xref: /openbmc/linux/fs/xfs/xfs_super.c (revision bc5aa3a0)
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 
51 #include <linux/namei.h>
52 #include <linux/init.h>
53 #include <linux/slab.h>
54 #include <linux/mount.h>
55 #include <linux/mempool.h>
56 #include <linux/writeback.h>
57 #include <linux/kthread.h>
58 #include <linux/freezer.h>
59 #include <linux/parser.h>
60 
61 static const struct super_operations xfs_super_operations;
62 struct bio_set *xfs_ioend_bioset;
63 
64 static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
65 #ifdef DEBUG
66 static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
67 #endif
68 
69 /*
70  * Table driven mount option parser.
71  */
72 enum {
73 	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
74 	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
75 	Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
76 	Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
77 	Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
78 	Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
79 	Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
80 	Opt_uquota, Opt_gquota, Opt_pquota,
81 	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
82 	Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
83 };
84 
85 static const match_table_t tokens = {
86 	{Opt_logbufs,	"logbufs=%u"},	/* number of XFS log buffers */
87 	{Opt_logbsize,	"logbsize=%s"},	/* size of XFS log buffers */
88 	{Opt_logdev,	"logdev=%s"},	/* log device */
89 	{Opt_rtdev,	"rtdev=%s"},	/* realtime I/O device */
90 	{Opt_biosize,	"biosize=%u"},	/* log2 of preferred buffered io size */
91 	{Opt_wsync,	"wsync"},	/* safe-mode nfs compatible mount */
92 	{Opt_noalign,	"noalign"},	/* turn off stripe alignment */
93 	{Opt_swalloc,	"swalloc"},	/* turn on stripe width allocation */
94 	{Opt_sunit,	"sunit=%u"},	/* data volume stripe unit */
95 	{Opt_swidth,	"swidth=%u"},	/* data volume stripe width */
96 	{Opt_nouuid,	"nouuid"},	/* ignore filesystem UUID */
97 	{Opt_mtpt,	"mtpt"},	/* filesystem mount point */
98 	{Opt_grpid,	"grpid"},	/* group-ID from parent directory */
99 	{Opt_nogrpid,	"nogrpid"},	/* group-ID from current process */
100 	{Opt_bsdgroups,	"bsdgroups"},	/* group-ID from parent directory */
101 	{Opt_sysvgroups,"sysvgroups"},	/* group-ID from current process */
102 	{Opt_allocsize,	"allocsize=%s"},/* preferred allocation size */
103 	{Opt_norecovery,"norecovery"},	/* don't run XFS recovery */
104 	{Opt_barrier,	"barrier"},	/* use writer barriers for log write and
105 					 * unwritten extent conversion */
106 	{Opt_nobarrier,	"nobarrier"},	/* .. disable */
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 	{Opt_err,	NULL},
135 };
136 
137 
138 STATIC int
139 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
140 {
141 	int	last, shift_left_factor = 0, _res;
142 	char	*value;
143 	int	ret = 0;
144 
145 	value = match_strdup(s);
146 	if (!value)
147 		return -ENOMEM;
148 
149 	last = strlen(value) - 1;
150 	if (value[last] == 'K' || value[last] == 'k') {
151 		shift_left_factor = 10;
152 		value[last] = '\0';
153 	}
154 	if (value[last] == 'M' || value[last] == 'm') {
155 		shift_left_factor = 20;
156 		value[last] = '\0';
157 	}
158 	if (value[last] == 'G' || value[last] == 'g') {
159 		shift_left_factor = 30;
160 		value[last] = '\0';
161 	}
162 
163 	if (kstrtoint(value, base, &_res))
164 		ret = -EINVAL;
165 	kfree(value);
166 	*res = _res << shift_left_factor;
167 	return ret;
168 }
169 
170 /*
171  * This function fills in xfs_mount_t fields based on mount args.
172  * Note: the superblock has _not_ yet been read in.
173  *
174  * Note that this function leaks the various device name allocations on
175  * failure.  The caller takes care of them.
176  *
177  * *sb is const because this is also used to test options on the remount
178  * path, and we don't want this to have any side effects at remount time.
179  * Today this function does not change *sb, but just to future-proof...
180  */
181 STATIC int
182 xfs_parseargs(
183 	struct xfs_mount	*mp,
184 	char			*options)
185 {
186 	const struct super_block *sb = mp->m_super;
187 	char			*p;
188 	substring_t		args[MAX_OPT_ARGS];
189 	int			dsunit = 0;
190 	int			dswidth = 0;
191 	int			iosize = 0;
192 	__uint8_t		iosizelog = 0;
193 
194 	/*
195 	 * set up the mount name first so all the errors will refer to the
196 	 * correct device.
197 	 */
198 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
199 	if (!mp->m_fsname)
200 		return -ENOMEM;
201 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
202 
203 	/*
204 	 * Copy binary VFS mount flags we are interested in.
205 	 */
206 	if (sb->s_flags & MS_RDONLY)
207 		mp->m_flags |= XFS_MOUNT_RDONLY;
208 	if (sb->s_flags & MS_DIRSYNC)
209 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
210 	if (sb->s_flags & MS_SYNCHRONOUS)
211 		mp->m_flags |= XFS_MOUNT_WSYNC;
212 
213 	/*
214 	 * Set some default flags that could be cleared by the mount option
215 	 * parsing.
216 	 */
217 	mp->m_flags |= XFS_MOUNT_BARRIER;
218 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
219 
220 	/*
221 	 * These can be overridden by the mount option parsing.
222 	 */
223 	mp->m_logbufs = -1;
224 	mp->m_logbsize = -1;
225 
226 	if (!options)
227 		goto done;
228 
229 	while ((p = strsep(&options, ",")) != NULL) {
230 		int		token;
231 
232 		if (!*p)
233 			continue;
234 
235 		token = match_token(p, tokens, args);
236 		switch (token) {
237 		case Opt_logbufs:
238 			if (match_int(args, &mp->m_logbufs))
239 				return -EINVAL;
240 			break;
241 		case Opt_logbsize:
242 			if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
243 				return -EINVAL;
244 			break;
245 		case Opt_logdev:
246 			mp->m_logname = match_strdup(args);
247 			if (!mp->m_logname)
248 				return -ENOMEM;
249 			break;
250 		case Opt_mtpt:
251 			xfs_warn(mp, "%s option not allowed on this system", p);
252 			return -EINVAL;
253 		case Opt_rtdev:
254 			mp->m_rtname = match_strdup(args);
255 			if (!mp->m_rtname)
256 				return -ENOMEM;
257 			break;
258 		case Opt_allocsize:
259 		case Opt_biosize:
260 			if (suffix_kstrtoint(args, 10, &iosize))
261 				return -EINVAL;
262 			iosizelog = ffs(iosize) - 1;
263 			break;
264 		case Opt_grpid:
265 		case Opt_bsdgroups:
266 			mp->m_flags |= XFS_MOUNT_GRPID;
267 			break;
268 		case Opt_nogrpid:
269 		case Opt_sysvgroups:
270 			mp->m_flags &= ~XFS_MOUNT_GRPID;
271 			break;
272 		case Opt_wsync:
273 			mp->m_flags |= XFS_MOUNT_WSYNC;
274 			break;
275 		case Opt_norecovery:
276 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
277 			break;
278 		case Opt_noalign:
279 			mp->m_flags |= XFS_MOUNT_NOALIGN;
280 			break;
281 		case Opt_swalloc:
282 			mp->m_flags |= XFS_MOUNT_SWALLOC;
283 			break;
284 		case Opt_sunit:
285 			if (match_int(args, &dsunit))
286 				return -EINVAL;
287 			break;
288 		case Opt_swidth:
289 			if (match_int(args, &dswidth))
290 				return -EINVAL;
291 			break;
292 		case Opt_inode32:
293 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
294 			break;
295 		case Opt_inode64:
296 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
297 			break;
298 		case Opt_nouuid:
299 			mp->m_flags |= XFS_MOUNT_NOUUID;
300 			break;
301 		case Opt_barrier:
302 			mp->m_flags |= XFS_MOUNT_BARRIER;
303 			break;
304 		case Opt_nobarrier:
305 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
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 		default:
375 			xfs_warn(mp, "unknown mount option [%s].", p);
376 			return -EINVAL;
377 		}
378 	}
379 
380 	/*
381 	 * no recovery flag requires a read-only mount
382 	 */
383 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
384 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
385 		xfs_warn(mp, "no-recovery mounts must be read-only.");
386 		return -EINVAL;
387 	}
388 
389 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
390 		xfs_warn(mp,
391 	"sunit and swidth options incompatible with the noalign option");
392 		return -EINVAL;
393 	}
394 
395 #ifndef CONFIG_XFS_QUOTA
396 	if (XFS_IS_QUOTA_RUNNING(mp)) {
397 		xfs_warn(mp, "quota support not available in this kernel.");
398 		return -EINVAL;
399 	}
400 #endif
401 
402 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
403 		xfs_warn(mp, "sunit and swidth must be specified together");
404 		return -EINVAL;
405 	}
406 
407 	if (dsunit && (dswidth % dsunit != 0)) {
408 		xfs_warn(mp,
409 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
410 			dswidth, dsunit);
411 		return -EINVAL;
412 	}
413 
414 done:
415 	if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
416 		/*
417 		 * At this point the superblock has not been read
418 		 * in, therefore we do not know the block size.
419 		 * Before the mount call ends we will convert
420 		 * these to FSBs.
421 		 */
422 		mp->m_dalign = dsunit;
423 		mp->m_swidth = dswidth;
424 	}
425 
426 	if (mp->m_logbufs != -1 &&
427 	    mp->m_logbufs != 0 &&
428 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
429 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
430 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
431 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
432 		return -EINVAL;
433 	}
434 	if (mp->m_logbsize != -1 &&
435 	    mp->m_logbsize !=  0 &&
436 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
437 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
438 	     !is_power_of_2(mp->m_logbsize))) {
439 		xfs_warn(mp,
440 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
441 			mp->m_logbsize);
442 		return -EINVAL;
443 	}
444 
445 	if (iosizelog) {
446 		if (iosizelog > XFS_MAX_IO_LOG ||
447 		    iosizelog < XFS_MIN_IO_LOG) {
448 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
449 				iosizelog, XFS_MIN_IO_LOG,
450 				XFS_MAX_IO_LOG);
451 			return -EINVAL;
452 		}
453 
454 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
455 		mp->m_readio_log = iosizelog;
456 		mp->m_writeio_log = iosizelog;
457 	}
458 
459 	return 0;
460 }
461 
462 struct proc_xfs_info {
463 	uint64_t	flag;
464 	char		*str;
465 };
466 
467 STATIC int
468 xfs_showargs(
469 	struct xfs_mount	*mp,
470 	struct seq_file		*m)
471 {
472 	static struct proc_xfs_info xfs_info_set[] = {
473 		/* the few simple ones we can get from the mount struct */
474 		{ XFS_MOUNT_IKEEP,		",ikeep" },
475 		{ XFS_MOUNT_WSYNC,		",wsync" },
476 		{ XFS_MOUNT_NOALIGN,		",noalign" },
477 		{ XFS_MOUNT_SWALLOC,		",swalloc" },
478 		{ XFS_MOUNT_NOUUID,		",nouuid" },
479 		{ XFS_MOUNT_NORECOVERY,		",norecovery" },
480 		{ XFS_MOUNT_ATTR2,		",attr2" },
481 		{ XFS_MOUNT_FILESTREAMS,	",filestreams" },
482 		{ XFS_MOUNT_GRPID,		",grpid" },
483 		{ XFS_MOUNT_DISCARD,		",discard" },
484 		{ XFS_MOUNT_SMALL_INUMS,	",inode32" },
485 		{ XFS_MOUNT_DAX,		",dax" },
486 		{ 0, NULL }
487 	};
488 	static struct proc_xfs_info xfs_info_unset[] = {
489 		/* the few simple ones we can get from the mount struct */
490 		{ XFS_MOUNT_COMPAT_IOSIZE,	",largeio" },
491 		{ XFS_MOUNT_BARRIER,		",nobarrier" },
492 		{ XFS_MOUNT_SMALL_INUMS,	",inode64" },
493 		{ 0, NULL }
494 	};
495 	struct proc_xfs_info	*xfs_infop;
496 
497 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
498 		if (mp->m_flags & xfs_infop->flag)
499 			seq_puts(m, xfs_infop->str);
500 	}
501 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
502 		if (!(mp->m_flags & xfs_infop->flag))
503 			seq_puts(m, xfs_infop->str);
504 	}
505 
506 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
507 		seq_printf(m, ",allocsize=%dk",
508 				(int)(1 << mp->m_writeio_log) >> 10);
509 
510 	if (mp->m_logbufs > 0)
511 		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
512 	if (mp->m_logbsize > 0)
513 		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
514 
515 	if (mp->m_logname)
516 		seq_show_option(m, "logdev", mp->m_logname);
517 	if (mp->m_rtname)
518 		seq_show_option(m, "rtdev", mp->m_rtname);
519 
520 	if (mp->m_dalign > 0)
521 		seq_printf(m, ",sunit=%d",
522 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
523 	if (mp->m_swidth > 0)
524 		seq_printf(m, ",swidth=%d",
525 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
526 
527 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
528 		seq_puts(m, ",usrquota");
529 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
530 		seq_puts(m, ",uqnoenforce");
531 
532 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
533 		if (mp->m_qflags & XFS_PQUOTA_ENFD)
534 			seq_puts(m, ",prjquota");
535 		else
536 			seq_puts(m, ",pqnoenforce");
537 	}
538 	if (mp->m_qflags & XFS_GQUOTA_ACCT) {
539 		if (mp->m_qflags & XFS_GQUOTA_ENFD)
540 			seq_puts(m, ",grpquota");
541 		else
542 			seq_puts(m, ",gqnoenforce");
543 	}
544 
545 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
546 		seq_puts(m, ",noquota");
547 
548 	return 0;
549 }
550 static __uint64_t
551 xfs_max_file_offset(
552 	unsigned int		blockshift)
553 {
554 	unsigned int		pagefactor = 1;
555 	unsigned int		bitshift = BITS_PER_LONG - 1;
556 
557 	/* Figure out maximum filesize, on Linux this can depend on
558 	 * the filesystem blocksize (on 32 bit platforms).
559 	 * __block_write_begin does this in an [unsigned] long...
560 	 *      page->index << (PAGE_SHIFT - bbits)
561 	 * So, for page sized blocks (4K on 32 bit platforms),
562 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
563 	 *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
564 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
565 	 * Note1: get_block_t takes a long (implicit cast from above)
566 	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
567 	 * can optionally convert the [unsigned] long from above into
568 	 * an [unsigned] long long.
569 	 */
570 
571 #if BITS_PER_LONG == 32
572 # if defined(CONFIG_LBDAF)
573 	ASSERT(sizeof(sector_t) == 8);
574 	pagefactor = PAGE_SIZE;
575 	bitshift = BITS_PER_LONG;
576 # else
577 	pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
578 # endif
579 #endif
580 
581 	return (((__uint64_t)pagefactor) << bitshift) - 1;
582 }
583 
584 /*
585  * Set parameters for inode allocation heuristics, taking into account
586  * filesystem size and inode32/inode64 mount options; i.e. specifically
587  * whether or not XFS_MOUNT_SMALL_INUMS is set.
588  *
589  * Inode allocation patterns are altered only if inode32 is requested
590  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
591  * If altered, XFS_MOUNT_32BITINODES is set as well.
592  *
593  * An agcount independent of that in the mount structure is provided
594  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
595  * to the potentially higher ag count.
596  *
597  * Returns the maximum AG index which may contain inodes.
598  */
599 xfs_agnumber_t
600 xfs_set_inode_alloc(
601 	struct xfs_mount *mp,
602 	xfs_agnumber_t	agcount)
603 {
604 	xfs_agnumber_t	index;
605 	xfs_agnumber_t	maxagi = 0;
606 	xfs_sb_t	*sbp = &mp->m_sb;
607 	xfs_agnumber_t	max_metadata;
608 	xfs_agino_t	agino;
609 	xfs_ino_t	ino;
610 
611 	/*
612 	 * Calculate how much should be reserved for inodes to meet
613 	 * the max inode percentage.  Used only for inode32.
614 	 */
615 	if (mp->m_maxicount) {
616 		__uint64_t	icount;
617 
618 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
619 		do_div(icount, 100);
620 		icount += sbp->sb_agblocks - 1;
621 		do_div(icount, sbp->sb_agblocks);
622 		max_metadata = icount;
623 	} else {
624 		max_metadata = agcount;
625 	}
626 
627 	/* Get the last possible inode in the filesystem */
628 	agino =	XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
629 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
630 
631 	/*
632 	 * If user asked for no more than 32-bit inodes, and the fs is
633 	 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
634 	 * the allocator to accommodate the request.
635 	 */
636 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
637 		mp->m_flags |= XFS_MOUNT_32BITINODES;
638 	else
639 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
640 
641 	for (index = 0; index < agcount; index++) {
642 		struct xfs_perag	*pag;
643 
644 		ino = XFS_AGINO_TO_INO(mp, index, agino);
645 
646 		pag = xfs_perag_get(mp, index);
647 
648 		if (mp->m_flags & XFS_MOUNT_32BITINODES) {
649 			if (ino > XFS_MAXINUMBER_32) {
650 				pag->pagi_inodeok = 0;
651 				pag->pagf_metadata = 0;
652 			} else {
653 				pag->pagi_inodeok = 1;
654 				maxagi++;
655 				if (index < max_metadata)
656 					pag->pagf_metadata = 1;
657 				else
658 					pag->pagf_metadata = 0;
659 			}
660 		} else {
661 			pag->pagi_inodeok = 1;
662 			pag->pagf_metadata = 0;
663 		}
664 
665 		xfs_perag_put(pag);
666 	}
667 
668 	return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
669 }
670 
671 STATIC int
672 xfs_blkdev_get(
673 	xfs_mount_t		*mp,
674 	const char		*name,
675 	struct block_device	**bdevp)
676 {
677 	int			error = 0;
678 
679 	*bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
680 				    mp);
681 	if (IS_ERR(*bdevp)) {
682 		error = PTR_ERR(*bdevp);
683 		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
684 	}
685 
686 	return error;
687 }
688 
689 STATIC void
690 xfs_blkdev_put(
691 	struct block_device	*bdev)
692 {
693 	if (bdev)
694 		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
695 }
696 
697 void
698 xfs_blkdev_issue_flush(
699 	xfs_buftarg_t		*buftarg)
700 {
701 	blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
702 }
703 
704 STATIC void
705 xfs_close_devices(
706 	struct xfs_mount	*mp)
707 {
708 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
709 		struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
710 		xfs_free_buftarg(mp, mp->m_logdev_targp);
711 		xfs_blkdev_put(logdev);
712 	}
713 	if (mp->m_rtdev_targp) {
714 		struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
715 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
716 		xfs_blkdev_put(rtdev);
717 	}
718 	xfs_free_buftarg(mp, mp->m_ddev_targp);
719 }
720 
721 /*
722  * The file system configurations are:
723  *	(1) device (partition) with data and internal log
724  *	(2) logical volume with data and log subvolumes.
725  *	(3) logical volume with data, log, and realtime subvolumes.
726  *
727  * We only have to handle opening the log and realtime volumes here if
728  * they are present.  The data subvolume has already been opened by
729  * get_sb_bdev() and is stored in sb->s_bdev.
730  */
731 STATIC int
732 xfs_open_devices(
733 	struct xfs_mount	*mp)
734 {
735 	struct block_device	*ddev = mp->m_super->s_bdev;
736 	struct block_device	*logdev = NULL, *rtdev = NULL;
737 	int			error;
738 
739 	/*
740 	 * Open real time and log devices - order is important.
741 	 */
742 	if (mp->m_logname) {
743 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
744 		if (error)
745 			goto out;
746 	}
747 
748 	if (mp->m_rtname) {
749 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
750 		if (error)
751 			goto out_close_logdev;
752 
753 		if (rtdev == ddev || rtdev == logdev) {
754 			xfs_warn(mp,
755 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
756 			error = -EINVAL;
757 			goto out_close_rtdev;
758 		}
759 	}
760 
761 	/*
762 	 * Setup xfs_mount buffer target pointers
763 	 */
764 	error = -ENOMEM;
765 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
766 	if (!mp->m_ddev_targp)
767 		goto out_close_rtdev;
768 
769 	if (rtdev) {
770 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
771 		if (!mp->m_rtdev_targp)
772 			goto out_free_ddev_targ;
773 	}
774 
775 	if (logdev && logdev != ddev) {
776 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
777 		if (!mp->m_logdev_targp)
778 			goto out_free_rtdev_targ;
779 	} else {
780 		mp->m_logdev_targp = mp->m_ddev_targp;
781 	}
782 
783 	return 0;
784 
785  out_free_rtdev_targ:
786 	if (mp->m_rtdev_targp)
787 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
788  out_free_ddev_targ:
789 	xfs_free_buftarg(mp, mp->m_ddev_targp);
790  out_close_rtdev:
791 	xfs_blkdev_put(rtdev);
792  out_close_logdev:
793 	if (logdev && logdev != ddev)
794 		xfs_blkdev_put(logdev);
795  out:
796 	return error;
797 }
798 
799 /*
800  * Setup xfs_mount buffer target pointers based on superblock
801  */
802 STATIC int
803 xfs_setup_devices(
804 	struct xfs_mount	*mp)
805 {
806 	int			error;
807 
808 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
809 	if (error)
810 		return error;
811 
812 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
813 		unsigned int	log_sector_size = BBSIZE;
814 
815 		if (xfs_sb_version_hassector(&mp->m_sb))
816 			log_sector_size = mp->m_sb.sb_logsectsize;
817 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
818 					    log_sector_size);
819 		if (error)
820 			return error;
821 	}
822 	if (mp->m_rtdev_targp) {
823 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
824 					    mp->m_sb.sb_sectsize);
825 		if (error)
826 			return error;
827 	}
828 
829 	return 0;
830 }
831 
832 STATIC int
833 xfs_init_mount_workqueues(
834 	struct xfs_mount	*mp)
835 {
836 	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
837 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
838 	if (!mp->m_buf_workqueue)
839 		goto out;
840 
841 	mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
842 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
843 	if (!mp->m_data_workqueue)
844 		goto out_destroy_buf;
845 
846 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
847 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
848 	if (!mp->m_unwritten_workqueue)
849 		goto out_destroy_data_iodone_queue;
850 
851 	mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
852 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
853 	if (!mp->m_cil_workqueue)
854 		goto out_destroy_unwritten;
855 
856 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
857 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
858 	if (!mp->m_reclaim_workqueue)
859 		goto out_destroy_cil;
860 
861 	mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
862 			WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
863 			mp->m_fsname);
864 	if (!mp->m_log_workqueue)
865 		goto out_destroy_reclaim;
866 
867 	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
868 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
869 	if (!mp->m_eofblocks_workqueue)
870 		goto out_destroy_log;
871 
872 	return 0;
873 
874 out_destroy_log:
875 	destroy_workqueue(mp->m_log_workqueue);
876 out_destroy_reclaim:
877 	destroy_workqueue(mp->m_reclaim_workqueue);
878 out_destroy_cil:
879 	destroy_workqueue(mp->m_cil_workqueue);
880 out_destroy_unwritten:
881 	destroy_workqueue(mp->m_unwritten_workqueue);
882 out_destroy_data_iodone_queue:
883 	destroy_workqueue(mp->m_data_workqueue);
884 out_destroy_buf:
885 	destroy_workqueue(mp->m_buf_workqueue);
886 out:
887 	return -ENOMEM;
888 }
889 
890 STATIC void
891 xfs_destroy_mount_workqueues(
892 	struct xfs_mount	*mp)
893 {
894 	destroy_workqueue(mp->m_eofblocks_workqueue);
895 	destroy_workqueue(mp->m_log_workqueue);
896 	destroy_workqueue(mp->m_reclaim_workqueue);
897 	destroy_workqueue(mp->m_cil_workqueue);
898 	destroy_workqueue(mp->m_data_workqueue);
899 	destroy_workqueue(mp->m_unwritten_workqueue);
900 	destroy_workqueue(mp->m_buf_workqueue);
901 }
902 
903 /*
904  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
905  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
906  * for IO to complete so that we effectively throttle multiple callers to the
907  * rate at which IO is completing.
908  */
909 void
910 xfs_flush_inodes(
911 	struct xfs_mount	*mp)
912 {
913 	struct super_block	*sb = mp->m_super;
914 
915 	if (down_read_trylock(&sb->s_umount)) {
916 		sync_inodes_sb(sb);
917 		up_read(&sb->s_umount);
918 	}
919 }
920 
921 /* Catch misguided souls that try to use this interface on XFS */
922 STATIC struct inode *
923 xfs_fs_alloc_inode(
924 	struct super_block	*sb)
925 {
926 	BUG();
927 	return NULL;
928 }
929 
930 /*
931  * Now that the generic code is guaranteed not to be accessing
932  * the linux inode, we can inactivate and reclaim the inode.
933  */
934 STATIC void
935 xfs_fs_destroy_inode(
936 	struct inode		*inode)
937 {
938 	struct xfs_inode	*ip = XFS_I(inode);
939 
940 	trace_xfs_destroy_inode(ip);
941 
942 	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
943 	XFS_STATS_INC(ip->i_mount, vn_rele);
944 	XFS_STATS_INC(ip->i_mount, vn_remove);
945 
946 	xfs_inactive(ip);
947 
948 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
949 	XFS_STATS_INC(ip->i_mount, vn_reclaim);
950 
951 	/*
952 	 * We should never get here with one of the reclaim flags already set.
953 	 */
954 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
955 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
956 
957 	/*
958 	 * We always use background reclaim here because even if the
959 	 * inode is clean, it still may be under IO and hence we have
960 	 * to take the flush lock. The background reclaim path handles
961 	 * this more efficiently than we can here, so simply let background
962 	 * reclaim tear down all inodes.
963 	 */
964 	xfs_inode_set_reclaim_tag(ip);
965 }
966 
967 /*
968  * Slab object creation initialisation for the XFS inode.
969  * This covers only the idempotent fields in the XFS inode;
970  * all other fields need to be initialised on allocation
971  * from the slab. This avoids the need to repeatedly initialise
972  * fields in the xfs inode that left in the initialise state
973  * when freeing the inode.
974  */
975 STATIC void
976 xfs_fs_inode_init_once(
977 	void			*inode)
978 {
979 	struct xfs_inode	*ip = inode;
980 
981 	memset(ip, 0, sizeof(struct xfs_inode));
982 
983 	/* vfs inode */
984 	inode_init_once(VFS_I(ip));
985 
986 	/* xfs inode */
987 	atomic_set(&ip->i_pincount, 0);
988 	spin_lock_init(&ip->i_flags_lock);
989 
990 	mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
991 		     "xfsino", ip->i_ino);
992 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
993 		     "xfsino", ip->i_ino);
994 }
995 
996 /*
997  * We do an unlocked check for XFS_IDONTCACHE here because we are already
998  * serialised against cache hits here via the inode->i_lock and igrab() in
999  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1000  * racing with us, and it avoids needing to grab a spinlock here for every inode
1001  * we drop the final reference on.
1002  */
1003 STATIC int
1004 xfs_fs_drop_inode(
1005 	struct inode		*inode)
1006 {
1007 	struct xfs_inode	*ip = XFS_I(inode);
1008 
1009 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1010 }
1011 
1012 STATIC void
1013 xfs_free_fsname(
1014 	struct xfs_mount	*mp)
1015 {
1016 	kfree(mp->m_fsname);
1017 	kfree(mp->m_rtname);
1018 	kfree(mp->m_logname);
1019 }
1020 
1021 STATIC int
1022 xfs_fs_sync_fs(
1023 	struct super_block	*sb,
1024 	int			wait)
1025 {
1026 	struct xfs_mount	*mp = XFS_M(sb);
1027 
1028 	/*
1029 	 * Doing anything during the async pass would be counterproductive.
1030 	 */
1031 	if (!wait)
1032 		return 0;
1033 
1034 	xfs_log_force(mp, XFS_LOG_SYNC);
1035 	if (laptop_mode) {
1036 		/*
1037 		 * The disk must be active because we're syncing.
1038 		 * We schedule log work now (now that the disk is
1039 		 * active) instead of later (when it might not be).
1040 		 */
1041 		flush_delayed_work(&mp->m_log->l_work);
1042 	}
1043 
1044 	return 0;
1045 }
1046 
1047 STATIC int
1048 xfs_fs_statfs(
1049 	struct dentry		*dentry,
1050 	struct kstatfs		*statp)
1051 {
1052 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1053 	xfs_sb_t		*sbp = &mp->m_sb;
1054 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1055 	__uint64_t		fakeinos, id;
1056 	__uint64_t		icount;
1057 	__uint64_t		ifree;
1058 	__uint64_t		fdblocks;
1059 	xfs_extlen_t		lsize;
1060 	__int64_t		ffree;
1061 
1062 	statp->f_type = XFS_SB_MAGIC;
1063 	statp->f_namelen = MAXNAMELEN - 1;
1064 
1065 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1066 	statp->f_fsid.val[0] = (u32)id;
1067 	statp->f_fsid.val[1] = (u32)(id >> 32);
1068 
1069 	icount = percpu_counter_sum(&mp->m_icount);
1070 	ifree = percpu_counter_sum(&mp->m_ifree);
1071 	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1072 
1073 	spin_lock(&mp->m_sb_lock);
1074 	statp->f_bsize = sbp->sb_blocksize;
1075 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1076 	statp->f_blocks = sbp->sb_dblocks - lsize;
1077 	spin_unlock(&mp->m_sb_lock);
1078 
1079 	statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1080 	statp->f_bavail = statp->f_bfree;
1081 
1082 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
1083 	statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1084 	if (mp->m_maxicount)
1085 		statp->f_files = min_t(typeof(statp->f_files),
1086 					statp->f_files,
1087 					mp->m_maxicount);
1088 
1089 	/* If sb_icount overshot maxicount, report actual allocation */
1090 	statp->f_files = max_t(typeof(statp->f_files),
1091 					statp->f_files,
1092 					sbp->sb_icount);
1093 
1094 	/* make sure statp->f_ffree does not underflow */
1095 	ffree = statp->f_files - (icount - ifree);
1096 	statp->f_ffree = max_t(__int64_t, ffree, 0);
1097 
1098 
1099 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1100 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1101 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1102 		xfs_qm_statvfs(ip, statp);
1103 	return 0;
1104 }
1105 
1106 STATIC void
1107 xfs_save_resvblks(struct xfs_mount *mp)
1108 {
1109 	__uint64_t resblks = 0;
1110 
1111 	mp->m_resblks_save = mp->m_resblks;
1112 	xfs_reserve_blocks(mp, &resblks, NULL);
1113 }
1114 
1115 STATIC void
1116 xfs_restore_resvblks(struct xfs_mount *mp)
1117 {
1118 	__uint64_t resblks;
1119 
1120 	if (mp->m_resblks_save) {
1121 		resblks = mp->m_resblks_save;
1122 		mp->m_resblks_save = 0;
1123 	} else
1124 		resblks = xfs_default_resblks(mp);
1125 
1126 	xfs_reserve_blocks(mp, &resblks, NULL);
1127 }
1128 
1129 /*
1130  * Trigger writeback of all the dirty metadata in the file system.
1131  *
1132  * This ensures that the metadata is written to their location on disk rather
1133  * than just existing in transactions in the log. This means after a quiesce
1134  * there is no log replay required to write the inodes to disk - this is the
1135  * primary difference between a sync and a quiesce.
1136  *
1137  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1138  * it is started again when appropriate.
1139  */
1140 static void
1141 xfs_quiesce_attr(
1142 	struct xfs_mount	*mp)
1143 {
1144 	int	error = 0;
1145 
1146 	/* wait for all modifications to complete */
1147 	while (atomic_read(&mp->m_active_trans) > 0)
1148 		delay(100);
1149 
1150 	/* force the log to unpin objects from the now complete transactions */
1151 	xfs_log_force(mp, XFS_LOG_SYNC);
1152 
1153 	/* reclaim inodes to do any IO before the freeze completes */
1154 	xfs_reclaim_inodes(mp, 0);
1155 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1156 
1157 	/* Push the superblock and write an unmount record */
1158 	error = xfs_log_sbcount(mp);
1159 	if (error)
1160 		xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1161 				"Frozen image may not be consistent.");
1162 	/*
1163 	 * Just warn here till VFS can correctly support
1164 	 * read-only remount without racing.
1165 	 */
1166 	WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1167 
1168 	xfs_log_quiesce(mp);
1169 }
1170 
1171 STATIC int
1172 xfs_test_remount_options(
1173 	struct super_block	*sb,
1174 	struct xfs_mount	*mp,
1175 	char			*options)
1176 {
1177 	int			error = 0;
1178 	struct xfs_mount	*tmp_mp;
1179 
1180 	tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1181 	if (!tmp_mp)
1182 		return -ENOMEM;
1183 
1184 	tmp_mp->m_super = sb;
1185 	error = xfs_parseargs(tmp_mp, options);
1186 	xfs_free_fsname(tmp_mp);
1187 	kfree(tmp_mp);
1188 
1189 	return error;
1190 }
1191 
1192 STATIC int
1193 xfs_fs_remount(
1194 	struct super_block	*sb,
1195 	int			*flags,
1196 	char			*options)
1197 {
1198 	struct xfs_mount	*mp = XFS_M(sb);
1199 	xfs_sb_t		*sbp = &mp->m_sb;
1200 	substring_t		args[MAX_OPT_ARGS];
1201 	char			*p;
1202 	int			error;
1203 
1204 	/* First, check for complete junk; i.e. invalid options */
1205 	error = xfs_test_remount_options(sb, mp, options);
1206 	if (error)
1207 		return error;
1208 
1209 	sync_filesystem(sb);
1210 	while ((p = strsep(&options, ",")) != NULL) {
1211 		int token;
1212 
1213 		if (!*p)
1214 			continue;
1215 
1216 		token = match_token(p, tokens, args);
1217 		switch (token) {
1218 		case Opt_barrier:
1219 			mp->m_flags |= XFS_MOUNT_BARRIER;
1220 			break;
1221 		case Opt_nobarrier:
1222 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
1223 			break;
1224 		case Opt_inode64:
1225 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1226 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1227 			break;
1228 		case Opt_inode32:
1229 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1230 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1231 			break;
1232 		default:
1233 			/*
1234 			 * Logically we would return an error here to prevent
1235 			 * users from believing they might have changed
1236 			 * mount options using remount which can't be changed.
1237 			 *
1238 			 * But unfortunately mount(8) adds all options from
1239 			 * mtab and fstab to the mount arguments in some cases
1240 			 * so we can't blindly reject options, but have to
1241 			 * check for each specified option if it actually
1242 			 * differs from the currently set option and only
1243 			 * reject it if that's the case.
1244 			 *
1245 			 * Until that is implemented we return success for
1246 			 * every remount request, and silently ignore all
1247 			 * options that we can't actually change.
1248 			 */
1249 #if 0
1250 			xfs_info(mp,
1251 		"mount option \"%s\" not supported for remount", p);
1252 			return -EINVAL;
1253 #else
1254 			break;
1255 #endif
1256 		}
1257 	}
1258 
1259 	/* ro -> rw */
1260 	if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1261 		if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1262 			xfs_warn(mp,
1263 		"ro->rw transition prohibited on norecovery mount");
1264 			return -EINVAL;
1265 		}
1266 
1267 		if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1268 		    xfs_sb_has_ro_compat_feature(sbp,
1269 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1270 			xfs_warn(mp,
1271 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1272 				(sbp->sb_features_ro_compat &
1273 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1274 			return -EINVAL;
1275 		}
1276 
1277 		mp->m_flags &= ~XFS_MOUNT_RDONLY;
1278 
1279 		/*
1280 		 * If this is the first remount to writeable state we
1281 		 * might have some superblock changes to update.
1282 		 */
1283 		if (mp->m_update_sb) {
1284 			error = xfs_sync_sb(mp, false);
1285 			if (error) {
1286 				xfs_warn(mp, "failed to write sb changes");
1287 				return error;
1288 			}
1289 			mp->m_update_sb = false;
1290 		}
1291 
1292 		/*
1293 		 * Fill out the reserve pool if it is empty. Use the stashed
1294 		 * value if it is non-zero, otherwise go with the default.
1295 		 */
1296 		xfs_restore_resvblks(mp);
1297 		xfs_log_work_queue(mp);
1298 		xfs_queue_eofblocks(mp);
1299 	}
1300 
1301 	/* rw -> ro */
1302 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1303 		/*
1304 		 * Before we sync the metadata, we need to free up the reserve
1305 		 * block pool so that the used block count in the superblock on
1306 		 * disk is correct at the end of the remount. Stash the current
1307 		 * reserve pool size so that if we get remounted rw, we can
1308 		 * return it to the same size.
1309 		 */
1310 		xfs_save_resvblks(mp);
1311 
1312 		/*
1313 		 * Cancel background eofb scanning so it cannot race with the
1314 		 * final log force+buftarg wait and deadlock the remount.
1315 		 */
1316 		cancel_delayed_work_sync(&mp->m_eofblocks_work);
1317 
1318 		xfs_quiesce_attr(mp);
1319 		mp->m_flags |= XFS_MOUNT_RDONLY;
1320 	}
1321 
1322 	return 0;
1323 }
1324 
1325 /*
1326  * Second stage of a freeze. The data is already frozen so we only
1327  * need to take care of the metadata. Once that's done sync the superblock
1328  * to the log to dirty it in case of a crash while frozen. This ensures that we
1329  * will recover the unlinked inode lists on the next mount.
1330  */
1331 STATIC int
1332 xfs_fs_freeze(
1333 	struct super_block	*sb)
1334 {
1335 	struct xfs_mount	*mp = XFS_M(sb);
1336 
1337 	xfs_save_resvblks(mp);
1338 	xfs_quiesce_attr(mp);
1339 	return xfs_sync_sb(mp, true);
1340 }
1341 
1342 STATIC int
1343 xfs_fs_unfreeze(
1344 	struct super_block	*sb)
1345 {
1346 	struct xfs_mount	*mp = XFS_M(sb);
1347 
1348 	xfs_restore_resvblks(mp);
1349 	xfs_log_work_queue(mp);
1350 	return 0;
1351 }
1352 
1353 STATIC int
1354 xfs_fs_show_options(
1355 	struct seq_file		*m,
1356 	struct dentry		*root)
1357 {
1358 	return xfs_showargs(XFS_M(root->d_sb), m);
1359 }
1360 
1361 /*
1362  * This function fills in xfs_mount_t fields based on mount args.
1363  * Note: the superblock _has_ now been read in.
1364  */
1365 STATIC int
1366 xfs_finish_flags(
1367 	struct xfs_mount	*mp)
1368 {
1369 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1370 
1371 	/* Fail a mount where the logbuf is smaller than the log stripe */
1372 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1373 		if (mp->m_logbsize <= 0 &&
1374 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1375 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1376 		} else if (mp->m_logbsize > 0 &&
1377 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1378 			xfs_warn(mp,
1379 		"logbuf size must be greater than or equal to log stripe size");
1380 			return -EINVAL;
1381 		}
1382 	} else {
1383 		/* Fail a mount if the logbuf is larger than 32K */
1384 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1385 			xfs_warn(mp,
1386 		"logbuf size for version 1 logs must be 16K or 32K");
1387 			return -EINVAL;
1388 		}
1389 	}
1390 
1391 	/*
1392 	 * V5 filesystems always use attr2 format for attributes.
1393 	 */
1394 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1395 	    (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1396 		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1397 			     "attr2 is always enabled for V5 filesystems.");
1398 		return -EINVAL;
1399 	}
1400 
1401 	/*
1402 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1403 	 * told by noattr2 to turn it off
1404 	 */
1405 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1406 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1407 		mp->m_flags |= XFS_MOUNT_ATTR2;
1408 
1409 	/*
1410 	 * prohibit r/w mounts of read-only filesystems
1411 	 */
1412 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1413 		xfs_warn(mp,
1414 			"cannot mount a read-only filesystem as read-write");
1415 		return -EROFS;
1416 	}
1417 
1418 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1419 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1420 	    !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1421 		xfs_warn(mp,
1422 		  "Super block does not support project and group quota together");
1423 		return -EINVAL;
1424 	}
1425 
1426 	return 0;
1427 }
1428 
1429 static int
1430 xfs_init_percpu_counters(
1431 	struct xfs_mount	*mp)
1432 {
1433 	int		error;
1434 
1435 	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1436 	if (error)
1437 		return -ENOMEM;
1438 
1439 	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1440 	if (error)
1441 		goto free_icount;
1442 
1443 	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1444 	if (error)
1445 		goto free_ifree;
1446 
1447 	return 0;
1448 
1449 free_ifree:
1450 	percpu_counter_destroy(&mp->m_ifree);
1451 free_icount:
1452 	percpu_counter_destroy(&mp->m_icount);
1453 	return -ENOMEM;
1454 }
1455 
1456 void
1457 xfs_reinit_percpu_counters(
1458 	struct xfs_mount	*mp)
1459 {
1460 	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1461 	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1462 	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1463 }
1464 
1465 static void
1466 xfs_destroy_percpu_counters(
1467 	struct xfs_mount	*mp)
1468 {
1469 	percpu_counter_destroy(&mp->m_icount);
1470 	percpu_counter_destroy(&mp->m_ifree);
1471 	percpu_counter_destroy(&mp->m_fdblocks);
1472 }
1473 
1474 STATIC int
1475 xfs_fs_fill_super(
1476 	struct super_block	*sb,
1477 	void			*data,
1478 	int			silent)
1479 {
1480 	struct inode		*root;
1481 	struct xfs_mount	*mp = NULL;
1482 	int			flags = 0, error = -ENOMEM;
1483 
1484 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1485 	if (!mp)
1486 		goto out;
1487 
1488 	spin_lock_init(&mp->m_sb_lock);
1489 	mutex_init(&mp->m_growlock);
1490 	atomic_set(&mp->m_active_trans, 0);
1491 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1492 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1493 	mp->m_kobj.kobject.kset = xfs_kset;
1494 
1495 	mp->m_super = sb;
1496 	sb->s_fs_info = mp;
1497 
1498 	error = xfs_parseargs(mp, (char *)data);
1499 	if (error)
1500 		goto out_free_fsname;
1501 
1502 	sb_min_blocksize(sb, BBSIZE);
1503 	sb->s_xattr = xfs_xattr_handlers;
1504 	sb->s_export_op = &xfs_export_operations;
1505 #ifdef CONFIG_XFS_QUOTA
1506 	sb->s_qcop = &xfs_quotactl_operations;
1507 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1508 #endif
1509 	sb->s_op = &xfs_super_operations;
1510 
1511 	if (silent)
1512 		flags |= XFS_MFSI_QUIET;
1513 
1514 	error = xfs_open_devices(mp);
1515 	if (error)
1516 		goto out_free_fsname;
1517 
1518 	error = xfs_init_mount_workqueues(mp);
1519 	if (error)
1520 		goto out_close_devices;
1521 
1522 	error = xfs_init_percpu_counters(mp);
1523 	if (error)
1524 		goto out_destroy_workqueues;
1525 
1526 	/* Allocate stats memory before we do operations that might use it */
1527 	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1528 	if (!mp->m_stats.xs_stats) {
1529 		error = -ENOMEM;
1530 		goto out_destroy_counters;
1531 	}
1532 
1533 	error = xfs_readsb(mp, flags);
1534 	if (error)
1535 		goto out_free_stats;
1536 
1537 	error = xfs_finish_flags(mp);
1538 	if (error)
1539 		goto out_free_sb;
1540 
1541 	error = xfs_setup_devices(mp);
1542 	if (error)
1543 		goto out_free_sb;
1544 
1545 	error = xfs_filestream_mount(mp);
1546 	if (error)
1547 		goto out_free_sb;
1548 
1549 	/*
1550 	 * we must configure the block size in the superblock before we run the
1551 	 * full mount process as the mount process can lookup and cache inodes.
1552 	 */
1553 	sb->s_magic = XFS_SB_MAGIC;
1554 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1555 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1556 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1557 	sb->s_max_links = XFS_MAXLINK;
1558 	sb->s_time_gran = 1;
1559 	set_posix_acl_flag(sb);
1560 
1561 	/* version 5 superblocks support inode version counters. */
1562 	if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1563 		sb->s_flags |= MS_I_VERSION;
1564 
1565 	if (mp->m_flags & XFS_MOUNT_DAX) {
1566 		xfs_warn(mp,
1567 		"DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1568 
1569 		error = bdev_dax_supported(sb, sb->s_blocksize);
1570 		if (error) {
1571 			xfs_alert(mp,
1572 			"DAX unsupported by block device. Turning off DAX.");
1573 			mp->m_flags &= ~XFS_MOUNT_DAX;
1574 		}
1575 	}
1576 
1577 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1578 		if (mp->m_sb.sb_rblocks) {
1579 			xfs_alert(mp,
1580 	"EXPERIMENTAL reverse mapping btree not compatible with realtime device!");
1581 			error = -EINVAL;
1582 			goto out_filestream_unmount;
1583 		}
1584 		xfs_alert(mp,
1585 	"EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
1586 	}
1587 
1588 	error = xfs_mountfs(mp);
1589 	if (error)
1590 		goto out_filestream_unmount;
1591 
1592 	root = igrab(VFS_I(mp->m_rootip));
1593 	if (!root) {
1594 		error = -ENOENT;
1595 		goto out_unmount;
1596 	}
1597 	sb->s_root = d_make_root(root);
1598 	if (!sb->s_root) {
1599 		error = -ENOMEM;
1600 		goto out_unmount;
1601 	}
1602 
1603 	return 0;
1604 
1605  out_filestream_unmount:
1606 	xfs_filestream_unmount(mp);
1607  out_free_sb:
1608 	xfs_freesb(mp);
1609  out_free_stats:
1610 	free_percpu(mp->m_stats.xs_stats);
1611  out_destroy_counters:
1612 	xfs_destroy_percpu_counters(mp);
1613  out_destroy_workqueues:
1614 	xfs_destroy_mount_workqueues(mp);
1615  out_close_devices:
1616 	xfs_close_devices(mp);
1617  out_free_fsname:
1618 	xfs_free_fsname(mp);
1619 	kfree(mp);
1620  out:
1621 	return error;
1622 
1623  out_unmount:
1624 	xfs_filestream_unmount(mp);
1625 	xfs_unmountfs(mp);
1626 	goto out_free_sb;
1627 }
1628 
1629 STATIC void
1630 xfs_fs_put_super(
1631 	struct super_block	*sb)
1632 {
1633 	struct xfs_mount	*mp = XFS_M(sb);
1634 
1635 	xfs_notice(mp, "Unmounting Filesystem");
1636 	xfs_filestream_unmount(mp);
1637 	xfs_unmountfs(mp);
1638 
1639 	xfs_freesb(mp);
1640 	free_percpu(mp->m_stats.xs_stats);
1641 	xfs_destroy_percpu_counters(mp);
1642 	xfs_destroy_mount_workqueues(mp);
1643 	xfs_close_devices(mp);
1644 	xfs_free_fsname(mp);
1645 	kfree(mp);
1646 }
1647 
1648 STATIC struct dentry *
1649 xfs_fs_mount(
1650 	struct file_system_type	*fs_type,
1651 	int			flags,
1652 	const char		*dev_name,
1653 	void			*data)
1654 {
1655 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1656 }
1657 
1658 static long
1659 xfs_fs_nr_cached_objects(
1660 	struct super_block	*sb,
1661 	struct shrink_control	*sc)
1662 {
1663 	return xfs_reclaim_inodes_count(XFS_M(sb));
1664 }
1665 
1666 static long
1667 xfs_fs_free_cached_objects(
1668 	struct super_block	*sb,
1669 	struct shrink_control	*sc)
1670 {
1671 	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1672 }
1673 
1674 static const struct super_operations xfs_super_operations = {
1675 	.alloc_inode		= xfs_fs_alloc_inode,
1676 	.destroy_inode		= xfs_fs_destroy_inode,
1677 	.drop_inode		= xfs_fs_drop_inode,
1678 	.put_super		= xfs_fs_put_super,
1679 	.sync_fs		= xfs_fs_sync_fs,
1680 	.freeze_fs		= xfs_fs_freeze,
1681 	.unfreeze_fs		= xfs_fs_unfreeze,
1682 	.statfs			= xfs_fs_statfs,
1683 	.remount_fs		= xfs_fs_remount,
1684 	.show_options		= xfs_fs_show_options,
1685 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1686 	.free_cached_objects	= xfs_fs_free_cached_objects,
1687 };
1688 
1689 static struct file_system_type xfs_fs_type = {
1690 	.owner			= THIS_MODULE,
1691 	.name			= "xfs",
1692 	.mount			= xfs_fs_mount,
1693 	.kill_sb		= kill_block_super,
1694 	.fs_flags		= FS_REQUIRES_DEV,
1695 };
1696 MODULE_ALIAS_FS("xfs");
1697 
1698 STATIC int __init
1699 xfs_init_zones(void)
1700 {
1701 	xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1702 			offsetof(struct xfs_ioend, io_inline_bio));
1703 	if (!xfs_ioend_bioset)
1704 		goto out;
1705 
1706 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1707 						"xfs_log_ticket");
1708 	if (!xfs_log_ticket_zone)
1709 		goto out_free_ioend_bioset;
1710 
1711 	xfs_bmap_free_item_zone = kmem_zone_init(
1712 			sizeof(struct xfs_extent_free_item),
1713 			"xfs_bmap_free_item");
1714 	if (!xfs_bmap_free_item_zone)
1715 		goto out_destroy_log_ticket_zone;
1716 
1717 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1718 						"xfs_btree_cur");
1719 	if (!xfs_btree_cur_zone)
1720 		goto out_destroy_bmap_free_item_zone;
1721 
1722 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1723 						"xfs_da_state");
1724 	if (!xfs_da_state_zone)
1725 		goto out_destroy_btree_cur_zone;
1726 
1727 	xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1728 	if (!xfs_ifork_zone)
1729 		goto out_destroy_da_state_zone;
1730 
1731 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1732 	if (!xfs_trans_zone)
1733 		goto out_destroy_ifork_zone;
1734 
1735 	xfs_log_item_desc_zone =
1736 		kmem_zone_init(sizeof(struct xfs_log_item_desc),
1737 			       "xfs_log_item_desc");
1738 	if (!xfs_log_item_desc_zone)
1739 		goto out_destroy_trans_zone;
1740 
1741 	/*
1742 	 * The size of the zone allocated buf log item is the maximum
1743 	 * size possible under XFS.  This wastes a little bit of memory,
1744 	 * but it is much faster.
1745 	 */
1746 	xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1747 					   "xfs_buf_item");
1748 	if (!xfs_buf_item_zone)
1749 		goto out_destroy_log_item_desc_zone;
1750 
1751 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1752 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1753 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1754 	if (!xfs_efd_zone)
1755 		goto out_destroy_buf_item_zone;
1756 
1757 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1758 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1759 				sizeof(xfs_extent_t))), "xfs_efi_item");
1760 	if (!xfs_efi_zone)
1761 		goto out_destroy_efd_zone;
1762 
1763 	xfs_inode_zone =
1764 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1765 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1766 			KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1767 	if (!xfs_inode_zone)
1768 		goto out_destroy_efi_zone;
1769 
1770 	xfs_ili_zone =
1771 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1772 					KM_ZONE_SPREAD, NULL);
1773 	if (!xfs_ili_zone)
1774 		goto out_destroy_inode_zone;
1775 	xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1776 					"xfs_icr");
1777 	if (!xfs_icreate_zone)
1778 		goto out_destroy_ili_zone;
1779 
1780 	xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1781 			"xfs_rud_item");
1782 	if (!xfs_rud_zone)
1783 		goto out_destroy_icreate_zone;
1784 
1785 	xfs_rui_zone = kmem_zone_init((sizeof(struct xfs_rui_log_item) +
1786 			((XFS_RUI_MAX_FAST_EXTENTS - 1) *
1787 				sizeof(struct xfs_map_extent))),
1788 			"xfs_rui_item");
1789 	if (!xfs_rui_zone)
1790 		goto out_destroy_rud_zone;
1791 
1792 	return 0;
1793 
1794  out_destroy_rud_zone:
1795 	kmem_zone_destroy(xfs_rud_zone);
1796  out_destroy_icreate_zone:
1797 	kmem_zone_destroy(xfs_icreate_zone);
1798  out_destroy_ili_zone:
1799 	kmem_zone_destroy(xfs_ili_zone);
1800  out_destroy_inode_zone:
1801 	kmem_zone_destroy(xfs_inode_zone);
1802  out_destroy_efi_zone:
1803 	kmem_zone_destroy(xfs_efi_zone);
1804  out_destroy_efd_zone:
1805 	kmem_zone_destroy(xfs_efd_zone);
1806  out_destroy_buf_item_zone:
1807 	kmem_zone_destroy(xfs_buf_item_zone);
1808  out_destroy_log_item_desc_zone:
1809 	kmem_zone_destroy(xfs_log_item_desc_zone);
1810  out_destroy_trans_zone:
1811 	kmem_zone_destroy(xfs_trans_zone);
1812  out_destroy_ifork_zone:
1813 	kmem_zone_destroy(xfs_ifork_zone);
1814  out_destroy_da_state_zone:
1815 	kmem_zone_destroy(xfs_da_state_zone);
1816  out_destroy_btree_cur_zone:
1817 	kmem_zone_destroy(xfs_btree_cur_zone);
1818  out_destroy_bmap_free_item_zone:
1819 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1820  out_destroy_log_ticket_zone:
1821 	kmem_zone_destroy(xfs_log_ticket_zone);
1822  out_free_ioend_bioset:
1823 	bioset_free(xfs_ioend_bioset);
1824  out:
1825 	return -ENOMEM;
1826 }
1827 
1828 STATIC void
1829 xfs_destroy_zones(void)
1830 {
1831 	/*
1832 	 * Make sure all delayed rcu free are flushed before we
1833 	 * destroy caches.
1834 	 */
1835 	rcu_barrier();
1836 	kmem_zone_destroy(xfs_rui_zone);
1837 	kmem_zone_destroy(xfs_rud_zone);
1838 	kmem_zone_destroy(xfs_icreate_zone);
1839 	kmem_zone_destroy(xfs_ili_zone);
1840 	kmem_zone_destroy(xfs_inode_zone);
1841 	kmem_zone_destroy(xfs_efi_zone);
1842 	kmem_zone_destroy(xfs_efd_zone);
1843 	kmem_zone_destroy(xfs_buf_item_zone);
1844 	kmem_zone_destroy(xfs_log_item_desc_zone);
1845 	kmem_zone_destroy(xfs_trans_zone);
1846 	kmem_zone_destroy(xfs_ifork_zone);
1847 	kmem_zone_destroy(xfs_da_state_zone);
1848 	kmem_zone_destroy(xfs_btree_cur_zone);
1849 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1850 	kmem_zone_destroy(xfs_log_ticket_zone);
1851 	bioset_free(xfs_ioend_bioset);
1852 }
1853 
1854 STATIC int __init
1855 xfs_init_workqueues(void)
1856 {
1857 	/*
1858 	 * The allocation workqueue can be used in memory reclaim situations
1859 	 * (writepage path), and parallelism is only limited by the number of
1860 	 * AGs in all the filesystems mounted. Hence use the default large
1861 	 * max_active value for this workqueue.
1862 	 */
1863 	xfs_alloc_wq = alloc_workqueue("xfsalloc",
1864 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1865 	if (!xfs_alloc_wq)
1866 		return -ENOMEM;
1867 
1868 	return 0;
1869 }
1870 
1871 STATIC void
1872 xfs_destroy_workqueues(void)
1873 {
1874 	destroy_workqueue(xfs_alloc_wq);
1875 }
1876 
1877 STATIC int __init
1878 init_xfs_fs(void)
1879 {
1880 	int			error;
1881 
1882 	xfs_check_ondisk_structs();
1883 
1884 	printk(KERN_INFO XFS_VERSION_STRING " with "
1885 			 XFS_BUILD_OPTIONS " enabled\n");
1886 
1887 	xfs_extent_free_init_defer_op();
1888 	xfs_rmap_update_init_defer_op();
1889 
1890 	xfs_dir_startup();
1891 
1892 	error = xfs_init_zones();
1893 	if (error)
1894 		goto out;
1895 
1896 	error = xfs_init_workqueues();
1897 	if (error)
1898 		goto out_destroy_zones;
1899 
1900 	error = xfs_mru_cache_init();
1901 	if (error)
1902 		goto out_destroy_wq;
1903 
1904 	error = xfs_buf_init();
1905 	if (error)
1906 		goto out_mru_cache_uninit;
1907 
1908 	error = xfs_init_procfs();
1909 	if (error)
1910 		goto out_buf_terminate;
1911 
1912 	error = xfs_sysctl_register();
1913 	if (error)
1914 		goto out_cleanup_procfs;
1915 
1916 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
1917 	if (!xfs_kset) {
1918 		error = -ENOMEM;
1919 		goto out_sysctl_unregister;
1920 	}
1921 
1922 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
1923 
1924 	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
1925 	if (!xfsstats.xs_stats) {
1926 		error = -ENOMEM;
1927 		goto out_kset_unregister;
1928 	}
1929 
1930 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
1931 			       "stats");
1932 	if (error)
1933 		goto out_free_stats;
1934 
1935 #ifdef DEBUG
1936 	xfs_dbg_kobj.kobject.kset = xfs_kset;
1937 	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
1938 	if (error)
1939 		goto out_remove_stats_kobj;
1940 #endif
1941 
1942 	error = xfs_qm_init();
1943 	if (error)
1944 		goto out_remove_dbg_kobj;
1945 
1946 	error = register_filesystem(&xfs_fs_type);
1947 	if (error)
1948 		goto out_qm_exit;
1949 	return 0;
1950 
1951  out_qm_exit:
1952 	xfs_qm_exit();
1953  out_remove_dbg_kobj:
1954 #ifdef DEBUG
1955 	xfs_sysfs_del(&xfs_dbg_kobj);
1956  out_remove_stats_kobj:
1957 #endif
1958 	xfs_sysfs_del(&xfsstats.xs_kobj);
1959  out_free_stats:
1960 	free_percpu(xfsstats.xs_stats);
1961  out_kset_unregister:
1962 	kset_unregister(xfs_kset);
1963  out_sysctl_unregister:
1964 	xfs_sysctl_unregister();
1965  out_cleanup_procfs:
1966 	xfs_cleanup_procfs();
1967  out_buf_terminate:
1968 	xfs_buf_terminate();
1969  out_mru_cache_uninit:
1970 	xfs_mru_cache_uninit();
1971  out_destroy_wq:
1972 	xfs_destroy_workqueues();
1973  out_destroy_zones:
1974 	xfs_destroy_zones();
1975  out:
1976 	return error;
1977 }
1978 
1979 STATIC void __exit
1980 exit_xfs_fs(void)
1981 {
1982 	xfs_qm_exit();
1983 	unregister_filesystem(&xfs_fs_type);
1984 #ifdef DEBUG
1985 	xfs_sysfs_del(&xfs_dbg_kobj);
1986 #endif
1987 	xfs_sysfs_del(&xfsstats.xs_kobj);
1988 	free_percpu(xfsstats.xs_stats);
1989 	kset_unregister(xfs_kset);
1990 	xfs_sysctl_unregister();
1991 	xfs_cleanup_procfs();
1992 	xfs_buf_terminate();
1993 	xfs_mru_cache_uninit();
1994 	xfs_destroy_workqueues();
1995 	xfs_destroy_zones();
1996 	xfs_uuid_table_free();
1997 }
1998 
1999 module_init(init_xfs_fs);
2000 module_exit(exit_xfs_fs);
2001 
2002 MODULE_AUTHOR("Silicon Graphics, Inc.");
2003 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2004 MODULE_LICENSE("GPL");
2005