfs_mount	*mp)
901 {
902 	destroy_workqueue(mp->m_eofblocks_workqueue);
903 	destroy_workqueue(mp->m_log_workqueue);
904 	destroy_workqueue(mp->m_reclaim_workqueue);
905 	destroy_workqueue(mp->m_cil_workqueue);
906 	destroy_workqueue(mp->m_data_workqueue);
907 	destroy_workqueue(mp->m_unwritten_workqueue);
908 	destroy_workqueue(mp->m_buf_workqueue);
909 }
910 
911 /*
912  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
913  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
914  * for IO to complete so that we effectively throttle multiple callers to the
915  * rate at which IO is completing.
916  */
917 void
918 xfs_flush_inodes(
919 	struct xfs_mount	*mp)
920 {
921 	struct super_block	*sb = mp->m_super;
922 
923 	if (down_read_trylock(&sb->s_umount)) {
924 		sync_inodes_sb(sb);
925 		up_read(&sb->s_umount);
926 	}
927 }
928 
929 /* Catch misguided souls that try to use this interface on XFS */
930 STATIC struct inode *
931 xfs_fs_alloc_inode(
932 	struct super_block	*sb)
933 {
934 	BUG();
935 	return NULL;
936 }
937 
938 /*
939  * Now that the generic code is guaranteed not to be accessing
940  * the linux inode, we can inactivate and reclaim the inode.
941  */
942 STATIC void
943 xfs_fs_destroy_inode(
944 	struct inode		*inode)
945 {
946 	struct xfs_inode	*ip = XFS_I(inode);
947 	int			error;
948 
949 	trace_xfs_destroy_inode(ip);
950 
951 	ASSERT(!rwsem_is_locked(&inode->i_rwsem));
952 	XFS_STATS_INC(ip->i_mount, vn_rele);
953 	XFS_STATS_INC(ip->i_mount, vn_remove);
954 
955 	if (xfs_is_reflink_inode(ip)) {
956 		error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
957 		if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
958 			xfs_warn(ip->i_mount,
959 "Error %d while evicting CoW blocks for inode %llu.",
960 					error, ip->i_ino);
961 	}
962 
963 	xfs_inactive(ip);
964 
965 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
966 	XFS_STATS_INC(ip->i_mount, vn_reclaim);
967 
968 	/*
969 	 * We should never get here with one of the reclaim flags already set.
970 	 */
971 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
972 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
973 
974 	/*
975 	 * We always use background reclaim here because even if the
976 	 * inode is clean, it still may be under IO and hence we have
977 	 * to take the flush lock. The background reclaim path handles
978 	 * this more efficiently than we can here, so simply let background
979 	 * reclaim tear down all inodes.
980 	 */
981 	xfs_inode_set_reclaim_tag(ip);
982 }
983 
984 /*
985  * Slab object creation initialisation for the XFS inode.
986  * This covers only the idempotent fields in the XFS inode;
987  * all other fields need to be initialised on allocation
988  * from the slab. This avoids the need to repeatedly initialise
989  * fields in the xfs inode that left in the initialise state
990  * when freeing the inode.
991  */
992 STATIC void
993 xfs_fs_inode_init_once(
994 	void			*inode)
995 {
996 	struct xfs_inode	*ip = inode;
997 
998 	memset(ip, 0, sizeof(struct xfs_inode));
999 
1000 	/* vfs inode */
1001 	inode_init_once(VFS_I(ip));
1002 
1003 	/* xfs inode */
1004 	atomic_set(&ip->i_pincount, 0);
1005 	spin_lock_init(&ip->i_flags_lock);
1006 
1007 	mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1008 		     "xfsino", ip->i_ino);
1009 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1010 		     "xfsino", ip->i_ino);
1011 }
1012 
1013 /*
1014  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1015  * serialised against cache hits here via the inode->i_lock and igrab() in
1016  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1017  * racing with us, and it avoids needing to grab a spinlock here for every inode
1018  * we drop the final reference on.
1019  */
1020 STATIC int
1021 xfs_fs_drop_inode(
1022 	struct inode		*inode)
1023 {
1024 	struct xfs_inode	*ip = XFS_I(inode);
1025 
1026 	/*
1027 	 * If this unlinked inode is in the middle of recovery, don't
1028 	 * drop the inode just yet; log recovery will take care of
1029 	 * that.  See the comment for this inode flag.
1030 	 */
1031 	if (ip->i_flags & XFS_IRECOVERY) {
1032 		ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1033 		return 0;
1034 	}
1035 
1036 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1037 }
1038 
1039 STATIC void
1040 xfs_free_fsname(
1041 	struct xfs_mount	*mp)
1042 {
1043 	kfree(mp->m_fsname);
1044 	kfree(mp->m_rtname);
1045 	kfree(mp->m_logname);
1046 }
1047 
1048 STATIC int
1049 xfs_fs_sync_fs(
1050 	struct super_block	*sb,
1051 	int			wait)
1052 {
1053 	struct xfs_mount	*mp = XFS_M(sb);
1054 
1055 	/*
1056 	 * Doing anything during the async pass would be counterproductive.
1057 	 */
1058 	if (!wait)
1059 		return 0;
1060 
1061 	xfs_log_force(mp, XFS_LOG_SYNC);
1062 	if (laptop_mode) {
1063 		/*
1064 		 * The disk must be active because we're syncing.
1065 		 * We schedule log work now (now that the disk is
1066 		 * active) instead of later (when it might not be).
1067 		 */
1068 		flush_delayed_work(&mp->m_log->l_work);
1069 	}
1070 
1071 	return 0;
1072 }
1073 
1074 STATIC int
1075 xfs_fs_statfs(
1076 	struct dentry		*dentry,
1077 	struct kstatfs		*statp)
1078 {
1079 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1080 	xfs_sb_t		*sbp = &mp->m_sb;
1081 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1082 	__uint64_t		fakeinos, id;
1083 	__uint64_t		icount;
1084 	__uint64_t		ifree;
1085 	__uint64_t		fdblocks;
1086 	xfs_extlen_t		lsize;
1087 	__int64_t		ffree;
1088 
1089 	statp->f_type = XFS_SB_MAGIC;
1090 	statp->f_namelen = MAXNAMELEN - 1;
1091 
1092 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1093 	statp->f_fsid.val[0] = (u32)id;
1094 	statp->f_fsid.val[1] = (u32)(id >> 32);
1095 
1096 	icount = percpu_counter_sum(&mp->m_icount);
1097 	ifree = percpu_counter_sum(&mp->m_ifree);
1098 	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1099 
1100 	spin_lock(&mp->m_sb_lock);
1101 	statp->f_bsize = sbp->sb_blocksize;
1102 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1103 	statp->f_blocks = sbp->sb_dblocks - lsize;
1104 	spin_unlock(&mp->m_sb_lock);
1105 
1106 	statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1107 	statp->f_bavail = statp->f_bfree;
1108 
1109 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
1110 	statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1111 	if (mp->m_maxicount)
1112 		statp->f_files = min_t(typeof(statp->f_files),
1113 					statp->f_files,
1114 					mp->m_maxicount);
1115 
1116 	/* If sb_icount overshot maxicount, report actual allocation */
1117 	statp->f_files = max_t(typeof(statp->f_files),
1118 					statp->f_files,
1119 					sbp->sb_icount);
1120 
1121 	/* make sure statp->f_ffree does not underflow */
1122 	ffree = statp->f_files - (icount - ifree);
1123 	statp->f_ffree = max_t(__int64_t, ffree, 0);
1124 
1125 
1126 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1127 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1128 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1129 		xfs_qm_statvfs(ip, statp);
1130 	return 0;
1131 }
1132 
1133 STATIC void
1134 xfs_save_resvblks(struct xfs_mount *mp)
1135 {
1136 	__uint64_t resblks = 0;
1137 
1138 	mp->m_resblks_save = mp->m_resblks;
1139 	xfs_reserve_blocks(mp, &resblks, NULL);
1140 }
1141 
1142 STATIC void
1143 xfs_restore_resvblks(struct xfs_mount *mp)
1144 {
1145 	__uint64_t resblks;
1146 
1147 	if (mp->m_resblks_save) {
1148 		resblks = mp->m_resblks_save;
1149 		mp->m_resblks_save = 0;
1150 	} else
1151 		resblks = xfs_default_resblks(mp);
1152 
1153 	xfs_reserve_blocks(mp, &resblks, NULL);
1154 }
1155 
1156 /*
1157  * Trigger writeback of all the dirty metadata in the file system.
1158  *
1159  * This ensures that the metadata is written to their location on disk rather
1160  * than just existing in transactions in the log. This means after a quiesce
1161  * there is no log replay required to write the inodes to disk - this is the
1162  * primary difference between a sync and a quiesce.
1163  *
1164  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1165  * it is started again when appropriate.
1166  */
1167 void
1168 xfs_quiesce_attr(
1169 	struct xfs_mount	*mp)
1170 {
1171 	int	error = 0;
1172 
1173 	/* wait for all modifications to complete */
1174 	while (atomic_read(&mp->m_active_trans) > 0)
1175 		delay(100);
1176 
1177 	/* force the log to unpin objects from the now complete transactions */
1178 	xfs_log_force(mp, XFS_LOG_SYNC);
1179 
1180 	/* reclaim inodes to do any IO before the freeze completes */
1181 	xfs_reclaim_inodes(mp, 0);
1182 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1183 
1184 	/* Push the superblock and write an unmount record */
1185 	error = xfs_log_sbcount(mp);
1186 	if (error)
1187 		xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1188 				"Frozen image may not be consistent.");
1189 	/*
1190 	 * Just warn here till VFS can correctly support
1191 	 * read-only remount without racing.
1192 	 */
1193 	WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1194 
1195 	xfs_log_quiesce(mp);
1196 }
1197 
1198 STATIC int
1199 xfs_test_remount_options(
1200 	struct super_block	*sb,
1201 	struct xfs_mount	*mp,
1202 	char			*options)
1203 {
1204 	int			error = 0;
1205 	struct xfs_mount	*tmp_mp;
1206 
1207 	tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1208 	if (!tmp_mp)
1209 		return -ENOMEM;
1210 
1211 	tmp_mp->m_super = sb;
1212 	error = xfs_parseargs(tmp_mp, options);
1213 	xfs_free_fsname(tmp_mp);
1214 	kfree(tmp_mp);
1215 
1216 	return error;
1217 }
1218 
1219 STATIC int
1220 xfs_fs_remount(
1221 	struct super_block	*sb,
1222 	int			*flags,
1223 	char			*options)
1224 {
1225 	struct xfs_mount	*mp = XFS_M(sb);
1226 	xfs_sb_t		*sbp = &mp->m_sb;
1227 	substring_t		args[MAX_OPT_ARGS];
1228 	char			*p;
1229 	int			error;
1230 
1231 	/* First, check for complete junk; i.e. invalid options */
1232 	error = xfs_test_remount_options(sb, mp, options);
1233 	if (error)
1234 		return error;
1235 
1236 	sync_filesystem(sb);
1237 	while ((p = strsep(&options, ",")) != NULL) {
1238 		int token;
1239 
1240 		if (!*p)
1241 			continue;
1242 
1243 		token = match_token(p, tokens, args);
1244 		switch (token) {
1245 		case Opt_barrier:
1246 			xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1247 			mp->m_flags |= XFS_MOUNT_BARRIER;
1248 			break;
1249 		case Opt_nobarrier:
1250 			xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1251 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
1252 			break;
1253 		case Opt_inode64:
1254 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1255 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1256 			break;
1257 		case Opt_inode32:
1258 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1259 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1260 			break;
1261 		default:
1262 			/*
1263 			 * Logically we would return an error here to prevent
1264 			 * users from believing they might have changed
1265 			 * mount options using remount which can't be changed.
1266 			 *
1267 			 * But unfortunately mount(8) adds all options from
1268 			 * mtab and fstab to the mount arguments in some cases
1269 			 * so we can't blindly reject options, but have to
1270 			 * check for each specified option if it actually
1271 			 * differs from the currently set option and only
1272 			 * reject it if that's the case.
1273 			 *
1274 			 * Until that is implemented we return success for
1275 			 * every remount request, and silently ignore all
1276 			 * options that we can't actually change.
1277 			 */
1278 #if 0
1279 			xfs_info(mp,
1280 		"mount option \"%s\" not supported for remount", p);
1281 			return -EINVAL;
1282 #else
1283 			break;
1284 #endif
1285 		}
1286 	}
1287 
1288 	/* ro -> rw */
1289 	if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1290 		if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1291 			xfs_warn(mp,
1292 		"ro->rw transition prohibited on norecovery mount");
1293 			return -EINVAL;
1294 		}
1295 
1296 		if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1297 		    xfs_sb_has_ro_compat_feature(sbp,
1298 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1299 			xfs_warn(mp,
1300 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1301 				(sbp->sb_features_ro_compat &
1302 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1303 			return -EINVAL;
1304 		}
1305 
1306 		mp->m_flags &= ~XFS_MOUNT_RDONLY;
1307 
1308 		/*
1309 		 * If this is the first remount to writeable state we
1310 		 * might have some superblock changes to update.
1311 		 */
1312 		if (mp->m_update_sb) {
1313 			error = xfs_sync_sb(mp, false);
1314 			if (error) {
1315 				xfs_warn(mp, "failed to write sb changes");
1316 				return error;
1317 			}
1318 			mp->m_update_sb = false;
1319 		}
1320 
1321 		/*
1322 		 * Fill out the reserve pool if it is empty. Use the stashed
1323 		 * value if it is non-zero, otherwise go with the default.
1324 		 */
1325 		xfs_restore_resvblks(mp);
1326 		xfs_log_work_queue(mp);
1327 		xfs_queue_eofblocks(mp);
1328 
1329 		/* Recover any CoW blocks that never got remapped. */
1330 		error = xfs_reflink_recover_cow(mp);
1331 		if (error) {
1332 			xfs_err(mp,
1333 	"Error %d recovering leftover CoW allocations.", error);
1334 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1335 			return error;
1336 		}
1337 
1338 		/* Create the per-AG metadata reservation pool .*/
1339 		error = xfs_fs_reserve_ag_blocks(mp);
1340 		if (error && error != -ENOSPC)
1341 			return error;
1342 	}
1343 
1344 	/* rw -> ro */
1345 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1346 		/* Free the per-AG metadata reservation pool. */
1347 		error = xfs_fs_unreserve_ag_blocks(mp);
1348 		if (error) {
1349 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1350 			return error;
1351 		}
1352 
1353 		/*
1354 		 * Before we sync the metadata, we need to free up the reserve
1355 		 * block pool so that the used block count in the superblock on
1356 		 * disk is correct at the end of the remount. Stash the current
1357 		 * reserve pool size so that if we get remounted rw, we can
1358 		 * return it to the same size.
1359 		 */
1360 		xfs_save_resvblks(mp);
1361 
1362 		/*
1363 		 * Cancel background eofb scanning so it cannot race with the
1364 		 * final log force+buftarg wait and deadlock the remount.
1365 		 */
1366 		cancel_delayed_work_sync(&mp->m_eofblocks_work);
1367 
1368 		xfs_quiesce_attr(mp);
1369 		mp->m_flags |= XFS_MOUNT_RDONLY;
1370 	}
1371 
1372 	return 0;
1373 }
1374 
1375 /*
1376  * Second stage of a freeze. The data is already frozen so we only
1377  * need to take care of the metadata. Once that's done sync the superblock
1378  * to the log to dirty it in case of a crash while frozen. This ensures that we
1379  * will recover the unlinked inode lists on the next mount.
1380  */
1381 STATIC int
1382 xfs_fs_freeze(
1383 	struct super_block	*sb)
1384 {
1385 	struct xfs_mount	*mp = XFS_M(sb);
1386 
1387 	xfs_save_resvblks(mp);
1388 	xfs_quiesce_attr(mp);
1389 	return xfs_sync_sb(mp, true);
1390 }
1391 
1392 STATIC int
1393 xfs_fs_unfreeze(
1394 	struct super_block	*sb)
1395 {
1396 	struct xfs_mount	*mp = XFS_M(sb);
1397 
1398 	xfs_restore_resvblks(mp);
1399 	xfs_log_work_queue(mp);
1400 	return 0;
1401 }
1402 
1403 STATIC int
1404 xfs_fs_show_options(
1405 	struct seq_file		*m,
1406 	struct dentry		*root)
1407 {
1408 	return xfs_showargs(XFS_M(root->d_sb), m);
1409 }
1410 
1411 /*
1412  * This function fills in xfs_mount_t fields based on mount args.
1413  * Note: the superblock _has_ now been read in.
1414  */
1415 STATIC int
1416 xfs_finish_flags(
1417 	struct xfs_mount	*mp)
1418 {
1419 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1420 
1421 	/* Fail a mount where the logbuf is smaller than the log stripe */
1422 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1423 		if (mp->m_logbsize <= 0 &&
1424 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1425 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1426 		} else if (mp->m_logbsize > 0 &&
1427 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1428 			xfs_warn(mp,
1429 		"logbuf size must be greater than or equal to log stripe size");
1430 			return -EINVAL;
1431 		}
1432 	} else {
1433 		/* Fail a mount if the logbuf is larger than 32K */
1434 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1435 			xfs_warn(mp,
1436 		"logbuf size for version 1 logs must be 16K or 32K");
1437 			return -EINVAL;
1438 		}
1439 	}
1440 
1441 	/*
1442 	 * V5 filesystems always use attr2 format for attributes.
1443 	 */
1444 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1445 	    (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1446 		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1447 			     "attr2 is always enabled for V5 filesystems.");
1448 		return -EINVAL;
1449 	}
1450 
1451 	/*
1452 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1453 	 * told by noattr2 to turn it off
1454 	 */
1455 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1456 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1457 		mp->m_flags |= XFS_MOUNT_ATTR2;
1458 
1459 	/*
1460 	 * prohibit r/w mounts of read-only filesystems
1461 	 */
1462 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1463 		xfs_warn(mp,
1464 			"cannot mount a read-only filesystem as read-write");
1465 		return -EROFS;
1466 	}
1467 
1468 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1469 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1470 	    !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1471 		xfs_warn(mp,
1472 		  "Super block does not support project and group quota together");
1473 		return -EINVAL;
1474 	}
1475 
1476 	return 0;
1477 }
1478 
1479 static int
1480 xfs_init_percpu_counters(
1481 	struct xfs_mount	*mp)
1482 {
1483 	int		error;
1484 
1485 	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1486 	if (error)
1487 		return -ENOMEM;
1488 
1489 	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1490 	if (error)
1491 		goto free_icount;
1492 
1493 	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1494 	if (error)
1495 		goto free_ifree;
1496 
1497 	return 0;
1498 
1499 free_ifree:
1500 	percpu_counter_destroy(&mp->m_ifree);
1501 free_icount:
1502 	percpu_counter_destroy(&mp->m_icount);
1503 	return -ENOMEM;
1504 }
1505 
1506 void
1507 xfs_reinit_percpu_counters(
1508 	struct xfs_mount	*mp)
1509 {
1510 	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1511 	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1512 	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1513 }
1514 
1515 static void
1516 xfs_destroy_percpu_counters(
1517 	struct xfs_mount	*mp)
1518 {
1519 	percpu_counter_destroy(&mp->m_icount);
1520 	percpu_counter_destroy(&mp->m_ifree);
1521 	percpu_counter_destroy(&mp->m_fdblocks);
1522 }
1523 
1524 STATIC int
1525 xfs_fs_fill_super(
1526 	struct super_block	*sb,
1527 	void			*data,
1528 	int			silent)
1529 {
1530 	struct inode		*root;
1531 	struct xfs_mount	*mp = NULL;
1532 	int			flags = 0, error = -ENOMEM;
1533 
1534 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1535 	if (!mp)
1536 		goto out;
1537 
1538 	spin_lock_init(&mp->m_sb_lock);
1539 	mutex_init(&mp->m_growlock);
1540 	atomic_set(&mp->m_active_trans, 0);
1541 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1542 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1543 	INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1544 	mp->m_kobj.kobject.kset = xfs_kset;
1545 
1546 	mp->m_super = sb;
1547 	sb->s_fs_info = mp;
1548 
1549 	error = xfs_parseargs(mp, (char *)data);
1550 	if (error)
1551 		goto out_free_fsname;
1552 
1553 	sb_min_blocksize(sb, BBSIZE);
1554 	sb->s_xattr = xfs_xattr_handlers;
1555 	sb->s_export_op = &xfs_export_operations;
1556 #ifdef CONFIG_XFS_QUOTA
1557 	sb->s_qcop = &xfs_quotactl_operations;
1558 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1559 #endif
1560 	sb->s_op = &xfs_super_operations;
1561 
1562 	if (silent)
1563 		flags |= XFS_MFSI_QUIET;
1564 
1565 	error = xfs_open_devices(mp);
1566 	if (error)
1567 		goto out_free_fsname;
1568 
1569 	error = xfs_init_mount_workqueues(mp);
1570 	if (error)
1571 		goto out_close_devices;
1572 
1573 	error = xfs_init_percpu_counters(mp);
1574 	if (error)
1575 		goto out_destroy_workqueues;
1576 
1577 	/* Allocate stats memory before we do operations that might use it */
1578 	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1579 	if (!mp->m_stats.xs_stats) {
1580 		error = -ENOMEM;
1581 		goto out_destroy_counters;
1582 	}
1583 
1584 	error = xfs_readsb(mp, flags);
1585 	if (error)
1586 		goto out_free_stats;
1587 
1588 	error = xfs_finish_flags(mp);
1589 	if (error)
1590 		goto out_free_sb;
1591 
1592 	error = xfs_setup_devices(mp);
1593 	if (error)
1594 		goto out_free_sb;
1595 
1596 	error = xfs_filestream_mount(mp);
1597 	if (error)
1598 		goto out_free_sb;
1599 
1600 	/*
1601 	 * we must configure the block size in the superblock before we run the
1602 	 * full mount process as the mount process can lookup and cache inodes.
1603 	 */
1604 	sb->s_magic = XFS_SB_MAGIC;
1605 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1606 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1607 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1608 	sb->s_max_links = XFS_MAXLINK;
1609 	sb->s_time_gran = 1;
1610 	set_posix_acl_flag(sb);
1611 
1612 	/* version 5 superblocks support inode version counters. */
1613 	if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1614 		sb->s_flags |= MS_I_VERSION;
1615 
1616 	if (mp->m_flags & XFS_MOUNT_DAX) {
1617 		xfs_warn(mp,
1618 		"DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1619 
1620 		error = bdev_dax_supported(sb, sb->s_blocksize);
1621 		if (error) {
1622 			xfs_alert(mp,
1623 			"DAX unsupported by block device. Turning off DAX.");
1624 			mp->m_flags &= ~XFS_MOUNT_DAX;
1625 		}
1626 		if (xfs_sb_version_hasreflink(&mp->m_sb))
1627 			xfs_alert(mp,
1628 		"DAX and reflink have not been tested together!");
1629 	}
1630 
1631 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1632 		if (mp->m_sb.sb_rblocks) {
1633 			xfs_alert(mp,
1634 	"EXPERIMENTAL reverse mapping btree not compatible with realtime device!");
1635 			error = -EINVAL;
1636 			goto out_filestream_unmount;
1637 		}
1638 		xfs_alert(mp,
1639 	"EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
1640 	}
1641 
1642 	if (xfs_sb_version_hasreflink(&mp->m_sb))
1643 		xfs_alert(mp,
1644 	"EXPERIMENTAL reflink feature enabled. Use at your own risk!");
1645 
1646 	error = xfs_mountfs(mp);
1647 	if (error)
1648 		goto out_filestream_unmount;
1649 
1650 	root = igrab(VFS_I(mp->m_rootip));
1651 	if (!root) {
1652 		error = -ENOENT;
1653 		goto out_unmount;
1654 	}
1655 	sb->s_root = d_make_root(root);
1656 	if (!sb->s_root) {
1657 		error = -ENOMEM;
1658 		goto out_unmount;
1659 	}
1660 
1661 	return 0;
1662 
1663  out_filestream_unmount:
1664 	xfs_filestream_unmount(mp);
1665  out_free_sb:
1666 	xfs_freesb(mp);
1667  out_free_stats:
1668 	free_percpu(mp->m_stats.xs_stats);
1669  out_destroy_counters:
1670 	xfs_destroy_percpu_counters(mp);
1671  out_destroy_workqueues:
1672 	xfs_destroy_mount_workqueues(mp);
1673  out_close_devices:
1674 	xfs_close_devices(mp);
1675  out_free_fsname:
1676 	xfs_free_fsname(mp);
1677 	kfree(mp);
1678  out:
1679 	return error;
1680 
1681  out_unmount:
1682 	xfs_filestream_unmount(mp);
1683 	xfs_unmountfs(mp);
1684 	goto out_free_sb;
1685 }
1686 
1687 STATIC void
1688 xfs_fs_put_super(
1689 	struct super_block	*sb)
1690 {
1691 	struct xfs_mount	*mp = XFS_M(sb);
1692 
1693 	xfs_notice(mp, "Unmounting Filesystem");
1694 	xfs_filestream_unmount(mp);
1695 	xfs_unmountfs(mp);
1696 
1697 	xfs_freesb(mp);
1698 	free_percpu(mp->m_stats.xs_stats);
1699 	xfs_destroy_percpu_counters(mp);
1700 	xfs_destroy_mount_workqueues(mp);
1701 	xfs_close_devices(mp);
1702 	xfs_free_fsname(mp);
1703 	kfree(mp);
1704 }
1705 
1706 STATIC struct dentry *
1707 xfs_fs_mount(
1708 	struct file_system_type	*fs_type,
1709 	int			flags,
1710 	const char		*dev_name,
1711 	void			*data)
1712 {
1713 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1714 }
1715 
1716 static long
1717 xfs_fs_nr_cached_objects(
1718 	struct super_block	*sb,
1719 	struct shrink_control	*sc)
1720 {
1721 	return xfs_reclaim_inodes_count(XFS_M(sb));
1722 }
1723 
1724 static long
1725 xfs_fs_free_cached_objects(
1726 	struct super_block	*sb,
1727 	struct shrink_control	*sc)
1728 {
1729 	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1730 }
1731 
1732 static const struct super_operations xfs_super_operations = {
1733 	.alloc_inode		= xfs_fs_alloc_inode,
1734 	.destroy_inode		= xfs_fs_destroy_inode,
1735 	.drop_inode		= xfs_fs_drop_inode,
1736 	.put_super		= xfs_fs_put_super,
1737 	.sync_fs		= xfs_fs_sync_fs,
1738 	.freeze_fs		= xfs_fs_freeze,
1739 	.unfreeze_fs		= xfs_fs_unfreeze,
1740 	.statfs			= xfs_fs_statfs,
1741 	.remount_fs		= xfs_fs_remount,
1742 	.show_options		= xfs_fs_show_options,
1743 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1744 	.free_cached_objects	= xfs_fs_free_cached_objects,
1745 };
1746 
1747 static struct file_system_type xfs_fs_type = {
1748 	.owner			= THIS_MODULE,
1749 	.name			= "xfs",
1750 	.mount			= xfs_fs_mount,
1751 	.kill_sb		= kill_block_super,
1752 	.fs_flags		= FS_REQUIRES_DEV,
1753 };
1754 MODULE_ALIAS_FS("xfs");
1755 
1756 STATIC int __init
1757 xfs_init_zones(void)
1758 {
1759 	xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1760 			offsetof(struct xfs_ioend, io_inline_bio));
1761 	if (!xfs_ioend_bioset)
1762 		goto out;
1763 
1764 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1765 						"xfs_log_ticket");
1766 	if (!xfs_log_ticket_zone)
1767 		goto out_free_ioend_bioset;
1768 
1769 	xfs_bmap_free_item_zone = kmem_zone_init(
1770 			sizeof(struct xfs_extent_free_item),
1771 			"xfs_bmap_free_item");
1772 	if (!xfs_bmap_free_item_zone)
1773 		goto out_destroy_log_ticket_zone;
1774 
1775 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1776 						"xfs_btree_cur");
1777 	if (!xfs_btree_cur_zone)
1778 		goto out_destroy_bmap_free_item_zone;
1779 
1780 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1781 						"xfs_da_state");
1782 	if (!xfs_da_state_zone)
1783 		goto out_destroy_btree_cur_zone;
1784 
1785 	xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1786 	if (!xfs_ifork_zone)
1787 		goto out_destroy_da_state_zone;
1788 
1789 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1790 	if (!xfs_trans_zone)
1791 		goto out_destroy_ifork_zone;
1792 
1793 	xfs_log_item_desc_zone =
1794 		kmem_zone_init(sizeof(struct xfs_log_item_desc),
1795 			       "xfs_log_item_desc");
1796 	if (!xfs_log_item_desc_zone)
1797 		goto out_destroy_trans_zone;
1798 
1799 	/*
1800 	 * The size of the zone allocated buf log item is the maximum
1801 	 * size possible under XFS.  This wastes a little bit of memory,
1802 	 * but it is much faster.
1803 	 */
1804 	xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1805 					   "xfs_buf_item");
1806 	if (!xfs_buf_item_zone)
1807 		goto out_destroy_log_item_desc_zone;
1808 
1809 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1810 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1811 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1812 	if (!xfs_efd_zone)
1813 		goto out_destroy_buf_item_zone;
1814 
1815 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1816 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1817 				sizeof(xfs_extent_t))), "xfs_efi_item");
1818 	if (!xfs_efi_zone)
1819 		goto out_destroy_efd_zone;
1820 
1821 	xfs_inode_zone =
1822 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1823 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1824 			KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1825 	if (!xfs_inode_zone)
1826 		goto out_destroy_efi_zone;
1827 
1828 	xfs_ili_zone =
1829 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1830 					KM_ZONE_SPREAD, NULL);
1831 	if (!xfs_ili_zone)
1832 		goto out_destroy_inode_zone;
1833 	xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1834 					"xfs_icr");
1835 	if (!xfs_icreate_zone)
1836 		goto out_destroy_ili_zone;
1837 
1838 	xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1839 			"xfs_rud_item");
1840 	if (!xfs_rud_zone)
1841 		goto out_destroy_icreate_zone;
1842 
1843 	xfs_rui_zone = kmem_zone_init(
1844 			xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1845 			"xfs_rui_item");
1846 	if (!xfs_rui_zone)
1847 		goto out_destroy_rud_zone;
1848 
1849 	xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1850 			"xfs_cud_item");
1851 	if (!xfs_cud_zone)
1852 		goto out_destroy_rui_zone;
1853 
1854 	xfs_cui_zone = kmem_zone_init(
1855 			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1856 			"xfs_cui_item");
1857 	if (!xfs_cui_zone)
1858 		goto out_destroy_cud_zone;
1859 
1860 	xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1861 			"xfs_bud_item");
1862 	if (!xfs_bud_zone)
1863 		goto out_destroy_cui_zone;
1864 
1865 	xfs_bui_zone = kmem_zone_init(
1866 			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1867 			"xfs_bui_item");
1868 	if (!xfs_bui_zone)
1869 		goto out_destroy_bud_zone;
1870 
1871 	return 0;
1872 
1873  out_destroy_bud_zone:
1874 	kmem_zone_destroy(xfs_bud_zone);
1875  out_destroy_cui_zone:
1876 	kmem_zone_destroy(xfs_cui_zone);
1877  out_destroy_cud_zone:
1878 	kmem_zone_destroy(xfs_cud_zone);
1879  out_destroy_rui_zone:
1880 	kmem_zone_destroy(xfs_rui_zone);
1881  out_destroy_rud_zone:
1882 	kmem_zone_destroy(xfs_rud_zone);
1883  out_destroy_icreate_zone:
1884 	kmem_zone_destroy(xfs_icreate_zone);
1885  out_destroy_ili_zone:
1886 	kmem_zone_destroy(xfs_ili_zone);
1887  out_destroy_inode_zone:
1888 	kmem_zone_destroy(xfs_inode_zone);
1889  out_destroy_efi_zone:
1890 	kmem_zone_destroy(xfs_efi_zone);
1891  out_destroy_efd_zone:
1892 	kmem_zone_destroy(xfs_efd_zone);
1893  out_destroy_buf_item_zone:
1894 	kmem_zone_destroy(xfs_buf_item_zone);
1895  out_destroy_log_item_desc_zone:
1896 	kmem_zone_destroy(xfs_log_item_desc_zone);
1897  out_destroy_trans_zone:
1898 	kmem_zone_destroy(xfs_trans_zone);
1899  out_destroy_ifork_zone:
1900 	kmem_zone_destroy(xfs_ifork_zone);
1901  out_destroy_da_state_zone:
1902 	kmem_zone_destroy(xfs_da_state_zone);
1903  out_destroy_btree_cur_zone:
1904 	kmem_zone_destroy(xfs_btree_cur_zone);
1905  out_destroy_bmap_free_item_zone:
1906 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1907  out_destroy_log_ticket_zone:
1908 	kmem_zone_destroy(xfs_log_ticket_zone);
1909  out_free_ioend_bioset:
1910 	bioset_free(xfs_ioend_bioset);
1911  out:
1912 	return -ENOMEM;
1913 }
1914 
1915 STATIC void
1916 xfs_destroy_zones(void)
1917 {
1918 	/*
1919 	 * Make sure all delayed rcu free are flushed before we
1920 	 * destroy caches.
1921 	 */
1922 	rcu_barrier();
1923 	kmem_zone_destroy(xfs_bui_zone);
1924 	kmem_zone_destroy(xfs_bud_zone);
1925 	kmem_zone_destroy(xfs_cui_zone);
1926 	kmem_zone_destroy(xfs_cud_zone);
1927 	kmem_zone_destroy(xfs_rui_zone);
1928 	kmem_zone_destroy(xfs_rud_zone);
1929 	kmem_zone_destroy(xfs_icreate_zone);
1930 	kmem_zone_destroy(xfs_ili_zone);
1931 	kmem_zone_destroy(xfs_inode_zone);
1932 	kmem_zone_destroy(xfs_efi_zone);
1933 	kmem_zone_destroy(xfs_efd_zone);
1934 	kmem_zone_destroy(xfs_buf_item_zone);
1935 	kmem_zone_destroy(xfs_log_item_desc_zone);
1936 	kmem_zone_destroy(xfs_trans_zone);
1937 	kmem_zone_destroy(xfs_ifork_zone);
1938 	kmem_zone_destroy(xfs_da_state_zone);
1939 	kmem_zone_destroy(xfs_btree_cur_zone);
1940 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1941 	kmem_zone_destroy(xfs_log_ticket_zone);
1942 	bioset_free(xfs_ioend_bioset);
1943 }
1944 
1945 STATIC int __init
1946 xfs_init_workqueues(void)
1947 {
1948 	/*
1949 	 * The allocation workqueue can be used in memory reclaim situations
1950 	 * (writepage path), and parallelism is only limited by the number of
1951 	 * AGs in all the filesystems mounted. Hence use the default large
1952 	 * max_active value for this workqueue.
1953 	 */
1954 	xfs_alloc_wq = alloc_workqueue("xfsalloc",
1955 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1956 	if (!xfs_alloc_wq)
1957 		return -ENOMEM;
1958 
1959 	xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
1960 	if (!xfs_discard_wq)
1961 		goto out_free_alloc_wq;
1962 
1963 	return 0;
1964 out_free_alloc_wq:
1965 	destroy_workqueue(xfs_alloc_wq);
1966 	return -ENOMEM;
1967 }
1968 
1969 STATIC void
1970 xfs_destroy_workqueues(void)
1971 {
1972 	destroy_workqueue(xfs_discard_wq);
1973 	destroy_workqueue(xfs_alloc_wq);
1974 }
1975 
1976 STATIC int __init
1977 init_xfs_fs(void)
1978 {
1979 	int			error;
1980 
1981 	xfs_check_ondisk_structs();
1982 
1983 	printk(KERN_INFO XFS_VERSION_STRING " with "
1984 			 XFS_BUILD_OPTIONS " enabled\n");
1985 
1986 	xfs_extent_free_init_defer_op();
1987 	xfs_rmap_update_init_defer_op();
1988 	xfs_refcount_update_init_defer_op();
1989 	xfs_bmap_update_init_defer_op();
1990 
1991 	xfs_dir_startup();
1992 
1993 	error = xfs_init_zones();
1994 	if (error)
1995 		goto out;
1996 
1997 	error = xfs_init_workqueues();
1998 	if (error)
1999 		goto out_destroy_zones;
2000 
2001 	error = xfs_mru_cache_init();
2002 	if (error)
2003 		goto out_destroy_wq;
2004 
2005 	error = xfs_buf_init();
2006 	if (error)
2007 		goto out_mru_cache_uninit;
2008 
2009 	error = xfs_init_procfs();
2010 	if (error)
2011 		goto out_buf_terminate;
2012 
2013 	error = xfs_sysctl_register();
2014 	if (error)
2015 		goto out_cleanup_procfs;
2016 
2017 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2018 	if (!xfs_kset) {
2019 		error = -ENOMEM;
2020 		goto out_sysctl_unregister;
2021 	}
2022 
2023 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
2024 
2025 	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2026 	if (!xfsstats.xs_stats) {
2027 		error = -ENOMEM;
2028 		goto out_kset_unregister;
2029 	}
2030 
2031 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2032 			       "stats");
2033 	if (error)
2034 		goto out_free_stats;
2035 
2036 #ifdef DEBUG
2037 	xfs_dbg_kobj.kobject.kset = xfs_kset;
2038 	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2039 	if (error)
2040 		goto out_remove_stats_kobj;
2041 #endif
2042 
2043 	error = xfs_qm_init();
2044 	if (error)
2045 		goto out_remove_dbg_kobj;
2046 
2047 	error = register_filesystem(&xfs_fs_type);
2048 	if (error)
2049 		goto out_qm_exit;
2050 	return 0;
2051 
2052  out_qm_exit:
2053 	xfs_qm_exit();
2054  out_remove_dbg_kobj:
2055 #ifdef DEBUG
2056 	xfs_sysfs_del(&xfs_dbg_kobj);
2057  out_remove_stats_kobj:
2058 #endif
2059 	xfs_sysfs_del(&xfsstats.xs_kobj);
2060  out_free_stats:
2061 	free_percpu(xfsstats.xs_stats);
2062  out_kset_unregister:
2063 	kset_unregister(xfs_kset);
2064  out_sysctl_unregister:
2065 	xfs_sysctl_unregister();
2066  out_cleanup_procfs:
2067 	xfs_cleanup_procfs();
2068  out_buf_terminate:
2069 	xfs_buf_terminate();
2070  out_mru_cache_uninit:
2071 	xfs_mru_cache_uninit();
2072  out_destroy_wq:
2073 	xfs_destroy_workqueues();
2074  out_destroy_zones:
2075 	xfs_destroy_zones();
2076  out:
2077 	return error;
2078 }
2079 
2080 STATIC void __exit
2081 exit_xfs_fs(void)
2082 {
2083 	xfs_qm_exit();
2084 	unregister_filesystem(&xfs_fs_type);
2085 #ifdef DEBUG
2086 	xfs_sysfs_del(&xfs_dbg_kobj);
2087 #endif
2088 	xfs_sysfs_del(&xfsstats.xs_kobj);
2089 	free_percpu(xfsstats.xs_stats);
2090 	kset_unregister(xfs_kset);
2091 	xfs_sysctl_unregister();
2092 	xfs_cleanup_procfs();
2093 	xfs_buf_terminate();
2094 	xfs_mru_cache_uninit();
2095 	xfs_destroy_workqueues();
2096 	xfs_destroy_zones();
2097 	xfs_uuid_table_free();
2098 }
2099 
2100 module_init(init_xfs_fs);
2101 module_exit(exit_xfs_fs);
2102 
2103 MODULE_AUTHOR("Silicon Graphics, Inc.");
2104 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2105 MODULE_LICENSE("GPL");
2106