RISC-V: Make mmap() with PROT_WRITE imply PROT_READ
[linux-2.6-microblaze.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
45 #include <linux/magic.h>
46 #include <linux/fs_context.h>
47 #include <linux/fs_parser.h>
48
49 static const struct super_operations xfs_super_operations;
50
51 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
52 #ifdef DEBUG
53 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
54 #endif
55
56 #ifdef CONFIG_HOTPLUG_CPU
57 static LIST_HEAD(xfs_mount_list);
58 static DEFINE_SPINLOCK(xfs_mount_list_lock);
59
60 static inline void xfs_mount_list_add(struct xfs_mount *mp)
61 {
62         spin_lock(&xfs_mount_list_lock);
63         list_add(&mp->m_mount_list, &xfs_mount_list);
64         spin_unlock(&xfs_mount_list_lock);
65 }
66
67 static inline void xfs_mount_list_del(struct xfs_mount *mp)
68 {
69         spin_lock(&xfs_mount_list_lock);
70         list_del(&mp->m_mount_list);
71         spin_unlock(&xfs_mount_list_lock);
72 }
73 #else /* !CONFIG_HOTPLUG_CPU */
74 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
75 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
76 #endif
77
78 enum xfs_dax_mode {
79         XFS_DAX_INODE = 0,
80         XFS_DAX_ALWAYS = 1,
81         XFS_DAX_NEVER = 2,
82 };
83
84 static void
85 xfs_mount_set_dax_mode(
86         struct xfs_mount        *mp,
87         enum xfs_dax_mode       mode)
88 {
89         switch (mode) {
90         case XFS_DAX_INODE:
91                 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
92                 break;
93         case XFS_DAX_ALWAYS:
94                 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
95                 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
96                 break;
97         case XFS_DAX_NEVER:
98                 mp->m_features |= XFS_FEAT_DAX_NEVER;
99                 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
100                 break;
101         }
102 }
103
104 static const struct constant_table dax_param_enums[] = {
105         {"inode",       XFS_DAX_INODE },
106         {"always",      XFS_DAX_ALWAYS },
107         {"never",       XFS_DAX_NEVER },
108         {}
109 };
110
111 /*
112  * Table driven mount option parser.
113  */
114 enum {
115         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
116         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
117         Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
118         Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
119         Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
120         Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
121         Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
122         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
123         Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
124 };
125
126 static const struct fs_parameter_spec xfs_fs_parameters[] = {
127         fsparam_u32("logbufs",          Opt_logbufs),
128         fsparam_string("logbsize",      Opt_logbsize),
129         fsparam_string("logdev",        Opt_logdev),
130         fsparam_string("rtdev",         Opt_rtdev),
131         fsparam_flag("wsync",           Opt_wsync),
132         fsparam_flag("noalign",         Opt_noalign),
133         fsparam_flag("swalloc",         Opt_swalloc),
134         fsparam_u32("sunit",            Opt_sunit),
135         fsparam_u32("swidth",           Opt_swidth),
136         fsparam_flag("nouuid",          Opt_nouuid),
137         fsparam_flag("grpid",           Opt_grpid),
138         fsparam_flag("nogrpid",         Opt_nogrpid),
139         fsparam_flag("bsdgroups",       Opt_bsdgroups),
140         fsparam_flag("sysvgroups",      Opt_sysvgroups),
141         fsparam_string("allocsize",     Opt_allocsize),
142         fsparam_flag("norecovery",      Opt_norecovery),
143         fsparam_flag("inode64",         Opt_inode64),
144         fsparam_flag("inode32",         Opt_inode32),
145         fsparam_flag("ikeep",           Opt_ikeep),
146         fsparam_flag("noikeep",         Opt_noikeep),
147         fsparam_flag("largeio",         Opt_largeio),
148         fsparam_flag("nolargeio",       Opt_nolargeio),
149         fsparam_flag("attr2",           Opt_attr2),
150         fsparam_flag("noattr2",         Opt_noattr2),
151         fsparam_flag("filestreams",     Opt_filestreams),
152         fsparam_flag("quota",           Opt_quota),
153         fsparam_flag("noquota",         Opt_noquota),
154         fsparam_flag("usrquota",        Opt_usrquota),
155         fsparam_flag("grpquota",        Opt_grpquota),
156         fsparam_flag("prjquota",        Opt_prjquota),
157         fsparam_flag("uquota",          Opt_uquota),
158         fsparam_flag("gquota",          Opt_gquota),
159         fsparam_flag("pquota",          Opt_pquota),
160         fsparam_flag("uqnoenforce",     Opt_uqnoenforce),
161         fsparam_flag("gqnoenforce",     Opt_gqnoenforce),
162         fsparam_flag("pqnoenforce",     Opt_pqnoenforce),
163         fsparam_flag("qnoenforce",      Opt_qnoenforce),
164         fsparam_flag("discard",         Opt_discard),
165         fsparam_flag("nodiscard",       Opt_nodiscard),
166         fsparam_flag("dax",             Opt_dax),
167         fsparam_enum("dax",             Opt_dax_enum, dax_param_enums),
168         {}
169 };
170
171 struct proc_xfs_info {
172         uint64_t        flag;
173         char            *str;
174 };
175
176 static int
177 xfs_fs_show_options(
178         struct seq_file         *m,
179         struct dentry           *root)
180 {
181         static struct proc_xfs_info xfs_info_set[] = {
182                 /* the few simple ones we can get from the mount struct */
183                 { XFS_FEAT_IKEEP,               ",ikeep" },
184                 { XFS_FEAT_WSYNC,               ",wsync" },
185                 { XFS_FEAT_NOALIGN,             ",noalign" },
186                 { XFS_FEAT_SWALLOC,             ",swalloc" },
187                 { XFS_FEAT_NOUUID,              ",nouuid" },
188                 { XFS_FEAT_NORECOVERY,          ",norecovery" },
189                 { XFS_FEAT_ATTR2,               ",attr2" },
190                 { XFS_FEAT_FILESTREAMS,         ",filestreams" },
191                 { XFS_FEAT_GRPID,               ",grpid" },
192                 { XFS_FEAT_DISCARD,             ",discard" },
193                 { XFS_FEAT_LARGE_IOSIZE,        ",largeio" },
194                 { XFS_FEAT_DAX_ALWAYS,          ",dax=always" },
195                 { XFS_FEAT_DAX_NEVER,           ",dax=never" },
196                 { 0, NULL }
197         };
198         struct xfs_mount        *mp = XFS_M(root->d_sb);
199         struct proc_xfs_info    *xfs_infop;
200
201         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
202                 if (mp->m_features & xfs_infop->flag)
203                         seq_puts(m, xfs_infop->str);
204         }
205
206         seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
207
208         if (xfs_has_allocsize(mp))
209                 seq_printf(m, ",allocsize=%dk",
210                            (1 << mp->m_allocsize_log) >> 10);
211
212         if (mp->m_logbufs > 0)
213                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
214         if (mp->m_logbsize > 0)
215                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
216
217         if (mp->m_logname)
218                 seq_show_option(m, "logdev", mp->m_logname);
219         if (mp->m_rtname)
220                 seq_show_option(m, "rtdev", mp->m_rtname);
221
222         if (mp->m_dalign > 0)
223                 seq_printf(m, ",sunit=%d",
224                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
225         if (mp->m_swidth > 0)
226                 seq_printf(m, ",swidth=%d",
227                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
228
229         if (mp->m_qflags & XFS_UQUOTA_ENFD)
230                 seq_puts(m, ",usrquota");
231         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
232                 seq_puts(m, ",uqnoenforce");
233
234         if (mp->m_qflags & XFS_PQUOTA_ENFD)
235                 seq_puts(m, ",prjquota");
236         else if (mp->m_qflags & XFS_PQUOTA_ACCT)
237                 seq_puts(m, ",pqnoenforce");
238
239         if (mp->m_qflags & XFS_GQUOTA_ENFD)
240                 seq_puts(m, ",grpquota");
241         else if (mp->m_qflags & XFS_GQUOTA_ACCT)
242                 seq_puts(m, ",gqnoenforce");
243
244         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
245                 seq_puts(m, ",noquota");
246
247         return 0;
248 }
249
250 /*
251  * Set parameters for inode allocation heuristics, taking into account
252  * filesystem size and inode32/inode64 mount options; i.e. specifically
253  * whether or not XFS_FEAT_SMALL_INUMS is set.
254  *
255  * Inode allocation patterns are altered only if inode32 is requested
256  * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
257  * If altered, XFS_OPSTATE_INODE32 is set as well.
258  *
259  * An agcount independent of that in the mount structure is provided
260  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
261  * to the potentially higher ag count.
262  *
263  * Returns the maximum AG index which may contain inodes.
264  */
265 xfs_agnumber_t
266 xfs_set_inode_alloc(
267         struct xfs_mount *mp,
268         xfs_agnumber_t  agcount)
269 {
270         xfs_agnumber_t  index;
271         xfs_agnumber_t  maxagi = 0;
272         xfs_sb_t        *sbp = &mp->m_sb;
273         xfs_agnumber_t  max_metadata;
274         xfs_agino_t     agino;
275         xfs_ino_t       ino;
276
277         /*
278          * Calculate how much should be reserved for inodes to meet
279          * the max inode percentage.  Used only for inode32.
280          */
281         if (M_IGEO(mp)->maxicount) {
282                 uint64_t        icount;
283
284                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
285                 do_div(icount, 100);
286                 icount += sbp->sb_agblocks - 1;
287                 do_div(icount, sbp->sb_agblocks);
288                 max_metadata = icount;
289         } else {
290                 max_metadata = agcount;
291         }
292
293         /* Get the last possible inode in the filesystem */
294         agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
295         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
296
297         /*
298          * If user asked for no more than 32-bit inodes, and the fs is
299          * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
300          * the allocator to accommodate the request.
301          */
302         if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
303                 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
304         else
305                 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
306
307         for (index = 0; index < agcount; index++) {
308                 struct xfs_perag        *pag;
309
310                 ino = XFS_AGINO_TO_INO(mp, index, agino);
311
312                 pag = xfs_perag_get(mp, index);
313
314                 if (xfs_is_inode32(mp)) {
315                         if (ino > XFS_MAXINUMBER_32) {
316                                 pag->pagi_inodeok = 0;
317                                 pag->pagf_metadata = 0;
318                         } else {
319                                 pag->pagi_inodeok = 1;
320                                 maxagi++;
321                                 if (index < max_metadata)
322                                         pag->pagf_metadata = 1;
323                                 else
324                                         pag->pagf_metadata = 0;
325                         }
326                 } else {
327                         pag->pagi_inodeok = 1;
328                         pag->pagf_metadata = 0;
329                 }
330
331                 xfs_perag_put(pag);
332         }
333
334         return xfs_is_inode32(mp) ? maxagi : agcount;
335 }
336
337 static int
338 xfs_setup_dax_always(
339         struct xfs_mount        *mp)
340 {
341         if (!mp->m_ddev_targp->bt_daxdev &&
342             (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
343                 xfs_alert(mp,
344                         "DAX unsupported by block device. Turning off DAX.");
345                 goto disable_dax;
346         }
347
348         if (mp->m_super->s_blocksize != PAGE_SIZE) {
349                 xfs_alert(mp,
350                         "DAX not supported for blocksize. Turning off DAX.");
351                 goto disable_dax;
352         }
353
354         if (xfs_has_reflink(mp) &&
355             bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
356                 xfs_alert(mp,
357                         "DAX and reflink cannot work with multi-partitions!");
358                 return -EINVAL;
359         }
360
361         xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
362         return 0;
363
364 disable_dax:
365         xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
366         return 0;
367 }
368
369 STATIC int
370 xfs_blkdev_get(
371         xfs_mount_t             *mp,
372         const char              *name,
373         struct block_device     **bdevp)
374 {
375         int                     error = 0;
376
377         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
378                                     mp);
379         if (IS_ERR(*bdevp)) {
380                 error = PTR_ERR(*bdevp);
381                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
382         }
383
384         return error;
385 }
386
387 STATIC void
388 xfs_blkdev_put(
389         struct block_device     *bdev)
390 {
391         if (bdev)
392                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
393 }
394
395 STATIC void
396 xfs_close_devices(
397         struct xfs_mount        *mp)
398 {
399         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
400                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
401
402                 xfs_free_buftarg(mp->m_logdev_targp);
403                 xfs_blkdev_put(logdev);
404         }
405         if (mp->m_rtdev_targp) {
406                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
407
408                 xfs_free_buftarg(mp->m_rtdev_targp);
409                 xfs_blkdev_put(rtdev);
410         }
411         xfs_free_buftarg(mp->m_ddev_targp);
412 }
413
414 /*
415  * The file system configurations are:
416  *      (1) device (partition) with data and internal log
417  *      (2) logical volume with data and log subvolumes.
418  *      (3) logical volume with data, log, and realtime subvolumes.
419  *
420  * We only have to handle opening the log and realtime volumes here if
421  * they are present.  The data subvolume has already been opened by
422  * get_sb_bdev() and is stored in sb->s_bdev.
423  */
424 STATIC int
425 xfs_open_devices(
426         struct xfs_mount        *mp)
427 {
428         struct block_device     *ddev = mp->m_super->s_bdev;
429         struct block_device     *logdev = NULL, *rtdev = NULL;
430         int                     error;
431
432         /*
433          * Open real time and log devices - order is important.
434          */
435         if (mp->m_logname) {
436                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
437                 if (error)
438                         return error;
439         }
440
441         if (mp->m_rtname) {
442                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
443                 if (error)
444                         goto out_close_logdev;
445
446                 if (rtdev == ddev || rtdev == logdev) {
447                         xfs_warn(mp,
448         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
449                         error = -EINVAL;
450                         goto out_close_rtdev;
451                 }
452         }
453
454         /*
455          * Setup xfs_mount buffer target pointers
456          */
457         error = -ENOMEM;
458         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
459         if (!mp->m_ddev_targp)
460                 goto out_close_rtdev;
461
462         if (rtdev) {
463                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
464                 if (!mp->m_rtdev_targp)
465                         goto out_free_ddev_targ;
466         }
467
468         if (logdev && logdev != ddev) {
469                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
470                 if (!mp->m_logdev_targp)
471                         goto out_free_rtdev_targ;
472         } else {
473                 mp->m_logdev_targp = mp->m_ddev_targp;
474         }
475
476         return 0;
477
478  out_free_rtdev_targ:
479         if (mp->m_rtdev_targp)
480                 xfs_free_buftarg(mp->m_rtdev_targp);
481  out_free_ddev_targ:
482         xfs_free_buftarg(mp->m_ddev_targp);
483  out_close_rtdev:
484         xfs_blkdev_put(rtdev);
485  out_close_logdev:
486         if (logdev && logdev != ddev)
487                 xfs_blkdev_put(logdev);
488         return error;
489 }
490
491 /*
492  * Setup xfs_mount buffer target pointers based on superblock
493  */
494 STATIC int
495 xfs_setup_devices(
496         struct xfs_mount        *mp)
497 {
498         int                     error;
499
500         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
501         if (error)
502                 return error;
503
504         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
505                 unsigned int    log_sector_size = BBSIZE;
506
507                 if (xfs_has_sector(mp))
508                         log_sector_size = mp->m_sb.sb_logsectsize;
509                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
510                                             log_sector_size);
511                 if (error)
512                         return error;
513         }
514         if (mp->m_rtdev_targp) {
515                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
516                                             mp->m_sb.sb_sectsize);
517                 if (error)
518                         return error;
519         }
520
521         return 0;
522 }
523
524 STATIC int
525 xfs_init_mount_workqueues(
526         struct xfs_mount        *mp)
527 {
528         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
529                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
530                         1, mp->m_super->s_id);
531         if (!mp->m_buf_workqueue)
532                 goto out;
533
534         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
535                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
536                         0, mp->m_super->s_id);
537         if (!mp->m_unwritten_workqueue)
538                 goto out_destroy_buf;
539
540         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
541                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
542                         0, mp->m_super->s_id);
543         if (!mp->m_reclaim_workqueue)
544                 goto out_destroy_unwritten;
545
546         mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
547                         XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
548                         0, mp->m_super->s_id);
549         if (!mp->m_blockgc_wq)
550                 goto out_destroy_reclaim;
551
552         mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
553                         XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
554                         1, mp->m_super->s_id);
555         if (!mp->m_inodegc_wq)
556                 goto out_destroy_blockgc;
557
558         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
559                         XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
560         if (!mp->m_sync_workqueue)
561                 goto out_destroy_inodegc;
562
563         return 0;
564
565 out_destroy_inodegc:
566         destroy_workqueue(mp->m_inodegc_wq);
567 out_destroy_blockgc:
568         destroy_workqueue(mp->m_blockgc_wq);
569 out_destroy_reclaim:
570         destroy_workqueue(mp->m_reclaim_workqueue);
571 out_destroy_unwritten:
572         destroy_workqueue(mp->m_unwritten_workqueue);
573 out_destroy_buf:
574         destroy_workqueue(mp->m_buf_workqueue);
575 out:
576         return -ENOMEM;
577 }
578
579 STATIC void
580 xfs_destroy_mount_workqueues(
581         struct xfs_mount        *mp)
582 {
583         destroy_workqueue(mp->m_sync_workqueue);
584         destroy_workqueue(mp->m_blockgc_wq);
585         destroy_workqueue(mp->m_inodegc_wq);
586         destroy_workqueue(mp->m_reclaim_workqueue);
587         destroy_workqueue(mp->m_unwritten_workqueue);
588         destroy_workqueue(mp->m_buf_workqueue);
589 }
590
591 static void
592 xfs_flush_inodes_worker(
593         struct work_struct      *work)
594 {
595         struct xfs_mount        *mp = container_of(work, struct xfs_mount,
596                                                    m_flush_inodes_work);
597         struct super_block      *sb = mp->m_super;
598
599         if (down_read_trylock(&sb->s_umount)) {
600                 sync_inodes_sb(sb);
601                 up_read(&sb->s_umount);
602         }
603 }
604
605 /*
606  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
607  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
608  * for IO to complete so that we effectively throttle multiple callers to the
609  * rate at which IO is completing.
610  */
611 void
612 xfs_flush_inodes(
613         struct xfs_mount        *mp)
614 {
615         /*
616          * If flush_work() returns true then that means we waited for a flush
617          * which was already in progress.  Don't bother running another scan.
618          */
619         if (flush_work(&mp->m_flush_inodes_work))
620                 return;
621
622         queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
623         flush_work(&mp->m_flush_inodes_work);
624 }
625
626 /* Catch misguided souls that try to use this interface on XFS */
627 STATIC struct inode *
628 xfs_fs_alloc_inode(
629         struct super_block      *sb)
630 {
631         BUG();
632         return NULL;
633 }
634
635 /*
636  * Now that the generic code is guaranteed not to be accessing
637  * the linux inode, we can inactivate and reclaim the inode.
638  */
639 STATIC void
640 xfs_fs_destroy_inode(
641         struct inode            *inode)
642 {
643         struct xfs_inode        *ip = XFS_I(inode);
644
645         trace_xfs_destroy_inode(ip);
646
647         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
648         XFS_STATS_INC(ip->i_mount, vn_rele);
649         XFS_STATS_INC(ip->i_mount, vn_remove);
650         xfs_inode_mark_reclaimable(ip);
651 }
652
653 static void
654 xfs_fs_dirty_inode(
655         struct inode                    *inode,
656         int                             flags)
657 {
658         struct xfs_inode                *ip = XFS_I(inode);
659         struct xfs_mount                *mp = ip->i_mount;
660         struct xfs_trans                *tp;
661
662         if (!(inode->i_sb->s_flags & SB_LAZYTIME))
663                 return;
664
665         /*
666          * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
667          * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
668          * in flags possibly together with I_DIRTY_SYNC.
669          */
670         if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
671                 return;
672
673         if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
674                 return;
675         xfs_ilock(ip, XFS_ILOCK_EXCL);
676         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
677         xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
678         xfs_trans_commit(tp);
679 }
680
681 /*
682  * Slab object creation initialisation for the XFS inode.
683  * This covers only the idempotent fields in the XFS inode;
684  * all other fields need to be initialised on allocation
685  * from the slab. This avoids the need to repeatedly initialise
686  * fields in the xfs inode that left in the initialise state
687  * when freeing the inode.
688  */
689 STATIC void
690 xfs_fs_inode_init_once(
691         void                    *inode)
692 {
693         struct xfs_inode        *ip = inode;
694
695         memset(ip, 0, sizeof(struct xfs_inode));
696
697         /* vfs inode */
698         inode_init_once(VFS_I(ip));
699
700         /* xfs inode */
701         atomic_set(&ip->i_pincount, 0);
702         spin_lock_init(&ip->i_flags_lock);
703
704         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
705                      "xfsino", ip->i_ino);
706 }
707
708 /*
709  * We do an unlocked check for XFS_IDONTCACHE here because we are already
710  * serialised against cache hits here via the inode->i_lock and igrab() in
711  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
712  * racing with us, and it avoids needing to grab a spinlock here for every inode
713  * we drop the final reference on.
714  */
715 STATIC int
716 xfs_fs_drop_inode(
717         struct inode            *inode)
718 {
719         struct xfs_inode        *ip = XFS_I(inode);
720
721         /*
722          * If this unlinked inode is in the middle of recovery, don't
723          * drop the inode just yet; log recovery will take care of
724          * that.  See the comment for this inode flag.
725          */
726         if (ip->i_flags & XFS_IRECOVERY) {
727                 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
728                 return 0;
729         }
730
731         return generic_drop_inode(inode);
732 }
733
734 static void
735 xfs_mount_free(
736         struct xfs_mount        *mp)
737 {
738         kfree(mp->m_rtname);
739         kfree(mp->m_logname);
740         kmem_free(mp);
741 }
742
743 STATIC int
744 xfs_fs_sync_fs(
745         struct super_block      *sb,
746         int                     wait)
747 {
748         struct xfs_mount        *mp = XFS_M(sb);
749         int                     error;
750
751         trace_xfs_fs_sync_fs(mp, __return_address);
752
753         /*
754          * Doing anything during the async pass would be counterproductive.
755          */
756         if (!wait)
757                 return 0;
758
759         error = xfs_log_force(mp, XFS_LOG_SYNC);
760         if (error)
761                 return error;
762
763         if (laptop_mode) {
764                 /*
765                  * The disk must be active because we're syncing.
766                  * We schedule log work now (now that the disk is
767                  * active) instead of later (when it might not be).
768                  */
769                 flush_delayed_work(&mp->m_log->l_work);
770         }
771
772         /*
773          * If we are called with page faults frozen out, it means we are about
774          * to freeze the transaction subsystem. Take the opportunity to shut
775          * down inodegc because once SB_FREEZE_FS is set it's too late to
776          * prevent inactivation races with freeze. The fs doesn't get called
777          * again by the freezing process until after SB_FREEZE_FS has been set,
778          * so it's now or never.  Same logic applies to speculative allocation
779          * garbage collection.
780          *
781          * We don't care if this is a normal syncfs call that does this or
782          * freeze that does this - we can run this multiple times without issue
783          * and we won't race with a restart because a restart can only occur
784          * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
785          */
786         if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
787                 xfs_inodegc_stop(mp);
788                 xfs_blockgc_stop(mp);
789         }
790
791         return 0;
792 }
793
794 STATIC int
795 xfs_fs_statfs(
796         struct dentry           *dentry,
797         struct kstatfs          *statp)
798 {
799         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
800         xfs_sb_t                *sbp = &mp->m_sb;
801         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
802         uint64_t                fakeinos, id;
803         uint64_t                icount;
804         uint64_t                ifree;
805         uint64_t                fdblocks;
806         xfs_extlen_t            lsize;
807         int64_t                 ffree;
808
809         /*
810          * Expedite background inodegc but don't wait. We do not want to block
811          * here waiting hours for a billion extent file to be truncated.
812          */
813         xfs_inodegc_push(mp);
814
815         statp->f_type = XFS_SUPER_MAGIC;
816         statp->f_namelen = MAXNAMELEN - 1;
817
818         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
819         statp->f_fsid = u64_to_fsid(id);
820
821         icount = percpu_counter_sum(&mp->m_icount);
822         ifree = percpu_counter_sum(&mp->m_ifree);
823         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
824
825         spin_lock(&mp->m_sb_lock);
826         statp->f_bsize = sbp->sb_blocksize;
827         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
828         statp->f_blocks = sbp->sb_dblocks - lsize;
829         spin_unlock(&mp->m_sb_lock);
830
831         /* make sure statp->f_bfree does not underflow */
832         statp->f_bfree = max_t(int64_t, 0,
833                                 fdblocks - xfs_fdblocks_unavailable(mp));
834         statp->f_bavail = statp->f_bfree;
835
836         fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
837         statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
838         if (M_IGEO(mp)->maxicount)
839                 statp->f_files = min_t(typeof(statp->f_files),
840                                         statp->f_files,
841                                         M_IGEO(mp)->maxicount);
842
843         /* If sb_icount overshot maxicount, report actual allocation */
844         statp->f_files = max_t(typeof(statp->f_files),
845                                         statp->f_files,
846                                         sbp->sb_icount);
847
848         /* make sure statp->f_ffree does not underflow */
849         ffree = statp->f_files - (icount - ifree);
850         statp->f_ffree = max_t(int64_t, ffree, 0);
851
852
853         if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
854             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
855                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
856                 xfs_qm_statvfs(ip, statp);
857
858         if (XFS_IS_REALTIME_MOUNT(mp) &&
859             (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
860                 s64     freertx;
861
862                 statp->f_blocks = sbp->sb_rblocks;
863                 freertx = percpu_counter_sum_positive(&mp->m_frextents);
864                 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
865         }
866
867         return 0;
868 }
869
870 STATIC void
871 xfs_save_resvblks(struct xfs_mount *mp)
872 {
873         uint64_t resblks = 0;
874
875         mp->m_resblks_save = mp->m_resblks;
876         xfs_reserve_blocks(mp, &resblks, NULL);
877 }
878
879 STATIC void
880 xfs_restore_resvblks(struct xfs_mount *mp)
881 {
882         uint64_t resblks;
883
884         if (mp->m_resblks_save) {
885                 resblks = mp->m_resblks_save;
886                 mp->m_resblks_save = 0;
887         } else
888                 resblks = xfs_default_resblks(mp);
889
890         xfs_reserve_blocks(mp, &resblks, NULL);
891 }
892
893 /*
894  * Second stage of a freeze. The data is already frozen so we only
895  * need to take care of the metadata. Once that's done sync the superblock
896  * to the log to dirty it in case of a crash while frozen. This ensures that we
897  * will recover the unlinked inode lists on the next mount.
898  */
899 STATIC int
900 xfs_fs_freeze(
901         struct super_block      *sb)
902 {
903         struct xfs_mount        *mp = XFS_M(sb);
904         unsigned int            flags;
905         int                     ret;
906
907         /*
908          * The filesystem is now frozen far enough that memory reclaim
909          * cannot safely operate on the filesystem. Hence we need to
910          * set a GFP_NOFS context here to avoid recursion deadlocks.
911          */
912         flags = memalloc_nofs_save();
913         xfs_save_resvblks(mp);
914         ret = xfs_log_quiesce(mp);
915         memalloc_nofs_restore(flags);
916
917         /*
918          * For read-write filesystems, we need to restart the inodegc on error
919          * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
920          * going to be run to restart it now.  We are at SB_FREEZE_FS level
921          * here, so we can restart safely without racing with a stop in
922          * xfs_fs_sync_fs().
923          */
924         if (ret && !xfs_is_readonly(mp)) {
925                 xfs_blockgc_start(mp);
926                 xfs_inodegc_start(mp);
927         }
928
929         return ret;
930 }
931
932 STATIC int
933 xfs_fs_unfreeze(
934         struct super_block      *sb)
935 {
936         struct xfs_mount        *mp = XFS_M(sb);
937
938         xfs_restore_resvblks(mp);
939         xfs_log_work_queue(mp);
940
941         /*
942          * Don't reactivate the inodegc worker on a readonly filesystem because
943          * inodes are sent directly to reclaim.  Don't reactivate the blockgc
944          * worker because there are no speculative preallocations on a readonly
945          * filesystem.
946          */
947         if (!xfs_is_readonly(mp)) {
948                 xfs_blockgc_start(mp);
949                 xfs_inodegc_start(mp);
950         }
951
952         return 0;
953 }
954
955 /*
956  * This function fills in xfs_mount_t fields based on mount args.
957  * Note: the superblock _has_ now been read in.
958  */
959 STATIC int
960 xfs_finish_flags(
961         struct xfs_mount        *mp)
962 {
963         /* Fail a mount where the logbuf is smaller than the log stripe */
964         if (xfs_has_logv2(mp)) {
965                 if (mp->m_logbsize <= 0 &&
966                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
967                         mp->m_logbsize = mp->m_sb.sb_logsunit;
968                 } else if (mp->m_logbsize > 0 &&
969                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
970                         xfs_warn(mp,
971                 "logbuf size must be greater than or equal to log stripe size");
972                         return -EINVAL;
973                 }
974         } else {
975                 /* Fail a mount if the logbuf is larger than 32K */
976                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
977                         xfs_warn(mp,
978                 "logbuf size for version 1 logs must be 16K or 32K");
979                         return -EINVAL;
980                 }
981         }
982
983         /*
984          * V5 filesystems always use attr2 format for attributes.
985          */
986         if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
987                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
988                              "attr2 is always enabled for V5 filesystems.");
989                 return -EINVAL;
990         }
991
992         /*
993          * prohibit r/w mounts of read-only filesystems
994          */
995         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
996                 xfs_warn(mp,
997                         "cannot mount a read-only filesystem as read-write");
998                 return -EROFS;
999         }
1000
1001         if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1002             (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1003             !xfs_has_pquotino(mp)) {
1004                 xfs_warn(mp,
1005                   "Super block does not support project and group quota together");
1006                 return -EINVAL;
1007         }
1008
1009         return 0;
1010 }
1011
1012 static int
1013 xfs_init_percpu_counters(
1014         struct xfs_mount        *mp)
1015 {
1016         int             error;
1017
1018         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1019         if (error)
1020                 return -ENOMEM;
1021
1022         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1023         if (error)
1024                 goto free_icount;
1025
1026         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1027         if (error)
1028                 goto free_ifree;
1029
1030         error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1031         if (error)
1032                 goto free_fdblocks;
1033
1034         error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1035         if (error)
1036                 goto free_delalloc;
1037
1038         return 0;
1039
1040 free_delalloc:
1041         percpu_counter_destroy(&mp->m_delalloc_blks);
1042 free_fdblocks:
1043         percpu_counter_destroy(&mp->m_fdblocks);
1044 free_ifree:
1045         percpu_counter_destroy(&mp->m_ifree);
1046 free_icount:
1047         percpu_counter_destroy(&mp->m_icount);
1048         return -ENOMEM;
1049 }
1050
1051 void
1052 xfs_reinit_percpu_counters(
1053         struct xfs_mount        *mp)
1054 {
1055         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1056         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1057         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1058         percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1059 }
1060
1061 static void
1062 xfs_destroy_percpu_counters(
1063         struct xfs_mount        *mp)
1064 {
1065         percpu_counter_destroy(&mp->m_icount);
1066         percpu_counter_destroy(&mp->m_ifree);
1067         percpu_counter_destroy(&mp->m_fdblocks);
1068         ASSERT(xfs_is_shutdown(mp) ||
1069                percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1070         percpu_counter_destroy(&mp->m_delalloc_blks);
1071         percpu_counter_destroy(&mp->m_frextents);
1072 }
1073
1074 static int
1075 xfs_inodegc_init_percpu(
1076         struct xfs_mount        *mp)
1077 {
1078         struct xfs_inodegc      *gc;
1079         int                     cpu;
1080
1081         mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1082         if (!mp->m_inodegc)
1083                 return -ENOMEM;
1084
1085         for_each_possible_cpu(cpu) {
1086                 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1087                 init_llist_head(&gc->list);
1088                 gc->items = 0;
1089                 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1090         }
1091         return 0;
1092 }
1093
1094 static void
1095 xfs_inodegc_free_percpu(
1096         struct xfs_mount        *mp)
1097 {
1098         if (!mp->m_inodegc)
1099                 return;
1100         free_percpu(mp->m_inodegc);
1101 }
1102
1103 static void
1104 xfs_fs_put_super(
1105         struct super_block      *sb)
1106 {
1107         struct xfs_mount        *mp = XFS_M(sb);
1108
1109         /* if ->fill_super failed, we have no mount to tear down */
1110         if (!sb->s_fs_info)
1111                 return;
1112
1113         xfs_notice(mp, "Unmounting Filesystem");
1114         xfs_filestream_unmount(mp);
1115         xfs_unmountfs(mp);
1116
1117         xfs_freesb(mp);
1118         free_percpu(mp->m_stats.xs_stats);
1119         xfs_mount_list_del(mp);
1120         xfs_inodegc_free_percpu(mp);
1121         xfs_destroy_percpu_counters(mp);
1122         xfs_destroy_mount_workqueues(mp);
1123         xfs_close_devices(mp);
1124
1125         sb->s_fs_info = NULL;
1126         xfs_mount_free(mp);
1127 }
1128
1129 static long
1130 xfs_fs_nr_cached_objects(
1131         struct super_block      *sb,
1132         struct shrink_control   *sc)
1133 {
1134         /* Paranoia: catch incorrect calls during mount setup or teardown */
1135         if (WARN_ON_ONCE(!sb->s_fs_info))
1136                 return 0;
1137         return xfs_reclaim_inodes_count(XFS_M(sb));
1138 }
1139
1140 static long
1141 xfs_fs_free_cached_objects(
1142         struct super_block      *sb,
1143         struct shrink_control   *sc)
1144 {
1145         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1146 }
1147
1148 static const struct super_operations xfs_super_operations = {
1149         .alloc_inode            = xfs_fs_alloc_inode,
1150         .destroy_inode          = xfs_fs_destroy_inode,
1151         .dirty_inode            = xfs_fs_dirty_inode,
1152         .drop_inode             = xfs_fs_drop_inode,
1153         .put_super              = xfs_fs_put_super,
1154         .sync_fs                = xfs_fs_sync_fs,
1155         .freeze_fs              = xfs_fs_freeze,
1156         .unfreeze_fs            = xfs_fs_unfreeze,
1157         .statfs                 = xfs_fs_statfs,
1158         .show_options           = xfs_fs_show_options,
1159         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1160         .free_cached_objects    = xfs_fs_free_cached_objects,
1161 };
1162
1163 static int
1164 suffix_kstrtoint(
1165         const char      *s,
1166         unsigned int    base,
1167         int             *res)
1168 {
1169         int             last, shift_left_factor = 0, _res;
1170         char            *value;
1171         int             ret = 0;
1172
1173         value = kstrdup(s, GFP_KERNEL);
1174         if (!value)
1175                 return -ENOMEM;
1176
1177         last = strlen(value) - 1;
1178         if (value[last] == 'K' || value[last] == 'k') {
1179                 shift_left_factor = 10;
1180                 value[last] = '\0';
1181         }
1182         if (value[last] == 'M' || value[last] == 'm') {
1183                 shift_left_factor = 20;
1184                 value[last] = '\0';
1185         }
1186         if (value[last] == 'G' || value[last] == 'g') {
1187                 shift_left_factor = 30;
1188                 value[last] = '\0';
1189         }
1190
1191         if (kstrtoint(value, base, &_res))
1192                 ret = -EINVAL;
1193         kfree(value);
1194         *res = _res << shift_left_factor;
1195         return ret;
1196 }
1197
1198 static inline void
1199 xfs_fs_warn_deprecated(
1200         struct fs_context       *fc,
1201         struct fs_parameter     *param,
1202         uint64_t                flag,
1203         bool                    value)
1204 {
1205         /* Don't print the warning if reconfiguring and current mount point
1206          * already had the flag set
1207          */
1208         if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1209             !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1210                 return;
1211         xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1212 }
1213
1214 /*
1215  * Set mount state from a mount option.
1216  *
1217  * NOTE: mp->m_super is NULL here!
1218  */
1219 static int
1220 xfs_fs_parse_param(
1221         struct fs_context       *fc,
1222         struct fs_parameter     *param)
1223 {
1224         struct xfs_mount        *parsing_mp = fc->s_fs_info;
1225         struct fs_parse_result  result;
1226         int                     size = 0;
1227         int                     opt;
1228
1229         opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1230         if (opt < 0)
1231                 return opt;
1232
1233         switch (opt) {
1234         case Opt_logbufs:
1235                 parsing_mp->m_logbufs = result.uint_32;
1236                 return 0;
1237         case Opt_logbsize:
1238                 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1239                         return -EINVAL;
1240                 return 0;
1241         case Opt_logdev:
1242                 kfree(parsing_mp->m_logname);
1243                 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1244                 if (!parsing_mp->m_logname)
1245                         return -ENOMEM;
1246                 return 0;
1247         case Opt_rtdev:
1248                 kfree(parsing_mp->m_rtname);
1249                 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1250                 if (!parsing_mp->m_rtname)
1251                         return -ENOMEM;
1252                 return 0;
1253         case Opt_allocsize:
1254                 if (suffix_kstrtoint(param->string, 10, &size))
1255                         return -EINVAL;
1256                 parsing_mp->m_allocsize_log = ffs(size) - 1;
1257                 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1258                 return 0;
1259         case Opt_grpid:
1260         case Opt_bsdgroups:
1261                 parsing_mp->m_features |= XFS_FEAT_GRPID;
1262                 return 0;
1263         case Opt_nogrpid:
1264         case Opt_sysvgroups:
1265                 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1266                 return 0;
1267         case Opt_wsync:
1268                 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1269                 return 0;
1270         case Opt_norecovery:
1271                 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1272                 return 0;
1273         case Opt_noalign:
1274                 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1275                 return 0;
1276         case Opt_swalloc:
1277                 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1278                 return 0;
1279         case Opt_sunit:
1280                 parsing_mp->m_dalign = result.uint_32;
1281                 return 0;
1282         case Opt_swidth:
1283                 parsing_mp->m_swidth = result.uint_32;
1284                 return 0;
1285         case Opt_inode32:
1286                 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1287                 return 0;
1288         case Opt_inode64:
1289                 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1290                 return 0;
1291         case Opt_nouuid:
1292                 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1293                 return 0;
1294         case Opt_largeio:
1295                 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1296                 return 0;
1297         case Opt_nolargeio:
1298                 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1299                 return 0;
1300         case Opt_filestreams:
1301                 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1302                 return 0;
1303         case Opt_noquota:
1304                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1305                 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1306                 return 0;
1307         case Opt_quota:
1308         case Opt_uquota:
1309         case Opt_usrquota:
1310                 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1311                 return 0;
1312         case Opt_qnoenforce:
1313         case Opt_uqnoenforce:
1314                 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1315                 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1316                 return 0;
1317         case Opt_pquota:
1318         case Opt_prjquota:
1319                 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1320                 return 0;
1321         case Opt_pqnoenforce:
1322                 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1323                 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1324                 return 0;
1325         case Opt_gquota:
1326         case Opt_grpquota:
1327                 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1328                 return 0;
1329         case Opt_gqnoenforce:
1330                 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1331                 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1332                 return 0;
1333         case Opt_discard:
1334                 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1335                 return 0;
1336         case Opt_nodiscard:
1337                 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1338                 return 0;
1339 #ifdef CONFIG_FS_DAX
1340         case Opt_dax:
1341                 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1342                 return 0;
1343         case Opt_dax_enum:
1344                 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1345                 return 0;
1346 #endif
1347         /* Following mount options will be removed in September 2025 */
1348         case Opt_ikeep:
1349                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1350                 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1351                 return 0;
1352         case Opt_noikeep:
1353                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1354                 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1355                 return 0;
1356         case Opt_attr2:
1357                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1358                 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1359                 return 0;
1360         case Opt_noattr2:
1361                 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1362                 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1363                 return 0;
1364         default:
1365                 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1366                 return -EINVAL;
1367         }
1368
1369         return 0;
1370 }
1371
1372 static int
1373 xfs_fs_validate_params(
1374         struct xfs_mount        *mp)
1375 {
1376         /* No recovery flag requires a read-only mount */
1377         if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1378                 xfs_warn(mp, "no-recovery mounts must be read-only.");
1379                 return -EINVAL;
1380         }
1381
1382         /*
1383          * We have not read the superblock at this point, so only the attr2
1384          * mount option can set the attr2 feature by this stage.
1385          */
1386         if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1387                 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1388                 return -EINVAL;
1389         }
1390
1391
1392         if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1393                 xfs_warn(mp,
1394         "sunit and swidth options incompatible with the noalign option");
1395                 return -EINVAL;
1396         }
1397
1398         if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1399                 xfs_warn(mp, "quota support not available in this kernel.");
1400                 return -EINVAL;
1401         }
1402
1403         if ((mp->m_dalign && !mp->m_swidth) ||
1404             (!mp->m_dalign && mp->m_swidth)) {
1405                 xfs_warn(mp, "sunit and swidth must be specified together");
1406                 return -EINVAL;
1407         }
1408
1409         if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1410                 xfs_warn(mp,
1411         "stripe width (%d) must be a multiple of the stripe unit (%d)",
1412                         mp->m_swidth, mp->m_dalign);
1413                 return -EINVAL;
1414         }
1415
1416         if (mp->m_logbufs != -1 &&
1417             mp->m_logbufs != 0 &&
1418             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1419              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1420                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1421                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1422                 return -EINVAL;
1423         }
1424
1425         if (mp->m_logbsize != -1 &&
1426             mp->m_logbsize !=  0 &&
1427             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1428              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1429              !is_power_of_2(mp->m_logbsize))) {
1430                 xfs_warn(mp,
1431                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1432                         mp->m_logbsize);
1433                 return -EINVAL;
1434         }
1435
1436         if (xfs_has_allocsize(mp) &&
1437             (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1438              mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1439                 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1440                         mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1441                 return -EINVAL;
1442         }
1443
1444         return 0;
1445 }
1446
1447 static int
1448 xfs_fs_fill_super(
1449         struct super_block      *sb,
1450         struct fs_context       *fc)
1451 {
1452         struct xfs_mount        *mp = sb->s_fs_info;
1453         struct inode            *root;
1454         int                     flags = 0, error;
1455
1456         mp->m_super = sb;
1457
1458         error = xfs_fs_validate_params(mp);
1459         if (error)
1460                 goto out_free_names;
1461
1462         sb_min_blocksize(sb, BBSIZE);
1463         sb->s_xattr = xfs_xattr_handlers;
1464         sb->s_export_op = &xfs_export_operations;
1465 #ifdef CONFIG_XFS_QUOTA
1466         sb->s_qcop = &xfs_quotactl_operations;
1467         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1468 #endif
1469         sb->s_op = &xfs_super_operations;
1470
1471         /*
1472          * Delay mount work if the debug hook is set. This is debug
1473          * instrumention to coordinate simulation of xfs mount failures with
1474          * VFS superblock operations
1475          */
1476         if (xfs_globals.mount_delay) {
1477                 xfs_notice(mp, "Delaying mount for %d seconds.",
1478                         xfs_globals.mount_delay);
1479                 msleep(xfs_globals.mount_delay * 1000);
1480         }
1481
1482         if (fc->sb_flags & SB_SILENT)
1483                 flags |= XFS_MFSI_QUIET;
1484
1485         error = xfs_open_devices(mp);
1486         if (error)
1487                 goto out_free_names;
1488
1489         error = xfs_init_mount_workqueues(mp);
1490         if (error)
1491                 goto out_close_devices;
1492
1493         error = xfs_init_percpu_counters(mp);
1494         if (error)
1495                 goto out_destroy_workqueues;
1496
1497         error = xfs_inodegc_init_percpu(mp);
1498         if (error)
1499                 goto out_destroy_counters;
1500
1501         /*
1502          * All percpu data structures requiring cleanup when a cpu goes offline
1503          * must be allocated before adding this @mp to the cpu-dead handler's
1504          * mount list.
1505          */
1506         xfs_mount_list_add(mp);
1507
1508         /* Allocate stats memory before we do operations that might use it */
1509         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1510         if (!mp->m_stats.xs_stats) {
1511                 error = -ENOMEM;
1512                 goto out_destroy_inodegc;
1513         }
1514
1515         error = xfs_readsb(mp, flags);
1516         if (error)
1517                 goto out_free_stats;
1518
1519         error = xfs_finish_flags(mp);
1520         if (error)
1521                 goto out_free_sb;
1522
1523         error = xfs_setup_devices(mp);
1524         if (error)
1525                 goto out_free_sb;
1526
1527         /* V4 support is undergoing deprecation. */
1528         if (!xfs_has_crc(mp)) {
1529 #ifdef CONFIG_XFS_SUPPORT_V4
1530                 xfs_warn_once(mp,
1531         "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1532 #else
1533                 xfs_warn(mp,
1534         "Deprecated V4 format (crc=0) not supported by kernel.");
1535                 error = -EINVAL;
1536                 goto out_free_sb;
1537 #endif
1538         }
1539
1540         /* Filesystem claims it needs repair, so refuse the mount. */
1541         if (xfs_has_needsrepair(mp)) {
1542                 xfs_warn(mp, "Filesystem needs repair.  Please run xfs_repair.");
1543                 error = -EFSCORRUPTED;
1544                 goto out_free_sb;
1545         }
1546
1547         /*
1548          * Don't touch the filesystem if a user tool thinks it owns the primary
1549          * superblock.  mkfs doesn't clear the flag from secondary supers, so
1550          * we don't check them at all.
1551          */
1552         if (mp->m_sb.sb_inprogress) {
1553                 xfs_warn(mp, "Offline file system operation in progress!");
1554                 error = -EFSCORRUPTED;
1555                 goto out_free_sb;
1556         }
1557
1558         /*
1559          * Until this is fixed only page-sized or smaller data blocks work.
1560          */
1561         if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1562                 xfs_warn(mp,
1563                 "File system with blocksize %d bytes. "
1564                 "Only pagesize (%ld) or less will currently work.",
1565                                 mp->m_sb.sb_blocksize, PAGE_SIZE);
1566                 error = -ENOSYS;
1567                 goto out_free_sb;
1568         }
1569
1570         /* Ensure this filesystem fits in the page cache limits */
1571         if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1572             xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1573                 xfs_warn(mp,
1574                 "file system too large to be mounted on this system.");
1575                 error = -EFBIG;
1576                 goto out_free_sb;
1577         }
1578
1579         /*
1580          * XFS block mappings use 54 bits to store the logical block offset.
1581          * This should suffice to handle the maximum file size that the VFS
1582          * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1583          * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1584          * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1585          * to check this assertion.
1586          *
1587          * Avoid integer overflow by comparing the maximum bmbt offset to the
1588          * maximum pagecache offset in units of fs blocks.
1589          */
1590         if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1591                 xfs_warn(mp,
1592 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1593                          XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1594                          XFS_MAX_FILEOFF);
1595                 error = -EINVAL;
1596                 goto out_free_sb;
1597         }
1598
1599         error = xfs_filestream_mount(mp);
1600         if (error)
1601                 goto out_free_sb;
1602
1603         /*
1604          * we must configure the block size in the superblock before we run the
1605          * full mount process as the mount process can lookup and cache inodes.
1606          */
1607         sb->s_magic = XFS_SUPER_MAGIC;
1608         sb->s_blocksize = mp->m_sb.sb_blocksize;
1609         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1610         sb->s_maxbytes = MAX_LFS_FILESIZE;
1611         sb->s_max_links = XFS_MAXLINK;
1612         sb->s_time_gran = 1;
1613         if (xfs_has_bigtime(mp)) {
1614                 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1615                 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1616         } else {
1617                 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1618                 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1619         }
1620         trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1621         sb->s_iflags |= SB_I_CGROUPWB;
1622
1623         set_posix_acl_flag(sb);
1624
1625         /* version 5 superblocks support inode version counters. */
1626         if (xfs_has_crc(mp))
1627                 sb->s_flags |= SB_I_VERSION;
1628
1629         if (xfs_has_dax_always(mp)) {
1630                 error = xfs_setup_dax_always(mp);
1631                 if (error)
1632                         goto out_filestream_unmount;
1633         }
1634
1635         if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1636                 xfs_warn(mp,
1637         "mounting with \"discard\" option, but the device does not support discard");
1638                 mp->m_features &= ~XFS_FEAT_DISCARD;
1639         }
1640
1641         if (xfs_has_reflink(mp)) {
1642                 if (mp->m_sb.sb_rblocks) {
1643                         xfs_alert(mp,
1644         "reflink not compatible with realtime device!");
1645                         error = -EINVAL;
1646                         goto out_filestream_unmount;
1647                 }
1648
1649                 if (xfs_globals.always_cow) {
1650                         xfs_info(mp, "using DEBUG-only always_cow mode.");
1651                         mp->m_always_cow = true;
1652                 }
1653         }
1654
1655         if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1656                 xfs_alert(mp,
1657         "reverse mapping btree not compatible with realtime device!");
1658                 error = -EINVAL;
1659                 goto out_filestream_unmount;
1660         }
1661
1662         if (xfs_has_large_extent_counts(mp))
1663                 xfs_warn(mp,
1664         "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1665
1666         error = xfs_mountfs(mp);
1667         if (error)
1668                 goto out_filestream_unmount;
1669
1670         root = igrab(VFS_I(mp->m_rootip));
1671         if (!root) {
1672                 error = -ENOENT;
1673                 goto out_unmount;
1674         }
1675         sb->s_root = d_make_root(root);
1676         if (!sb->s_root) {
1677                 error = -ENOMEM;
1678                 goto out_unmount;
1679         }
1680
1681         return 0;
1682
1683  out_filestream_unmount:
1684         xfs_filestream_unmount(mp);
1685  out_free_sb:
1686         xfs_freesb(mp);
1687  out_free_stats:
1688         free_percpu(mp->m_stats.xs_stats);
1689  out_destroy_inodegc:
1690         xfs_mount_list_del(mp);
1691         xfs_inodegc_free_percpu(mp);
1692  out_destroy_counters:
1693         xfs_destroy_percpu_counters(mp);
1694  out_destroy_workqueues:
1695         xfs_destroy_mount_workqueues(mp);
1696  out_close_devices:
1697         xfs_close_devices(mp);
1698  out_free_names:
1699         sb->s_fs_info = NULL;
1700         xfs_mount_free(mp);
1701         return error;
1702
1703  out_unmount:
1704         xfs_filestream_unmount(mp);
1705         xfs_unmountfs(mp);
1706         goto out_free_sb;
1707 }
1708
1709 static int
1710 xfs_fs_get_tree(
1711         struct fs_context       *fc)
1712 {
1713         return get_tree_bdev(fc, xfs_fs_fill_super);
1714 }
1715
1716 static int
1717 xfs_remount_rw(
1718         struct xfs_mount        *mp)
1719 {
1720         struct xfs_sb           *sbp = &mp->m_sb;
1721         int error;
1722
1723         if (xfs_has_norecovery(mp)) {
1724                 xfs_warn(mp,
1725                         "ro->rw transition prohibited on norecovery mount");
1726                 return -EINVAL;
1727         }
1728
1729         if (xfs_sb_is_v5(sbp) &&
1730             xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1731                 xfs_warn(mp,
1732         "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1733                         (sbp->sb_features_ro_compat &
1734                                 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1735                 return -EINVAL;
1736         }
1737
1738         clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1739
1740         /*
1741          * If this is the first remount to writeable state we might have some
1742          * superblock changes to update.
1743          */
1744         if (mp->m_update_sb) {
1745                 error = xfs_sync_sb(mp, false);
1746                 if (error) {
1747                         xfs_warn(mp, "failed to write sb changes");
1748                         return error;
1749                 }
1750                 mp->m_update_sb = false;
1751         }
1752
1753         /*
1754          * Fill out the reserve pool if it is empty. Use the stashed value if
1755          * it is non-zero, otherwise go with the default.
1756          */
1757         xfs_restore_resvblks(mp);
1758         xfs_log_work_queue(mp);
1759         xfs_blockgc_start(mp);
1760
1761         /* Create the per-AG metadata reservation pool .*/
1762         error = xfs_fs_reserve_ag_blocks(mp);
1763         if (error && error != -ENOSPC)
1764                 return error;
1765
1766         /* Re-enable the background inode inactivation worker. */
1767         xfs_inodegc_start(mp);
1768
1769         return 0;
1770 }
1771
1772 static int
1773 xfs_remount_ro(
1774         struct xfs_mount        *mp)
1775 {
1776         struct xfs_icwalk       icw = {
1777                 .icw_flags      = XFS_ICWALK_FLAG_SYNC,
1778         };
1779         int                     error;
1780
1781         /* Flush all the dirty data to disk. */
1782         error = sync_filesystem(mp->m_super);
1783         if (error)
1784                 return error;
1785
1786         /*
1787          * Cancel background eofb scanning so it cannot race with the final
1788          * log force+buftarg wait and deadlock the remount.
1789          */
1790         xfs_blockgc_stop(mp);
1791
1792         /*
1793          * Clear out all remaining COW staging extents and speculative post-EOF
1794          * preallocations so that we don't leave inodes requiring inactivation
1795          * cleanups during reclaim on a read-only mount.  We must process every
1796          * cached inode, so this requires a synchronous cache scan.
1797          */
1798         error = xfs_blockgc_free_space(mp, &icw);
1799         if (error) {
1800                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1801                 return error;
1802         }
1803
1804         /*
1805          * Stop the inodegc background worker.  xfs_fs_reconfigure already
1806          * flushed all pending inodegc work when it sync'd the filesystem.
1807          * The VFS holds s_umount, so we know that inodes cannot enter
1808          * xfs_fs_destroy_inode during a remount operation.  In readonly mode
1809          * we send inodes straight to reclaim, so no inodes will be queued.
1810          */
1811         xfs_inodegc_stop(mp);
1812
1813         /* Free the per-AG metadata reservation pool. */
1814         error = xfs_fs_unreserve_ag_blocks(mp);
1815         if (error) {
1816                 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1817                 return error;
1818         }
1819
1820         /*
1821          * Before we sync the metadata, we need to free up the reserve block
1822          * pool so that the used block count in the superblock on disk is
1823          * correct at the end of the remount. Stash the current* reserve pool
1824          * size so that if we get remounted rw, we can return it to the same
1825          * size.
1826          */
1827         xfs_save_resvblks(mp);
1828
1829         xfs_log_clean(mp);
1830         set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1831
1832         return 0;
1833 }
1834
1835 /*
1836  * Logically we would return an error here to prevent users from believing
1837  * they might have changed mount options using remount which can't be changed.
1838  *
1839  * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1840  * arguments in some cases so we can't blindly reject options, but have to
1841  * check for each specified option if it actually differs from the currently
1842  * set option and only reject it if that's the case.
1843  *
1844  * Until that is implemented we return success for every remount request, and
1845  * silently ignore all options that we can't actually change.
1846  */
1847 static int
1848 xfs_fs_reconfigure(
1849         struct fs_context *fc)
1850 {
1851         struct xfs_mount        *mp = XFS_M(fc->root->d_sb);
1852         struct xfs_mount        *new_mp = fc->s_fs_info;
1853         int                     flags = fc->sb_flags;
1854         int                     error;
1855
1856         /* version 5 superblocks always support version counters. */
1857         if (xfs_has_crc(mp))
1858                 fc->sb_flags |= SB_I_VERSION;
1859
1860         error = xfs_fs_validate_params(new_mp);
1861         if (error)
1862                 return error;
1863
1864         /* inode32 -> inode64 */
1865         if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1866                 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1867                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1868         }
1869
1870         /* inode64 -> inode32 */
1871         if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1872                 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1873                 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1874         }
1875
1876         /* ro -> rw */
1877         if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1878                 error = xfs_remount_rw(mp);
1879                 if (error)
1880                         return error;
1881         }
1882
1883         /* rw -> ro */
1884         if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1885                 error = xfs_remount_ro(mp);
1886                 if (error)
1887                         return error;
1888         }
1889
1890         return 0;
1891 }
1892
1893 static void xfs_fs_free(
1894         struct fs_context       *fc)
1895 {
1896         struct xfs_mount        *mp = fc->s_fs_info;
1897
1898         /*
1899          * mp is stored in the fs_context when it is initialized.
1900          * mp is transferred to the superblock on a successful mount,
1901          * but if an error occurs before the transfer we have to free
1902          * it here.
1903          */
1904         if (mp)
1905                 xfs_mount_free(mp);
1906 }
1907
1908 static const struct fs_context_operations xfs_context_ops = {
1909         .parse_param = xfs_fs_parse_param,
1910         .get_tree    = xfs_fs_get_tree,
1911         .reconfigure = xfs_fs_reconfigure,
1912         .free        = xfs_fs_free,
1913 };
1914
1915 static int xfs_init_fs_context(
1916         struct fs_context       *fc)
1917 {
1918         struct xfs_mount        *mp;
1919
1920         mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1921         if (!mp)
1922                 return -ENOMEM;
1923
1924         spin_lock_init(&mp->m_sb_lock);
1925         spin_lock_init(&mp->m_agirotor_lock);
1926         INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1927         spin_lock_init(&mp->m_perag_lock);
1928         mutex_init(&mp->m_growlock);
1929         INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1930         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1931         mp->m_kobj.kobject.kset = xfs_kset;
1932         /*
1933          * We don't create the finobt per-ag space reservation until after log
1934          * recovery, so we must set this to true so that an ifree transaction
1935          * started during log recovery will not depend on space reservations
1936          * for finobt expansion.
1937          */
1938         mp->m_finobt_nores = true;
1939
1940         /*
1941          * These can be overridden by the mount option parsing.
1942          */
1943         mp->m_logbufs = -1;
1944         mp->m_logbsize = -1;
1945         mp->m_allocsize_log = 16; /* 64k */
1946
1947         /*
1948          * Copy binary VFS mount flags we are interested in.
1949          */
1950         if (fc->sb_flags & SB_RDONLY)
1951                 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1952         if (fc->sb_flags & SB_DIRSYNC)
1953                 mp->m_features |= XFS_FEAT_DIRSYNC;
1954         if (fc->sb_flags & SB_SYNCHRONOUS)
1955                 mp->m_features |= XFS_FEAT_WSYNC;
1956
1957         fc->s_fs_info = mp;
1958         fc->ops = &xfs_context_ops;
1959
1960         return 0;
1961 }
1962
1963 static struct file_system_type xfs_fs_type = {
1964         .owner                  = THIS_MODULE,
1965         .name                   = "xfs",
1966         .init_fs_context        = xfs_init_fs_context,
1967         .parameters             = xfs_fs_parameters,
1968         .kill_sb                = kill_block_super,
1969         .fs_flags               = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1970 };
1971 MODULE_ALIAS_FS("xfs");
1972
1973 STATIC int __init
1974 xfs_init_caches(void)
1975 {
1976         int             error;
1977
1978         xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
1979                                          SLAB_HWCACHE_ALIGN |
1980                                          SLAB_RECLAIM_ACCOUNT |
1981                                          SLAB_MEM_SPREAD,
1982                                          NULL);
1983         if (!xfs_buf_cache)
1984                 goto out;
1985
1986         xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1987                                                 sizeof(struct xlog_ticket),
1988                                                 0, 0, NULL);
1989         if (!xfs_log_ticket_cache)
1990                 goto out_destroy_buf_cache;
1991
1992         error = xfs_btree_init_cur_caches();
1993         if (error)
1994                 goto out_destroy_log_ticket_cache;
1995
1996         error = xfs_defer_init_item_caches();
1997         if (error)
1998                 goto out_destroy_btree_cur_cache;
1999
2000         xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2001                                               sizeof(struct xfs_da_state),
2002                                               0, 0, NULL);
2003         if (!xfs_da_state_cache)
2004                 goto out_destroy_defer_item_cache;
2005
2006         xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2007                                            sizeof(struct xfs_ifork),
2008                                            0, 0, NULL);
2009         if (!xfs_ifork_cache)
2010                 goto out_destroy_da_state_cache;
2011
2012         xfs_trans_cache = kmem_cache_create("xfs_trans",
2013                                            sizeof(struct xfs_trans),
2014                                            0, 0, NULL);
2015         if (!xfs_trans_cache)
2016                 goto out_destroy_ifork_cache;
2017
2018
2019         /*
2020          * The size of the cache-allocated buf log item is the maximum
2021          * size possible under XFS.  This wastes a little bit of memory,
2022          * but it is much faster.
2023          */
2024         xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2025                                               sizeof(struct xfs_buf_log_item),
2026                                               0, 0, NULL);
2027         if (!xfs_buf_item_cache)
2028                 goto out_destroy_trans_cache;
2029
2030         xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2031                                         (sizeof(struct xfs_efd_log_item) +
2032                                         (XFS_EFD_MAX_FAST_EXTENTS - 1) *
2033                                         sizeof(struct xfs_extent)),
2034                                         0, 0, NULL);
2035         if (!xfs_efd_cache)
2036                 goto out_destroy_buf_item_cache;
2037
2038         xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2039                                          (sizeof(struct xfs_efi_log_item) +
2040                                          (XFS_EFI_MAX_FAST_EXTENTS - 1) *
2041                                          sizeof(struct xfs_extent)),
2042                                          0, 0, NULL);
2043         if (!xfs_efi_cache)
2044                 goto out_destroy_efd_cache;
2045
2046         xfs_inode_cache = kmem_cache_create("xfs_inode",
2047                                            sizeof(struct xfs_inode), 0,
2048                                            (SLAB_HWCACHE_ALIGN |
2049                                             SLAB_RECLAIM_ACCOUNT |
2050                                             SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2051                                            xfs_fs_inode_init_once);
2052         if (!xfs_inode_cache)
2053                 goto out_destroy_efi_cache;
2054
2055         xfs_ili_cache = kmem_cache_create("xfs_ili",
2056                                          sizeof(struct xfs_inode_log_item), 0,
2057                                          SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2058                                          NULL);
2059         if (!xfs_ili_cache)
2060                 goto out_destroy_inode_cache;
2061
2062         xfs_icreate_cache = kmem_cache_create("xfs_icr",
2063                                              sizeof(struct xfs_icreate_item),
2064                                              0, 0, NULL);
2065         if (!xfs_icreate_cache)
2066                 goto out_destroy_ili_cache;
2067
2068         xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2069                                          sizeof(struct xfs_rud_log_item),
2070                                          0, 0, NULL);
2071         if (!xfs_rud_cache)
2072                 goto out_destroy_icreate_cache;
2073
2074         xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2075                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2076                         0, 0, NULL);
2077         if (!xfs_rui_cache)
2078                 goto out_destroy_rud_cache;
2079
2080         xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2081                                          sizeof(struct xfs_cud_log_item),
2082                                          0, 0, NULL);
2083         if (!xfs_cud_cache)
2084                 goto out_destroy_rui_cache;
2085
2086         xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2087                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2088                         0, 0, NULL);
2089         if (!xfs_cui_cache)
2090                 goto out_destroy_cud_cache;
2091
2092         xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2093                                          sizeof(struct xfs_bud_log_item),
2094                                          0, 0, NULL);
2095         if (!xfs_bud_cache)
2096                 goto out_destroy_cui_cache;
2097
2098         xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2099                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2100                         0, 0, NULL);
2101         if (!xfs_bui_cache)
2102                 goto out_destroy_bud_cache;
2103
2104         xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2105                                             sizeof(struct xfs_attrd_log_item),
2106                                             0, 0, NULL);
2107         if (!xfs_attrd_cache)
2108                 goto out_destroy_bui_cache;
2109
2110         xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2111                                             sizeof(struct xfs_attri_log_item),
2112                                             0, 0, NULL);
2113         if (!xfs_attri_cache)
2114                 goto out_destroy_attrd_cache;
2115
2116         xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2117                                              sizeof(struct xfs_iunlink_item),
2118                                              0, 0, NULL);
2119         if (!xfs_iunlink_cache)
2120                 goto out_destroy_attri_cache;
2121
2122         return 0;
2123
2124  out_destroy_attri_cache:
2125         kmem_cache_destroy(xfs_attri_cache);
2126  out_destroy_attrd_cache:
2127         kmem_cache_destroy(xfs_attrd_cache);
2128  out_destroy_bui_cache:
2129         kmem_cache_destroy(xfs_bui_cache);
2130  out_destroy_bud_cache:
2131         kmem_cache_destroy(xfs_bud_cache);
2132  out_destroy_cui_cache:
2133         kmem_cache_destroy(xfs_cui_cache);
2134  out_destroy_cud_cache:
2135         kmem_cache_destroy(xfs_cud_cache);
2136  out_destroy_rui_cache:
2137         kmem_cache_destroy(xfs_rui_cache);
2138  out_destroy_rud_cache:
2139         kmem_cache_destroy(xfs_rud_cache);
2140  out_destroy_icreate_cache:
2141         kmem_cache_destroy(xfs_icreate_cache);
2142  out_destroy_ili_cache:
2143         kmem_cache_destroy(xfs_ili_cache);
2144  out_destroy_inode_cache:
2145         kmem_cache_destroy(xfs_inode_cache);
2146  out_destroy_efi_cache:
2147         kmem_cache_destroy(xfs_efi_cache);
2148  out_destroy_efd_cache:
2149         kmem_cache_destroy(xfs_efd_cache);
2150  out_destroy_buf_item_cache:
2151         kmem_cache_destroy(xfs_buf_item_cache);
2152  out_destroy_trans_cache:
2153         kmem_cache_destroy(xfs_trans_cache);
2154  out_destroy_ifork_cache:
2155         kmem_cache_destroy(xfs_ifork_cache);
2156  out_destroy_da_state_cache:
2157         kmem_cache_destroy(xfs_da_state_cache);
2158  out_destroy_defer_item_cache:
2159         xfs_defer_destroy_item_caches();
2160  out_destroy_btree_cur_cache:
2161         xfs_btree_destroy_cur_caches();
2162  out_destroy_log_ticket_cache:
2163         kmem_cache_destroy(xfs_log_ticket_cache);
2164  out_destroy_buf_cache:
2165         kmem_cache_destroy(xfs_buf_cache);
2166  out:
2167         return -ENOMEM;
2168 }
2169
2170 STATIC void
2171 xfs_destroy_caches(void)
2172 {
2173         /*
2174          * Make sure all delayed rcu free are flushed before we
2175          * destroy caches.
2176          */
2177         rcu_barrier();
2178         kmem_cache_destroy(xfs_iunlink_cache);
2179         kmem_cache_destroy(xfs_attri_cache);
2180         kmem_cache_destroy(xfs_attrd_cache);
2181         kmem_cache_destroy(xfs_bui_cache);
2182         kmem_cache_destroy(xfs_bud_cache);
2183         kmem_cache_destroy(xfs_cui_cache);
2184         kmem_cache_destroy(xfs_cud_cache);
2185         kmem_cache_destroy(xfs_rui_cache);
2186         kmem_cache_destroy(xfs_rud_cache);
2187         kmem_cache_destroy(xfs_icreate_cache);
2188         kmem_cache_destroy(xfs_ili_cache);
2189         kmem_cache_destroy(xfs_inode_cache);
2190         kmem_cache_destroy(xfs_efi_cache);
2191         kmem_cache_destroy(xfs_efd_cache);
2192         kmem_cache_destroy(xfs_buf_item_cache);
2193         kmem_cache_destroy(xfs_trans_cache);
2194         kmem_cache_destroy(xfs_ifork_cache);
2195         kmem_cache_destroy(xfs_da_state_cache);
2196         xfs_defer_destroy_item_caches();
2197         xfs_btree_destroy_cur_caches();
2198         kmem_cache_destroy(xfs_log_ticket_cache);
2199         kmem_cache_destroy(xfs_buf_cache);
2200 }
2201
2202 STATIC int __init
2203 xfs_init_workqueues(void)
2204 {
2205         /*
2206          * The allocation workqueue can be used in memory reclaim situations
2207          * (writepage path), and parallelism is only limited by the number of
2208          * AGs in all the filesystems mounted. Hence use the default large
2209          * max_active value for this workqueue.
2210          */
2211         xfs_alloc_wq = alloc_workqueue("xfsalloc",
2212                         XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2213         if (!xfs_alloc_wq)
2214                 return -ENOMEM;
2215
2216         xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2217                         0);
2218         if (!xfs_discard_wq)
2219                 goto out_free_alloc_wq;
2220
2221         return 0;
2222 out_free_alloc_wq:
2223         destroy_workqueue(xfs_alloc_wq);
2224         return -ENOMEM;
2225 }
2226
2227 STATIC void
2228 xfs_destroy_workqueues(void)
2229 {
2230         destroy_workqueue(xfs_discard_wq);
2231         destroy_workqueue(xfs_alloc_wq);
2232 }
2233
2234 #ifdef CONFIG_HOTPLUG_CPU
2235 static int
2236 xfs_cpu_dead(
2237         unsigned int            cpu)
2238 {
2239         struct xfs_mount        *mp, *n;
2240
2241         spin_lock(&xfs_mount_list_lock);
2242         list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2243                 spin_unlock(&xfs_mount_list_lock);
2244                 xfs_inodegc_cpu_dead(mp, cpu);
2245                 xlog_cil_pcp_dead(mp->m_log, cpu);
2246                 spin_lock(&xfs_mount_list_lock);
2247         }
2248         spin_unlock(&xfs_mount_list_lock);
2249         return 0;
2250 }
2251
2252 static int __init
2253 xfs_cpu_hotplug_init(void)
2254 {
2255         int     error;
2256
2257         error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2258                         xfs_cpu_dead);
2259         if (error < 0)
2260                 xfs_alert(NULL,
2261 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2262                         error);
2263         return error;
2264 }
2265
2266 static void
2267 xfs_cpu_hotplug_destroy(void)
2268 {
2269         cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2270 }
2271
2272 #else /* !CONFIG_HOTPLUG_CPU */
2273 static inline int xfs_cpu_hotplug_init(void) { return 0; }
2274 static inline void xfs_cpu_hotplug_destroy(void) {}
2275 #endif
2276
2277 STATIC int __init
2278 init_xfs_fs(void)
2279 {
2280         int                     error;
2281
2282         xfs_check_ondisk_structs();
2283
2284         printk(KERN_INFO XFS_VERSION_STRING " with "
2285                          XFS_BUILD_OPTIONS " enabled\n");
2286
2287         xfs_dir_startup();
2288
2289         error = xfs_cpu_hotplug_init();
2290         if (error)
2291                 goto out;
2292
2293         error = xfs_init_caches();
2294         if (error)
2295                 goto out_destroy_hp;
2296
2297         error = xfs_init_workqueues();
2298         if (error)
2299                 goto out_destroy_caches;
2300
2301         error = xfs_mru_cache_init();
2302         if (error)
2303                 goto out_destroy_wq;
2304
2305         error = xfs_init_procfs();
2306         if (error)
2307                 goto out_mru_cache_uninit;
2308
2309         error = xfs_sysctl_register();
2310         if (error)
2311                 goto out_cleanup_procfs;
2312
2313         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2314         if (!xfs_kset) {
2315                 error = -ENOMEM;
2316                 goto out_sysctl_unregister;
2317         }
2318
2319         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2320
2321         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2322         if (!xfsstats.xs_stats) {
2323                 error = -ENOMEM;
2324                 goto out_kset_unregister;
2325         }
2326
2327         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2328                                "stats");
2329         if (error)
2330                 goto out_free_stats;
2331
2332 #ifdef DEBUG
2333         xfs_dbg_kobj.kobject.kset = xfs_kset;
2334         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2335         if (error)
2336                 goto out_remove_stats_kobj;
2337 #endif
2338
2339         error = xfs_qm_init();
2340         if (error)
2341                 goto out_remove_dbg_kobj;
2342
2343         error = register_filesystem(&xfs_fs_type);
2344         if (error)
2345                 goto out_qm_exit;
2346         return 0;
2347
2348  out_qm_exit:
2349         xfs_qm_exit();
2350  out_remove_dbg_kobj:
2351 #ifdef DEBUG
2352         xfs_sysfs_del(&xfs_dbg_kobj);
2353  out_remove_stats_kobj:
2354 #endif
2355         xfs_sysfs_del(&xfsstats.xs_kobj);
2356  out_free_stats:
2357         free_percpu(xfsstats.xs_stats);
2358  out_kset_unregister:
2359         kset_unregister(xfs_kset);
2360  out_sysctl_unregister:
2361         xfs_sysctl_unregister();
2362  out_cleanup_procfs:
2363         xfs_cleanup_procfs();
2364  out_mru_cache_uninit:
2365         xfs_mru_cache_uninit();
2366  out_destroy_wq:
2367         xfs_destroy_workqueues();
2368  out_destroy_caches:
2369         xfs_destroy_caches();
2370  out_destroy_hp:
2371         xfs_cpu_hotplug_destroy();
2372  out:
2373         return error;
2374 }
2375
2376 STATIC void __exit
2377 exit_xfs_fs(void)
2378 {
2379         xfs_qm_exit();
2380         unregister_filesystem(&xfs_fs_type);
2381 #ifdef DEBUG
2382         xfs_sysfs_del(&xfs_dbg_kobj);
2383 #endif
2384         xfs_sysfs_del(&xfsstats.xs_kobj);
2385         free_percpu(xfsstats.xs_stats);
2386         kset_unregister(xfs_kset);
2387         xfs_sysctl_unregister();
2388         xfs_cleanup_procfs();
2389         xfs_mru_cache_uninit();
2390         xfs_destroy_workqueues();
2391         xfs_destroy_caches();
2392         xfs_uuid_table_free();
2393         xfs_cpu_hotplug_destroy();
2394 }
2395
2396 module_init(init_xfs_fs);
2397 module_exit(exit_xfs_fs);
2398
2399 MODULE_AUTHOR("Silicon Graphics, Inc.");
2400 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2401 MODULE_LICENSE("GPL");