4d2e87462ac4a19499ad45ea17f2246379beb577
[linux-block.git] / fs / xfs / xfs_super.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6
7 #include "xfs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_sb.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
21 #include "xfs_log.h"
22 #include "xfs_log_priv.h"
23 #include "xfs_dir2.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_mru_cache.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_icache.h"
28 #include "xfs_trace.h"
29 #include "xfs_icreate_item.h"
30 #include "xfs_filestream.h"
31 #include "xfs_quota.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_ondisk.h"
34 #include "xfs_rmap_item.h"
35 #include "xfs_refcount_item.h"
36 #include "xfs_bmap_item.h"
37 #include "xfs_reflink.h"
38 #include "xfs_pwork.h"
39 #include "xfs_ag.h"
40 #include "xfs_defer.h"
41 #include "xfs_attr_item.h"
42 #include "xfs_xattr.h"
43 #include "xfs_iunlink_item.h"
44 #include "xfs_dahash_test.h"
45
46 #include <linux/magic.h>
47 #include <linux/fs_context.h>
48 #include <linux/fs_parser.h>
49
50 static const struct super_operations xfs_super_operations;
51
52 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
53 #ifdef DEBUG
54 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
55 #endif
56
57 #ifdef CONFIG_HOTPLUG_CPU
58 static LIST_HEAD(xfs_mount_list);
59 static DEFINE_SPINLOCK(xfs_mount_list_lock);
60
61 static inline void xfs_mount_list_add(struct xfs_mount *mp)
62 {
63         spin_lock(&xfs_mount_list_lock);
64         list_add(&mp->m_mount_list, &xfs_mount_list);
65         spin_unlock(&xfs_mount_list_lock);
66 }
67
68 static inline void xfs_mount_list_del(struct xfs_mount *mp)
69 {
70         spin_lock(&xfs_mount_list_lock);
71         list_del(&mp->m_mount_list);
72         spin_unlock(&xfs_mount_list_lock);
73 }
74 #else /* !CONFIG_HOTPLUG_CPU */
75 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
76 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
77 #endif
78
79 enum xfs_dax_mode {
80         XFS_DAX_INODE = 0,
81         XFS_DAX_ALWAYS = 1,
82         XFS_DAX_NEVER = 2,
83 };
84
85 static void
86 xfs_mount_set_dax_mode(
87         struct xfs_mount        *mp,
88         enum xfs_dax_mode       mode)
89 {
90         switch (mode) {
91         case XFS_DAX_INODE:
92                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
93                 break;
94         case XFS_DAX_ALWAYS:
95                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
96                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
97                 break;
98         case XFS_DAX_NEVER:
99                 mp->m_features |= XFS_FEAT_DAX_NEVER;
100                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
101                 break;
102         }
103 }
104
105 static const struct constant_table dax_param_enums[] = {
106         {"inode",       XFS_DAX_INODE },
107         {"always",      XFS_DAX_ALWAYS },
108         {"never",       XFS_DAX_NEVER },
109         {}
110 };
111
112 /*
113  * Table driven mount option parser.
114  */
115 enum {
116         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
117         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
118         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
119         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
120         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
121         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
122         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
123         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
124         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
125 };
126
127 static const struct fs_parameter_spec xfs_fs_parameters[] = {
128         fsparam_u32("logbufs",          Opt_logbufs),
129         fsparam_string("logbsize",      Opt_logbsize),
130         fsparam_string("logdev",        Opt_logdev),
131         fsparam_string("rtdev",         Opt_rtdev),
132         fsparam_flag("wsync",           Opt_wsync),
133         fsparam_flag("noalign",         Opt_noalign),
134         fsparam_flag("swalloc",         Opt_swalloc),
135         fsparam_u32("sunit",            Opt_sunit),
136         fsparam_u32("swidth",           Opt_swidth),
137         fsparam_flag("nouuid",          Opt_nouuid),
138         fsparam_flag("grpid",           Opt_grpid),
139         fsparam_flag("nogrpid",         Opt_nogrpid),
140         fsparam_flag("bsdgroups",       Opt_bsdgroups),
141         fsparam_flag("sysvgroups",      Opt_sysvgroups),
142         fsparam_string("allocsize",     Opt_allocsize),
143         fsparam_flag("norecovery",      Opt_norecovery),
144         fsparam_flag("inode64",         Opt_inode64),
145         fsparam_flag("inode32",         Opt_inode32),
146         fsparam_flag("ikeep",           Opt_ikeep),
147         fsparam_flag("noikeep",         Opt_noikeep),
148         fsparam_flag("largeio",         Opt_largeio),
149         fsparam_flag("nolargeio",       Opt_nolargeio),
150         fsparam_flag("attr2",           Opt_attr2),
151         fsparam_flag("noattr2",         Opt_noattr2),
152         fsparam_flag("filestreams",     Opt_filestreams),
153         fsparam_flag("quota",           Opt_quota),
154         fsparam_flag("noquota",         Opt_noquota),
155         fsparam_flag("usrquota",        Opt_usrquota),
156         fsparam_flag("grpquota",        Opt_grpquota),
157         fsparam_flag("prjquota",        Opt_prjquota),
158         fsparam_flag("uquota",          Opt_uquota),
159         fsparam_flag("gquota",          Opt_gquota),
160         fsparam_flag("pquota",          Opt_pquota),
161         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
162         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
163         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
164         fsparam_flag("qnoenforce",      Opt_qnoenforce),
165         fsparam_flag("discard",         Opt_discard),
166         fsparam_flag("nodiscard",       Opt_nodiscard),
167         fsparam_flag("dax",             Opt_dax),
168         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
169         {}
170 };
171
172 struct proc_xfs_info {
173         uint64_t        flag;
174         char            *str;
175 };
176
177 static int
178 xfs_fs_show_options(
179         struct seq_file         *m,
180         struct dentry           *root)
181 {
182         static struct proc_xfs_info xfs_info_set[] = {
183                 /* the few simple ones we can get from the mount struct */
184                 { XFS_FEAT_IKEEP,               ",ikeep" },
185                 { XFS_FEAT_WSYNC,               ",wsync" },
186                 { XFS_FEAT_NOALIGN,             ",noalign" },
187                 { XFS_FEAT_SWALLOC,             ",swalloc" },
188                 { XFS_FEAT_NOUUID,              ",nouuid" },
189                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
190                 { XFS_FEAT_ATTR2,               ",attr2" },
191                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
192                 { XFS_FEAT_GRPID,               ",grpid" },
193                 { XFS_FEAT_DISCARD,             ",discard" },
194                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
195                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
196                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
197                 { 0, NULL }
198         };
199         struct xfs_mount        *mp = XFS_M(root->d_sb);
200         struct proc_xfs_info    *xfs_infop;
201
202         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
203                 if (mp->m_features & xfs_infop->flag)
204                         seq_puts(m, xfs_infop->str);
205         }
206
207         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
208
209         if (xfs_has_allocsize(mp))
210                 seq_printf(m, ",allocsize=%dk",
211                            (1 << mp->m_allocsize_log) >> 10);
212
213         if (mp->m_logbufs > 0)
214                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
215         if (mp->m_logbsize > 0)
216                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
217
218         if (mp->m_logname)
219                 seq_show_option(m, "logdev", mp->m_logname);
220         if (mp->m_rtname)
221                 seq_show_option(m, "rtdev", mp->m_rtname);
222
223         if (mp->m_dalign > 0)
224                 seq_printf(m, ",sunit=%d",
225                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
226         if (mp->m_swidth > 0)
227                 seq_printf(m, ",swidth=%d",
228                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
229
230         if (mp->m_qflags & XFS_UQUOTA_ENFD)
231                 seq_puts(m, ",usrquota");
232         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
233                 seq_puts(m, ",uqnoenforce");
234
235         if (mp->m_qflags & XFS_PQUOTA_ENFD)
236                 seq_puts(m, ",prjquota");
237         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
238                 seq_puts(m, ",pqnoenforce");
239
240         if (mp->m_qflags & XFS_GQUOTA_ENFD)
241                 seq_puts(m, ",grpquota");
242         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
243                 seq_puts(m, ",gqnoenforce");
244
245         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
246                 seq_puts(m, ",noquota");
247
248         return 0;
249 }
250
251 static bool
252 xfs_set_inode_alloc_perag(
253         struct xfs_perag        *pag,
254         xfs_ino_t               ino,
255         xfs_agnumber_t          max_metadata)
256 {
257         if (!xfs_is_inode32(pag->pag_mount)) {
258                 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
259                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
260                 return false;
261         }
262
263         if (ino > XFS_MAXINUMBER_32) {
264                 clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
265                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
266                 return false;
267         }
268
269         set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
270         if (pag->pag_agno < max_metadata)
271                 set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
272         else
273                 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
274         return true;
275 }
276
277 /*
278  * Set parameters for inode allocation heuristics, taking into account
279  * filesystem size and inode32/inode64 mount options; i.e. specifically
280  * whether or not XFS_FEAT_SMALL_INUMS is set.
281  *
282  * Inode allocation patterns are altered only if inode32 is requested
283  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
284  * If altered, XFS_OPSTATE_INODE32 is set as well.
285  *
286  * An agcount independent of that in the mount structure is provided
287  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
288  * to the potentially higher ag count.
289  *
290  * Returns the maximum AG index which may contain inodes.
291  */
292 xfs_agnumber_t
293 xfs_set_inode_alloc(
294         struct xfs_mount *mp,
295         xfs_agnumber_t  agcount)
296 {
297         xfs_agnumber_t  index;
298         xfs_agnumber_t  maxagi = 0;
299         xfs_sb_t        *sbp = &mp->m_sb;
300         xfs_agnumber_t  max_metadata;
301         xfs_agino_t     agino;
302         xfs_ino_t       ino;
303
304         /*
305          * Calculate how much should be reserved for inodes to meet
306          * the max inode percentage.  Used only for inode32.
307          */
308         if (M_IGEO(mp)->maxicount) {
309                 uint64_t        icount;
310
311                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
312                 do_div(icount, 100);
313                 icount += sbp->sb_agblocks - 1;
314                 do_div(icount, sbp->sb_agblocks);
315                 max_metadata = icount;
316         } else {
317                 max_metadata = agcount;
318         }
319
320         /* Get the last possible inode in the filesystem */
321         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
322         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
323
324         /*
325          * If user asked for no more than 32-bit inodes, and the fs is
326          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
327          * the allocator to accommodate the request.
328          */
329         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
330                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
331         else
332                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
333
334         for (index = 0; index < agcount; index++) {
335                 struct xfs_perag        *pag;
336
337                 ino = XFS_AGINO_TO_INO(mp, index, agino);
338
339                 pag = xfs_perag_get(mp, index);
340                 if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
341                         maxagi++;
342                 xfs_perag_put(pag);
343         }
344
345         return xfs_is_inode32(mp) ? maxagi : agcount;
346 }
347
348 static int
349 xfs_setup_dax_always(
350         struct xfs_mount        *mp)
351 {
352         if (!mp->m_ddev_targp->bt_daxdev &&
353             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
354                 xfs_alert(mp,
355                         "DAX unsupported by block device. Turning off DAX.");
356                 goto disable_dax;
357         }
358
359         if (mp->m_super->s_blocksize != PAGE_SIZE) {
360                 xfs_alert(mp,
361                         "DAX not supported for blocksize. Turning off DAX.");
362                 goto disable_dax;
363         }
364
365         if (xfs_has_reflink(mp) &&
366             bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
367                 xfs_alert(mp,
368                         "DAX and reflink cannot work with multi-partitions!");
369                 return -EINVAL;
370         }
371
372         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
373         return 0;
374
375 disable_dax:
376         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
377         return 0;
378 }
379
380 STATIC int
381 xfs_blkdev_get(
382         xfs_mount_t             *mp,
383         const char              *name,
384         struct block_device     **bdevp)
385 {
386         int                     error = 0;
387
388         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
389                                     mp);
390         if (IS_ERR(*bdevp)) {
391                 error = PTR_ERR(*bdevp);
392                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
393         }
394
395         return error;
396 }
397
398 STATIC void
399 xfs_blkdev_put(
400         struct block_device     *bdev)
401 {
402         if (bdev)
403                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
404 }
405
406 STATIC void
407 xfs_close_devices(
408         struct xfs_mount        *mp)
409 {
410         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
411                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
412
413                 xfs_free_buftarg(mp->m_logdev_targp);
414                 xfs_blkdev_put(logdev);
415         }
416         if (mp->m_rtdev_targp) {
417                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
418
419                 xfs_free_buftarg(mp->m_rtdev_targp);
420                 xfs_blkdev_put(rtdev);
421         }
422         xfs_free_buftarg(mp->m_ddev_targp);
423 }
424
425 /*
426  * The file system configurations are:
427  *      (1) device (partition) with data and internal log
428  *      (2) logical volume with data and log subvolumes.
429  *      (3) logical volume with data, log, and realtime subvolumes.
430  *
431  * We only have to handle opening the log and realtime volumes here if
432  * they are present.  The data subvolume has already been opened by
433  * get_sb_bdev() and is stored in sb->s_bdev.
434  */
435 STATIC int
436 xfs_open_devices(
437         struct xfs_mount        *mp)
438 {
439         struct block_device     *ddev = mp->m_super->s_bdev;
440         struct block_device     *logdev = NULL, *rtdev = NULL;
441         int                     error;
442
443         /*
444          * Open real time and log devices - order is important.
445          */
446         if (mp->m_logname) {
447                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
448                 if (error)
449                         return error;
450         }
451
452         if (mp->m_rtname) {
453                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
454                 if (error)
455                         goto out_close_logdev;
456
457                 if (rtdev == ddev || rtdev == logdev) {
458                         xfs_warn(mp,
459         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
460                         error = -EINVAL;
461                         goto out_close_rtdev;
462                 }
463         }
464
465         /*
466          * Setup xfs_mount buffer target pointers
467          */
468         error = -ENOMEM;
469         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
470         if (!mp->m_ddev_targp)
471                 goto out_close_rtdev;
472
473         if (rtdev) {
474                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
475                 if (!mp->m_rtdev_targp)
476                         goto out_free_ddev_targ;
477         }
478
479         if (logdev && logdev != ddev) {
480                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
481                 if (!mp->m_logdev_targp)
482                         goto out_free_rtdev_targ;
483         } else {
484                 mp->m_logdev_targp = mp->m_ddev_targp;
485         }
486
487         return 0;
488
489  out_free_rtdev_targ:
490         if (mp->m_rtdev_targp)
491                 xfs_free_buftarg(mp->m_rtdev_targp);
492  out_free_ddev_targ:
493         xfs_free_buftarg(mp->m_ddev_targp);
494  out_close_rtdev:
495         xfs_blkdev_put(rtdev);
496  out_close_logdev:
497         if (logdev && logdev != ddev)
498                 xfs_blkdev_put(logdev);
499         return error;
500 }
501
502 /*
503  * Setup xfs_mount buffer target pointers based on superblock
504  */
505 STATIC int
506 xfs_setup_devices(
507         struct xfs_mount        *mp)
508 {
509         int                     error;
510
511         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
512         if (error)
513                 return error;
514
515         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
516                 unsigned int    log_sector_size = BBSIZE;
517
518                 if (xfs_has_sector(mp))
519                         log_sector_size = mp->m_sb.sb_logsectsize;
520                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
521                                             log_sector_size);
522                 if (error)
523                         return error;
524         }
525         if (mp->m_rtdev_targp) {
526                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
527                                             mp->m_sb.sb_sectsize);
528                 if (error)
529                         return error;
530         }
531
532         return 0;
533 }
534
535 STATIC int
536 xfs_init_mount_workqueues(
537         struct xfs_mount        *mp)
538 {
539         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
540                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
541                         1, mp->m_super->s_id);
542         if (!mp->m_buf_workqueue)
543                 goto out;
544
545         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
546                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
547                         0, mp->m_super->s_id);
548         if (!mp->m_unwritten_workqueue)
549                 goto out_destroy_buf;
550
551         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
552                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
553                         0, mp->m_super->s_id);
554         if (!mp->m_reclaim_workqueue)
555                 goto out_destroy_unwritten;
556
557         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
558                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
559                         0, mp->m_super->s_id);
560         if (!mp->m_blockgc_wq)
561                 goto out_destroy_reclaim;
562
563         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
564                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
565                         1, mp->m_super->s_id);
566         if (!mp->m_inodegc_wq)
567                 goto out_destroy_blockgc;
568
569         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
570                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
571         if (!mp->m_sync_workqueue)
572                 goto out_destroy_inodegc;
573
574         return 0;
575
576 out_destroy_inodegc:
577         destroy_workqueue(mp->m_inodegc_wq);
578 out_destroy_blockgc:
579         destroy_workqueue(mp->m_blockgc_wq);
580 out_destroy_reclaim:
581         destroy_workqueue(mp->m_reclaim_workqueue);
582 out_destroy_unwritten:
583         destroy_workqueue(mp->m_unwritten_workqueue);
584 out_destroy_buf:
585         destroy_workqueue(mp->m_buf_workqueue);
586 out:
587         return -ENOMEM;
588 }
589
590 STATIC void
591 xfs_destroy_mount_workqueues(
592         struct xfs_mount        *mp)
593 {
594         destroy_workqueue(mp->m_sync_workqueue);
595         destroy_workqueue(mp->m_blockgc_wq);
596         destroy_workqueue(mp->m_inodegc_wq);
597         destroy_workqueue(mp->m_reclaim_workqueue);
598         destroy_workqueue(mp->m_unwritten_workqueue);
599         destroy_workqueue(mp->m_buf_workqueue);
600 }
601
602 static void
603 xfs_flush_inodes_worker(
604         struct work_struct      *work)
605 {
606         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
607                                                    m_flush_inodes_work);
608         struct super_block      *sb = mp->m_super;
609
610         if (down_read_trylock(&sb->s_umount)) {
611                 sync_inodes_sb(sb);
612                 up_read(&sb->s_umount);
613         }
614 }
615
616 /*
617  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
618  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
619  * for IO to complete so that we effectively throttle multiple callers to the
620  * rate at which IO is completing.
621  */
622 void
623 xfs_flush_inodes(
624         struct xfs_mount        *mp)
625 {
626         /*
627          * If flush_work() returns true then that means we waited for a flush
628          * which was already in progress.  Don't bother running another scan.
629          */
630         if (flush_work(&mp->m_flush_inodes_work))
631                 return;
632
633         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
634         flush_work(&mp->m_flush_inodes_work);
635 }
636
637 /* Catch misguided souls that try to use this interface on XFS */
638 STATIC struct inode *
639 xfs_fs_alloc_inode(
640         struct super_block      *sb)
641 {
642         BUG();
643         return NULL;
644 }
645
646 /*
647  * Now that the generic code is guaranteed not to be accessing
648  * the linux inode, we can inactivate and reclaim the inode.
649  */
650 STATIC void
651 xfs_fs_destroy_inode(
652         struct inode            *inode)
653 {
654         struct xfs_inode        *ip = XFS_I(inode);
655
656         trace_xfs_destroy_inode(ip);
657
658         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
659         XFS_STATS_INC(ip->i_mount, vn_rele);
660         XFS_STATS_INC(ip->i_mount, vn_remove);
661         xfs_inode_mark_reclaimable(ip);
662 }
663
664 static void
665 xfs_fs_dirty_inode(
666         struct inode                    *inode,
667         int                             flags)
668 {
669         struct xfs_inode                *ip = XFS_I(inode);
670         struct xfs_mount                *mp = ip->i_mount;
671         struct xfs_trans                *tp;
672
673         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
674                 return;
675
676         /*
677          * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
678          * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
679          * in flags possibly together with I_DIRTY_SYNC.
680          */
681         if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
682                 return;
683
684         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
685                 return;
686         xfs_ilock(ip, XFS_ILOCK_EXCL);
687         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
688         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
689         xfs_trans_commit(tp);
690 }
691
692 /*
693  * Slab object creation initialisation for the XFS inode.
694  * This covers only the idempotent fields in the XFS inode;
695  * all other fields need to be initialised on allocation
696  * from the slab. This avoids the need to repeatedly initialise
697  * fields in the xfs inode that left in the initialise state
698  * when freeing the inode.
699  */
700 STATIC void
701 xfs_fs_inode_init_once(
702         void                    *inode)
703 {
704         struct xfs_inode        *ip = inode;
705
706         memset(ip, 0, sizeof(struct xfs_inode));
707
708         /* vfs inode */
709         inode_init_once(VFS_I(ip));
710
711         /* xfs inode */
712         atomic_set(&ip->i_pincount, 0);
713         spin_lock_init(&ip->i_flags_lock);
714
715         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
716                      "xfsino", ip->i_ino);
717 }
718
719 /*
720  * We do an unlocked check for XFS_IDONTCACHE here because we are already
721  * serialised against cache hits here via the inode->i_lock and igrab() in
722  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
723  * racing with us, and it avoids needing to grab a spinlock here for every inode
724  * we drop the final reference on.
725  */
726 STATIC int
727 xfs_fs_drop_inode(
728         struct inode            *inode)
729 {
730         struct xfs_inode        *ip = XFS_I(inode);
731
732         /*
733          * If this unlinked inode is in the middle of recovery, don't
734          * drop the inode just yet; log recovery will take care of
735          * that.  See the comment for this inode flag.
736          */
737         if (ip->i_flags & XFS_IRECOVERY) {
738                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
739                 return 0;
740         }
741
742         return generic_drop_inode(inode);
743 }
744
745 static void
746 xfs_mount_free(
747         struct xfs_mount        *mp)
748 {
749         kfree(mp->m_rtname);
750         kfree(mp->m_logname);
751         kmem_free(mp);
752 }
753
754 STATIC int
755 xfs_fs_sync_fs(
756         struct super_block      *sb,
757         int                     wait)
758 {
759         struct xfs_mount        *mp = XFS_M(sb);
760         int                     error;
761
762         trace_xfs_fs_sync_fs(mp, __return_address);
763
764         /*
765          * Doing anything during the async pass would be counterproductive.
766          */
767         if (!wait)
768                 return 0;
769
770         error = xfs_log_force(mp, XFS_LOG_SYNC);
771         if (error)
772                 return error;
773
774         if (laptop_mode) {
775                 /*
776                  * The disk must be active because we're syncing.
777                  * We schedule log work now (now that the disk is
778                  * active) instead of later (when it might not be).
779                  */
780                 flush_delayed_work(&mp->m_log->l_work);
781         }
782
783         /*
784          * If we are called with page faults frozen out, it means we are about
785          * to freeze the transaction subsystem. Take the opportunity to shut
786          * down inodegc because once SB_FREEZE_FS is set it's too late to
787          * prevent inactivation races with freeze. The fs doesn't get called
788          * again by the freezing process until after SB_FREEZE_FS has been set,
789          * so it's now or never.  Same logic applies to speculative allocation
790          * garbage collection.
791          *
792          * We don't care if this is a normal syncfs call that does this or
793          * freeze that does this - we can run this multiple times without issue
794          * and we won't race with a restart because a restart can only occur
795          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
796          */
797         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
798                 xfs_inodegc_stop(mp);
799                 xfs_blockgc_stop(mp);
800         }
801
802         return 0;
803 }
804
805 STATIC int
806 xfs_fs_statfs(
807         struct dentry           *dentry,
808         struct kstatfs          *statp)
809 {
810         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
811         xfs_sb_t                *sbp = &mp->m_sb;
812         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
813         uint64_t                fakeinos, id;
814         uint64_t                icount;
815         uint64_t                ifree;
816         uint64_t                fdblocks;
817         xfs_extlen_t            lsize;
818         int64_t                 ffree;
819
820         /*
821          * Expedite background inodegc but don't wait. We do not want to block
822          * here waiting hours for a billion extent file to be truncated.
823          */
824         xfs_inodegc_push(mp);
825
826         statp->f_type = XFS_SUPER_MAGIC;
827         statp->f_namelen = MAXNAMELEN - 1;
828
829         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
830         statp->f_fsid = u64_to_fsid(id);
831
832         icount = percpu_counter_sum(&mp->m_icount);
833         ifree = percpu_counter_sum(&mp->m_ifree);
834         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
835
836         spin_lock(&mp->m_sb_lock);
837         statp->f_bsize = sbp->sb_blocksize;
838         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
839         statp->f_blocks = sbp->sb_dblocks - lsize;
840         spin_unlock(&mp->m_sb_lock);
841
842         /* make sure statp->f_bfree does not underflow */
843         statp->f_bfree = max_t(int64_t, 0,
844                                 fdblocks - xfs_fdblocks_unavailable(mp));
845         statp->f_bavail = statp->f_bfree;
846
847         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
848         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
849         if (M_IGEO(mp)->maxicount)
850                 statp->f_files = min_t(typeof(statp->f_files),
851                                         statp->f_files,
852                                         M_IGEO(mp)->maxicount);
853
854         /* If sb_icount overshot maxicount, report actual allocation */
855         statp->f_files = max_t(typeof(statp->f_files),
856                                         statp->f_files,
857                                         sbp->sb_icount);
858
859         /* make sure statp->f_ffree does not underflow */
860         ffree = statp->f_files - (icount - ifree);
861         statp->f_ffree = max_t(int64_t, ffree, 0);
862
863
864         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
865             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
866                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
867                 xfs_qm_statvfs(ip, statp);
868
869         if (XFS_IS_REALTIME_MOUNT(mp) &&
870             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
871                 s64     freertx;
872
873                 statp->f_blocks = sbp->sb_rblocks;
874                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
875                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
876         }
877
878         return 0;
879 }
880
881 STATIC void
882 xfs_save_resvblks(struct xfs_mount *mp)
883 {
884         uint64_t resblks = 0;
885
886         mp->m_resblks_save = mp->m_resblks;
887         xfs_reserve_blocks(mp, &resblks, NULL);
888 }
889
890 STATIC void
891 xfs_restore_resvblks(struct xfs_mount *mp)
892 {
893         uint64_t resblks;
894
895         if (mp->m_resblks_save) {
896                 resblks = mp->m_resblks_save;
897                 mp->m_resblks_save = 0;
898         } else
899                 resblks = xfs_default_resblks(mp);
900
901         xfs_reserve_blocks(mp, &resblks, NULL);
902 }
903
904 /*
905  * Second stage of a freeze. The data is already frozen so we only
906  * need to take care of the metadata. Once that's done sync the superblock
907  * to the log to dirty it in case of a crash while frozen. This ensures that we
908  * will recover the unlinked inode lists on the next mount.
909  */
910 STATIC int
911 xfs_fs_freeze(
912         struct super_block      *sb)
913 {
914         struct xfs_mount        *mp = XFS_M(sb);
915         unsigned int            flags;
916         int                     ret;
917
918         /*
919          * The filesystem is now frozen far enough that memory reclaim
920          * cannot safely operate on the filesystem. Hence we need to
921          * set a GFP_NOFS context here to avoid recursion deadlocks.
922          */
923         flags = memalloc_nofs_save();
924         xfs_save_resvblks(mp);
925         ret = xfs_log_quiesce(mp);
926         memalloc_nofs_restore(flags);
927
928         /*
929          * For read-write filesystems, we need to restart the inodegc on error
930          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
931          * going to be run to restart it now.  We are at SB_FREEZE_FS level
932          * here, so we can restart safely without racing with a stop in
933          * xfs_fs_sync_fs().
934          */
935         if (ret && !xfs_is_readonly(mp)) {
936                 xfs_blockgc_start(mp);
937                 xfs_inodegc_start(mp);
938         }
939
940         return ret;
941 }
942
943 STATIC int
944 xfs_fs_unfreeze(
945         struct super_block      *sb)
946 {
947         struct xfs_mount        *mp = XFS_M(sb);
948
949         xfs_restore_resvblks(mp);
950         xfs_log_work_queue(mp);
951
952         /*
953          * Don't reactivate the inodegc worker on a readonly filesystem because
954          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
955          * worker because there are no speculative preallocations on a readonly
956          * filesystem.
957          */
958         if (!xfs_is_readonly(mp)) {
959                 xfs_blockgc_start(mp);
960                 xfs_inodegc_start(mp);
961         }
962
963         return 0;
964 }
965
966 /*
967  * This function fills in xfs_mount_t fields based on mount args.
968  * Note: the superblock _has_ now been read in.
969  */
970 STATIC int
971 xfs_finish_flags(
972         struct xfs_mount        *mp)
973 {
974         /* Fail a mount where the logbuf is smaller than the log stripe */
975         if (xfs_has_logv2(mp)) {
976                 if (mp->m_logbsize <= 0 &&
977                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
978                         mp->m_logbsize = mp->m_sb.sb_logsunit;
979                 } else if (mp->m_logbsize > 0 &&
980                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
981                         xfs_warn(mp,
982                 "logbuf size must be greater than or equal to log stripe size");
983                         return -EINVAL;
984                 }
985         } else {
986                 /* Fail a mount if the logbuf is larger than 32K */
987                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
988                         xfs_warn(mp,
989                 "logbuf size for version 1 logs must be 16K or 32K");
990                         return -EINVAL;
991                 }
992         }
993
994         /*
995          * V5 filesystems always use attr2 format for attributes.
996          */
997         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
998                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
999                              "attr2 is always enabled for V5 filesystems.");
1000                 return -EINVAL;
1001         }
1002
1003         /*
1004          * prohibit r/w mounts of read-only filesystems
1005          */
1006         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1007                 xfs_warn(mp,
1008                         "cannot mount a read-only filesystem as read-write");
1009                 return -EROFS;
1010         }
1011
1012         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1013             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1014             !xfs_has_pquotino(mp)) {
1015                 xfs_warn(mp,
1016                   "Super block does not support project and group quota together");
1017                 return -EINVAL;
1018         }
1019
1020         return 0;
1021 }
1022
1023 static int
1024 xfs_init_percpu_counters(
1025         struct xfs_mount        *mp)
1026 {
1027         int             error;
1028
1029         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1030         if (error)
1031                 return -ENOMEM;
1032
1033         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1034         if (error)
1035                 goto free_icount;
1036
1037         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1038         if (error)
1039                 goto free_ifree;
1040
1041         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1042         if (error)
1043                 goto free_fdblocks;
1044
1045         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1046         if (error)
1047                 goto free_delalloc;
1048
1049         return 0;
1050
1051 free_delalloc:
1052         percpu_counter_destroy(&mp->m_delalloc_blks);
1053 free_fdblocks:
1054         percpu_counter_destroy(&mp->m_fdblocks);
1055 free_ifree:
1056         percpu_counter_destroy(&mp->m_ifree);
1057 free_icount:
1058         percpu_counter_destroy(&mp->m_icount);
1059         return -ENOMEM;
1060 }
1061
1062 void
1063 xfs_reinit_percpu_counters(
1064         struct xfs_mount        *mp)
1065 {
1066         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1067         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1068         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1069         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1070 }
1071
1072 static void
1073 xfs_destroy_percpu_counters(
1074         struct xfs_mount        *mp)
1075 {
1076         percpu_counter_destroy(&mp->m_icount);
1077         percpu_counter_destroy(&mp->m_ifree);
1078         percpu_counter_destroy(&mp->m_fdblocks);
1079         ASSERT(xfs_is_shutdown(mp) ||
1080                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1081         percpu_counter_destroy(&mp->m_delalloc_blks);
1082         percpu_counter_destroy(&mp->m_frextents);
1083 }
1084
1085 static int
1086 xfs_inodegc_init_percpu(
1087         struct xfs_mount        *mp)
1088 {
1089         struct xfs_inodegc      *gc;
1090         int                     cpu;
1091
1092         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1093         if (!mp->m_inodegc)
1094                 return -ENOMEM;
1095
1096         for_each_possible_cpu(cpu) {
1097                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1098                 init_llist_head(&gc->list);
1099                 gc->items = 0;
1100                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1101         }
1102         return 0;
1103 }
1104
1105 static void
1106 xfs_inodegc_free_percpu(
1107         struct xfs_mount        *mp)
1108 {
1109         if (!mp->m_inodegc)
1110                 return;
1111         free_percpu(mp->m_inodegc);
1112 }
1113
1114 static void
1115 xfs_fs_put_super(
1116         struct super_block      *sb)
1117 {
1118         struct xfs_mount        *mp = XFS_M(sb);
1119
1120         /* if ->fill_super failed, we have no mount to tear down */
1121         if (!sb->s_fs_info)
1122                 return;
1123
1124         xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1125         xfs_filestream_unmount(mp);
1126         xfs_unmountfs(mp);
1127
1128         xfs_freesb(mp);
1129         free_percpu(mp->m_stats.xs_stats);
1130         xfs_mount_list_del(mp);
1131         xfs_inodegc_free_percpu(mp);
1132         xfs_destroy_percpu_counters(mp);
1133         xfs_destroy_mount_workqueues(mp);
1134         xfs_close_devices(mp);
1135
1136         sb->s_fs_info = NULL;
1137         xfs_mount_free(mp);
1138 }
1139
1140 static long
1141 xfs_fs_nr_cached_objects(
1142         struct super_block      *sb,
1143         struct shrink_control   *sc)
1144 {
1145         /* Paranoia: catch incorrect calls during mount setup or teardown */
1146         if (WARN_ON_ONCE(!sb->s_fs_info))
1147                 return 0;
1148         return xfs_reclaim_inodes_count(XFS_M(sb));
1149 }
1150
1151 static long
1152 xfs_fs_free_cached_objects(
1153         struct super_block      *sb,
1154         struct shrink_control   *sc)
1155 {
1156         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1157 }
1158
1159 static const struct super_operations xfs_super_operations = {
1160         .alloc_inode            = xfs_fs_alloc_inode,
1161         .destroy_inode          = xfs_fs_destroy_inode,
1162         .dirty_inode            = xfs_fs_dirty_inode,
1163         .drop_inode             = xfs_fs_drop_inode,
1164         .put_super              = xfs_fs_put_super,
1165         .sync_fs                = xfs_fs_sync_fs,
1166         .freeze_fs              = xfs_fs_freeze,
1167         .unfreeze_fs            = xfs_fs_unfreeze,
1168         .statfs                 = xfs_fs_statfs,
1169         .show_options           = xfs_fs_show_options,
1170         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1171         .free_cached_objects    = xfs_fs_free_cached_objects,
1172 };
1173
1174 static int
1175 suffix_kstrtoint(
1176         const char      *s,
1177         unsigned int    base,
1178         int             *res)
1179 {
1180         int             last, shift_left_factor = 0, _res;
1181         char            *value;
1182         int             ret = 0;
1183
1184         value = kstrdup(s, GFP_KERNEL);
1185         if (!value)
1186                 return -ENOMEM;
1187
1188         last = strlen(value) - 1;
1189         if (value[last] == 'K' || value[last] == 'k') {
1190                 shift_left_factor = 10;
1191                 value[last] = '\0';
1192         }
1193         if (value[last] == 'M' || value[last] == 'm') {
1194                 shift_left_factor = 20;
1195                 value[last] = '\0';
1196         }
1197         if (value[last] == 'G' || value[last] == 'g') {
1198                 shift_left_factor = 30;
1199                 value[last] = '\0';
1200         }
1201
1202         if (kstrtoint(value, base, &_res))
1203                 ret = -EINVAL;
1204         kfree(value);
1205         *res = _res << shift_left_factor;
1206         return ret;
1207 }
1208
1209 static inline void
1210 xfs_fs_warn_deprecated(
1211         struct fs_context       *fc,
1212         struct fs_parameter     *param,
1213         uint64_t                flag,
1214         bool                    value)
1215 {
1216         /* Don't print the warning if reconfiguring and current mount point
1217          * already had the flag set
1218          */
1219         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1220             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1221                 return;
1222         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1223 }
1224
1225 /*
1226  * Set mount state from a mount option.
1227  *
1228  * NOTE: mp->m_super is NULL here!
1229  */
1230 static int
1231 xfs_fs_parse_param(
1232         struct fs_context       *fc,
1233         struct fs_parameter     *param)
1234 {
1235         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1236         struct fs_parse_result  result;
1237         int                     size = 0;
1238         int                     opt;
1239
1240         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1241         if (opt < 0)
1242                 return opt;
1243
1244         switch (opt) {
1245         case Opt_logbufs:
1246                 parsing_mp->m_logbufs = result.uint_32;
1247                 return 0;
1248         case Opt_logbsize:
1249                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1250                         return -EINVAL;
1251                 return 0;
1252         case Opt_logdev:
1253                 kfree(parsing_mp->m_logname);
1254                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1255                 if (!parsing_mp->m_logname)
1256                         return -ENOMEM;
1257                 return 0;
1258         case Opt_rtdev:
1259                 kfree(parsing_mp->m_rtname);
1260                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1261                 if (!parsing_mp->m_rtname)
1262                         return -ENOMEM;
1263                 return 0;
1264         case Opt_allocsize:
1265                 if (suffix_kstrtoint(param->string, 10, &size))
1266                         return -EINVAL;
1267                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1268                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1269                 return 0;
1270         case Opt_grpid:
1271         case Opt_bsdgroups:
1272                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1273                 return 0;
1274         case Opt_nogrpid:
1275         case Opt_sysvgroups:
1276                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1277                 return 0;
1278         case Opt_wsync:
1279                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1280                 return 0;
1281         case Opt_norecovery:
1282                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1283                 return 0;
1284         case Opt_noalign:
1285                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1286                 return 0;
1287         case Opt_swalloc:
1288                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1289                 return 0;
1290         case Opt_sunit:
1291                 parsing_mp->m_dalign = result.uint_32;
1292                 return 0;
1293         case Opt_swidth:
1294                 parsing_mp->m_swidth = result.uint_32;
1295                 return 0;
1296         case Opt_inode32:
1297                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1298                 return 0;
1299         case Opt_inode64:
1300                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1301                 return 0;
1302         case Opt_nouuid:
1303                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1304                 return 0;
1305         case Opt_largeio:
1306                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1307                 return 0;
1308         case Opt_nolargeio:
1309                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1310                 return 0;
1311         case Opt_filestreams:
1312                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1313                 return 0;
1314         case Opt_noquota:
1315                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1316                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1317                 return 0;
1318         case Opt_quota:
1319         case Opt_uquota:
1320         case Opt_usrquota:
1321                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1322                 return 0;
1323         case Opt_qnoenforce:
1324         case Opt_uqnoenforce:
1325                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1326                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1327                 return 0;
1328         case Opt_pquota:
1329         case Opt_prjquota:
1330                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1331                 return 0;
1332         case Opt_pqnoenforce:
1333                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1334                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1335                 return 0;
1336         case Opt_gquota:
1337         case Opt_grpquota:
1338                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1339                 return 0;
1340         case Opt_gqnoenforce:
1341                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1342                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1343                 return 0;
1344         case Opt_discard:
1345                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1346                 return 0;
1347         case Opt_nodiscard:
1348                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1349                 return 0;
1350 #ifdef CONFIG_FS_DAX
1351         case Opt_dax:
1352                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1353                 return 0;
1354         case Opt_dax_enum:
1355                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1356                 return 0;
1357 #endif
1358         /* Following mount options will be removed in September 2025 */
1359         case Opt_ikeep:
1360                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1361                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1362                 return 0;
1363         case Opt_noikeep:
1364                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1365                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1366                 return 0;
1367         case Opt_attr2:
1368                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1369                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1370                 return 0;
1371         case Opt_noattr2:
1372                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1373                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1374                 return 0;
1375         default:
1376                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1377                 return -EINVAL;
1378         }
1379
1380         return 0;
1381 }
1382
1383 static int
1384 xfs_fs_validate_params(
1385         struct xfs_mount        *mp)
1386 {
1387         /* No recovery flag requires a read-only mount */
1388         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1389                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1390                 return -EINVAL;
1391         }
1392
1393         /*
1394          * We have not read the superblock at this point, so only the attr2
1395          * mount option can set the attr2 feature by this stage.
1396          */
1397         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1398                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1399                 return -EINVAL;
1400         }
1401
1402
1403         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1404                 xfs_warn(mp,
1405         "sunit and swidth options incompatible with the noalign option");
1406                 return -EINVAL;
1407         }
1408
1409         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1410                 xfs_warn(mp, "quota support not available in this kernel.");
1411                 return -EINVAL;
1412         }
1413
1414         if ((mp->m_dalign && !mp->m_swidth) ||
1415             (!mp->m_dalign && mp->m_swidth)) {
1416                 xfs_warn(mp, "sunit and swidth must be specified together");
1417                 return -EINVAL;
1418         }
1419
1420         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1421                 xfs_warn(mp,
1422         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1423                         mp->m_swidth, mp->m_dalign);
1424                 return -EINVAL;
1425         }
1426
1427         if (mp->m_logbufs != -1 &&
1428             mp->m_logbufs != 0 &&
1429             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1430              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1431                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1432                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1433                 return -EINVAL;
1434         }
1435
1436         if (mp->m_logbsize != -1 &&
1437             mp->m_logbsize !=  0 &&
1438             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1439              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1440              !is_power_of_2(mp->m_logbsize))) {
1441                 xfs_warn(mp,
1442                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1443                         mp->m_logbsize);
1444                 return -EINVAL;
1445         }
1446
1447         if (xfs_has_allocsize(mp) &&
1448             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1449              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1450                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1451                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1452                 return -EINVAL;
1453         }
1454
1455         return 0;
1456 }
1457
1458 static int
1459 xfs_fs_fill_super(
1460         struct super_block      *sb,
1461         struct fs_context       *fc)
1462 {
1463         struct xfs_mount        *mp = sb->s_fs_info;
1464         struct inode            *root;
1465         int                     flags = 0, error;
1466
1467         mp->m_super = sb;
1468
1469         error = xfs_fs_validate_params(mp);
1470         if (error)
1471                 goto out_free_names;
1472
1473         sb_min_blocksize(sb, BBSIZE);
1474         sb->s_xattr = xfs_xattr_handlers;
1475         sb->s_export_op = &xfs_export_operations;
1476 #ifdef CONFIG_XFS_QUOTA
1477         sb->s_qcop = &xfs_quotactl_operations;
1478         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1479 #endif
1480         sb->s_op = &xfs_super_operations;
1481
1482         /*
1483          * Delay mount work if the debug hook is set. This is debug
1484          * instrumention to coordinate simulation of xfs mount failures with
1485          * VFS superblock operations
1486          */
1487         if (xfs_globals.mount_delay) {
1488                 xfs_notice(mp, "Delaying mount for %d seconds.",
1489                         xfs_globals.mount_delay);
1490                 msleep(xfs_globals.mount_delay * 1000);
1491         }
1492
1493         if (fc->sb_flags & SB_SILENT)
1494                 flags |= XFS_MFSI_QUIET;
1495
1496         error = xfs_open_devices(mp);
1497         if (error)
1498                 goto out_free_names;
1499
1500         error = xfs_init_mount_workqueues(mp);
1501         if (error)
1502                 goto out_close_devices;
1503
1504         error = xfs_init_percpu_counters(mp);
1505         if (error)
1506                 goto out_destroy_workqueues;
1507
1508         error = xfs_inodegc_init_percpu(mp);
1509         if (error)
1510                 goto out_destroy_counters;
1511
1512         /*
1513          * All percpu data structures requiring cleanup when a cpu goes offline
1514          * must be allocated before adding this @mp to the cpu-dead handler's
1515          * mount list.
1516          */
1517         xfs_mount_list_add(mp);
1518
1519         /* Allocate stats memory before we do operations that might use it */
1520         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1521         if (!mp->m_stats.xs_stats) {
1522                 error = -ENOMEM;
1523                 goto out_destroy_inodegc;
1524         }
1525
1526         error = xfs_readsb(mp, flags);
1527         if (error)
1528                 goto out_free_stats;
1529
1530         error = xfs_finish_flags(mp);
1531         if (error)
1532                 goto out_free_sb;
1533
1534         error = xfs_setup_devices(mp);
1535         if (error)
1536                 goto out_free_sb;
1537
1538         /* V4 support is undergoing deprecation. */
1539         if (!xfs_has_crc(mp)) {
1540 #ifdef CONFIG_XFS_SUPPORT_V4
1541                 xfs_warn_once(mp,
1542         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1543 #else
1544                 xfs_warn(mp,
1545         "Deprecated V4 format (crc=0) not supported by kernel.");
1546                 error = -EINVAL;
1547                 goto out_free_sb;
1548 #endif
1549         }
1550
1551         /* ASCII case insensitivity is undergoing deprecation. */
1552         if (xfs_has_asciici(mp)) {
1553 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1554                 xfs_warn_once(mp,
1555         "Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1556 #else
1557                 xfs_warn(mp,
1558         "Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1559                 error = -EINVAL;
1560                 goto out_free_sb;
1561 #endif
1562         }
1563
1564         /* Filesystem claims it needs repair, so refuse the mount. */
1565         if (xfs_has_needsrepair(mp)) {
1566                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1567                 error = -EFSCORRUPTED;
1568                 goto out_free_sb;
1569         }
1570
1571         /*
1572          * Don't touch the filesystem if a user tool thinks it owns the primary
1573          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1574          * we don't check them at all.
1575          */
1576         if (mp->m_sb.sb_inprogress) {
1577                 xfs_warn(mp, "Offline file system operation in progress!");
1578                 error = -EFSCORRUPTED;
1579                 goto out_free_sb;
1580         }
1581
1582         /*
1583          * Until this is fixed only page-sized or smaller data blocks work.
1584          */
1585         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1586                 xfs_warn(mp,
1587                 "File system with blocksize %d bytes. "
1588                 "Only pagesize (%ld) or less will currently work.",
1589                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1590                 error = -ENOSYS;
1591                 goto out_free_sb;
1592         }
1593
1594         /* Ensure this filesystem fits in the page cache limits */
1595         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1596             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1597                 xfs_warn(mp,
1598                 "file system too large to be mounted on this system.");
1599                 error = -EFBIG;
1600                 goto out_free_sb;
1601         }
1602
1603         /*
1604          * XFS block mappings use 54 bits to store the logical block offset.
1605          * This should suffice to handle the maximum file size that the VFS
1606          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1607          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1608          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1609          * to check this assertion.
1610          *
1611          * Avoid integer overflow by comparing the maximum bmbt offset to the
1612          * maximum pagecache offset in units of fs blocks.
1613          */
1614         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1615                 xfs_warn(mp,
1616 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1617                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1618                          XFS_MAX_FILEOFF);
1619                 error = -EINVAL;
1620                 goto out_free_sb;
1621         }
1622
1623         error = xfs_filestream_mount(mp);
1624         if (error)
1625                 goto out_free_sb;
1626
1627         /*
1628          * we must configure the block size in the superblock before we run the
1629          * full mount process as the mount process can lookup and cache inodes.
1630          */
1631         sb->s_magic = XFS_SUPER_MAGIC;
1632         sb->s_blocksize = mp->m_sb.sb_blocksize;
1633         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1634         sb->s_maxbytes = MAX_LFS_FILESIZE;
1635         sb->s_max_links = XFS_MAXLINK;
1636         sb->s_time_gran = 1;
1637         if (xfs_has_bigtime(mp)) {
1638                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1639                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1640         } else {
1641                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1642                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1643         }
1644         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1645         sb->s_iflags |= SB_I_CGROUPWB;
1646
1647         set_posix_acl_flag(sb);
1648
1649         /* version 5 superblocks support inode version counters. */
1650         if (xfs_has_crc(mp))
1651                 sb->s_flags |= SB_I_VERSION;
1652
1653         if (xfs_has_dax_always(mp)) {
1654                 error = xfs_setup_dax_always(mp);
1655                 if (error)
1656                         goto out_filestream_unmount;
1657         }
1658
1659         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1660                 xfs_warn(mp,
1661         "mounting with \"discard\" option, but the device does not support discard");
1662                 mp->m_features &= ~XFS_FEAT_DISCARD;
1663         }
1664
1665         if (xfs_has_reflink(mp)) {
1666                 if (mp->m_sb.sb_rblocks) {
1667                         xfs_alert(mp,
1668         "reflink not compatible with realtime device!");
1669                         error = -EINVAL;
1670                         goto out_filestream_unmount;
1671                 }
1672
1673                 if (xfs_globals.always_cow) {
1674                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1675                         mp->m_always_cow = true;
1676                 }
1677         }
1678
1679         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1680                 xfs_alert(mp,
1681         "reverse mapping btree not compatible with realtime device!");
1682                 error = -EINVAL;
1683                 goto out_filestream_unmount;
1684         }
1685
1686         if (xfs_has_large_extent_counts(mp))
1687                 xfs_warn(mp,
1688         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1689
1690         error = xfs_mountfs(mp);
1691         if (error)
1692                 goto out_filestream_unmount;
1693
1694         root = igrab(VFS_I(mp->m_rootip));
1695         if (!root) {
1696                 error = -ENOENT;
1697                 goto out_unmount;
1698         }
1699         sb->s_root = d_make_root(root);
1700         if (!sb->s_root) {
1701                 error = -ENOMEM;
1702                 goto out_unmount;
1703         }
1704
1705         return 0;
1706
1707  out_filestream_unmount:
1708         xfs_filestream_unmount(mp);
1709  out_free_sb:
1710         xfs_freesb(mp);
1711  out_free_stats:
1712         free_percpu(mp->m_stats.xs_stats);
1713  out_destroy_inodegc:
1714         xfs_mount_list_del(mp);
1715         xfs_inodegc_free_percpu(mp);
1716  out_destroy_counters:
1717         xfs_destroy_percpu_counters(mp);
1718  out_destroy_workqueues:
1719         xfs_destroy_mount_workqueues(mp);
1720  out_close_devices:
1721         xfs_close_devices(mp);
1722  out_free_names:
1723         sb->s_fs_info = NULL;
1724         xfs_mount_free(mp);
1725         return error;
1726
1727  out_unmount:
1728         xfs_filestream_unmount(mp);
1729         xfs_unmountfs(mp);
1730         goto out_free_sb;
1731 }
1732
1733 static int
1734 xfs_fs_get_tree(
1735         struct fs_context       *fc)
1736 {
1737         return get_tree_bdev(fc, xfs_fs_fill_super);
1738 }
1739
1740 static int
1741 xfs_remount_rw(
1742         struct xfs_mount        *mp)
1743 {
1744         struct xfs_sb           *sbp = &mp->m_sb;
1745         int error;
1746
1747         if (xfs_has_norecovery(mp)) {
1748                 xfs_warn(mp,
1749                         "ro->rw transition prohibited on norecovery mount");
1750                 return -EINVAL;
1751         }
1752
1753         if (xfs_sb_is_v5(sbp) &&
1754             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1755                 xfs_warn(mp,
1756         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1757                         (sbp->sb_features_ro_compat &
1758                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1759                 return -EINVAL;
1760         }
1761
1762         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1763
1764         /*
1765          * If this is the first remount to writeable state we might have some
1766          * superblock changes to update.
1767          */
1768         if (mp->m_update_sb) {
1769                 error = xfs_sync_sb(mp, false);
1770                 if (error) {
1771                         xfs_warn(mp, "failed to write sb changes");
1772                         return error;
1773                 }
1774                 mp->m_update_sb = false;
1775         }
1776
1777         /*
1778          * Fill out the reserve pool if it is empty. Use the stashed value if
1779          * it is non-zero, otherwise go with the default.
1780          */
1781         xfs_restore_resvblks(mp);
1782         xfs_log_work_queue(mp);
1783         xfs_blockgc_start(mp);
1784
1785         /* Create the per-AG metadata reservation pool .*/
1786         error = xfs_fs_reserve_ag_blocks(mp);
1787         if (error && error != -ENOSPC)
1788                 return error;
1789
1790         /* Re-enable the background inode inactivation worker. */
1791         xfs_inodegc_start(mp);
1792
1793         return 0;
1794 }
1795
1796 static int
1797 xfs_remount_ro(
1798         struct xfs_mount        *mp)
1799 {
1800         struct xfs_icwalk       icw = {
1801                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1802         };
1803         int                     error;
1804
1805         /* Flush all the dirty data to disk. */
1806         error = sync_filesystem(mp->m_super);
1807         if (error)
1808                 return error;
1809
1810         /*
1811          * Cancel background eofb scanning so it cannot race with the final
1812          * log force+buftarg wait and deadlock the remount.
1813          */
1814         xfs_blockgc_stop(mp);
1815
1816         /*
1817          * Clear out all remaining COW staging extents and speculative post-EOF
1818          * preallocations so that we don't leave inodes requiring inactivation
1819          * cleanups during reclaim on a read-only mount.  We must process every
1820          * cached inode, so this requires a synchronous cache scan.
1821          */
1822         error = xfs_blockgc_free_space(mp, &icw);
1823         if (error) {
1824                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1825                 return error;
1826         }
1827
1828         /*
1829          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1830          * flushed all pending inodegc work when it sync'd the filesystem.
1831          * The VFS holds s_umount, so we know that inodes cannot enter
1832          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1833          * we send inodes straight to reclaim, so no inodes will be queued.
1834          */
1835         xfs_inodegc_stop(mp);
1836
1837         /* Free the per-AG metadata reservation pool. */
1838         error = xfs_fs_unreserve_ag_blocks(mp);
1839         if (error) {
1840                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1841                 return error;
1842         }
1843
1844         /*
1845          * Before we sync the metadata, we need to free up the reserve block
1846          * pool so that the used block count in the superblock on disk is
1847          * correct at the end of the remount. Stash the current* reserve pool
1848          * size so that if we get remounted rw, we can return it to the same
1849          * size.
1850          */
1851         xfs_save_resvblks(mp);
1852
1853         xfs_log_clean(mp);
1854         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1855
1856         return 0;
1857 }
1858
1859 /*
1860  * Logically we would return an error here to prevent users from believing
1861  * they might have changed mount options using remount which can't be changed.
1862  *
1863  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1864  * arguments in some cases so we can't blindly reject options, but have to
1865  * check for each specified option if it actually differs from the currently
1866  * set option and only reject it if that's the case.
1867  *
1868  * Until that is implemented we return success for every remount request, and
1869  * silently ignore all options that we can't actually change.
1870  */
1871 static int
1872 xfs_fs_reconfigure(
1873         struct fs_context *fc)
1874 {
1875         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1876         struct xfs_mount        *new_mp = fc->s_fs_info;
1877         int                     flags = fc->sb_flags;
1878         int                     error;
1879
1880         /* version 5 superblocks always support version counters. */
1881         if (xfs_has_crc(mp))
1882                 fc->sb_flags |= SB_I_VERSION;
1883
1884         error = xfs_fs_validate_params(new_mp);
1885         if (error)
1886                 return error;
1887
1888         /* inode32 -> inode64 */
1889         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1890                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1891                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1892         }
1893
1894         /* inode64 -> inode32 */
1895         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1896                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1897                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1898         }
1899
1900         /* ro -> rw */
1901         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1902                 error = xfs_remount_rw(mp);
1903                 if (error)
1904                         return error;
1905         }
1906
1907         /* rw -> ro */
1908         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1909                 error = xfs_remount_ro(mp);
1910                 if (error)
1911                         return error;
1912         }
1913
1914         return 0;
1915 }
1916
1917 static void xfs_fs_free(
1918         struct fs_context       *fc)
1919 {
1920         struct xfs_mount        *mp = fc->s_fs_info;
1921
1922         /*
1923          * mp is stored in the fs_context when it is initialized.
1924          * mp is transferred to the superblock on a successful mount,
1925          * but if an error occurs before the transfer we have to free
1926          * it here.
1927          */
1928         if (mp)
1929                 xfs_mount_free(mp);
1930 }
1931
1932 static const struct fs_context_operations xfs_context_ops = {
1933         .parse_param = xfs_fs_parse_param,
1934         .get_tree    = xfs_fs_get_tree,
1935         .reconfigure = xfs_fs_reconfigure,
1936         .free        = xfs_fs_free,
1937 };
1938
1939 static int xfs_init_fs_context(
1940         struct fs_context       *fc)
1941 {
1942         struct xfs_mount        *mp;
1943
1944         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1945         if (!mp)
1946                 return -ENOMEM;
1947
1948         spin_lock_init(&mp->m_sb_lock);
1949         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1950         spin_lock_init(&mp->m_perag_lock);
1951         mutex_init(&mp->m_growlock);
1952         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1953         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1954         mp->m_kobj.kobject.kset = xfs_kset;
1955         /*
1956          * We don't create the finobt per-ag space reservation until after log
1957          * recovery, so we must set this to true so that an ifree transaction
1958          * started during log recovery will not depend on space reservations
1959          * for finobt expansion.
1960          */
1961         mp->m_finobt_nores = true;
1962
1963         /*
1964          * These can be overridden by the mount option parsing.
1965          */
1966         mp->m_logbufs = -1;
1967         mp->m_logbsize = -1;
1968         mp->m_allocsize_log = 16; /* 64k */
1969
1970         /*
1971          * Copy binary VFS mount flags we are interested in.
1972          */
1973         if (fc->sb_flags & SB_RDONLY)
1974                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1975         if (fc->sb_flags & SB_DIRSYNC)
1976                 mp->m_features |= XFS_FEAT_DIRSYNC;
1977         if (fc->sb_flags & SB_SYNCHRONOUS)
1978                 mp->m_features |= XFS_FEAT_WSYNC;
1979
1980         fc->s_fs_info = mp;
1981         fc->ops = &xfs_context_ops;
1982
1983         return 0;
1984 }
1985
1986 static struct file_system_type xfs_fs_type = {
1987         .owner                  = THIS_MODULE,
1988         .name                   = "xfs",
1989         .init_fs_context        = xfs_init_fs_context,
1990         .parameters             = xfs_fs_parameters,
1991         .kill_sb                = kill_block_super,
1992         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1993 };
1994 MODULE_ALIAS_FS("xfs");
1995
1996 STATIC int __init
1997 xfs_init_caches(void)
1998 {
1999         int             error;
2000
2001         xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2002                                          SLAB_HWCACHE_ALIGN |
2003                                          SLAB_RECLAIM_ACCOUNT |
2004                                          SLAB_MEM_SPREAD,
2005                                          NULL);
2006         if (!xfs_buf_cache)
2007                 goto out;
2008
2009         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
2010                                                 sizeof(struct xlog_ticket),
2011                                                 0, 0, NULL);
2012         if (!xfs_log_ticket_cache)
2013                 goto out_destroy_buf_cache;
2014
2015         error = xfs_btree_init_cur_caches();
2016         if (error)
2017                 goto out_destroy_log_ticket_cache;
2018
2019         error = xfs_defer_init_item_caches();
2020         if (error)
2021                 goto out_destroy_btree_cur_cache;
2022
2023         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2024                                               sizeof(struct xfs_da_state),
2025                                               0, 0, NULL);
2026         if (!xfs_da_state_cache)
2027                 goto out_destroy_defer_item_cache;
2028
2029         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2030                                            sizeof(struct xfs_ifork),
2031                                            0, 0, NULL);
2032         if (!xfs_ifork_cache)
2033                 goto out_destroy_da_state_cache;
2034
2035         xfs_trans_cache = kmem_cache_create("xfs_trans",
2036                                            sizeof(struct xfs_trans),
2037                                            0, 0, NULL);
2038         if (!xfs_trans_cache)
2039                 goto out_destroy_ifork_cache;
2040
2041
2042         /*
2043          * The size of the cache-allocated buf log item is the maximum
2044          * size possible under XFS.  This wastes a little bit of memory,
2045          * but it is much faster.
2046          */
2047         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2048                                               sizeof(struct xfs_buf_log_item),
2049                                               0, 0, NULL);
2050         if (!xfs_buf_item_cache)
2051                 goto out_destroy_trans_cache;
2052
2053         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2054                         xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2055                         0, 0, NULL);
2056         if (!xfs_efd_cache)
2057                 goto out_destroy_buf_item_cache;
2058
2059         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2060                         xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2061                         0, 0, NULL);
2062         if (!xfs_efi_cache)
2063                 goto out_destroy_efd_cache;
2064
2065         xfs_inode_cache = kmem_cache_create("xfs_inode",
2066                                            sizeof(struct xfs_inode), 0,
2067                                            (SLAB_HWCACHE_ALIGN |
2068                                             SLAB_RECLAIM_ACCOUNT |
2069                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2070                                            xfs_fs_inode_init_once);
2071         if (!xfs_inode_cache)
2072                 goto out_destroy_efi_cache;
2073
2074         xfs_ili_cache = kmem_cache_create("xfs_ili",
2075                                          sizeof(struct xfs_inode_log_item), 0,
2076                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2077                                          NULL);
2078         if (!xfs_ili_cache)
2079                 goto out_destroy_inode_cache;
2080
2081         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2082                                              sizeof(struct xfs_icreate_item),
2083                                              0, 0, NULL);
2084         if (!xfs_icreate_cache)
2085                 goto out_destroy_ili_cache;
2086
2087         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2088                                          sizeof(struct xfs_rud_log_item),
2089                                          0, 0, NULL);
2090         if (!xfs_rud_cache)
2091                 goto out_destroy_icreate_cache;
2092
2093         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2094                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2095                         0, 0, NULL);
2096         if (!xfs_rui_cache)
2097                 goto out_destroy_rud_cache;
2098
2099         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2100                                          sizeof(struct xfs_cud_log_item),
2101                                          0, 0, NULL);
2102         if (!xfs_cud_cache)
2103                 goto out_destroy_rui_cache;
2104
2105         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2106                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2107                         0, 0, NULL);
2108         if (!xfs_cui_cache)
2109                 goto out_destroy_cud_cache;
2110
2111         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2112                                          sizeof(struct xfs_bud_log_item),
2113                                          0, 0, NULL);
2114         if (!xfs_bud_cache)
2115                 goto out_destroy_cui_cache;
2116
2117         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2118                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2119                         0, 0, NULL);
2120         if (!xfs_bui_cache)
2121                 goto out_destroy_bud_cache;
2122
2123         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2124                                             sizeof(struct xfs_attrd_log_item),
2125                                             0, 0, NULL);
2126         if (!xfs_attrd_cache)
2127                 goto out_destroy_bui_cache;
2128
2129         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2130                                             sizeof(struct xfs_attri_log_item),
2131                                             0, 0, NULL);
2132         if (!xfs_attri_cache)
2133                 goto out_destroy_attrd_cache;
2134
2135         xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2136                                              sizeof(struct xfs_iunlink_item),
2137                                              0, 0, NULL);
2138         if (!xfs_iunlink_cache)
2139                 goto out_destroy_attri_cache;
2140
2141         return 0;
2142
2143  out_destroy_attri_cache:
2144         kmem_cache_destroy(xfs_attri_cache);
2145  out_destroy_attrd_cache:
2146         kmem_cache_destroy(xfs_attrd_cache);
2147  out_destroy_bui_cache:
2148         kmem_cache_destroy(xfs_bui_cache);
2149  out_destroy_bud_cache:
2150         kmem_cache_destroy(xfs_bud_cache);
2151  out_destroy_cui_cache:
2152         kmem_cache_destroy(xfs_cui_cache);
2153  out_destroy_cud_cache:
2154         kmem_cache_destroy(xfs_cud_cache);
2155  out_destroy_rui_cache:
2156         kmem_cache_destroy(xfs_rui_cache);
2157  out_destroy_rud_cache:
2158         kmem_cache_destroy(xfs_rud_cache);
2159  out_destroy_icreate_cache:
2160         kmem_cache_destroy(xfs_icreate_cache);
2161  out_destroy_ili_cache:
2162         kmem_cache_destroy(xfs_ili_cache);
2163  out_destroy_inode_cache:
2164         kmem_cache_destroy(xfs_inode_cache);
2165  out_destroy_efi_cache:
2166         kmem_cache_destroy(xfs_efi_cache);
2167  out_destroy_efd_cache:
2168         kmem_cache_destroy(xfs_efd_cache);
2169  out_destroy_buf_item_cache:
2170         kmem_cache_destroy(xfs_buf_item_cache);
2171  out_destroy_trans_cache:
2172         kmem_cache_destroy(xfs_trans_cache);
2173  out_destroy_ifork_cache:
2174         kmem_cache_destroy(xfs_ifork_cache);
2175  out_destroy_da_state_cache:
2176         kmem_cache_destroy(xfs_da_state_cache);
2177  out_destroy_defer_item_cache:
2178         xfs_defer_destroy_item_caches();
2179  out_destroy_btree_cur_cache:
2180         xfs_btree_destroy_cur_caches();
2181  out_destroy_log_ticket_cache:
2182         kmem_cache_destroy(xfs_log_ticket_cache);
2183  out_destroy_buf_cache:
2184         kmem_cache_destroy(xfs_buf_cache);
2185  out:
2186         return -ENOMEM;
2187 }
2188
2189 STATIC void
2190 xfs_destroy_caches(void)
2191 {
2192         /*
2193          * Make sure all delayed rcu free are flushed before we
2194          * destroy caches.
2195          */
2196         rcu_barrier();
2197         kmem_cache_destroy(xfs_iunlink_cache);
2198         kmem_cache_destroy(xfs_attri_cache);
2199         kmem_cache_destroy(xfs_attrd_cache);
2200         kmem_cache_destroy(xfs_bui_cache);
2201         kmem_cache_destroy(xfs_bud_cache);
2202         kmem_cache_destroy(xfs_cui_cache);
2203         kmem_cache_destroy(xfs_cud_cache);
2204         kmem_cache_destroy(xfs_rui_cache);
2205         kmem_cache_destroy(xfs_rud_cache);
2206         kmem_cache_destroy(xfs_icreate_cache);
2207         kmem_cache_destroy(xfs_ili_cache);
2208         kmem_cache_destroy(xfs_inode_cache);
2209         kmem_cache_destroy(xfs_efi_cache);
2210         kmem_cache_destroy(xfs_efd_cache);
2211         kmem_cache_destroy(xfs_buf_item_cache);
2212         kmem_cache_destroy(xfs_trans_cache);
2213         kmem_cache_destroy(xfs_ifork_cache);
2214         kmem_cache_destroy(xfs_da_state_cache);
2215         xfs_defer_destroy_item_caches();
2216         xfs_btree_destroy_cur_caches();
2217         kmem_cache_destroy(xfs_log_ticket_cache);
2218         kmem_cache_destroy(xfs_buf_cache);
2219 }
2220
2221 STATIC int __init
2222 xfs_init_workqueues(void)
2223 {
2224         /*
2225          * The allocation workqueue can be used in memory reclaim situations
2226          * (writepage path), and parallelism is only limited by the number of
2227          * AGs in all the filesystems mounted. Hence use the default large
2228          * max_active value for this workqueue.
2229          */
2230         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2231                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2232         if (!xfs_alloc_wq)
2233                 return -ENOMEM;
2234
2235         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2236                         0);
2237         if (!xfs_discard_wq)
2238                 goto out_free_alloc_wq;
2239
2240         return 0;
2241 out_free_alloc_wq:
2242         destroy_workqueue(xfs_alloc_wq);
2243         return -ENOMEM;
2244 }
2245
2246 STATIC void
2247 xfs_destroy_workqueues(void)
2248 {
2249         destroy_workqueue(xfs_discard_wq);
2250         destroy_workqueue(xfs_alloc_wq);
2251 }
2252
2253 #ifdef CONFIG_HOTPLUG_CPU
2254 static int
2255 xfs_cpu_dead(
2256         unsigned int            cpu)
2257 {
2258         struct xfs_mount        *mp, *n;
2259
2260         spin_lock(&xfs_mount_list_lock);
2261         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2262                 spin_unlock(&xfs_mount_list_lock);
2263                 xfs_inodegc_cpu_dead(mp, cpu);
2264                 xlog_cil_pcp_dead(mp->m_log, cpu);
2265                 spin_lock(&xfs_mount_list_lock);
2266         }
2267         spin_unlock(&xfs_mount_list_lock);
2268         return 0;
2269 }
2270
2271 static int __init
2272 xfs_cpu_hotplug_init(void)
2273 {
2274         int     error;
2275
2276         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2277                         xfs_cpu_dead);
2278         if (error < 0)
2279                 xfs_alert(NULL,
2280 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2281                         error);
2282         return error;
2283 }
2284
2285 static void
2286 xfs_cpu_hotplug_destroy(void)
2287 {
2288         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2289 }
2290
2291 #else /* !CONFIG_HOTPLUG_CPU */
2292 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2293 static inline void xfs_cpu_hotplug_destroy(void) {}
2294 #endif
2295
2296 STATIC int __init
2297 init_xfs_fs(void)
2298 {
2299         int                     error;
2300
2301         xfs_check_ondisk_structs();
2302
2303         error = xfs_dahash_test();
2304         if (error)
2305                 return error;
2306
2307         printk(KERN_INFO XFS_VERSION_STRING " with "
2308                          XFS_BUILD_OPTIONS " enabled\n");
2309
2310         xfs_dir_startup();
2311
2312         error = xfs_cpu_hotplug_init();
2313         if (error)
2314                 goto out;
2315
2316         error = xfs_init_caches();
2317         if (error)
2318                 goto out_destroy_hp;
2319
2320         error = xfs_init_workqueues();
2321         if (error)
2322                 goto out_destroy_caches;
2323
2324         error = xfs_mru_cache_init();
2325         if (error)
2326                 goto out_destroy_wq;
2327
2328         error = xfs_init_procfs();
2329         if (error)
2330                 goto out_mru_cache_uninit;
2331
2332         error = xfs_sysctl_register();
2333         if (error)
2334                 goto out_cleanup_procfs;
2335
2336         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2337         if (!xfs_kset) {
2338                 error = -ENOMEM;
2339                 goto out_sysctl_unregister;
2340         }
2341
2342         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2343
2344         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2345         if (!xfsstats.xs_stats) {
2346                 error = -ENOMEM;
2347                 goto out_kset_unregister;
2348         }
2349
2350         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2351                                "stats");
2352         if (error)
2353                 goto out_free_stats;
2354
2355 #ifdef DEBUG
2356         xfs_dbg_kobj.kobject.kset = xfs_kset;
2357         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2358         if (error)
2359                 goto out_remove_stats_kobj;
2360 #endif
2361
2362         error = xfs_qm_init();
2363         if (error)
2364                 goto out_remove_dbg_kobj;
2365
2366         error = register_filesystem(&xfs_fs_type);
2367         if (error)
2368                 goto out_qm_exit;
2369         return 0;
2370
2371  out_qm_exit:
2372         xfs_qm_exit();
2373  out_remove_dbg_kobj:
2374 #ifdef DEBUG
2375         xfs_sysfs_del(&xfs_dbg_kobj);
2376  out_remove_stats_kobj:
2377 #endif
2378         xfs_sysfs_del(&xfsstats.xs_kobj);
2379  out_free_stats:
2380         free_percpu(xfsstats.xs_stats);
2381  out_kset_unregister:
2382         kset_unregister(xfs_kset);
2383  out_sysctl_unregister:
2384         xfs_sysctl_unregister();
2385  out_cleanup_procfs:
2386         xfs_cleanup_procfs();
2387  out_mru_cache_uninit:
2388         xfs_mru_cache_uninit();
2389  out_destroy_wq:
2390         xfs_destroy_workqueues();
2391  out_destroy_caches:
2392         xfs_destroy_caches();
2393  out_destroy_hp:
2394         xfs_cpu_hotplug_destroy();
2395  out:
2396         return error;
2397 }
2398
2399 STATIC void __exit
2400 exit_xfs_fs(void)
2401 {
2402         xfs_qm_exit();
2403         unregister_filesystem(&xfs_fs_type);
2404 #ifdef DEBUG
2405         xfs_sysfs_del(&xfs_dbg_kobj);
2406 #endif
2407         xfs_sysfs_del(&xfsstats.xs_kobj);
2408         free_percpu(xfsstats.xs_stats);
2409         kset_unregister(xfs_kset);
2410         xfs_sysctl_unregister();
2411         xfs_cleanup_procfs();
2412         xfs_mru_cache_uninit();
2413         xfs_destroy_workqueues();
2414         xfs_destroy_caches();
2415         xfs_uuid_table_free();
2416         xfs_cpu_hotplug_destroy();
2417 }
2418
2419 module_init(init_xfs_fs);
2420 module_exit(exit_xfs_fs);
2421
2422 MODULE_AUTHOR("Silicon Graphics, Inc.");
2423 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2424 MODULE_LICENSE("GPL");