GNU Linux-libre 5.19-rc6-gnu
[releases.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
44 #include <linux/magic.h>
45 #include <linux/fs_context.h>
46 #include <linux/fs_parser.h>
47
48 static const struct super_operations xfs_super_operations;
49
50 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
51 #ifdef DEBUG
52 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
53 #endif
54
55 #ifdef CONFIG_HOTPLUG_CPU
56 static LIST_HEAD(xfs_mount_list);
57 static DEFINE_SPINLOCK(xfs_mount_list_lock);
58
59 static inline void xfs_mount_list_add(struct xfs_mount *mp)
60 {
61         spin_lock(&xfs_mount_list_lock);
62         list_add(&mp->m_mount_list, &xfs_mount_list);
63         spin_unlock(&xfs_mount_list_lock);
64 }
65
66 static inline void xfs_mount_list_del(struct xfs_mount *mp)
67 {
68         spin_lock(&xfs_mount_list_lock);
69         list_del(&mp->m_mount_list);
70         spin_unlock(&xfs_mount_list_lock);
71 }
72 #else /* !CONFIG_HOTPLUG_CPU */
73 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
74 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
75 #endif
76
77 enum xfs_dax_mode {
78         XFS_DAX_INODE = 0,
79         XFS_DAX_ALWAYS = 1,
80         XFS_DAX_NEVER = 2,
81 };
82
83 static void
84 xfs_mount_set_dax_mode(
85         struct xfs_mount        *mp,
86         enum xfs_dax_mode       mode)
87 {
88         switch (mode) {
89         case XFS_DAX_INODE:
90                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
91                 break;
92         case XFS_DAX_ALWAYS:
93                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
94                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
95                 break;
96         case XFS_DAX_NEVER:
97                 mp->m_features |= XFS_FEAT_DAX_NEVER;
98                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
99                 break;
100         }
101 }
102
103 static const struct constant_table dax_param_enums[] = {
104         {"inode",       XFS_DAX_INODE },
105         {"always",      XFS_DAX_ALWAYS },
106         {"never",       XFS_DAX_NEVER },
107         {}
108 };
109
110 /*
111  * Table driven mount option parser.
112  */
113 enum {
114         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
115         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
116         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
117         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
118         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
119         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
120         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
121         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
122         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
123 };
124
125 static const struct fs_parameter_spec xfs_fs_parameters[] = {
126         fsparam_u32("logbufs",          Opt_logbufs),
127         fsparam_string("logbsize",      Opt_logbsize),
128         fsparam_string("logdev",        Opt_logdev),
129         fsparam_string("rtdev",         Opt_rtdev),
130         fsparam_flag("wsync",           Opt_wsync),
131         fsparam_flag("noalign",         Opt_noalign),
132         fsparam_flag("swalloc",         Opt_swalloc),
133         fsparam_u32("sunit",            Opt_sunit),
134         fsparam_u32("swidth",           Opt_swidth),
135         fsparam_flag("nouuid",          Opt_nouuid),
136         fsparam_flag("grpid",           Opt_grpid),
137         fsparam_flag("nogrpid",         Opt_nogrpid),
138         fsparam_flag("bsdgroups",       Opt_bsdgroups),
139         fsparam_flag("sysvgroups",      Opt_sysvgroups),
140         fsparam_string("allocsize",     Opt_allocsize),
141         fsparam_flag("norecovery",      Opt_norecovery),
142         fsparam_flag("inode64",         Opt_inode64),
143         fsparam_flag("inode32",         Opt_inode32),
144         fsparam_flag("ikeep",           Opt_ikeep),
145         fsparam_flag("noikeep",         Opt_noikeep),
146         fsparam_flag("largeio",         Opt_largeio),
147         fsparam_flag("nolargeio",       Opt_nolargeio),
148         fsparam_flag("attr2",           Opt_attr2),
149         fsparam_flag("noattr2",         Opt_noattr2),
150         fsparam_flag("filestreams",     Opt_filestreams),
151         fsparam_flag("quota",           Opt_quota),
152         fsparam_flag("noquota",         Opt_noquota),
153         fsparam_flag("usrquota",        Opt_usrquota),
154         fsparam_flag("grpquota",        Opt_grpquota),
155         fsparam_flag("prjquota",        Opt_prjquota),
156         fsparam_flag("uquota",          Opt_uquota),
157         fsparam_flag("gquota",          Opt_gquota),
158         fsparam_flag("pquota",          Opt_pquota),
159         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
160         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
161         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
162         fsparam_flag("qnoenforce",      Opt_qnoenforce),
163         fsparam_flag("discard",         Opt_discard),
164         fsparam_flag("nodiscard",       Opt_nodiscard),
165         fsparam_flag("dax",             Opt_dax),
166         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
167         {}
168 };
169
170 struct proc_xfs_info {
171         uint64_t        flag;
172         char            *str;
173 };
174
175 static int
176 xfs_fs_show_options(
177         struct seq_file         *m,
178         struct dentry           *root)
179 {
180         static struct proc_xfs_info xfs_info_set[] = {
181                 /* the few simple ones we can get from the mount struct */
182                 { XFS_FEAT_IKEEP,               ",ikeep" },
183                 { XFS_FEAT_WSYNC,               ",wsync" },
184                 { XFS_FEAT_NOALIGN,             ",noalign" },
185                 { XFS_FEAT_SWALLOC,             ",swalloc" },
186                 { XFS_FEAT_NOUUID,              ",nouuid" },
187                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
188                 { XFS_FEAT_ATTR2,               ",attr2" },
189                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
190                 { XFS_FEAT_GRPID,               ",grpid" },
191                 { XFS_FEAT_DISCARD,             ",discard" },
192                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
193                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
194                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
195                 { 0, NULL }
196         };
197         struct xfs_mount        *mp = XFS_M(root->d_sb);
198         struct proc_xfs_info    *xfs_infop;
199
200         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
201                 if (mp->m_features & xfs_infop->flag)
202                         seq_puts(m, xfs_infop->str);
203         }
204
205         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
206
207         if (xfs_has_allocsize(mp))
208                 seq_printf(m, ",allocsize=%dk",
209                            (1 << mp->m_allocsize_log) >> 10);
210
211         if (mp->m_logbufs > 0)
212                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
213         if (mp->m_logbsize > 0)
214                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
215
216         if (mp->m_logname)
217                 seq_show_option(m, "logdev", mp->m_logname);
218         if (mp->m_rtname)
219                 seq_show_option(m, "rtdev", mp->m_rtname);
220
221         if (mp->m_dalign > 0)
222                 seq_printf(m, ",sunit=%d",
223                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
224         if (mp->m_swidth > 0)
225                 seq_printf(m, ",swidth=%d",
226                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
227
228         if (mp->m_qflags & XFS_UQUOTA_ENFD)
229                 seq_puts(m, ",usrquota");
230         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
231                 seq_puts(m, ",uqnoenforce");
232
233         if (mp->m_qflags & XFS_PQUOTA_ENFD)
234                 seq_puts(m, ",prjquota");
235         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
236                 seq_puts(m, ",pqnoenforce");
237
238         if (mp->m_qflags & XFS_GQUOTA_ENFD)
239                 seq_puts(m, ",grpquota");
240         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
241                 seq_puts(m, ",gqnoenforce");
242
243         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
244                 seq_puts(m, ",noquota");
245
246         return 0;
247 }
248
249 /*
250  * Set parameters for inode allocation heuristics, taking into account
251  * filesystem size and inode32/inode64 mount options; i.e. specifically
252  * whether or not XFS_FEAT_SMALL_INUMS is set.
253  *
254  * Inode allocation patterns are altered only if inode32 is requested
255  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
256  * If altered, XFS_OPSTATE_INODE32 is set as well.
257  *
258  * An agcount independent of that in the mount structure is provided
259  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
260  * to the potentially higher ag count.
261  *
262  * Returns the maximum AG index which may contain inodes.
263  */
264 xfs_agnumber_t
265 xfs_set_inode_alloc(
266         struct xfs_mount *mp,
267         xfs_agnumber_t  agcount)
268 {
269         xfs_agnumber_t  index;
270         xfs_agnumber_t  maxagi = 0;
271         xfs_sb_t        *sbp = &mp->m_sb;
272         xfs_agnumber_t  max_metadata;
273         xfs_agino_t     agino;
274         xfs_ino_t       ino;
275
276         /*
277          * Calculate how much should be reserved for inodes to meet
278          * the max inode percentage.  Used only for inode32.
279          */
280         if (M_IGEO(mp)->maxicount) {
281                 uint64_t        icount;
282
283                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
284                 do_div(icount, 100);
285                 icount += sbp->sb_agblocks - 1;
286                 do_div(icount, sbp->sb_agblocks);
287                 max_metadata = icount;
288         } else {
289                 max_metadata = agcount;
290         }
291
292         /* Get the last possible inode in the filesystem */
293         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
294         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
295
296         /*
297          * If user asked for no more than 32-bit inodes, and the fs is
298          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
299          * the allocator to accommodate the request.
300          */
301         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
302                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
303         else
304                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
305
306         for (index = 0; index < agcount; index++) {
307                 struct xfs_perag        *pag;
308
309                 ino = XFS_AGINO_TO_INO(mp, index, agino);
310
311                 pag = xfs_perag_get(mp, index);
312
313                 if (xfs_is_inode32(mp)) {
314                         if (ino > XFS_MAXINUMBER_32) {
315                                 pag->pagi_inodeok = 0;
316                                 pag->pagf_metadata = 0;
317                         } else {
318                                 pag->pagi_inodeok = 1;
319                                 maxagi++;
320                                 if (index < max_metadata)
321                                         pag->pagf_metadata = 1;
322                                 else
323                                         pag->pagf_metadata = 0;
324                         }
325                 } else {
326                         pag->pagi_inodeok = 1;
327                         pag->pagf_metadata = 0;
328                 }
329
330                 xfs_perag_put(pag);
331         }
332
333         return xfs_is_inode32(mp) ? maxagi : agcount;
334 }
335
336 static int
337 xfs_setup_dax_always(
338         struct xfs_mount        *mp)
339 {
340         if (!mp->m_ddev_targp->bt_daxdev &&
341             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
342                 xfs_alert(mp,
343                         "DAX unsupported by block device. Turning off DAX.");
344                 goto disable_dax;
345         }
346
347         if (mp->m_super->s_blocksize != PAGE_SIZE) {
348                 xfs_alert(mp,
349                         "DAX not supported for blocksize. Turning off DAX.");
350                 goto disable_dax;
351         }
352
353         if (xfs_has_reflink(mp)) {
354                 xfs_alert(mp, "DAX and reflink cannot be used together!");
355                 return -EINVAL;
356         }
357
358         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
359         return 0;
360
361 disable_dax:
362         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
363         return 0;
364 }
365
366 STATIC int
367 xfs_blkdev_get(
368         xfs_mount_t             *mp,
369         const char              *name,
370         struct block_device     **bdevp)
371 {
372         int                     error = 0;
373
374         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
375                                     mp);
376         if (IS_ERR(*bdevp)) {
377                 error = PTR_ERR(*bdevp);
378                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
379         }
380
381         return error;
382 }
383
384 STATIC void
385 xfs_blkdev_put(
386         struct block_device     *bdev)
387 {
388         if (bdev)
389                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
390 }
391
392 STATIC void
393 xfs_close_devices(
394         struct xfs_mount        *mp)
395 {
396         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
397                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
398
399                 xfs_free_buftarg(mp->m_logdev_targp);
400                 xfs_blkdev_put(logdev);
401         }
402         if (mp->m_rtdev_targp) {
403                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
404
405                 xfs_free_buftarg(mp->m_rtdev_targp);
406                 xfs_blkdev_put(rtdev);
407         }
408         xfs_free_buftarg(mp->m_ddev_targp);
409 }
410
411 /*
412  * The file system configurations are:
413  *      (1) device (partition) with data and internal log
414  *      (2) logical volume with data and log subvolumes.
415  *      (3) logical volume with data, log, and realtime subvolumes.
416  *
417  * We only have to handle opening the log and realtime volumes here if
418  * they are present.  The data subvolume has already been opened by
419  * get_sb_bdev() and is stored in sb->s_bdev.
420  */
421 STATIC int
422 xfs_open_devices(
423         struct xfs_mount        *mp)
424 {
425         struct block_device     *ddev = mp->m_super->s_bdev;
426         struct block_device     *logdev = NULL, *rtdev = NULL;
427         int                     error;
428
429         /*
430          * Open real time and log devices - order is important.
431          */
432         if (mp->m_logname) {
433                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
434                 if (error)
435                         return error;
436         }
437
438         if (mp->m_rtname) {
439                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
440                 if (error)
441                         goto out_close_logdev;
442
443                 if (rtdev == ddev || rtdev == logdev) {
444                         xfs_warn(mp,
445         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
446                         error = -EINVAL;
447                         goto out_close_rtdev;
448                 }
449         }
450
451         /*
452          * Setup xfs_mount buffer target pointers
453          */
454         error = -ENOMEM;
455         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
456         if (!mp->m_ddev_targp)
457                 goto out_close_rtdev;
458
459         if (rtdev) {
460                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
461                 if (!mp->m_rtdev_targp)
462                         goto out_free_ddev_targ;
463         }
464
465         if (logdev && logdev != ddev) {
466                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
467                 if (!mp->m_logdev_targp)
468                         goto out_free_rtdev_targ;
469         } else {
470                 mp->m_logdev_targp = mp->m_ddev_targp;
471         }
472
473         return 0;
474
475  out_free_rtdev_targ:
476         if (mp->m_rtdev_targp)
477                 xfs_free_buftarg(mp->m_rtdev_targp);
478  out_free_ddev_targ:
479         xfs_free_buftarg(mp->m_ddev_targp);
480  out_close_rtdev:
481         xfs_blkdev_put(rtdev);
482  out_close_logdev:
483         if (logdev && logdev != ddev)
484                 xfs_blkdev_put(logdev);
485         return error;
486 }
487
488 /*
489  * Setup xfs_mount buffer target pointers based on superblock
490  */
491 STATIC int
492 xfs_setup_devices(
493         struct xfs_mount        *mp)
494 {
495         int                     error;
496
497         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
498         if (error)
499                 return error;
500
501         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
502                 unsigned int    log_sector_size = BBSIZE;
503
504                 if (xfs_has_sector(mp))
505                         log_sector_size = mp->m_sb.sb_logsectsize;
506                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
507                                             log_sector_size);
508                 if (error)
509                         return error;
510         }
511         if (mp->m_rtdev_targp) {
512                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
513                                             mp->m_sb.sb_sectsize);
514                 if (error)
515                         return error;
516         }
517
518         return 0;
519 }
520
521 STATIC int
522 xfs_init_mount_workqueues(
523         struct xfs_mount        *mp)
524 {
525         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
526                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
527                         1, mp->m_super->s_id);
528         if (!mp->m_buf_workqueue)
529                 goto out;
530
531         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
532                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
533                         0, mp->m_super->s_id);
534         if (!mp->m_unwritten_workqueue)
535                 goto out_destroy_buf;
536
537         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
538                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
539                         0, mp->m_super->s_id);
540         if (!mp->m_reclaim_workqueue)
541                 goto out_destroy_unwritten;
542
543         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
544                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
545                         0, mp->m_super->s_id);
546         if (!mp->m_blockgc_wq)
547                 goto out_destroy_reclaim;
548
549         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
550                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
551                         1, mp->m_super->s_id);
552         if (!mp->m_inodegc_wq)
553                 goto out_destroy_blockgc;
554
555         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
556                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
557         if (!mp->m_sync_workqueue)
558                 goto out_destroy_inodegc;
559
560         return 0;
561
562 out_destroy_inodegc:
563         destroy_workqueue(mp->m_inodegc_wq);
564 out_destroy_blockgc:
565         destroy_workqueue(mp->m_blockgc_wq);
566 out_destroy_reclaim:
567         destroy_workqueue(mp->m_reclaim_workqueue);
568 out_destroy_unwritten:
569         destroy_workqueue(mp->m_unwritten_workqueue);
570 out_destroy_buf:
571         destroy_workqueue(mp->m_buf_workqueue);
572 out:
573         return -ENOMEM;
574 }
575
576 STATIC void
577 xfs_destroy_mount_workqueues(
578         struct xfs_mount        *mp)
579 {
580         destroy_workqueue(mp->m_sync_workqueue);
581         destroy_workqueue(mp->m_blockgc_wq);
582         destroy_workqueue(mp->m_inodegc_wq);
583         destroy_workqueue(mp->m_reclaim_workqueue);
584         destroy_workqueue(mp->m_unwritten_workqueue);
585         destroy_workqueue(mp->m_buf_workqueue);
586 }
587
588 static void
589 xfs_flush_inodes_worker(
590         struct work_struct      *work)
591 {
592         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
593                                                    m_flush_inodes_work);
594         struct super_block      *sb = mp->m_super;
595
596         if (down_read_trylock(&sb->s_umount)) {
597                 sync_inodes_sb(sb);
598                 up_read(&sb->s_umount);
599         }
600 }
601
602 /*
603  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
604  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
605  * for IO to complete so that we effectively throttle multiple callers to the
606  * rate at which IO is completing.
607  */
608 void
609 xfs_flush_inodes(
610         struct xfs_mount        *mp)
611 {
612         /*
613          * If flush_work() returns true then that means we waited for a flush
614          * which was already in progress.  Don't bother running another scan.
615          */
616         if (flush_work(&mp->m_flush_inodes_work))
617                 return;
618
619         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
620         flush_work(&mp->m_flush_inodes_work);
621 }
622
623 /* Catch misguided souls that try to use this interface on XFS */
624 STATIC struct inode *
625 xfs_fs_alloc_inode(
626         struct super_block      *sb)
627 {
628         BUG();
629         return NULL;
630 }
631
632 /*
633  * Now that the generic code is guaranteed not to be accessing
634  * the linux inode, we can inactivate and reclaim the inode.
635  */
636 STATIC void
637 xfs_fs_destroy_inode(
638         struct inode            *inode)
639 {
640         struct xfs_inode        *ip = XFS_I(inode);
641
642         trace_xfs_destroy_inode(ip);
643
644         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
645         XFS_STATS_INC(ip->i_mount, vn_rele);
646         XFS_STATS_INC(ip->i_mount, vn_remove);
647         xfs_inode_mark_reclaimable(ip);
648 }
649
650 static void
651 xfs_fs_dirty_inode(
652         struct inode                    *inode,
653         int                             flag)
654 {
655         struct xfs_inode                *ip = XFS_I(inode);
656         struct xfs_mount                *mp = ip->i_mount;
657         struct xfs_trans                *tp;
658
659         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
660                 return;
661         if (flag != I_DIRTY_SYNC || !(inode->i_state & I_DIRTY_TIME))
662                 return;
663
664         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
665                 return;
666         xfs_ilock(ip, XFS_ILOCK_EXCL);
667         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
668         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
669         xfs_trans_commit(tp);
670 }
671
672 /*
673  * Slab object creation initialisation for the XFS inode.
674  * This covers only the idempotent fields in the XFS inode;
675  * all other fields need to be initialised on allocation
676  * from the slab. This avoids the need to repeatedly initialise
677  * fields in the xfs inode that left in the initialise state
678  * when freeing the inode.
679  */
680 STATIC void
681 xfs_fs_inode_init_once(
682         void                    *inode)
683 {
684         struct xfs_inode        *ip = inode;
685
686         memset(ip, 0, sizeof(struct xfs_inode));
687
688         /* vfs inode */
689         inode_init_once(VFS_I(ip));
690
691         /* xfs inode */
692         atomic_set(&ip->i_pincount, 0);
693         spin_lock_init(&ip->i_flags_lock);
694
695         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
696                      "xfsino", ip->i_ino);
697 }
698
699 /*
700  * We do an unlocked check for XFS_IDONTCACHE here because we are already
701  * serialised against cache hits here via the inode->i_lock and igrab() in
702  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
703  * racing with us, and it avoids needing to grab a spinlock here for every inode
704  * we drop the final reference on.
705  */
706 STATIC int
707 xfs_fs_drop_inode(
708         struct inode            *inode)
709 {
710         struct xfs_inode        *ip = XFS_I(inode);
711
712         /*
713          * If this unlinked inode is in the middle of recovery, don't
714          * drop the inode just yet; log recovery will take care of
715          * that.  See the comment for this inode flag.
716          */
717         if (ip->i_flags & XFS_IRECOVERY) {
718                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
719                 return 0;
720         }
721
722         return generic_drop_inode(inode);
723 }
724
725 static void
726 xfs_mount_free(
727         struct xfs_mount        *mp)
728 {
729         kfree(mp->m_rtname);
730         kfree(mp->m_logname);
731         kmem_free(mp);
732 }
733
734 STATIC int
735 xfs_fs_sync_fs(
736         struct super_block      *sb,
737         int                     wait)
738 {
739         struct xfs_mount        *mp = XFS_M(sb);
740         int                     error;
741
742         trace_xfs_fs_sync_fs(mp, __return_address);
743
744         /*
745          * Doing anything during the async pass would be counterproductive.
746          */
747         if (!wait)
748                 return 0;
749
750         error = xfs_log_force(mp, XFS_LOG_SYNC);
751         if (error)
752                 return error;
753
754         if (laptop_mode) {
755                 /*
756                  * The disk must be active because we're syncing.
757                  * We schedule log work now (now that the disk is
758                  * active) instead of later (when it might not be).
759                  */
760                 flush_delayed_work(&mp->m_log->l_work);
761         }
762
763         /*
764          * If we are called with page faults frozen out, it means we are about
765          * to freeze the transaction subsystem. Take the opportunity to shut
766          * down inodegc because once SB_FREEZE_FS is set it's too late to
767          * prevent inactivation races with freeze. The fs doesn't get called
768          * again by the freezing process until after SB_FREEZE_FS has been set,
769          * so it's now or never.  Same logic applies to speculative allocation
770          * garbage collection.
771          *
772          * We don't care if this is a normal syncfs call that does this or
773          * freeze that does this - we can run this multiple times without issue
774          * and we won't race with a restart because a restart can only occur
775          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
776          */
777         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
778                 xfs_inodegc_stop(mp);
779                 xfs_blockgc_stop(mp);
780         }
781
782         return 0;
783 }
784
785 STATIC int
786 xfs_fs_statfs(
787         struct dentry           *dentry,
788         struct kstatfs          *statp)
789 {
790         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
791         xfs_sb_t                *sbp = &mp->m_sb;
792         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
793         uint64_t                fakeinos, id;
794         uint64_t                icount;
795         uint64_t                ifree;
796         uint64_t                fdblocks;
797         xfs_extlen_t            lsize;
798         int64_t                 ffree;
799
800         /*
801          * Expedite background inodegc but don't wait. We do not want to block
802          * here waiting hours for a billion extent file to be truncated.
803          */
804         xfs_inodegc_push(mp);
805
806         statp->f_type = XFS_SUPER_MAGIC;
807         statp->f_namelen = MAXNAMELEN - 1;
808
809         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
810         statp->f_fsid = u64_to_fsid(id);
811
812         icount = percpu_counter_sum(&mp->m_icount);
813         ifree = percpu_counter_sum(&mp->m_ifree);
814         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
815
816         spin_lock(&mp->m_sb_lock);
817         statp->f_bsize = sbp->sb_blocksize;
818         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
819         statp->f_blocks = sbp->sb_dblocks - lsize;
820         spin_unlock(&mp->m_sb_lock);
821
822         /* make sure statp->f_bfree does not underflow */
823         statp->f_bfree = max_t(int64_t, 0,
824                                 fdblocks - xfs_fdblocks_unavailable(mp));
825         statp->f_bavail = statp->f_bfree;
826
827         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
828         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
829         if (M_IGEO(mp)->maxicount)
830                 statp->f_files = min_t(typeof(statp->f_files),
831                                         statp->f_files,
832                                         M_IGEO(mp)->maxicount);
833
834         /* If sb_icount overshot maxicount, report actual allocation */
835         statp->f_files = max_t(typeof(statp->f_files),
836                                         statp->f_files,
837                                         sbp->sb_icount);
838
839         /* make sure statp->f_ffree does not underflow */
840         ffree = statp->f_files - (icount - ifree);
841         statp->f_ffree = max_t(int64_t, ffree, 0);
842
843
844         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
845             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
846                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
847                 xfs_qm_statvfs(ip, statp);
848
849         if (XFS_IS_REALTIME_MOUNT(mp) &&
850             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
851                 s64     freertx;
852
853                 statp->f_blocks = sbp->sb_rblocks;
854                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
855                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
856         }
857
858         return 0;
859 }
860
861 STATIC void
862 xfs_save_resvblks(struct xfs_mount *mp)
863 {
864         uint64_t resblks = 0;
865
866         mp->m_resblks_save = mp->m_resblks;
867         xfs_reserve_blocks(mp, &resblks, NULL);
868 }
869
870 STATIC void
871 xfs_restore_resvblks(struct xfs_mount *mp)
872 {
873         uint64_t resblks;
874
875         if (mp->m_resblks_save) {
876                 resblks = mp->m_resblks_save;
877                 mp->m_resblks_save = 0;
878         } else
879                 resblks = xfs_default_resblks(mp);
880
881         xfs_reserve_blocks(mp, &resblks, NULL);
882 }
883
884 /*
885  * Second stage of a freeze. The data is already frozen so we only
886  * need to take care of the metadata. Once that's done sync the superblock
887  * to the log to dirty it in case of a crash while frozen. This ensures that we
888  * will recover the unlinked inode lists on the next mount.
889  */
890 STATIC int
891 xfs_fs_freeze(
892         struct super_block      *sb)
893 {
894         struct xfs_mount        *mp = XFS_M(sb);
895         unsigned int            flags;
896         int                     ret;
897
898         /*
899          * The filesystem is now frozen far enough that memory reclaim
900          * cannot safely operate on the filesystem. Hence we need to
901          * set a GFP_NOFS context here to avoid recursion deadlocks.
902          */
903         flags = memalloc_nofs_save();
904         xfs_save_resvblks(mp);
905         ret = xfs_log_quiesce(mp);
906         memalloc_nofs_restore(flags);
907
908         /*
909          * For read-write filesystems, we need to restart the inodegc on error
910          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
911          * going to be run to restart it now.  We are at SB_FREEZE_FS level
912          * here, so we can restart safely without racing with a stop in
913          * xfs_fs_sync_fs().
914          */
915         if (ret && !xfs_is_readonly(mp)) {
916                 xfs_blockgc_start(mp);
917                 xfs_inodegc_start(mp);
918         }
919
920         return ret;
921 }
922
923 STATIC int
924 xfs_fs_unfreeze(
925         struct super_block      *sb)
926 {
927         struct xfs_mount        *mp = XFS_M(sb);
928
929         xfs_restore_resvblks(mp);
930         xfs_log_work_queue(mp);
931
932         /*
933          * Don't reactivate the inodegc worker on a readonly filesystem because
934          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
935          * worker because there are no speculative preallocations on a readonly
936          * filesystem.
937          */
938         if (!xfs_is_readonly(mp)) {
939                 xfs_blockgc_start(mp);
940                 xfs_inodegc_start(mp);
941         }
942
943         return 0;
944 }
945
946 /*
947  * This function fills in xfs_mount_t fields based on mount args.
948  * Note: the superblock _has_ now been read in.
949  */
950 STATIC int
951 xfs_finish_flags(
952         struct xfs_mount        *mp)
953 {
954         /* Fail a mount where the logbuf is smaller than the log stripe */
955         if (xfs_has_logv2(mp)) {
956                 if (mp->m_logbsize <= 0 &&
957                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
958                         mp->m_logbsize = mp->m_sb.sb_logsunit;
959                 } else if (mp->m_logbsize > 0 &&
960                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
961                         xfs_warn(mp,
962                 "logbuf size must be greater than or equal to log stripe size");
963                         return -EINVAL;
964                 }
965         } else {
966                 /* Fail a mount if the logbuf is larger than 32K */
967                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
968                         xfs_warn(mp,
969                 "logbuf size for version 1 logs must be 16K or 32K");
970                         return -EINVAL;
971                 }
972         }
973
974         /*
975          * V5 filesystems always use attr2 format for attributes.
976          */
977         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
978                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
979                              "attr2 is always enabled for V5 filesystems.");
980                 return -EINVAL;
981         }
982
983         /*
984          * prohibit r/w mounts of read-only filesystems
985          */
986         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
987                 xfs_warn(mp,
988                         "cannot mount a read-only filesystem as read-write");
989                 return -EROFS;
990         }
991
992         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
993             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
994             !xfs_has_pquotino(mp)) {
995                 xfs_warn(mp,
996                   "Super block does not support project and group quota together");
997                 return -EINVAL;
998         }
999
1000         return 0;
1001 }
1002
1003 static int
1004 xfs_init_percpu_counters(
1005         struct xfs_mount        *mp)
1006 {
1007         int             error;
1008
1009         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1010         if (error)
1011                 return -ENOMEM;
1012
1013         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1014         if (error)
1015                 goto free_icount;
1016
1017         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1018         if (error)
1019                 goto free_ifree;
1020
1021         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1022         if (error)
1023                 goto free_fdblocks;
1024
1025         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1026         if (error)
1027                 goto free_delalloc;
1028
1029         return 0;
1030
1031 free_delalloc:
1032         percpu_counter_destroy(&mp->m_delalloc_blks);
1033 free_fdblocks:
1034         percpu_counter_destroy(&mp->m_fdblocks);
1035 free_ifree:
1036         percpu_counter_destroy(&mp->m_ifree);
1037 free_icount:
1038         percpu_counter_destroy(&mp->m_icount);
1039         return -ENOMEM;
1040 }
1041
1042 void
1043 xfs_reinit_percpu_counters(
1044         struct xfs_mount        *mp)
1045 {
1046         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1047         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1048         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1049         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1050 }
1051
1052 static void
1053 xfs_destroy_percpu_counters(
1054         struct xfs_mount        *mp)
1055 {
1056         percpu_counter_destroy(&mp->m_icount);
1057         percpu_counter_destroy(&mp->m_ifree);
1058         percpu_counter_destroy(&mp->m_fdblocks);
1059         ASSERT(xfs_is_shutdown(mp) ||
1060                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1061         percpu_counter_destroy(&mp->m_delalloc_blks);
1062         percpu_counter_destroy(&mp->m_frextents);
1063 }
1064
1065 static int
1066 xfs_inodegc_init_percpu(
1067         struct xfs_mount        *mp)
1068 {
1069         struct xfs_inodegc      *gc;
1070         int                     cpu;
1071
1072         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1073         if (!mp->m_inodegc)
1074                 return -ENOMEM;
1075
1076         for_each_possible_cpu(cpu) {
1077                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1078                 init_llist_head(&gc->list);
1079                 gc->items = 0;
1080                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1081         }
1082         return 0;
1083 }
1084
1085 static void
1086 xfs_inodegc_free_percpu(
1087         struct xfs_mount        *mp)
1088 {
1089         if (!mp->m_inodegc)
1090                 return;
1091         free_percpu(mp->m_inodegc);
1092 }
1093
1094 static void
1095 xfs_fs_put_super(
1096         struct super_block      *sb)
1097 {
1098         struct xfs_mount        *mp = XFS_M(sb);
1099
1100         /* if ->fill_super failed, we have no mount to tear down */
1101         if (!sb->s_fs_info)
1102                 return;
1103
1104         xfs_notice(mp, "Unmounting Filesystem");
1105         xfs_filestream_unmount(mp);
1106         xfs_unmountfs(mp);
1107
1108         xfs_freesb(mp);
1109         free_percpu(mp->m_stats.xs_stats);
1110         xfs_mount_list_del(mp);
1111         xfs_inodegc_free_percpu(mp);
1112         xfs_destroy_percpu_counters(mp);
1113         xfs_destroy_mount_workqueues(mp);
1114         xfs_close_devices(mp);
1115
1116         sb->s_fs_info = NULL;
1117         xfs_mount_free(mp);
1118 }
1119
1120 static long
1121 xfs_fs_nr_cached_objects(
1122         struct super_block      *sb,
1123         struct shrink_control   *sc)
1124 {
1125         /* Paranoia: catch incorrect calls during mount setup or teardown */
1126         if (WARN_ON_ONCE(!sb->s_fs_info))
1127                 return 0;
1128         return xfs_reclaim_inodes_count(XFS_M(sb));
1129 }
1130
1131 static long
1132 xfs_fs_free_cached_objects(
1133         struct super_block      *sb,
1134         struct shrink_control   *sc)
1135 {
1136         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1137 }
1138
1139 static const struct super_operations xfs_super_operations = {
1140         .alloc_inode            = xfs_fs_alloc_inode,
1141         .destroy_inode          = xfs_fs_destroy_inode,
1142         .dirty_inode            = xfs_fs_dirty_inode,
1143         .drop_inode             = xfs_fs_drop_inode,
1144         .put_super              = xfs_fs_put_super,
1145         .sync_fs                = xfs_fs_sync_fs,
1146         .freeze_fs              = xfs_fs_freeze,
1147         .unfreeze_fs            = xfs_fs_unfreeze,
1148         .statfs                 = xfs_fs_statfs,
1149         .show_options           = xfs_fs_show_options,
1150         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1151         .free_cached_objects    = xfs_fs_free_cached_objects,
1152 };
1153
1154 static int
1155 suffix_kstrtoint(
1156         const char      *s,
1157         unsigned int    base,
1158         int             *res)
1159 {
1160         int             last, shift_left_factor = 0, _res;
1161         char            *value;
1162         int             ret = 0;
1163
1164         value = kstrdup(s, GFP_KERNEL);
1165         if (!value)
1166                 return -ENOMEM;
1167
1168         last = strlen(value) - 1;
1169         if (value[last] == 'K' || value[last] == 'k') {
1170                 shift_left_factor = 10;
1171                 value[last] = '\0';
1172         }
1173         if (value[last] == 'M' || value[last] == 'm') {
1174                 shift_left_factor = 20;
1175                 value[last] = '\0';
1176         }
1177         if (value[last] == 'G' || value[last] == 'g') {
1178                 shift_left_factor = 30;
1179                 value[last] = '\0';
1180         }
1181
1182         if (kstrtoint(value, base, &_res))
1183                 ret = -EINVAL;
1184         kfree(value);
1185         *res = _res << shift_left_factor;
1186         return ret;
1187 }
1188
1189 static inline void
1190 xfs_fs_warn_deprecated(
1191         struct fs_context       *fc,
1192         struct fs_parameter     *param,
1193         uint64_t                flag,
1194         bool                    value)
1195 {
1196         /* Don't print the warning if reconfiguring and current mount point
1197          * already had the flag set
1198          */
1199         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1200             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1201                 return;
1202         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1203 }
1204
1205 /*
1206  * Set mount state from a mount option.
1207  *
1208  * NOTE: mp->m_super is NULL here!
1209  */
1210 static int
1211 xfs_fs_parse_param(
1212         struct fs_context       *fc,
1213         struct fs_parameter     *param)
1214 {
1215         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1216         struct fs_parse_result  result;
1217         int                     size = 0;
1218         int                     opt;
1219
1220         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1221         if (opt < 0)
1222                 return opt;
1223
1224         switch (opt) {
1225         case Opt_logbufs:
1226                 parsing_mp->m_logbufs = result.uint_32;
1227                 return 0;
1228         case Opt_logbsize:
1229                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1230                         return -EINVAL;
1231                 return 0;
1232         case Opt_logdev:
1233                 kfree(parsing_mp->m_logname);
1234                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1235                 if (!parsing_mp->m_logname)
1236                         return -ENOMEM;
1237                 return 0;
1238         case Opt_rtdev:
1239                 kfree(parsing_mp->m_rtname);
1240                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1241                 if (!parsing_mp->m_rtname)
1242                         return -ENOMEM;
1243                 return 0;
1244         case Opt_allocsize:
1245                 if (suffix_kstrtoint(param->string, 10, &size))
1246                         return -EINVAL;
1247                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1248                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1249                 return 0;
1250         case Opt_grpid:
1251         case Opt_bsdgroups:
1252                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1253                 return 0;
1254         case Opt_nogrpid:
1255         case Opt_sysvgroups:
1256                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1257                 return 0;
1258         case Opt_wsync:
1259                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1260                 return 0;
1261         case Opt_norecovery:
1262                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1263                 return 0;
1264         case Opt_noalign:
1265                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1266                 return 0;
1267         case Opt_swalloc:
1268                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1269                 return 0;
1270         case Opt_sunit:
1271                 parsing_mp->m_dalign = result.uint_32;
1272                 return 0;
1273         case Opt_swidth:
1274                 parsing_mp->m_swidth = result.uint_32;
1275                 return 0;
1276         case Opt_inode32:
1277                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1278                 return 0;
1279         case Opt_inode64:
1280                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1281                 return 0;
1282         case Opt_nouuid:
1283                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1284                 return 0;
1285         case Opt_largeio:
1286                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1287                 return 0;
1288         case Opt_nolargeio:
1289                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1290                 return 0;
1291         case Opt_filestreams:
1292                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1293                 return 0;
1294         case Opt_noquota:
1295                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1296                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1297                 return 0;
1298         case Opt_quota:
1299         case Opt_uquota:
1300         case Opt_usrquota:
1301                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1302                 return 0;
1303         case Opt_qnoenforce:
1304         case Opt_uqnoenforce:
1305                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1306                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1307                 return 0;
1308         case Opt_pquota:
1309         case Opt_prjquota:
1310                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1311                 return 0;
1312         case Opt_pqnoenforce:
1313                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1314                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1315                 return 0;
1316         case Opt_gquota:
1317         case Opt_grpquota:
1318                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1319                 return 0;
1320         case Opt_gqnoenforce:
1321                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1322                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1323                 return 0;
1324         case Opt_discard:
1325                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1326                 return 0;
1327         case Opt_nodiscard:
1328                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1329                 return 0;
1330 #ifdef CONFIG_FS_DAX
1331         case Opt_dax:
1332                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1333                 return 0;
1334         case Opt_dax_enum:
1335                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1336                 return 0;
1337 #endif
1338         /* Following mount options will be removed in September 2025 */
1339         case Opt_ikeep:
1340                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1341                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1342                 return 0;
1343         case Opt_noikeep:
1344                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1345                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1346                 return 0;
1347         case Opt_attr2:
1348                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1349                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1350                 return 0;
1351         case Opt_noattr2:
1352                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1353                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1354                 return 0;
1355         default:
1356                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1357                 return -EINVAL;
1358         }
1359
1360         return 0;
1361 }
1362
1363 static int
1364 xfs_fs_validate_params(
1365         struct xfs_mount        *mp)
1366 {
1367         /* No recovery flag requires a read-only mount */
1368         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1369                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1370                 return -EINVAL;
1371         }
1372
1373         /*
1374          * We have not read the superblock at this point, so only the attr2
1375          * mount option can set the attr2 feature by this stage.
1376          */
1377         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1378                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1379                 return -EINVAL;
1380         }
1381
1382
1383         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1384                 xfs_warn(mp,
1385         "sunit and swidth options incompatible with the noalign option");
1386                 return -EINVAL;
1387         }
1388
1389         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1390                 xfs_warn(mp, "quota support not available in this kernel.");
1391                 return -EINVAL;
1392         }
1393
1394         if ((mp->m_dalign && !mp->m_swidth) ||
1395             (!mp->m_dalign && mp->m_swidth)) {
1396                 xfs_warn(mp, "sunit and swidth must be specified together");
1397                 return -EINVAL;
1398         }
1399
1400         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1401                 xfs_warn(mp,
1402         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1403                         mp->m_swidth, mp->m_dalign);
1404                 return -EINVAL;
1405         }
1406
1407         if (mp->m_logbufs != -1 &&
1408             mp->m_logbufs != 0 &&
1409             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1410              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1411                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1412                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1413                 return -EINVAL;
1414         }
1415
1416         if (mp->m_logbsize != -1 &&
1417             mp->m_logbsize !=  0 &&
1418             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1419              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1420              !is_power_of_2(mp->m_logbsize))) {
1421                 xfs_warn(mp,
1422                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1423                         mp->m_logbsize);
1424                 return -EINVAL;
1425         }
1426
1427         if (xfs_has_allocsize(mp) &&
1428             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1429              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1430                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1431                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1432                 return -EINVAL;
1433         }
1434
1435         return 0;
1436 }
1437
1438 static int
1439 xfs_fs_fill_super(
1440         struct super_block      *sb,
1441         struct fs_context       *fc)
1442 {
1443         struct xfs_mount        *mp = sb->s_fs_info;
1444         struct inode            *root;
1445         int                     flags = 0, error;
1446
1447         mp->m_super = sb;
1448
1449         error = xfs_fs_validate_params(mp);
1450         if (error)
1451                 goto out_free_names;
1452
1453         sb_min_blocksize(sb, BBSIZE);
1454         sb->s_xattr = xfs_xattr_handlers;
1455         sb->s_export_op = &xfs_export_operations;
1456 #ifdef CONFIG_XFS_QUOTA
1457         sb->s_qcop = &xfs_quotactl_operations;
1458         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1459 #endif
1460         sb->s_op = &xfs_super_operations;
1461
1462         /*
1463          * Delay mount work if the debug hook is set. This is debug
1464          * instrumention to coordinate simulation of xfs mount failures with
1465          * VFS superblock operations
1466          */
1467         if (xfs_globals.mount_delay) {
1468                 xfs_notice(mp, "Delaying mount for %d seconds.",
1469                         xfs_globals.mount_delay);
1470                 msleep(xfs_globals.mount_delay * 1000);
1471         }
1472
1473         if (fc->sb_flags & SB_SILENT)
1474                 flags |= XFS_MFSI_QUIET;
1475
1476         error = xfs_open_devices(mp);
1477         if (error)
1478                 goto out_free_names;
1479
1480         error = xfs_init_mount_workqueues(mp);
1481         if (error)
1482                 goto out_close_devices;
1483
1484         error = xfs_init_percpu_counters(mp);
1485         if (error)
1486                 goto out_destroy_workqueues;
1487
1488         error = xfs_inodegc_init_percpu(mp);
1489         if (error)
1490                 goto out_destroy_counters;
1491
1492         /*
1493          * All percpu data structures requiring cleanup when a cpu goes offline
1494          * must be allocated before adding this @mp to the cpu-dead handler's
1495          * mount list.
1496          */
1497         xfs_mount_list_add(mp);
1498
1499         /* Allocate stats memory before we do operations that might use it */
1500         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1501         if (!mp->m_stats.xs_stats) {
1502                 error = -ENOMEM;
1503                 goto out_destroy_inodegc;
1504         }
1505
1506         error = xfs_readsb(mp, flags);
1507         if (error)
1508                 goto out_free_stats;
1509
1510         error = xfs_finish_flags(mp);
1511         if (error)
1512                 goto out_free_sb;
1513
1514         error = xfs_setup_devices(mp);
1515         if (error)
1516                 goto out_free_sb;
1517
1518         /* V4 support is undergoing deprecation. */
1519         if (!xfs_has_crc(mp)) {
1520 #ifdef CONFIG_XFS_SUPPORT_V4
1521                 xfs_warn_once(mp,
1522         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1523 #else
1524                 xfs_warn(mp,
1525         "Deprecated V4 format (crc=0) not supported by kernel.");
1526                 error = -EINVAL;
1527                 goto out_free_sb;
1528 #endif
1529         }
1530
1531         /* Filesystem claims it needs repair, so refuse the mount. */
1532         if (xfs_has_needsrepair(mp)) {
1533                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1534                 error = -EFSCORRUPTED;
1535                 goto out_free_sb;
1536         }
1537
1538         /*
1539          * Don't touch the filesystem if a user tool thinks it owns the primary
1540          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1541          * we don't check them at all.
1542          */
1543         if (mp->m_sb.sb_inprogress) {
1544                 xfs_warn(mp, "Offline file system operation in progress!");
1545                 error = -EFSCORRUPTED;
1546                 goto out_free_sb;
1547         }
1548
1549         /*
1550          * Until this is fixed only page-sized or smaller data blocks work.
1551          */
1552         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1553                 xfs_warn(mp,
1554                 "File system with blocksize %d bytes. "
1555                 "Only pagesize (%ld) or less will currently work.",
1556                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1557                 error = -ENOSYS;
1558                 goto out_free_sb;
1559         }
1560
1561         /* Ensure this filesystem fits in the page cache limits */
1562         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1563             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1564                 xfs_warn(mp,
1565                 "file system too large to be mounted on this system.");
1566                 error = -EFBIG;
1567                 goto out_free_sb;
1568         }
1569
1570         /*
1571          * XFS block mappings use 54 bits to store the logical block offset.
1572          * This should suffice to handle the maximum file size that the VFS
1573          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1574          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1575          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1576          * to check this assertion.
1577          *
1578          * Avoid integer overflow by comparing the maximum bmbt offset to the
1579          * maximum pagecache offset in units of fs blocks.
1580          */
1581         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1582                 xfs_warn(mp,
1583 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1584                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1585                          XFS_MAX_FILEOFF);
1586                 error = -EINVAL;
1587                 goto out_free_sb;
1588         }
1589
1590         error = xfs_filestream_mount(mp);
1591         if (error)
1592                 goto out_free_sb;
1593
1594         /*
1595          * we must configure the block size in the superblock before we run the
1596          * full mount process as the mount process can lookup and cache inodes.
1597          */
1598         sb->s_magic = XFS_SUPER_MAGIC;
1599         sb->s_blocksize = mp->m_sb.sb_blocksize;
1600         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1601         sb->s_maxbytes = MAX_LFS_FILESIZE;
1602         sb->s_max_links = XFS_MAXLINK;
1603         sb->s_time_gran = 1;
1604         if (xfs_has_bigtime(mp)) {
1605                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1606                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1607         } else {
1608                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1609                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1610         }
1611         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1612         sb->s_iflags |= SB_I_CGROUPWB;
1613
1614         set_posix_acl_flag(sb);
1615
1616         /* version 5 superblocks support inode version counters. */
1617         if (xfs_has_crc(mp))
1618                 sb->s_flags |= SB_I_VERSION;
1619
1620         if (xfs_has_dax_always(mp)) {
1621                 error = xfs_setup_dax_always(mp);
1622                 if (error)
1623                         goto out_filestream_unmount;
1624         }
1625
1626         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1627                 xfs_warn(mp,
1628         "mounting with \"discard\" option, but the device does not support discard");
1629                 mp->m_features &= ~XFS_FEAT_DISCARD;
1630         }
1631
1632         if (xfs_has_reflink(mp)) {
1633                 if (mp->m_sb.sb_rblocks) {
1634                         xfs_alert(mp,
1635         "reflink not compatible with realtime device!");
1636                         error = -EINVAL;
1637                         goto out_filestream_unmount;
1638                 }
1639
1640                 if (xfs_globals.always_cow) {
1641                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1642                         mp->m_always_cow = true;
1643                 }
1644         }
1645
1646         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1647                 xfs_alert(mp,
1648         "reverse mapping btree not compatible with realtime device!");
1649                 error = -EINVAL;
1650                 goto out_filestream_unmount;
1651         }
1652
1653         if (xfs_has_large_extent_counts(mp))
1654                 xfs_warn(mp,
1655         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1656
1657         error = xfs_mountfs(mp);
1658         if (error)
1659                 goto out_filestream_unmount;
1660
1661         root = igrab(VFS_I(mp->m_rootip));
1662         if (!root) {
1663                 error = -ENOENT;
1664                 goto out_unmount;
1665         }
1666         sb->s_root = d_make_root(root);
1667         if (!sb->s_root) {
1668                 error = -ENOMEM;
1669                 goto out_unmount;
1670         }
1671
1672         return 0;
1673
1674  out_filestream_unmount:
1675         xfs_filestream_unmount(mp);
1676  out_free_sb:
1677         xfs_freesb(mp);
1678  out_free_stats:
1679         free_percpu(mp->m_stats.xs_stats);
1680  out_destroy_inodegc:
1681         xfs_mount_list_del(mp);
1682         xfs_inodegc_free_percpu(mp);
1683  out_destroy_counters:
1684         xfs_destroy_percpu_counters(mp);
1685  out_destroy_workqueues:
1686         xfs_destroy_mount_workqueues(mp);
1687  out_close_devices:
1688         xfs_close_devices(mp);
1689  out_free_names:
1690         sb->s_fs_info = NULL;
1691         xfs_mount_free(mp);
1692         return error;
1693
1694  out_unmount:
1695         xfs_filestream_unmount(mp);
1696         xfs_unmountfs(mp);
1697         goto out_free_sb;
1698 }
1699
1700 static int
1701 xfs_fs_get_tree(
1702         struct fs_context       *fc)
1703 {
1704         return get_tree_bdev(fc, xfs_fs_fill_super);
1705 }
1706
1707 static int
1708 xfs_remount_rw(
1709         struct xfs_mount        *mp)
1710 {
1711         struct xfs_sb           *sbp = &mp->m_sb;
1712         int error;
1713
1714         if (xfs_has_norecovery(mp)) {
1715                 xfs_warn(mp,
1716                         "ro->rw transition prohibited on norecovery mount");
1717                 return -EINVAL;
1718         }
1719
1720         if (xfs_sb_is_v5(sbp) &&
1721             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1722                 xfs_warn(mp,
1723         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1724                         (sbp->sb_features_ro_compat &
1725                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1726                 return -EINVAL;
1727         }
1728
1729         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1730
1731         /*
1732          * If this is the first remount to writeable state we might have some
1733          * superblock changes to update.
1734          */
1735         if (mp->m_update_sb) {
1736                 error = xfs_sync_sb(mp, false);
1737                 if (error) {
1738                         xfs_warn(mp, "failed to write sb changes");
1739                         return error;
1740                 }
1741                 mp->m_update_sb = false;
1742         }
1743
1744         /*
1745          * Fill out the reserve pool if it is empty. Use the stashed value if
1746          * it is non-zero, otherwise go with the default.
1747          */
1748         xfs_restore_resvblks(mp);
1749         xfs_log_work_queue(mp);
1750         xfs_blockgc_start(mp);
1751
1752         /* Create the per-AG metadata reservation pool .*/
1753         error = xfs_fs_reserve_ag_blocks(mp);
1754         if (error && error != -ENOSPC)
1755                 return error;
1756
1757         /* Re-enable the background inode inactivation worker. */
1758         xfs_inodegc_start(mp);
1759
1760         return 0;
1761 }
1762
1763 static int
1764 xfs_remount_ro(
1765         struct xfs_mount        *mp)
1766 {
1767         struct xfs_icwalk       icw = {
1768                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1769         };
1770         int                     error;
1771
1772         /* Flush all the dirty data to disk. */
1773         error = sync_filesystem(mp->m_super);
1774         if (error)
1775                 return error;
1776
1777         /*
1778          * Cancel background eofb scanning so it cannot race with the final
1779          * log force+buftarg wait and deadlock the remount.
1780          */
1781         xfs_blockgc_stop(mp);
1782
1783         /*
1784          * Clear out all remaining COW staging extents and speculative post-EOF
1785          * preallocations so that we don't leave inodes requiring inactivation
1786          * cleanups during reclaim on a read-only mount.  We must process every
1787          * cached inode, so this requires a synchronous cache scan.
1788          */
1789         error = xfs_blockgc_free_space(mp, &icw);
1790         if (error) {
1791                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1792                 return error;
1793         }
1794
1795         /*
1796          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1797          * flushed all pending inodegc work when it sync'd the filesystem.
1798          * The VFS holds s_umount, so we know that inodes cannot enter
1799          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1800          * we send inodes straight to reclaim, so no inodes will be queued.
1801          */
1802         xfs_inodegc_stop(mp);
1803
1804         /* Free the per-AG metadata reservation pool. */
1805         error = xfs_fs_unreserve_ag_blocks(mp);
1806         if (error) {
1807                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1808                 return error;
1809         }
1810
1811         /*
1812          * Before we sync the metadata, we need to free up the reserve block
1813          * pool so that the used block count in the superblock on disk is
1814          * correct at the end of the remount. Stash the current* reserve pool
1815          * size so that if we get remounted rw, we can return it to the same
1816          * size.
1817          */
1818         xfs_save_resvblks(mp);
1819
1820         xfs_log_clean(mp);
1821         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1822
1823         return 0;
1824 }
1825
1826 /*
1827  * Logically we would return an error here to prevent users from believing
1828  * they might have changed mount options using remount which can't be changed.
1829  *
1830  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1831  * arguments in some cases so we can't blindly reject options, but have to
1832  * check for each specified option if it actually differs from the currently
1833  * set option and only reject it if that's the case.
1834  *
1835  * Until that is implemented we return success for every remount request, and
1836  * silently ignore all options that we can't actually change.
1837  */
1838 static int
1839 xfs_fs_reconfigure(
1840         struct fs_context *fc)
1841 {
1842         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1843         struct xfs_mount        *new_mp = fc->s_fs_info;
1844         int                     flags = fc->sb_flags;
1845         int                     error;
1846
1847         /* version 5 superblocks always support version counters. */
1848         if (xfs_has_crc(mp))
1849                 fc->sb_flags |= SB_I_VERSION;
1850
1851         error = xfs_fs_validate_params(new_mp);
1852         if (error)
1853                 return error;
1854
1855         /* inode32 -> inode64 */
1856         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1857                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1858                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1859         }
1860
1861         /* inode64 -> inode32 */
1862         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1863                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1864                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1865         }
1866
1867         /* ro -> rw */
1868         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1869                 error = xfs_remount_rw(mp);
1870                 if (error)
1871                         return error;
1872         }
1873
1874         /* rw -> ro */
1875         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1876                 error = xfs_remount_ro(mp);
1877                 if (error)
1878                         return error;
1879         }
1880
1881         return 0;
1882 }
1883
1884 static void xfs_fs_free(
1885         struct fs_context       *fc)
1886 {
1887         struct xfs_mount        *mp = fc->s_fs_info;
1888
1889         /*
1890          * mp is stored in the fs_context when it is initialized.
1891          * mp is transferred to the superblock on a successful mount,
1892          * but if an error occurs before the transfer we have to free
1893          * it here.
1894          */
1895         if (mp)
1896                 xfs_mount_free(mp);
1897 }
1898
1899 static const struct fs_context_operations xfs_context_ops = {
1900         .parse_param = xfs_fs_parse_param,
1901         .get_tree    = xfs_fs_get_tree,
1902         .reconfigure = xfs_fs_reconfigure,
1903         .free        = xfs_fs_free,
1904 };
1905
1906 static int xfs_init_fs_context(
1907         struct fs_context       *fc)
1908 {
1909         struct xfs_mount        *mp;
1910
1911         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1912         if (!mp)
1913                 return -ENOMEM;
1914
1915         spin_lock_init(&mp->m_sb_lock);
1916         spin_lock_init(&mp->m_agirotor_lock);
1917         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1918         spin_lock_init(&mp->m_perag_lock);
1919         mutex_init(&mp->m_growlock);
1920         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1921         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1922         mp->m_kobj.kobject.kset = xfs_kset;
1923         /*
1924          * We don't create the finobt per-ag space reservation until after log
1925          * recovery, so we must set this to true so that an ifree transaction
1926          * started during log recovery will not depend on space reservations
1927          * for finobt expansion.
1928          */
1929         mp->m_finobt_nores = true;
1930
1931         /*
1932          * These can be overridden by the mount option parsing.
1933          */
1934         mp->m_logbufs = -1;
1935         mp->m_logbsize = -1;
1936         mp->m_allocsize_log = 16; /* 64k */
1937
1938         /*
1939          * Copy binary VFS mount flags we are interested in.
1940          */
1941         if (fc->sb_flags & SB_RDONLY)
1942                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1943         if (fc->sb_flags & SB_DIRSYNC)
1944                 mp->m_features |= XFS_FEAT_DIRSYNC;
1945         if (fc->sb_flags & SB_SYNCHRONOUS)
1946                 mp->m_features |= XFS_FEAT_WSYNC;
1947
1948         fc->s_fs_info = mp;
1949         fc->ops = &xfs_context_ops;
1950
1951         return 0;
1952 }
1953
1954 static struct file_system_type xfs_fs_type = {
1955         .owner                  = THIS_MODULE,
1956         .name                   = "xfs",
1957         .init_fs_context        = xfs_init_fs_context,
1958         .parameters             = xfs_fs_parameters,
1959         .kill_sb                = kill_block_super,
1960         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1961 };
1962 MODULE_ALIAS_FS("xfs");
1963
1964 STATIC int __init
1965 xfs_init_caches(void)
1966 {
1967         int             error;
1968
1969         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1970                                                 sizeof(struct xlog_ticket),
1971                                                 0, 0, NULL);
1972         if (!xfs_log_ticket_cache)
1973                 goto out;
1974
1975         error = xfs_btree_init_cur_caches();
1976         if (error)
1977                 goto out_destroy_log_ticket_cache;
1978
1979         error = xfs_defer_init_item_caches();
1980         if (error)
1981                 goto out_destroy_btree_cur_cache;
1982
1983         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
1984                                               sizeof(struct xfs_da_state),
1985                                               0, 0, NULL);
1986         if (!xfs_da_state_cache)
1987                 goto out_destroy_defer_item_cache;
1988
1989         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
1990                                            sizeof(struct xfs_ifork),
1991                                            0, 0, NULL);
1992         if (!xfs_ifork_cache)
1993                 goto out_destroy_da_state_cache;
1994
1995         xfs_trans_cache = kmem_cache_create("xfs_trans",
1996                                            sizeof(struct xfs_trans),
1997                                            0, 0, NULL);
1998         if (!xfs_trans_cache)
1999                 goto out_destroy_ifork_cache;
2000
2001
2002         /*
2003          * The size of the cache-allocated buf log item is the maximum
2004          * size possible under XFS.  This wastes a little bit of memory,
2005          * but it is much faster.
2006          */
2007         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2008                                               sizeof(struct xfs_buf_log_item),
2009                                               0, 0, NULL);
2010         if (!xfs_buf_item_cache)
2011                 goto out_destroy_trans_cache;
2012
2013         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2014                                         (sizeof(struct xfs_efd_log_item) +
2015                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2016                                         sizeof(struct xfs_extent)),
2017                                         0, 0, NULL);
2018         if (!xfs_efd_cache)
2019                 goto out_destroy_buf_item_cache;
2020
2021         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2022                                          (sizeof(struct xfs_efi_log_item) +
2023                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2024                                          sizeof(struct xfs_extent)),
2025                                          0, 0, NULL);
2026         if (!xfs_efi_cache)
2027                 goto out_destroy_efd_cache;
2028
2029         xfs_inode_cache = kmem_cache_create("xfs_inode",
2030                                            sizeof(struct xfs_inode), 0,
2031                                            (SLAB_HWCACHE_ALIGN |
2032                                             SLAB_RECLAIM_ACCOUNT |
2033                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2034                                            xfs_fs_inode_init_once);
2035         if (!xfs_inode_cache)
2036                 goto out_destroy_efi_cache;
2037
2038         xfs_ili_cache = kmem_cache_create("xfs_ili",
2039                                          sizeof(struct xfs_inode_log_item), 0,
2040                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2041                                          NULL);
2042         if (!xfs_ili_cache)
2043                 goto out_destroy_inode_cache;
2044
2045         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2046                                              sizeof(struct xfs_icreate_item),
2047                                              0, 0, NULL);
2048         if (!xfs_icreate_cache)
2049                 goto out_destroy_ili_cache;
2050
2051         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2052                                          sizeof(struct xfs_rud_log_item),
2053                                          0, 0, NULL);
2054         if (!xfs_rud_cache)
2055                 goto out_destroy_icreate_cache;
2056
2057         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2058                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2059                         0, 0, NULL);
2060         if (!xfs_rui_cache)
2061                 goto out_destroy_rud_cache;
2062
2063         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2064                                          sizeof(struct xfs_cud_log_item),
2065                                          0, 0, NULL);
2066         if (!xfs_cud_cache)
2067                 goto out_destroy_rui_cache;
2068
2069         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2070                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2071                         0, 0, NULL);
2072         if (!xfs_cui_cache)
2073                 goto out_destroy_cud_cache;
2074
2075         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2076                                          sizeof(struct xfs_bud_log_item),
2077                                          0, 0, NULL);
2078         if (!xfs_bud_cache)
2079                 goto out_destroy_cui_cache;
2080
2081         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2082                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2083                         0, 0, NULL);
2084         if (!xfs_bui_cache)
2085                 goto out_destroy_bud_cache;
2086
2087         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2088                                             sizeof(struct xfs_attrd_log_item),
2089                                             0, 0, NULL);
2090         if (!xfs_attrd_cache)
2091                 goto out_destroy_bui_cache;
2092
2093         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2094                                             sizeof(struct xfs_attri_log_item),
2095                                             0, 0, NULL);
2096         if (!xfs_attri_cache)
2097                 goto out_destroy_attrd_cache;
2098
2099         return 0;
2100
2101  out_destroy_attrd_cache:
2102         kmem_cache_destroy(xfs_attrd_cache);
2103  out_destroy_bui_cache:
2104         kmem_cache_destroy(xfs_bui_cache);
2105  out_destroy_bud_cache:
2106         kmem_cache_destroy(xfs_bud_cache);
2107  out_destroy_cui_cache:
2108         kmem_cache_destroy(xfs_cui_cache);
2109  out_destroy_cud_cache:
2110         kmem_cache_destroy(xfs_cud_cache);
2111  out_destroy_rui_cache:
2112         kmem_cache_destroy(xfs_rui_cache);
2113  out_destroy_rud_cache:
2114         kmem_cache_destroy(xfs_rud_cache);
2115  out_destroy_icreate_cache:
2116         kmem_cache_destroy(xfs_icreate_cache);
2117  out_destroy_ili_cache:
2118         kmem_cache_destroy(xfs_ili_cache);
2119  out_destroy_inode_cache:
2120         kmem_cache_destroy(xfs_inode_cache);
2121  out_destroy_efi_cache:
2122         kmem_cache_destroy(xfs_efi_cache);
2123  out_destroy_efd_cache:
2124         kmem_cache_destroy(xfs_efd_cache);
2125  out_destroy_buf_item_cache:
2126         kmem_cache_destroy(xfs_buf_item_cache);
2127  out_destroy_trans_cache:
2128         kmem_cache_destroy(xfs_trans_cache);
2129  out_destroy_ifork_cache:
2130         kmem_cache_destroy(xfs_ifork_cache);
2131  out_destroy_da_state_cache:
2132         kmem_cache_destroy(xfs_da_state_cache);
2133  out_destroy_defer_item_cache:
2134         xfs_defer_destroy_item_caches();
2135  out_destroy_btree_cur_cache:
2136         xfs_btree_destroy_cur_caches();
2137  out_destroy_log_ticket_cache:
2138         kmem_cache_destroy(xfs_log_ticket_cache);
2139  out:
2140         return -ENOMEM;
2141 }
2142
2143 STATIC void
2144 xfs_destroy_caches(void)
2145 {
2146         /*
2147          * Make sure all delayed rcu free are flushed before we
2148          * destroy caches.
2149          */
2150         rcu_barrier();
2151         kmem_cache_destroy(xfs_attri_cache);
2152         kmem_cache_destroy(xfs_attrd_cache);
2153         kmem_cache_destroy(xfs_bui_cache);
2154         kmem_cache_destroy(xfs_bud_cache);
2155         kmem_cache_destroy(xfs_cui_cache);
2156         kmem_cache_destroy(xfs_cud_cache);
2157         kmem_cache_destroy(xfs_rui_cache);
2158         kmem_cache_destroy(xfs_rud_cache);
2159         kmem_cache_destroy(xfs_icreate_cache);
2160         kmem_cache_destroy(xfs_ili_cache);
2161         kmem_cache_destroy(xfs_inode_cache);
2162         kmem_cache_destroy(xfs_efi_cache);
2163         kmem_cache_destroy(xfs_efd_cache);
2164         kmem_cache_destroy(xfs_buf_item_cache);
2165         kmem_cache_destroy(xfs_trans_cache);
2166         kmem_cache_destroy(xfs_ifork_cache);
2167         kmem_cache_destroy(xfs_da_state_cache);
2168         xfs_defer_destroy_item_caches();
2169         xfs_btree_destroy_cur_caches();
2170         kmem_cache_destroy(xfs_log_ticket_cache);
2171 }
2172
2173 STATIC int __init
2174 xfs_init_workqueues(void)
2175 {
2176         /*
2177          * The allocation workqueue can be used in memory reclaim situations
2178          * (writepage path), and parallelism is only limited by the number of
2179          * AGs in all the filesystems mounted. Hence use the default large
2180          * max_active value for this workqueue.
2181          */
2182         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2183                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2184         if (!xfs_alloc_wq)
2185                 return -ENOMEM;
2186
2187         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2188                         0);
2189         if (!xfs_discard_wq)
2190                 goto out_free_alloc_wq;
2191
2192         return 0;
2193 out_free_alloc_wq:
2194         destroy_workqueue(xfs_alloc_wq);
2195         return -ENOMEM;
2196 }
2197
2198 STATIC void
2199 xfs_destroy_workqueues(void)
2200 {
2201         destroy_workqueue(xfs_discard_wq);
2202         destroy_workqueue(xfs_alloc_wq);
2203 }
2204
2205 #ifdef CONFIG_HOTPLUG_CPU
2206 static int
2207 xfs_cpu_dead(
2208         unsigned int            cpu)
2209 {
2210         struct xfs_mount        *mp, *n;
2211
2212         spin_lock(&xfs_mount_list_lock);
2213         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2214                 spin_unlock(&xfs_mount_list_lock);
2215                 xfs_inodegc_cpu_dead(mp, cpu);
2216                 spin_lock(&xfs_mount_list_lock);
2217         }
2218         spin_unlock(&xfs_mount_list_lock);
2219         return 0;
2220 }
2221
2222 static int __init
2223 xfs_cpu_hotplug_init(void)
2224 {
2225         int     error;
2226
2227         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2228                         xfs_cpu_dead);
2229         if (error < 0)
2230                 xfs_alert(NULL,
2231 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2232                         error);
2233         return error;
2234 }
2235
2236 static void
2237 xfs_cpu_hotplug_destroy(void)
2238 {
2239         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2240 }
2241
2242 #else /* !CONFIG_HOTPLUG_CPU */
2243 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2244 static inline void xfs_cpu_hotplug_destroy(void) {}
2245 #endif
2246
2247 STATIC int __init
2248 init_xfs_fs(void)
2249 {
2250         int                     error;
2251
2252         xfs_check_ondisk_structs();
2253
2254         printk(KERN_INFO XFS_VERSION_STRING " with "
2255                          XFS_BUILD_OPTIONS " enabled\n");
2256
2257         xfs_dir_startup();
2258
2259         error = xfs_cpu_hotplug_init();
2260         if (error)
2261                 goto out;
2262
2263         error = xfs_init_caches();
2264         if (error)
2265                 goto out_destroy_hp;
2266
2267         error = xfs_init_workqueues();
2268         if (error)
2269                 goto out_destroy_caches;
2270
2271         error = xfs_mru_cache_init();
2272         if (error)
2273                 goto out_destroy_wq;
2274
2275         error = xfs_buf_init();
2276         if (error)
2277                 goto out_mru_cache_uninit;
2278
2279         error = xfs_init_procfs();
2280         if (error)
2281                 goto out_buf_terminate;
2282
2283         error = xfs_sysctl_register();
2284         if (error)
2285                 goto out_cleanup_procfs;
2286
2287         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2288         if (!xfs_kset) {
2289                 error = -ENOMEM;
2290                 goto out_sysctl_unregister;
2291         }
2292
2293         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2294
2295         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2296         if (!xfsstats.xs_stats) {
2297                 error = -ENOMEM;
2298                 goto out_kset_unregister;
2299         }
2300
2301         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2302                                "stats");
2303         if (error)
2304                 goto out_free_stats;
2305
2306 #ifdef DEBUG
2307         xfs_dbg_kobj.kobject.kset = xfs_kset;
2308         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2309         if (error)
2310                 goto out_remove_stats_kobj;
2311 #endif
2312
2313         error = xfs_qm_init();
2314         if (error)
2315                 goto out_remove_dbg_kobj;
2316
2317         error = register_filesystem(&xfs_fs_type);
2318         if (error)
2319                 goto out_qm_exit;
2320         return 0;
2321
2322  out_qm_exit:
2323         xfs_qm_exit();
2324  out_remove_dbg_kobj:
2325 #ifdef DEBUG
2326         xfs_sysfs_del(&xfs_dbg_kobj);
2327  out_remove_stats_kobj:
2328 #endif
2329         xfs_sysfs_del(&xfsstats.xs_kobj);
2330  out_free_stats:
2331         free_percpu(xfsstats.xs_stats);
2332  out_kset_unregister:
2333         kset_unregister(xfs_kset);
2334  out_sysctl_unregister:
2335         xfs_sysctl_unregister();
2336  out_cleanup_procfs:
2337         xfs_cleanup_procfs();
2338  out_buf_terminate:
2339         xfs_buf_terminate();
2340  out_mru_cache_uninit:
2341         xfs_mru_cache_uninit();
2342  out_destroy_wq:
2343         xfs_destroy_workqueues();
2344  out_destroy_caches:
2345         xfs_destroy_caches();
2346  out_destroy_hp:
2347         xfs_cpu_hotplug_destroy();
2348  out:
2349         return error;
2350 }
2351
2352 STATIC void __exit
2353 exit_xfs_fs(void)
2354 {
2355         xfs_qm_exit();
2356         unregister_filesystem(&xfs_fs_type);
2357 #ifdef DEBUG
2358         xfs_sysfs_del(&xfs_dbg_kobj);
2359 #endif
2360         xfs_sysfs_del(&xfsstats.xs_kobj);
2361         free_percpu(xfsstats.xs_stats);
2362         kset_unregister(xfs_kset);
2363         xfs_sysctl_unregister();
2364         xfs_cleanup_procfs();
2365         xfs_buf_terminate();
2366         xfs_mru_cache_uninit();
2367         xfs_destroy_workqueues();
2368         xfs_destroy_caches();
2369         xfs_uuid_table_free();
2370         xfs_cpu_hotplug_destroy();
2371 }
2372
2373 module_init(init_xfs_fs);
2374 module_exit(exit_xfs_fs);
2375
2376 MODULE_AUTHOR("Silicon Graphics, Inc.");
2377 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2378 MODULE_LICENSE("GPL");