xref: /linux/fs/xfs/xfs_super.c (revision 2c97b5ae83dca56718774e7b4bf9640f05d11867)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 
7 #include "xfs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_sb.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
21 #include "xfs_log.h"
22 #include "xfs_log_priv.h"
23 #include "xfs_dir2.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_mru_cache.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_icache.h"
28 #include "xfs_trace.h"
29 #include "xfs_icreate_item.h"
30 #include "xfs_filestream.h"
31 #include "xfs_quota.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_ondisk.h"
34 #include "xfs_rmap_item.h"
35 #include "xfs_refcount_item.h"
36 #include "xfs_bmap_item.h"
37 #include "xfs_reflink.h"
38 
39 #include <linux/magic.h>
40 #include <linux/parser.h>
41 
42 static const struct super_operations xfs_super_operations;
43 
44 static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
45 #ifdef DEBUG
46 static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
47 #endif
48 
49 /*
50  * Table driven mount option parser.
51  */
52 enum {
53 	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
54 	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
55 	Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
56 	Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
57 	Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
58 	Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
59 	Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
60 	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
61 	Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
62 };
63 
64 static const match_table_t tokens = {
65 	{Opt_logbufs,	"logbufs=%u"},	/* number of XFS log buffers */
66 	{Opt_logbsize,	"logbsize=%s"},	/* size of XFS log buffers */
67 	{Opt_logdev,	"logdev=%s"},	/* log device */
68 	{Opt_rtdev,	"rtdev=%s"},	/* realtime I/O device */
69 	{Opt_biosize,	"biosize=%u"},	/* log2 of preferred buffered io size */
70 	{Opt_wsync,	"wsync"},	/* safe-mode nfs compatible mount */
71 	{Opt_noalign,	"noalign"},	/* turn off stripe alignment */
72 	{Opt_swalloc,	"swalloc"},	/* turn on stripe width allocation */
73 	{Opt_sunit,	"sunit=%u"},	/* data volume stripe unit */
74 	{Opt_swidth,	"swidth=%u"},	/* data volume stripe width */
75 	{Opt_nouuid,	"nouuid"},	/* ignore filesystem UUID */
76 	{Opt_grpid,	"grpid"},	/* group-ID from parent directory */
77 	{Opt_nogrpid,	"nogrpid"},	/* group-ID from current process */
78 	{Opt_bsdgroups,	"bsdgroups"},	/* group-ID from parent directory */
79 	{Opt_sysvgroups,"sysvgroups"},	/* group-ID from current process */
80 	{Opt_allocsize,	"allocsize=%s"},/* preferred allocation size */
81 	{Opt_norecovery,"norecovery"},	/* don't run XFS recovery */
82 	{Opt_inode64,	"inode64"},	/* inodes can be allocated anywhere */
83 	{Opt_inode32,   "inode32"},	/* inode allocation limited to
84 					 * XFS_MAXINUMBER_32 */
85 	{Opt_ikeep,	"ikeep"},	/* do not free empty inode clusters */
86 	{Opt_noikeep,	"noikeep"},	/* free empty inode clusters */
87 	{Opt_largeio,	"largeio"},	/* report large I/O sizes in stat() */
88 	{Opt_nolargeio,	"nolargeio"},	/* do not report large I/O sizes
89 					 * in stat(). */
90 	{Opt_attr2,	"attr2"},	/* do use attr2 attribute format */
91 	{Opt_noattr2,	"noattr2"},	/* do not use attr2 attribute format */
92 	{Opt_filestreams,"filestreams"},/* use filestreams allocator */
93 	{Opt_quota,	"quota"},	/* disk quotas (user) */
94 	{Opt_noquota,	"noquota"},	/* no quotas */
95 	{Opt_usrquota,	"usrquota"},	/* user quota enabled */
96 	{Opt_grpquota,	"grpquota"},	/* group quota enabled */
97 	{Opt_prjquota,	"prjquota"},	/* project quota enabled */
98 	{Opt_uquota,	"uquota"},	/* user quota (IRIX variant) */
99 	{Opt_gquota,	"gquota"},	/* group quota (IRIX variant) */
100 	{Opt_pquota,	"pquota"},	/* project quota (IRIX variant) */
101 	{Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
102 	{Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
103 	{Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
104 	{Opt_qnoenforce, "qnoenforce"},	/* same as uqnoenforce */
105 	{Opt_discard,	"discard"},	/* Discard unused blocks */
106 	{Opt_nodiscard,	"nodiscard"},	/* Do not discard unused blocks */
107 	{Opt_dax,	"dax"},		/* Enable direct access to bdev pages */
108 	{Opt_err,	NULL},
109 };
110 
111 
112 STATIC int
113 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
114 {
115 	int	last, shift_left_factor = 0, _res;
116 	char	*value;
117 	int	ret = 0;
118 
119 	value = match_strdup(s);
120 	if (!value)
121 		return -ENOMEM;
122 
123 	last = strlen(value) - 1;
124 	if (value[last] == 'K' || value[last] == 'k') {
125 		shift_left_factor = 10;
126 		value[last] = '\0';
127 	}
128 	if (value[last] == 'M' || value[last] == 'm') {
129 		shift_left_factor = 20;
130 		value[last] = '\0';
131 	}
132 	if (value[last] == 'G' || value[last] == 'g') {
133 		shift_left_factor = 30;
134 		value[last] = '\0';
135 	}
136 
137 	if (kstrtoint(value, base, &_res))
138 		ret = -EINVAL;
139 	kfree(value);
140 	*res = _res << shift_left_factor;
141 	return ret;
142 }
143 
144 /*
145  * This function fills in xfs_mount_t fields based on mount args.
146  * Note: the superblock has _not_ yet been read in.
147  *
148  * Note that this function leaks the various device name allocations on
149  * failure.  The caller takes care of them.
150  *
151  * *sb is const because this is also used to test options on the remount
152  * path, and we don't want this to have any side effects at remount time.
153  * Today this function does not change *sb, but just to future-proof...
154  */
155 STATIC int
156 xfs_parseargs(
157 	struct xfs_mount	*mp,
158 	char			*options)
159 {
160 	const struct super_block *sb = mp->m_super;
161 	char			*p;
162 	substring_t		args[MAX_OPT_ARGS];
163 	int			dsunit = 0;
164 	int			dswidth = 0;
165 	int			iosize = 0;
166 	uint8_t			iosizelog = 0;
167 
168 	/*
169 	 * set up the mount name first so all the errors will refer to the
170 	 * correct device.
171 	 */
172 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
173 	if (!mp->m_fsname)
174 		return -ENOMEM;
175 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
176 
177 	/*
178 	 * Copy binary VFS mount flags we are interested in.
179 	 */
180 	if (sb_rdonly(sb))
181 		mp->m_flags |= XFS_MOUNT_RDONLY;
182 	if (sb->s_flags & SB_DIRSYNC)
183 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
184 	if (sb->s_flags & SB_SYNCHRONOUS)
185 		mp->m_flags |= XFS_MOUNT_WSYNC;
186 
187 	/*
188 	 * Set some default flags that could be cleared by the mount option
189 	 * parsing.
190 	 */
191 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
192 
193 	/*
194 	 * These can be overridden by the mount option parsing.
195 	 */
196 	mp->m_logbufs = -1;
197 	mp->m_logbsize = -1;
198 
199 	if (!options)
200 		goto done;
201 
202 	while ((p = strsep(&options, ",")) != NULL) {
203 		int		token;
204 
205 		if (!*p)
206 			continue;
207 
208 		token = match_token(p, tokens, args);
209 		switch (token) {
210 		case Opt_logbufs:
211 			if (match_int(args, &mp->m_logbufs))
212 				return -EINVAL;
213 			break;
214 		case Opt_logbsize:
215 			if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
216 				return -EINVAL;
217 			break;
218 		case Opt_logdev:
219 			kfree(mp->m_logname);
220 			mp->m_logname = match_strdup(args);
221 			if (!mp->m_logname)
222 				return -ENOMEM;
223 			break;
224 		case Opt_rtdev:
225 			kfree(mp->m_rtname);
226 			mp->m_rtname = match_strdup(args);
227 			if (!mp->m_rtname)
228 				return -ENOMEM;
229 			break;
230 		case Opt_allocsize:
231 		case Opt_biosize:
232 			if (suffix_kstrtoint(args, 10, &iosize))
233 				return -EINVAL;
234 			iosizelog = ffs(iosize) - 1;
235 			break;
236 		case Opt_grpid:
237 		case Opt_bsdgroups:
238 			mp->m_flags |= XFS_MOUNT_GRPID;
239 			break;
240 		case Opt_nogrpid:
241 		case Opt_sysvgroups:
242 			mp->m_flags &= ~XFS_MOUNT_GRPID;
243 			break;
244 		case Opt_wsync:
245 			mp->m_flags |= XFS_MOUNT_WSYNC;
246 			break;
247 		case Opt_norecovery:
248 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
249 			break;
250 		case Opt_noalign:
251 			mp->m_flags |= XFS_MOUNT_NOALIGN;
252 			break;
253 		case Opt_swalloc:
254 			mp->m_flags |= XFS_MOUNT_SWALLOC;
255 			break;
256 		case Opt_sunit:
257 			if (match_int(args, &dsunit))
258 				return -EINVAL;
259 			break;
260 		case Opt_swidth:
261 			if (match_int(args, &dswidth))
262 				return -EINVAL;
263 			break;
264 		case Opt_inode32:
265 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
266 			break;
267 		case Opt_inode64:
268 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
269 			break;
270 		case Opt_nouuid:
271 			mp->m_flags |= XFS_MOUNT_NOUUID;
272 			break;
273 		case Opt_ikeep:
274 			mp->m_flags |= XFS_MOUNT_IKEEP;
275 			break;
276 		case Opt_noikeep:
277 			mp->m_flags &= ~XFS_MOUNT_IKEEP;
278 			break;
279 		case Opt_largeio:
280 			mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
281 			break;
282 		case Opt_nolargeio:
283 			mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
284 			break;
285 		case Opt_attr2:
286 			mp->m_flags |= XFS_MOUNT_ATTR2;
287 			break;
288 		case Opt_noattr2:
289 			mp->m_flags &= ~XFS_MOUNT_ATTR2;
290 			mp->m_flags |= XFS_MOUNT_NOATTR2;
291 			break;
292 		case Opt_filestreams:
293 			mp->m_flags |= XFS_MOUNT_FILESTREAMS;
294 			break;
295 		case Opt_noquota:
296 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
297 			mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
298 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
299 			break;
300 		case Opt_quota:
301 		case Opt_uquota:
302 		case Opt_usrquota:
303 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
304 					 XFS_UQUOTA_ENFD);
305 			break;
306 		case Opt_qnoenforce:
307 		case Opt_uqnoenforce:
308 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
309 			mp->m_qflags &= ~XFS_UQUOTA_ENFD;
310 			break;
311 		case Opt_pquota:
312 		case Opt_prjquota:
313 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
314 					 XFS_PQUOTA_ENFD);
315 			break;
316 		case Opt_pqnoenforce:
317 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
318 			mp->m_qflags &= ~XFS_PQUOTA_ENFD;
319 			break;
320 		case Opt_gquota:
321 		case Opt_grpquota:
322 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
323 					 XFS_GQUOTA_ENFD);
324 			break;
325 		case Opt_gqnoenforce:
326 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
327 			mp->m_qflags &= ~XFS_GQUOTA_ENFD;
328 			break;
329 		case Opt_discard:
330 			mp->m_flags |= XFS_MOUNT_DISCARD;
331 			break;
332 		case Opt_nodiscard:
333 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
334 			break;
335 #ifdef CONFIG_FS_DAX
336 		case Opt_dax:
337 			mp->m_flags |= XFS_MOUNT_DAX;
338 			break;
339 #endif
340 		default:
341 			xfs_warn(mp, "unknown mount option [%s].", p);
342 			return -EINVAL;
343 		}
344 	}
345 
346 	/*
347 	 * no recovery flag requires a read-only mount
348 	 */
349 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
350 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
351 		xfs_warn(mp, "no-recovery mounts must be read-only.");
352 		return -EINVAL;
353 	}
354 
355 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
356 		xfs_warn(mp,
357 	"sunit and swidth options incompatible with the noalign option");
358 		return -EINVAL;
359 	}
360 
361 #ifndef CONFIG_XFS_QUOTA
362 	if (XFS_IS_QUOTA_RUNNING(mp)) {
363 		xfs_warn(mp, "quota support not available in this kernel.");
364 		return -EINVAL;
365 	}
366 #endif
367 
368 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
369 		xfs_warn(mp, "sunit and swidth must be specified together");
370 		return -EINVAL;
371 	}
372 
373 	if (dsunit && (dswidth % dsunit != 0)) {
374 		xfs_warn(mp,
375 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
376 			dswidth, dsunit);
377 		return -EINVAL;
378 	}
379 
380 done:
381 	if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
382 		/*
383 		 * At this point the superblock has not been read
384 		 * in, therefore we do not know the block size.
385 		 * Before the mount call ends we will convert
386 		 * these to FSBs.
387 		 */
388 		mp->m_dalign = dsunit;
389 		mp->m_swidth = dswidth;
390 	}
391 
392 	if (mp->m_logbufs != -1 &&
393 	    mp->m_logbufs != 0 &&
394 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
395 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
396 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
397 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
398 		return -EINVAL;
399 	}
400 	if (mp->m_logbsize != -1 &&
401 	    mp->m_logbsize !=  0 &&
402 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
403 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
404 	     !is_power_of_2(mp->m_logbsize))) {
405 		xfs_warn(mp,
406 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
407 			mp->m_logbsize);
408 		return -EINVAL;
409 	}
410 
411 	if (iosizelog) {
412 		if (iosizelog > XFS_MAX_IO_LOG ||
413 		    iosizelog < XFS_MIN_IO_LOG) {
414 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
415 				iosizelog, XFS_MIN_IO_LOG,
416 				XFS_MAX_IO_LOG);
417 			return -EINVAL;
418 		}
419 
420 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
421 		mp->m_readio_log = iosizelog;
422 		mp->m_writeio_log = iosizelog;
423 	}
424 
425 	return 0;
426 }
427 
428 struct proc_xfs_info {
429 	uint64_t	flag;
430 	char		*str;
431 };
432 
433 STATIC void
434 xfs_showargs(
435 	struct xfs_mount	*mp,
436 	struct seq_file		*m)
437 {
438 	static struct proc_xfs_info xfs_info_set[] = {
439 		/* the few simple ones we can get from the mount struct */
440 		{ XFS_MOUNT_IKEEP,		",ikeep" },
441 		{ XFS_MOUNT_WSYNC,		",wsync" },
442 		{ XFS_MOUNT_NOALIGN,		",noalign" },
443 		{ XFS_MOUNT_SWALLOC,		",swalloc" },
444 		{ XFS_MOUNT_NOUUID,		",nouuid" },
445 		{ XFS_MOUNT_NORECOVERY,		",norecovery" },
446 		{ XFS_MOUNT_ATTR2,		",attr2" },
447 		{ XFS_MOUNT_FILESTREAMS,	",filestreams" },
448 		{ XFS_MOUNT_GRPID,		",grpid" },
449 		{ XFS_MOUNT_DISCARD,		",discard" },
450 		{ XFS_MOUNT_SMALL_INUMS,	",inode32" },
451 		{ XFS_MOUNT_DAX,		",dax" },
452 		{ 0, NULL }
453 	};
454 	static struct proc_xfs_info xfs_info_unset[] = {
455 		/* the few simple ones we can get from the mount struct */
456 		{ XFS_MOUNT_COMPAT_IOSIZE,	",largeio" },
457 		{ XFS_MOUNT_SMALL_INUMS,	",inode64" },
458 		{ 0, NULL }
459 	};
460 	struct proc_xfs_info	*xfs_infop;
461 
462 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
463 		if (mp->m_flags & xfs_infop->flag)
464 			seq_puts(m, xfs_infop->str);
465 	}
466 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
467 		if (!(mp->m_flags & xfs_infop->flag))
468 			seq_puts(m, xfs_infop->str);
469 	}
470 
471 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
472 		seq_printf(m, ",allocsize=%dk",
473 				(int)(1 << mp->m_writeio_log) >> 10);
474 
475 	if (mp->m_logbufs > 0)
476 		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
477 	if (mp->m_logbsize > 0)
478 		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
479 
480 	if (mp->m_logname)
481 		seq_show_option(m, "logdev", mp->m_logname);
482 	if (mp->m_rtname)
483 		seq_show_option(m, "rtdev", mp->m_rtname);
484 
485 	if (mp->m_dalign > 0)
486 		seq_printf(m, ",sunit=%d",
487 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
488 	if (mp->m_swidth > 0)
489 		seq_printf(m, ",swidth=%d",
490 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
491 
492 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
493 		seq_puts(m, ",usrquota");
494 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
495 		seq_puts(m, ",uqnoenforce");
496 
497 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
498 		if (mp->m_qflags & XFS_PQUOTA_ENFD)
499 			seq_puts(m, ",prjquota");
500 		else
501 			seq_puts(m, ",pqnoenforce");
502 	}
503 	if (mp->m_qflags & XFS_GQUOTA_ACCT) {
504 		if (mp->m_qflags & XFS_GQUOTA_ENFD)
505 			seq_puts(m, ",grpquota");
506 		else
507 			seq_puts(m, ",gqnoenforce");
508 	}
509 
510 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
511 		seq_puts(m, ",noquota");
512 }
513 
514 static uint64_t
515 xfs_max_file_offset(
516 	unsigned int		blockshift)
517 {
518 	unsigned int		pagefactor = 1;
519 	unsigned int		bitshift = BITS_PER_LONG - 1;
520 
521 	/* Figure out maximum filesize, on Linux this can depend on
522 	 * the filesystem blocksize (on 32 bit platforms).
523 	 * __block_write_begin does this in an [unsigned] long long...
524 	 *      page->index << (PAGE_SHIFT - bbits)
525 	 * So, for page sized blocks (4K on 32 bit platforms),
526 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
527 	 *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
528 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
529 	 */
530 
531 #if BITS_PER_LONG == 32
532 	ASSERT(sizeof(sector_t) == 8);
533 	pagefactor = PAGE_SIZE;
534 	bitshift = BITS_PER_LONG;
535 #endif
536 
537 	return (((uint64_t)pagefactor) << bitshift) - 1;
538 }
539 
540 /*
541  * Set parameters for inode allocation heuristics, taking into account
542  * filesystem size and inode32/inode64 mount options; i.e. specifically
543  * whether or not XFS_MOUNT_SMALL_INUMS is set.
544  *
545  * Inode allocation patterns are altered only if inode32 is requested
546  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
547  * If altered, XFS_MOUNT_32BITINODES is set as well.
548  *
549  * An agcount independent of that in the mount structure is provided
550  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
551  * to the potentially higher ag count.
552  *
553  * Returns the maximum AG index which may contain inodes.
554  */
555 xfs_agnumber_t
556 xfs_set_inode_alloc(
557 	struct xfs_mount *mp,
558 	xfs_agnumber_t	agcount)
559 {
560 	xfs_agnumber_t	index;
561 	xfs_agnumber_t	maxagi = 0;
562 	xfs_sb_t	*sbp = &mp->m_sb;
563 	xfs_agnumber_t	max_metadata;
564 	xfs_agino_t	agino;
565 	xfs_ino_t	ino;
566 
567 	/*
568 	 * Calculate how much should be reserved for inodes to meet
569 	 * the max inode percentage.  Used only for inode32.
570 	 */
571 	if (M_IGEO(mp)->maxicount) {
572 		uint64_t	icount;
573 
574 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
575 		do_div(icount, 100);
576 		icount += sbp->sb_agblocks - 1;
577 		do_div(icount, sbp->sb_agblocks);
578 		max_metadata = icount;
579 	} else {
580 		max_metadata = agcount;
581 	}
582 
583 	/* Get the last possible inode in the filesystem */
584 	agino =	XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
585 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
586 
587 	/*
588 	 * If user asked for no more than 32-bit inodes, and the fs is
589 	 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
590 	 * the allocator to accommodate the request.
591 	 */
592 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
593 		mp->m_flags |= XFS_MOUNT_32BITINODES;
594 	else
595 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
596 
597 	for (index = 0; index < agcount; index++) {
598 		struct xfs_perag	*pag;
599 
600 		ino = XFS_AGINO_TO_INO(mp, index, agino);
601 
602 		pag = xfs_perag_get(mp, index);
603 
604 		if (mp->m_flags & XFS_MOUNT_32BITINODES) {
605 			if (ino > XFS_MAXINUMBER_32) {
606 				pag->pagi_inodeok = 0;
607 				pag->pagf_metadata = 0;
608 			} else {
609 				pag->pagi_inodeok = 1;
610 				maxagi++;
611 				if (index < max_metadata)
612 					pag->pagf_metadata = 1;
613 				else
614 					pag->pagf_metadata = 0;
615 			}
616 		} else {
617 			pag->pagi_inodeok = 1;
618 			pag->pagf_metadata = 0;
619 		}
620 
621 		xfs_perag_put(pag);
622 	}
623 
624 	return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
625 }
626 
627 STATIC int
628 xfs_blkdev_get(
629 	xfs_mount_t		*mp,
630 	const char		*name,
631 	struct block_device	**bdevp)
632 {
633 	int			error = 0;
634 
635 	*bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
636 				    mp);
637 	if (IS_ERR(*bdevp)) {
638 		error = PTR_ERR(*bdevp);
639 		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
640 	}
641 
642 	return error;
643 }
644 
645 STATIC void
646 xfs_blkdev_put(
647 	struct block_device	*bdev)
648 {
649 	if (bdev)
650 		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
651 }
652 
653 void
654 xfs_blkdev_issue_flush(
655 	xfs_buftarg_t		*buftarg)
656 {
657 	blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
658 }
659 
660 STATIC void
661 xfs_close_devices(
662 	struct xfs_mount	*mp)
663 {
664 	struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
665 
666 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
667 		struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
668 		struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
669 
670 		xfs_free_buftarg(mp->m_logdev_targp);
671 		xfs_blkdev_put(logdev);
672 		fs_put_dax(dax_logdev);
673 	}
674 	if (mp->m_rtdev_targp) {
675 		struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
676 		struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
677 
678 		xfs_free_buftarg(mp->m_rtdev_targp);
679 		xfs_blkdev_put(rtdev);
680 		fs_put_dax(dax_rtdev);
681 	}
682 	xfs_free_buftarg(mp->m_ddev_targp);
683 	fs_put_dax(dax_ddev);
684 }
685 
686 /*
687  * The file system configurations are:
688  *	(1) device (partition) with data and internal log
689  *	(2) logical volume with data and log subvolumes.
690  *	(3) logical volume with data, log, and realtime subvolumes.
691  *
692  * We only have to handle opening the log and realtime volumes here if
693  * they are present.  The data subvolume has already been opened by
694  * get_sb_bdev() and is stored in sb->s_bdev.
695  */
696 STATIC int
697 xfs_open_devices(
698 	struct xfs_mount	*mp)
699 {
700 	struct block_device	*ddev = mp->m_super->s_bdev;
701 	struct dax_device	*dax_ddev = fs_dax_get_by_bdev(ddev);
702 	struct dax_device	*dax_logdev = NULL, *dax_rtdev = NULL;
703 	struct block_device	*logdev = NULL, *rtdev = NULL;
704 	int			error;
705 
706 	/*
707 	 * Open real time and log devices - order is important.
708 	 */
709 	if (mp->m_logname) {
710 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
711 		if (error)
712 			goto out;
713 		dax_logdev = fs_dax_get_by_bdev(logdev);
714 	}
715 
716 	if (mp->m_rtname) {
717 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
718 		if (error)
719 			goto out_close_logdev;
720 
721 		if (rtdev == ddev || rtdev == logdev) {
722 			xfs_warn(mp,
723 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
724 			error = -EINVAL;
725 			goto out_close_rtdev;
726 		}
727 		dax_rtdev = fs_dax_get_by_bdev(rtdev);
728 	}
729 
730 	/*
731 	 * Setup xfs_mount buffer target pointers
732 	 */
733 	error = -ENOMEM;
734 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
735 	if (!mp->m_ddev_targp)
736 		goto out_close_rtdev;
737 
738 	if (rtdev) {
739 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
740 		if (!mp->m_rtdev_targp)
741 			goto out_free_ddev_targ;
742 	}
743 
744 	if (logdev && logdev != ddev) {
745 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
746 		if (!mp->m_logdev_targp)
747 			goto out_free_rtdev_targ;
748 	} else {
749 		mp->m_logdev_targp = mp->m_ddev_targp;
750 	}
751 
752 	return 0;
753 
754  out_free_rtdev_targ:
755 	if (mp->m_rtdev_targp)
756 		xfs_free_buftarg(mp->m_rtdev_targp);
757  out_free_ddev_targ:
758 	xfs_free_buftarg(mp->m_ddev_targp);
759  out_close_rtdev:
760 	xfs_blkdev_put(rtdev);
761 	fs_put_dax(dax_rtdev);
762  out_close_logdev:
763 	if (logdev && logdev != ddev) {
764 		xfs_blkdev_put(logdev);
765 		fs_put_dax(dax_logdev);
766 	}
767  out:
768 	fs_put_dax(dax_ddev);
769 	return error;
770 }
771 
772 /*
773  * Setup xfs_mount buffer target pointers based on superblock
774  */
775 STATIC int
776 xfs_setup_devices(
777 	struct xfs_mount	*mp)
778 {
779 	int			error;
780 
781 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
782 	if (error)
783 		return error;
784 
785 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
786 		unsigned int	log_sector_size = BBSIZE;
787 
788 		if (xfs_sb_version_hassector(&mp->m_sb))
789 			log_sector_size = mp->m_sb.sb_logsectsize;
790 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
791 					    log_sector_size);
792 		if (error)
793 			return error;
794 	}
795 	if (mp->m_rtdev_targp) {
796 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
797 					    mp->m_sb.sb_sectsize);
798 		if (error)
799 			return error;
800 	}
801 
802 	return 0;
803 }
804 
805 STATIC int
806 xfs_init_mount_workqueues(
807 	struct xfs_mount	*mp)
808 {
809 	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
810 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
811 	if (!mp->m_buf_workqueue)
812 		goto out;
813 
814 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
815 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
816 	if (!mp->m_unwritten_workqueue)
817 		goto out_destroy_buf;
818 
819 	mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
820 			WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND,
821 			0, mp->m_fsname);
822 	if (!mp->m_cil_workqueue)
823 		goto out_destroy_unwritten;
824 
825 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
826 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
827 	if (!mp->m_reclaim_workqueue)
828 		goto out_destroy_cil;
829 
830 	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
831 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
832 	if (!mp->m_eofblocks_workqueue)
833 		goto out_destroy_reclaim;
834 
835 	mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
836 					       mp->m_fsname);
837 	if (!mp->m_sync_workqueue)
838 		goto out_destroy_eofb;
839 
840 	return 0;
841 
842 out_destroy_eofb:
843 	destroy_workqueue(mp->m_eofblocks_workqueue);
844 out_destroy_reclaim:
845 	destroy_workqueue(mp->m_reclaim_workqueue);
846 out_destroy_cil:
847 	destroy_workqueue(mp->m_cil_workqueue);
848 out_destroy_unwritten:
849 	destroy_workqueue(mp->m_unwritten_workqueue);
850 out_destroy_buf:
851 	destroy_workqueue(mp->m_buf_workqueue);
852 out:
853 	return -ENOMEM;
854 }
855 
856 STATIC void
857 xfs_destroy_mount_workqueues(
858 	struct xfs_mount	*mp)
859 {
860 	destroy_workqueue(mp->m_sync_workqueue);
861 	destroy_workqueue(mp->m_eofblocks_workqueue);
862 	destroy_workqueue(mp->m_reclaim_workqueue);
863 	destroy_workqueue(mp->m_cil_workqueue);
864 	destroy_workqueue(mp->m_unwritten_workqueue);
865 	destroy_workqueue(mp->m_buf_workqueue);
866 }
867 
868 /*
869  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
870  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
871  * for IO to complete so that we effectively throttle multiple callers to the
872  * rate at which IO is completing.
873  */
874 void
875 xfs_flush_inodes(
876 	struct xfs_mount	*mp)
877 {
878 	struct super_block	*sb = mp->m_super;
879 
880 	if (down_read_trylock(&sb->s_umount)) {
881 		sync_inodes_sb(sb);
882 		up_read(&sb->s_umount);
883 	}
884 }
885 
886 /* Catch misguided souls that try to use this interface on XFS */
887 STATIC struct inode *
888 xfs_fs_alloc_inode(
889 	struct super_block	*sb)
890 {
891 	BUG();
892 	return NULL;
893 }
894 
895 #ifdef DEBUG
896 static void
897 xfs_check_delalloc(
898 	struct xfs_inode	*ip,
899 	int			whichfork)
900 {
901 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
902 	struct xfs_bmbt_irec	got;
903 	struct xfs_iext_cursor	icur;
904 
905 	if (!ifp || !xfs_iext_lookup_extent(ip, ifp, 0, &icur, &got))
906 		return;
907 	do {
908 		if (isnullstartblock(got.br_startblock)) {
909 			xfs_warn(ip->i_mount,
910 	"ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
911 				ip->i_ino,
912 				whichfork == XFS_DATA_FORK ? "data" : "cow",
913 				got.br_startoff, got.br_blockcount);
914 		}
915 	} while (xfs_iext_next_extent(ifp, &icur, &got));
916 }
917 #else
918 #define xfs_check_delalloc(ip, whichfork)	do { } while (0)
919 #endif
920 
921 /*
922  * Now that the generic code is guaranteed not to be accessing
923  * the linux inode, we can inactivate and reclaim the inode.
924  */
925 STATIC void
926 xfs_fs_destroy_inode(
927 	struct inode		*inode)
928 {
929 	struct xfs_inode	*ip = XFS_I(inode);
930 
931 	trace_xfs_destroy_inode(ip);
932 
933 	ASSERT(!rwsem_is_locked(&inode->i_rwsem));
934 	XFS_STATS_INC(ip->i_mount, vn_rele);
935 	XFS_STATS_INC(ip->i_mount, vn_remove);
936 
937 	xfs_inactive(ip);
938 
939 	if (!XFS_FORCED_SHUTDOWN(ip->i_mount) && ip->i_delayed_blks) {
940 		xfs_check_delalloc(ip, XFS_DATA_FORK);
941 		xfs_check_delalloc(ip, XFS_COW_FORK);
942 		ASSERT(0);
943 	}
944 
945 	XFS_STATS_INC(ip->i_mount, vn_reclaim);
946 
947 	/*
948 	 * We should never get here with one of the reclaim flags already set.
949 	 */
950 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
951 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
952 
953 	/*
954 	 * We always use background reclaim here because even if the
955 	 * inode is clean, it still may be under IO and hence we have
956 	 * to take the flush lock. The background reclaim path handles
957 	 * this more efficiently than we can here, so simply let background
958 	 * reclaim tear down all inodes.
959 	 */
960 	xfs_inode_set_reclaim_tag(ip);
961 }
962 
963 static void
964 xfs_fs_dirty_inode(
965 	struct inode			*inode,
966 	int				flag)
967 {
968 	struct xfs_inode		*ip = XFS_I(inode);
969 	struct xfs_mount		*mp = ip->i_mount;
970 	struct xfs_trans		*tp;
971 
972 	if (!(inode->i_sb->s_flags & SB_LAZYTIME))
973 		return;
974 	if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
975 		return;
976 
977 	if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
978 		return;
979 	xfs_ilock(ip, XFS_ILOCK_EXCL);
980 	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
981 	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
982 	xfs_trans_commit(tp);
983 }
984 
985 /*
986  * Slab object creation initialisation for the XFS inode.
987  * This covers only the idempotent fields in the XFS inode;
988  * all other fields need to be initialised on allocation
989  * from the slab. This avoids the need to repeatedly initialise
990  * fields in the xfs inode that left in the initialise state
991  * when freeing the inode.
992  */
993 STATIC void
994 xfs_fs_inode_init_once(
995 	void			*inode)
996 {
997 	struct xfs_inode	*ip = inode;
998 
999 	memset(ip, 0, sizeof(struct xfs_inode));
1000 
1001 	/* vfs inode */
1002 	inode_init_once(VFS_I(ip));
1003 
1004 	/* xfs inode */
1005 	atomic_set(&ip->i_pincount, 0);
1006 	spin_lock_init(&ip->i_flags_lock);
1007 
1008 	mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1009 		     "xfsino", ip->i_ino);
1010 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1011 		     "xfsino", ip->i_ino);
1012 }
1013 
1014 /*
1015  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1016  * serialised against cache hits here via the inode->i_lock and igrab() in
1017  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1018  * racing with us, and it avoids needing to grab a spinlock here for every inode
1019  * we drop the final reference on.
1020  */
1021 STATIC int
1022 xfs_fs_drop_inode(
1023 	struct inode		*inode)
1024 {
1025 	struct xfs_inode	*ip = XFS_I(inode);
1026 
1027 	/*
1028 	 * If this unlinked inode is in the middle of recovery, don't
1029 	 * drop the inode just yet; log recovery will take care of
1030 	 * that.  See the comment for this inode flag.
1031 	 */
1032 	if (ip->i_flags & XFS_IRECOVERY) {
1033 		ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1034 		return 0;
1035 	}
1036 
1037 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1038 }
1039 
1040 STATIC void
1041 xfs_free_fsname(
1042 	struct xfs_mount	*mp)
1043 {
1044 	kfree(mp->m_fsname);
1045 	kfree(mp->m_rtname);
1046 	kfree(mp->m_logname);
1047 }
1048 
1049 STATIC int
1050 xfs_fs_sync_fs(
1051 	struct super_block	*sb,
1052 	int			wait)
1053 {
1054 	struct xfs_mount	*mp = XFS_M(sb);
1055 
1056 	/*
1057 	 * Doing anything during the async pass would be counterproductive.
1058 	 */
1059 	if (!wait)
1060 		return 0;
1061 
1062 	xfs_log_force(mp, XFS_LOG_SYNC);
1063 	if (laptop_mode) {
1064 		/*
1065 		 * The disk must be active because we're syncing.
1066 		 * We schedule log work now (now that the disk is
1067 		 * active) instead of later (when it might not be).
1068 		 */
1069 		flush_delayed_work(&mp->m_log->l_work);
1070 	}
1071 
1072 	return 0;
1073 }
1074 
1075 STATIC int
1076 xfs_fs_statfs(
1077 	struct dentry		*dentry,
1078 	struct kstatfs		*statp)
1079 {
1080 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1081 	xfs_sb_t		*sbp = &mp->m_sb;
1082 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1083 	uint64_t		fakeinos, id;
1084 	uint64_t		icount;
1085 	uint64_t		ifree;
1086 	uint64_t		fdblocks;
1087 	xfs_extlen_t		lsize;
1088 	int64_t			ffree;
1089 
1090 	statp->f_type = XFS_SUPER_MAGIC;
1091 	statp->f_namelen = MAXNAMELEN - 1;
1092 
1093 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1094 	statp->f_fsid.val[0] = (u32)id;
1095 	statp->f_fsid.val[1] = (u32)(id >> 32);
1096 
1097 	icount = percpu_counter_sum(&mp->m_icount);
1098 	ifree = percpu_counter_sum(&mp->m_ifree);
1099 	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1100 
1101 	spin_lock(&mp->m_sb_lock);
1102 	statp->f_bsize = sbp->sb_blocksize;
1103 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1104 	statp->f_blocks = sbp->sb_dblocks - lsize;
1105 	spin_unlock(&mp->m_sb_lock);
1106 
1107 	statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1108 	statp->f_bavail = statp->f_bfree;
1109 
1110 	fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
1111 	statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1112 	if (M_IGEO(mp)->maxicount)
1113 		statp->f_files = min_t(typeof(statp->f_files),
1114 					statp->f_files,
1115 					M_IGEO(mp)->maxicount);
1116 
1117 	/* If sb_icount overshot maxicount, report actual allocation */
1118 	statp->f_files = max_t(typeof(statp->f_files),
1119 					statp->f_files,
1120 					sbp->sb_icount);
1121 
1122 	/* make sure statp->f_ffree does not underflow */
1123 	ffree = statp->f_files - (icount - ifree);
1124 	statp->f_ffree = max_t(int64_t, ffree, 0);
1125 
1126 
1127 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1128 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1129 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1130 		xfs_qm_statvfs(ip, statp);
1131 
1132 	if (XFS_IS_REALTIME_MOUNT(mp) &&
1133 	    (ip->i_d.di_flags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
1134 		statp->f_blocks = sbp->sb_rblocks;
1135 		statp->f_bavail = statp->f_bfree =
1136 			sbp->sb_frextents * sbp->sb_rextsize;
1137 	}
1138 
1139 	return 0;
1140 }
1141 
1142 STATIC void
1143 xfs_save_resvblks(struct xfs_mount *mp)
1144 {
1145 	uint64_t resblks = 0;
1146 
1147 	mp->m_resblks_save = mp->m_resblks;
1148 	xfs_reserve_blocks(mp, &resblks, NULL);
1149 }
1150 
1151 STATIC void
1152 xfs_restore_resvblks(struct xfs_mount *mp)
1153 {
1154 	uint64_t resblks;
1155 
1156 	if (mp->m_resblks_save) {
1157 		resblks = mp->m_resblks_save;
1158 		mp->m_resblks_save = 0;
1159 	} else
1160 		resblks = xfs_default_resblks(mp);
1161 
1162 	xfs_reserve_blocks(mp, &resblks, NULL);
1163 }
1164 
1165 /*
1166  * Trigger writeback of all the dirty metadata in the file system.
1167  *
1168  * This ensures that the metadata is written to their location on disk rather
1169  * than just existing in transactions in the log. This means after a quiesce
1170  * there is no log replay required to write the inodes to disk - this is the
1171  * primary difference between a sync and a quiesce.
1172  *
1173  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1174  * it is started again when appropriate.
1175  */
1176 void
1177 xfs_quiesce_attr(
1178 	struct xfs_mount	*mp)
1179 {
1180 	int	error = 0;
1181 
1182 	/* wait for all modifications to complete */
1183 	while (atomic_read(&mp->m_active_trans) > 0)
1184 		delay(100);
1185 
1186 	/* force the log to unpin objects from the now complete transactions */
1187 	xfs_log_force(mp, XFS_LOG_SYNC);
1188 
1189 	/* reclaim inodes to do any IO before the freeze completes */
1190 	xfs_reclaim_inodes(mp, 0);
1191 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1192 
1193 	/* Push the superblock and write an unmount record */
1194 	error = xfs_log_sbcount(mp);
1195 	if (error)
1196 		xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1197 				"Frozen image may not be consistent.");
1198 	/*
1199 	 * Just warn here till VFS can correctly support
1200 	 * read-only remount without racing.
1201 	 */
1202 	WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1203 
1204 	xfs_log_quiesce(mp);
1205 }
1206 
1207 STATIC int
1208 xfs_test_remount_options(
1209 	struct super_block	*sb,
1210 	char			*options)
1211 {
1212 	int			error = 0;
1213 	struct xfs_mount	*tmp_mp;
1214 
1215 	tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1216 	if (!tmp_mp)
1217 		return -ENOMEM;
1218 
1219 	tmp_mp->m_super = sb;
1220 	error = xfs_parseargs(tmp_mp, options);
1221 	xfs_free_fsname(tmp_mp);
1222 	kmem_free(tmp_mp);
1223 
1224 	return error;
1225 }
1226 
1227 STATIC int
1228 xfs_fs_remount(
1229 	struct super_block	*sb,
1230 	int			*flags,
1231 	char			*options)
1232 {
1233 	struct xfs_mount	*mp = XFS_M(sb);
1234 	xfs_sb_t		*sbp = &mp->m_sb;
1235 	substring_t		args[MAX_OPT_ARGS];
1236 	char			*p;
1237 	int			error;
1238 
1239 	/* First, check for complete junk; i.e. invalid options */
1240 	error = xfs_test_remount_options(sb, options);
1241 	if (error)
1242 		return error;
1243 
1244 	sync_filesystem(sb);
1245 	while ((p = strsep(&options, ",")) != NULL) {
1246 		int token;
1247 
1248 		if (!*p)
1249 			continue;
1250 
1251 		token = match_token(p, tokens, args);
1252 		switch (token) {
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 & SB_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 
1328 		/* Recover any CoW blocks that never got remapped. */
1329 		error = xfs_reflink_recover_cow(mp);
1330 		if (error) {
1331 			xfs_err(mp,
1332 	"Error %d recovering leftover CoW allocations.", error);
1333 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1334 			return error;
1335 		}
1336 		xfs_start_block_reaping(mp);
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 & SB_RDONLY)) {
1346 		/*
1347 		 * Cancel background eofb scanning so it cannot race with the
1348 		 * final log force+buftarg wait and deadlock the remount.
1349 		 */
1350 		xfs_stop_block_reaping(mp);
1351 
1352 		/* Get rid of any leftover CoW reservations... */
1353 		error = xfs_icache_free_cowblocks(mp, NULL);
1354 		if (error) {
1355 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1356 			return error;
1357 		}
1358 
1359 		/* Free the per-AG metadata reservation pool. */
1360 		error = xfs_fs_unreserve_ag_blocks(mp);
1361 		if (error) {
1362 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1363 			return error;
1364 		}
1365 
1366 		/*
1367 		 * Before we sync the metadata, we need to free up the reserve
1368 		 * block pool so that the used block count in the superblock on
1369 		 * disk is correct at the end of the remount. Stash the current
1370 		 * reserve pool size so that if we get remounted rw, we can
1371 		 * return it to the same size.
1372 		 */
1373 		xfs_save_resvblks(mp);
1374 
1375 		xfs_quiesce_attr(mp);
1376 		mp->m_flags |= XFS_MOUNT_RDONLY;
1377 	}
1378 
1379 	return 0;
1380 }
1381 
1382 /*
1383  * Second stage of a freeze. The data is already frozen so we only
1384  * need to take care of the metadata. Once that's done sync the superblock
1385  * to the log to dirty it in case of a crash while frozen. This ensures that we
1386  * will recover the unlinked inode lists on the next mount.
1387  */
1388 STATIC int
1389 xfs_fs_freeze(
1390 	struct super_block	*sb)
1391 {
1392 	struct xfs_mount	*mp = XFS_M(sb);
1393 
1394 	xfs_stop_block_reaping(mp);
1395 	xfs_save_resvblks(mp);
1396 	xfs_quiesce_attr(mp);
1397 	return xfs_sync_sb(mp, true);
1398 }
1399 
1400 STATIC int
1401 xfs_fs_unfreeze(
1402 	struct super_block	*sb)
1403 {
1404 	struct xfs_mount	*mp = XFS_M(sb);
1405 
1406 	xfs_restore_resvblks(mp);
1407 	xfs_log_work_queue(mp);
1408 	xfs_start_block_reaping(mp);
1409 	return 0;
1410 }
1411 
1412 STATIC int
1413 xfs_fs_show_options(
1414 	struct seq_file		*m,
1415 	struct dentry		*root)
1416 {
1417 	xfs_showargs(XFS_M(root->d_sb), m);
1418 	return 0;
1419 }
1420 
1421 /*
1422  * This function fills in xfs_mount_t fields based on mount args.
1423  * Note: the superblock _has_ now been read in.
1424  */
1425 STATIC int
1426 xfs_finish_flags(
1427 	struct xfs_mount	*mp)
1428 {
1429 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1430 
1431 	/* Fail a mount where the logbuf is smaller than the log stripe */
1432 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1433 		if (mp->m_logbsize <= 0 &&
1434 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1435 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1436 		} else if (mp->m_logbsize > 0 &&
1437 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1438 			xfs_warn(mp,
1439 		"logbuf size must be greater than or equal to log stripe size");
1440 			return -EINVAL;
1441 		}
1442 	} else {
1443 		/* Fail a mount if the logbuf is larger than 32K */
1444 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1445 			xfs_warn(mp,
1446 		"logbuf size for version 1 logs must be 16K or 32K");
1447 			return -EINVAL;
1448 		}
1449 	}
1450 
1451 	/*
1452 	 * V5 filesystems always use attr2 format for attributes.
1453 	 */
1454 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1455 	    (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1456 		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1457 			     "attr2 is always enabled for V5 filesystems.");
1458 		return -EINVAL;
1459 	}
1460 
1461 	/*
1462 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1463 	 * told by noattr2 to turn it off
1464 	 */
1465 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1466 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1467 		mp->m_flags |= XFS_MOUNT_ATTR2;
1468 
1469 	/*
1470 	 * prohibit r/w mounts of read-only filesystems
1471 	 */
1472 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1473 		xfs_warn(mp,
1474 			"cannot mount a read-only filesystem as read-write");
1475 		return -EROFS;
1476 	}
1477 
1478 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1479 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1480 	    !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1481 		xfs_warn(mp,
1482 		  "Super block does not support project and group quota together");
1483 		return -EINVAL;
1484 	}
1485 
1486 	return 0;
1487 }
1488 
1489 static int
1490 xfs_init_percpu_counters(
1491 	struct xfs_mount	*mp)
1492 {
1493 	int		error;
1494 
1495 	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1496 	if (error)
1497 		return -ENOMEM;
1498 
1499 	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1500 	if (error)
1501 		goto free_icount;
1502 
1503 	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1504 	if (error)
1505 		goto free_ifree;
1506 
1507 	error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1508 	if (error)
1509 		goto free_fdblocks;
1510 
1511 	return 0;
1512 
1513 free_fdblocks:
1514 	percpu_counter_destroy(&mp->m_fdblocks);
1515 free_ifree:
1516 	percpu_counter_destroy(&mp->m_ifree);
1517 free_icount:
1518 	percpu_counter_destroy(&mp->m_icount);
1519 	return -ENOMEM;
1520 }
1521 
1522 void
1523 xfs_reinit_percpu_counters(
1524 	struct xfs_mount	*mp)
1525 {
1526 	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1527 	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1528 	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1529 }
1530 
1531 static void
1532 xfs_destroy_percpu_counters(
1533 	struct xfs_mount	*mp)
1534 {
1535 	percpu_counter_destroy(&mp->m_icount);
1536 	percpu_counter_destroy(&mp->m_ifree);
1537 	percpu_counter_destroy(&mp->m_fdblocks);
1538 	ASSERT(XFS_FORCED_SHUTDOWN(mp) ||
1539 	       percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1540 	percpu_counter_destroy(&mp->m_delalloc_blks);
1541 }
1542 
1543 static struct xfs_mount *
1544 xfs_mount_alloc(
1545 	struct super_block	*sb)
1546 {
1547 	struct xfs_mount	*mp;
1548 
1549 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1550 	if (!mp)
1551 		return NULL;
1552 
1553 	mp->m_super = sb;
1554 	spin_lock_init(&mp->m_sb_lock);
1555 	spin_lock_init(&mp->m_agirotor_lock);
1556 	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1557 	spin_lock_init(&mp->m_perag_lock);
1558 	mutex_init(&mp->m_growlock);
1559 	atomic_set(&mp->m_active_trans, 0);
1560 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1561 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1562 	INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1563 	mp->m_kobj.kobject.kset = xfs_kset;
1564 	/*
1565 	 * We don't create the finobt per-ag space reservation until after log
1566 	 * recovery, so we must set this to true so that an ifree transaction
1567 	 * started during log recovery will not depend on space reservations
1568 	 * for finobt expansion.
1569 	 */
1570 	mp->m_finobt_nores = true;
1571 	return mp;
1572 }
1573 
1574 
1575 STATIC int
1576 xfs_fs_fill_super(
1577 	struct super_block	*sb,
1578 	void			*data,
1579 	int			silent)
1580 {
1581 	struct inode		*root;
1582 	struct xfs_mount	*mp = NULL;
1583 	int			flags = 0, error = -ENOMEM;
1584 
1585 	/*
1586 	 * allocate mp and do all low-level struct initializations before we
1587 	 * attach it to the super
1588 	 */
1589 	mp = xfs_mount_alloc(sb);
1590 	if (!mp)
1591 		goto out;
1592 	sb->s_fs_info = mp;
1593 
1594 	error = xfs_parseargs(mp, (char *)data);
1595 	if (error)
1596 		goto out_free_fsname;
1597 
1598 	sb_min_blocksize(sb, BBSIZE);
1599 	sb->s_xattr = xfs_xattr_handlers;
1600 	sb->s_export_op = &xfs_export_operations;
1601 #ifdef CONFIG_XFS_QUOTA
1602 	sb->s_qcop = &xfs_quotactl_operations;
1603 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1604 #endif
1605 	sb->s_op = &xfs_super_operations;
1606 
1607 	/*
1608 	 * Delay mount work if the debug hook is set. This is debug
1609 	 * instrumention to coordinate simulation of xfs mount failures with
1610 	 * VFS superblock operations
1611 	 */
1612 	if (xfs_globals.mount_delay) {
1613 		xfs_notice(mp, "Delaying mount for %d seconds.",
1614 			xfs_globals.mount_delay);
1615 		msleep(xfs_globals.mount_delay * 1000);
1616 	}
1617 
1618 	if (silent)
1619 		flags |= XFS_MFSI_QUIET;
1620 
1621 	error = xfs_open_devices(mp);
1622 	if (error)
1623 		goto out_free_fsname;
1624 
1625 	error = xfs_init_mount_workqueues(mp);
1626 	if (error)
1627 		goto out_close_devices;
1628 
1629 	error = xfs_init_percpu_counters(mp);
1630 	if (error)
1631 		goto out_destroy_workqueues;
1632 
1633 	/* Allocate stats memory before we do operations that might use it */
1634 	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1635 	if (!mp->m_stats.xs_stats) {
1636 		error = -ENOMEM;
1637 		goto out_destroy_counters;
1638 	}
1639 
1640 	error = xfs_readsb(mp, flags);
1641 	if (error)
1642 		goto out_free_stats;
1643 
1644 	error = xfs_finish_flags(mp);
1645 	if (error)
1646 		goto out_free_sb;
1647 
1648 	error = xfs_setup_devices(mp);
1649 	if (error)
1650 		goto out_free_sb;
1651 
1652 	error = xfs_filestream_mount(mp);
1653 	if (error)
1654 		goto out_free_sb;
1655 
1656 	/*
1657 	 * we must configure the block size in the superblock before we run the
1658 	 * full mount process as the mount process can lookup and cache inodes.
1659 	 */
1660 	sb->s_magic = XFS_SUPER_MAGIC;
1661 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1662 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1663 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1664 	sb->s_max_links = XFS_MAXLINK;
1665 	sb->s_time_gran = 1;
1666 	sb->s_time_min = S32_MIN;
1667 	sb->s_time_max = S32_MAX;
1668 	sb->s_iflags |= SB_I_CGROUPWB;
1669 
1670 	set_posix_acl_flag(sb);
1671 
1672 	/* version 5 superblocks support inode version counters. */
1673 	if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1674 		sb->s_flags |= SB_I_VERSION;
1675 
1676 	if (mp->m_flags & XFS_MOUNT_DAX) {
1677 		bool rtdev_is_dax = false, datadev_is_dax;
1678 
1679 		xfs_warn(mp,
1680 		"DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1681 
1682 		datadev_is_dax = bdev_dax_supported(mp->m_ddev_targp->bt_bdev,
1683 			sb->s_blocksize);
1684 		if (mp->m_rtdev_targp)
1685 			rtdev_is_dax = bdev_dax_supported(
1686 				mp->m_rtdev_targp->bt_bdev, sb->s_blocksize);
1687 		if (!rtdev_is_dax && !datadev_is_dax) {
1688 			xfs_alert(mp,
1689 			"DAX unsupported by block device. Turning off DAX.");
1690 			mp->m_flags &= ~XFS_MOUNT_DAX;
1691 		}
1692 		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1693 			xfs_alert(mp,
1694 		"DAX and reflink cannot be used together!");
1695 			error = -EINVAL;
1696 			goto out_filestream_unmount;
1697 		}
1698 	}
1699 
1700 	if (mp->m_flags & XFS_MOUNT_DISCARD) {
1701 		struct request_queue *q = bdev_get_queue(sb->s_bdev);
1702 
1703 		if (!blk_queue_discard(q)) {
1704 			xfs_warn(mp, "mounting with \"discard\" option, but "
1705 					"the device does not support discard");
1706 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
1707 		}
1708 	}
1709 
1710 	if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1711 		if (mp->m_sb.sb_rblocks) {
1712 			xfs_alert(mp,
1713 	"reflink not compatible with realtime device!");
1714 			error = -EINVAL;
1715 			goto out_filestream_unmount;
1716 		}
1717 
1718 		if (xfs_globals.always_cow) {
1719 			xfs_info(mp, "using DEBUG-only always_cow mode.");
1720 			mp->m_always_cow = true;
1721 		}
1722 	}
1723 
1724 	if (xfs_sb_version_hasrmapbt(&mp->m_sb) && mp->m_sb.sb_rblocks) {
1725 		xfs_alert(mp,
1726 	"reverse mapping btree not compatible with realtime device!");
1727 		error = -EINVAL;
1728 		goto out_filestream_unmount;
1729 	}
1730 
1731 	error = xfs_mountfs(mp);
1732 	if (error)
1733 		goto out_filestream_unmount;
1734 
1735 	root = igrab(VFS_I(mp->m_rootip));
1736 	if (!root) {
1737 		error = -ENOENT;
1738 		goto out_unmount;
1739 	}
1740 	sb->s_root = d_make_root(root);
1741 	if (!sb->s_root) {
1742 		error = -ENOMEM;
1743 		goto out_unmount;
1744 	}
1745 
1746 	return 0;
1747 
1748  out_filestream_unmount:
1749 	xfs_filestream_unmount(mp);
1750  out_free_sb:
1751 	xfs_freesb(mp);
1752  out_free_stats:
1753 	free_percpu(mp->m_stats.xs_stats);
1754  out_destroy_counters:
1755 	xfs_destroy_percpu_counters(mp);
1756  out_destroy_workqueues:
1757 	xfs_destroy_mount_workqueues(mp);
1758  out_close_devices:
1759 	xfs_close_devices(mp);
1760  out_free_fsname:
1761 	sb->s_fs_info = NULL;
1762 	xfs_free_fsname(mp);
1763 	kfree(mp);
1764  out:
1765 	return error;
1766 
1767  out_unmount:
1768 	xfs_filestream_unmount(mp);
1769 	xfs_unmountfs(mp);
1770 	goto out_free_sb;
1771 }
1772 
1773 STATIC void
1774 xfs_fs_put_super(
1775 	struct super_block	*sb)
1776 {
1777 	struct xfs_mount	*mp = XFS_M(sb);
1778 
1779 	/* if ->fill_super failed, we have no mount to tear down */
1780 	if (!sb->s_fs_info)
1781 		return;
1782 
1783 	xfs_notice(mp, "Unmounting Filesystem");
1784 	xfs_filestream_unmount(mp);
1785 	xfs_unmountfs(mp);
1786 
1787 	xfs_freesb(mp);
1788 	free_percpu(mp->m_stats.xs_stats);
1789 	xfs_destroy_percpu_counters(mp);
1790 	xfs_destroy_mount_workqueues(mp);
1791 	xfs_close_devices(mp);
1792 
1793 	sb->s_fs_info = NULL;
1794 	xfs_free_fsname(mp);
1795 	kfree(mp);
1796 }
1797 
1798 STATIC struct dentry *
1799 xfs_fs_mount(
1800 	struct file_system_type	*fs_type,
1801 	int			flags,
1802 	const char		*dev_name,
1803 	void			*data)
1804 {
1805 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1806 }
1807 
1808 static long
1809 xfs_fs_nr_cached_objects(
1810 	struct super_block	*sb,
1811 	struct shrink_control	*sc)
1812 {
1813 	/* Paranoia: catch incorrect calls during mount setup or teardown */
1814 	if (WARN_ON_ONCE(!sb->s_fs_info))
1815 		return 0;
1816 	return xfs_reclaim_inodes_count(XFS_M(sb));
1817 }
1818 
1819 static long
1820 xfs_fs_free_cached_objects(
1821 	struct super_block	*sb,
1822 	struct shrink_control	*sc)
1823 {
1824 	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1825 }
1826 
1827 static const struct super_operations xfs_super_operations = {
1828 	.alloc_inode		= xfs_fs_alloc_inode,
1829 	.destroy_inode		= xfs_fs_destroy_inode,
1830 	.dirty_inode		= xfs_fs_dirty_inode,
1831 	.drop_inode		= xfs_fs_drop_inode,
1832 	.put_super		= xfs_fs_put_super,
1833 	.sync_fs		= xfs_fs_sync_fs,
1834 	.freeze_fs		= xfs_fs_freeze,
1835 	.unfreeze_fs		= xfs_fs_unfreeze,
1836 	.statfs			= xfs_fs_statfs,
1837 	.remount_fs		= xfs_fs_remount,
1838 	.show_options		= xfs_fs_show_options,
1839 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1840 	.free_cached_objects	= xfs_fs_free_cached_objects,
1841 };
1842 
1843 static struct file_system_type xfs_fs_type = {
1844 	.owner			= THIS_MODULE,
1845 	.name			= "xfs",
1846 	.mount			= xfs_fs_mount,
1847 	.kill_sb		= kill_block_super,
1848 	.fs_flags		= FS_REQUIRES_DEV,
1849 };
1850 MODULE_ALIAS_FS("xfs");
1851 
1852 STATIC int __init
1853 xfs_init_zones(void)
1854 {
1855 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1856 						"xfs_log_ticket");
1857 	if (!xfs_log_ticket_zone)
1858 		goto out;
1859 
1860 	xfs_bmap_free_item_zone = kmem_zone_init(
1861 			sizeof(struct xfs_extent_free_item),
1862 			"xfs_bmap_free_item");
1863 	if (!xfs_bmap_free_item_zone)
1864 		goto out_destroy_log_ticket_zone;
1865 
1866 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1867 						"xfs_btree_cur");
1868 	if (!xfs_btree_cur_zone)
1869 		goto out_destroy_bmap_free_item_zone;
1870 
1871 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1872 						"xfs_da_state");
1873 	if (!xfs_da_state_zone)
1874 		goto out_destroy_btree_cur_zone;
1875 
1876 	xfs_ifork_zone = kmem_zone_init(sizeof(struct xfs_ifork), "xfs_ifork");
1877 	if (!xfs_ifork_zone)
1878 		goto out_destroy_da_state_zone;
1879 
1880 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1881 	if (!xfs_trans_zone)
1882 		goto out_destroy_ifork_zone;
1883 
1884 
1885 	/*
1886 	 * The size of the zone allocated buf log item is the maximum
1887 	 * size possible under XFS.  This wastes a little bit of memory,
1888 	 * but it is much faster.
1889 	 */
1890 	xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1891 					   "xfs_buf_item");
1892 	if (!xfs_buf_item_zone)
1893 		goto out_destroy_trans_zone;
1894 
1895 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1896 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1897 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1898 	if (!xfs_efd_zone)
1899 		goto out_destroy_buf_item_zone;
1900 
1901 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1902 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1903 				sizeof(xfs_extent_t))), "xfs_efi_item");
1904 	if (!xfs_efi_zone)
1905 		goto out_destroy_efd_zone;
1906 
1907 	xfs_inode_zone =
1908 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1909 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1910 			KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1911 	if (!xfs_inode_zone)
1912 		goto out_destroy_efi_zone;
1913 
1914 	xfs_ili_zone =
1915 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1916 					KM_ZONE_SPREAD, NULL);
1917 	if (!xfs_ili_zone)
1918 		goto out_destroy_inode_zone;
1919 	xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1920 					"xfs_icr");
1921 	if (!xfs_icreate_zone)
1922 		goto out_destroy_ili_zone;
1923 
1924 	xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1925 			"xfs_rud_item");
1926 	if (!xfs_rud_zone)
1927 		goto out_destroy_icreate_zone;
1928 
1929 	xfs_rui_zone = kmem_zone_init(
1930 			xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1931 			"xfs_rui_item");
1932 	if (!xfs_rui_zone)
1933 		goto out_destroy_rud_zone;
1934 
1935 	xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1936 			"xfs_cud_item");
1937 	if (!xfs_cud_zone)
1938 		goto out_destroy_rui_zone;
1939 
1940 	xfs_cui_zone = kmem_zone_init(
1941 			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1942 			"xfs_cui_item");
1943 	if (!xfs_cui_zone)
1944 		goto out_destroy_cud_zone;
1945 
1946 	xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1947 			"xfs_bud_item");
1948 	if (!xfs_bud_zone)
1949 		goto out_destroy_cui_zone;
1950 
1951 	xfs_bui_zone = kmem_zone_init(
1952 			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1953 			"xfs_bui_item");
1954 	if (!xfs_bui_zone)
1955 		goto out_destroy_bud_zone;
1956 
1957 	return 0;
1958 
1959  out_destroy_bud_zone:
1960 	kmem_zone_destroy(xfs_bud_zone);
1961  out_destroy_cui_zone:
1962 	kmem_zone_destroy(xfs_cui_zone);
1963  out_destroy_cud_zone:
1964 	kmem_zone_destroy(xfs_cud_zone);
1965  out_destroy_rui_zone:
1966 	kmem_zone_destroy(xfs_rui_zone);
1967  out_destroy_rud_zone:
1968 	kmem_zone_destroy(xfs_rud_zone);
1969  out_destroy_icreate_zone:
1970 	kmem_zone_destroy(xfs_icreate_zone);
1971  out_destroy_ili_zone:
1972 	kmem_zone_destroy(xfs_ili_zone);
1973  out_destroy_inode_zone:
1974 	kmem_zone_destroy(xfs_inode_zone);
1975  out_destroy_efi_zone:
1976 	kmem_zone_destroy(xfs_efi_zone);
1977  out_destroy_efd_zone:
1978 	kmem_zone_destroy(xfs_efd_zone);
1979  out_destroy_buf_item_zone:
1980 	kmem_zone_destroy(xfs_buf_item_zone);
1981  out_destroy_trans_zone:
1982 	kmem_zone_destroy(xfs_trans_zone);
1983  out_destroy_ifork_zone:
1984 	kmem_zone_destroy(xfs_ifork_zone);
1985  out_destroy_da_state_zone:
1986 	kmem_zone_destroy(xfs_da_state_zone);
1987  out_destroy_btree_cur_zone:
1988 	kmem_zone_destroy(xfs_btree_cur_zone);
1989  out_destroy_bmap_free_item_zone:
1990 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1991  out_destroy_log_ticket_zone:
1992 	kmem_zone_destroy(xfs_log_ticket_zone);
1993  out:
1994 	return -ENOMEM;
1995 }
1996 
1997 STATIC void
1998 xfs_destroy_zones(void)
1999 {
2000 	/*
2001 	 * Make sure all delayed rcu free are flushed before we
2002 	 * destroy caches.
2003 	 */
2004 	rcu_barrier();
2005 	kmem_zone_destroy(xfs_bui_zone);
2006 	kmem_zone_destroy(xfs_bud_zone);
2007 	kmem_zone_destroy(xfs_cui_zone);
2008 	kmem_zone_destroy(xfs_cud_zone);
2009 	kmem_zone_destroy(xfs_rui_zone);
2010 	kmem_zone_destroy(xfs_rud_zone);
2011 	kmem_zone_destroy(xfs_icreate_zone);
2012 	kmem_zone_destroy(xfs_ili_zone);
2013 	kmem_zone_destroy(xfs_inode_zone);
2014 	kmem_zone_destroy(xfs_efi_zone);
2015 	kmem_zone_destroy(xfs_efd_zone);
2016 	kmem_zone_destroy(xfs_buf_item_zone);
2017 	kmem_zone_destroy(xfs_trans_zone);
2018 	kmem_zone_destroy(xfs_ifork_zone);
2019 	kmem_zone_destroy(xfs_da_state_zone);
2020 	kmem_zone_destroy(xfs_btree_cur_zone);
2021 	kmem_zone_destroy(xfs_bmap_free_item_zone);
2022 	kmem_zone_destroy(xfs_log_ticket_zone);
2023 }
2024 
2025 STATIC int __init
2026 xfs_init_workqueues(void)
2027 {
2028 	/*
2029 	 * The allocation workqueue can be used in memory reclaim situations
2030 	 * (writepage path), and parallelism is only limited by the number of
2031 	 * AGs in all the filesystems mounted. Hence use the default large
2032 	 * max_active value for this workqueue.
2033 	 */
2034 	xfs_alloc_wq = alloc_workqueue("xfsalloc",
2035 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
2036 	if (!xfs_alloc_wq)
2037 		return -ENOMEM;
2038 
2039 	xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
2040 	if (!xfs_discard_wq)
2041 		goto out_free_alloc_wq;
2042 
2043 	return 0;
2044 out_free_alloc_wq:
2045 	destroy_workqueue(xfs_alloc_wq);
2046 	return -ENOMEM;
2047 }
2048 
2049 STATIC void
2050 xfs_destroy_workqueues(void)
2051 {
2052 	destroy_workqueue(xfs_discard_wq);
2053 	destroy_workqueue(xfs_alloc_wq);
2054 }
2055 
2056 STATIC int __init
2057 init_xfs_fs(void)
2058 {
2059 	int			error;
2060 
2061 	xfs_check_ondisk_structs();
2062 
2063 	printk(KERN_INFO XFS_VERSION_STRING " with "
2064 			 XFS_BUILD_OPTIONS " enabled\n");
2065 
2066 	xfs_dir_startup();
2067 
2068 	error = xfs_init_zones();
2069 	if (error)
2070 		goto out;
2071 
2072 	error = xfs_init_workqueues();
2073 	if (error)
2074 		goto out_destroy_zones;
2075 
2076 	error = xfs_mru_cache_init();
2077 	if (error)
2078 		goto out_destroy_wq;
2079 
2080 	error = xfs_buf_init();
2081 	if (error)
2082 		goto out_mru_cache_uninit;
2083 
2084 	error = xfs_init_procfs();
2085 	if (error)
2086 		goto out_buf_terminate;
2087 
2088 	error = xfs_sysctl_register();
2089 	if (error)
2090 		goto out_cleanup_procfs;
2091 
2092 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2093 	if (!xfs_kset) {
2094 		error = -ENOMEM;
2095 		goto out_sysctl_unregister;
2096 	}
2097 
2098 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
2099 
2100 	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2101 	if (!xfsstats.xs_stats) {
2102 		error = -ENOMEM;
2103 		goto out_kset_unregister;
2104 	}
2105 
2106 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2107 			       "stats");
2108 	if (error)
2109 		goto out_free_stats;
2110 
2111 #ifdef DEBUG
2112 	xfs_dbg_kobj.kobject.kset = xfs_kset;
2113 	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2114 	if (error)
2115 		goto out_remove_stats_kobj;
2116 #endif
2117 
2118 	error = xfs_qm_init();
2119 	if (error)
2120 		goto out_remove_dbg_kobj;
2121 
2122 	error = register_filesystem(&xfs_fs_type);
2123 	if (error)
2124 		goto out_qm_exit;
2125 	return 0;
2126 
2127  out_qm_exit:
2128 	xfs_qm_exit();
2129  out_remove_dbg_kobj:
2130 #ifdef DEBUG
2131 	xfs_sysfs_del(&xfs_dbg_kobj);
2132  out_remove_stats_kobj:
2133 #endif
2134 	xfs_sysfs_del(&xfsstats.xs_kobj);
2135  out_free_stats:
2136 	free_percpu(xfsstats.xs_stats);
2137  out_kset_unregister:
2138 	kset_unregister(xfs_kset);
2139  out_sysctl_unregister:
2140 	xfs_sysctl_unregister();
2141  out_cleanup_procfs:
2142 	xfs_cleanup_procfs();
2143  out_buf_terminate:
2144 	xfs_buf_terminate();
2145  out_mru_cache_uninit:
2146 	xfs_mru_cache_uninit();
2147  out_destroy_wq:
2148 	xfs_destroy_workqueues();
2149  out_destroy_zones:
2150 	xfs_destroy_zones();
2151  out:
2152 	return error;
2153 }
2154 
2155 STATIC void __exit
2156 exit_xfs_fs(void)
2157 {
2158 	xfs_qm_exit();
2159 	unregister_filesystem(&xfs_fs_type);
2160 #ifdef DEBUG
2161 	xfs_sysfs_del(&xfs_dbg_kobj);
2162 #endif
2163 	xfs_sysfs_del(&xfsstats.xs_kobj);
2164 	free_percpu(xfsstats.xs_stats);
2165 	kset_unregister(xfs_kset);
2166 	xfs_sysctl_unregister();
2167 	xfs_cleanup_procfs();
2168 	xfs_buf_terminate();
2169 	xfs_mru_cache_uninit();
2170 	xfs_destroy_workqueues();
2171 	xfs_destroy_zones();
2172 	xfs_uuid_table_free();
2173 }
2174 
2175 module_init(init_xfs_fs);
2176 module_exit(exit_xfs_fs);
2177 
2178 MODULE_AUTHOR("Silicon Graphics, Inc.");
2179 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2180 MODULE_LICENSE("GPL");
2181