1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
22 #include "xfs_log_priv.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"
39 #include <linux/magic.h>
40 #include <linux/fs_context.h>
41 #include <linux/fs_parser.h>
43 static const struct super_operations xfs_super_operations
;
45 static struct kset
*xfs_kset
; /* top-level xfs sysfs dir */
47 static struct xfs_kobj xfs_dbg_kobj
; /* global debug sysfs attrs */
57 xfs_mount_set_dax_mode(
59 enum xfs_dax_mode mode
)
63 mp
->m_flags
&= ~(XFS_MOUNT_DAX_ALWAYS
| XFS_MOUNT_DAX_NEVER
);
66 mp
->m_flags
|= XFS_MOUNT_DAX_ALWAYS
;
67 mp
->m_flags
&= ~XFS_MOUNT_DAX_NEVER
;
70 mp
->m_flags
|= XFS_MOUNT_DAX_NEVER
;
71 mp
->m_flags
&= ~XFS_MOUNT_DAX_ALWAYS
;
76 static const struct constant_table dax_param_enums
[] = {
77 {"inode", XFS_DAX_INODE
},
78 {"always", XFS_DAX_ALWAYS
},
79 {"never", XFS_DAX_NEVER
},
84 * Table driven mount option parser.
87 Opt_logbufs
, Opt_logbsize
, Opt_logdev
, Opt_rtdev
,
88 Opt_wsync
, Opt_noalign
, Opt_swalloc
, Opt_sunit
, Opt_swidth
, Opt_nouuid
,
89 Opt_grpid
, Opt_nogrpid
, Opt_bsdgroups
, Opt_sysvgroups
,
90 Opt_allocsize
, Opt_norecovery
, Opt_inode64
, Opt_inode32
, Opt_ikeep
,
91 Opt_noikeep
, Opt_largeio
, Opt_nolargeio
, Opt_attr2
, Opt_noattr2
,
92 Opt_filestreams
, Opt_quota
, Opt_noquota
, Opt_usrquota
, Opt_grpquota
,
93 Opt_prjquota
, Opt_uquota
, Opt_gquota
, Opt_pquota
,
94 Opt_uqnoenforce
, Opt_gqnoenforce
, Opt_pqnoenforce
, Opt_qnoenforce
,
95 Opt_discard
, Opt_nodiscard
, Opt_dax
, Opt_dax_enum
,
98 static const struct fs_parameter_spec xfs_fs_parameters
[] = {
99 fsparam_u32("logbufs", Opt_logbufs
),
100 fsparam_string("logbsize", Opt_logbsize
),
101 fsparam_string("logdev", Opt_logdev
),
102 fsparam_string("rtdev", Opt_rtdev
),
103 fsparam_flag("wsync", Opt_wsync
),
104 fsparam_flag("noalign", Opt_noalign
),
105 fsparam_flag("swalloc", Opt_swalloc
),
106 fsparam_u32("sunit", Opt_sunit
),
107 fsparam_u32("swidth", Opt_swidth
),
108 fsparam_flag("nouuid", Opt_nouuid
),
109 fsparam_flag("grpid", Opt_grpid
),
110 fsparam_flag("nogrpid", Opt_nogrpid
),
111 fsparam_flag("bsdgroups", Opt_bsdgroups
),
112 fsparam_flag("sysvgroups", Opt_sysvgroups
),
113 fsparam_string("allocsize", Opt_allocsize
),
114 fsparam_flag("norecovery", Opt_norecovery
),
115 fsparam_flag("inode64", Opt_inode64
),
116 fsparam_flag("inode32", Opt_inode32
),
117 fsparam_flag("ikeep", Opt_ikeep
),
118 fsparam_flag("noikeep", Opt_noikeep
),
119 fsparam_flag("largeio", Opt_largeio
),
120 fsparam_flag("nolargeio", Opt_nolargeio
),
121 fsparam_flag("attr2", Opt_attr2
),
122 fsparam_flag("noattr2", Opt_noattr2
),
123 fsparam_flag("filestreams", Opt_filestreams
),
124 fsparam_flag("quota", Opt_quota
),
125 fsparam_flag("noquota", Opt_noquota
),
126 fsparam_flag("usrquota", Opt_usrquota
),
127 fsparam_flag("grpquota", Opt_grpquota
),
128 fsparam_flag("prjquota", Opt_prjquota
),
129 fsparam_flag("uquota", Opt_uquota
),
130 fsparam_flag("gquota", Opt_gquota
),
131 fsparam_flag("pquota", Opt_pquota
),
132 fsparam_flag("uqnoenforce", Opt_uqnoenforce
),
133 fsparam_flag("gqnoenforce", Opt_gqnoenforce
),
134 fsparam_flag("pqnoenforce", Opt_pqnoenforce
),
135 fsparam_flag("qnoenforce", Opt_qnoenforce
),
136 fsparam_flag("discard", Opt_discard
),
137 fsparam_flag("nodiscard", Opt_nodiscard
),
138 fsparam_flag("dax", Opt_dax
),
139 fsparam_enum("dax", Opt_dax_enum
, dax_param_enums
),
143 struct proc_xfs_info
{
153 static struct proc_xfs_info xfs_info_set
[] = {
154 /* the few simple ones we can get from the mount struct */
155 { XFS_MOUNT_IKEEP
, ",ikeep" },
156 { XFS_MOUNT_WSYNC
, ",wsync" },
157 { XFS_MOUNT_NOALIGN
, ",noalign" },
158 { XFS_MOUNT_SWALLOC
, ",swalloc" },
159 { XFS_MOUNT_NOUUID
, ",nouuid" },
160 { XFS_MOUNT_NORECOVERY
, ",norecovery" },
161 { XFS_MOUNT_ATTR2
, ",attr2" },
162 { XFS_MOUNT_FILESTREAMS
, ",filestreams" },
163 { XFS_MOUNT_GRPID
, ",grpid" },
164 { XFS_MOUNT_DISCARD
, ",discard" },
165 { XFS_MOUNT_LARGEIO
, ",largeio" },
166 { XFS_MOUNT_DAX_ALWAYS
, ",dax=always" },
167 { XFS_MOUNT_DAX_NEVER
, ",dax=never" },
170 struct xfs_mount
*mp
= XFS_M(root
->d_sb
);
171 struct proc_xfs_info
*xfs_infop
;
173 for (xfs_infop
= xfs_info_set
; xfs_infop
->flag
; xfs_infop
++) {
174 if (mp
->m_flags
& xfs_infop
->flag
)
175 seq_puts(m
, xfs_infop
->str
);
178 seq_printf(m
, ",inode%d",
179 (mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) ? 32 : 64);
181 if (mp
->m_flags
& XFS_MOUNT_ALLOCSIZE
)
182 seq_printf(m
, ",allocsize=%dk",
183 (1 << mp
->m_allocsize_log
) >> 10);
185 if (mp
->m_logbufs
> 0)
186 seq_printf(m
, ",logbufs=%d", mp
->m_logbufs
);
187 if (mp
->m_logbsize
> 0)
188 seq_printf(m
, ",logbsize=%dk", mp
->m_logbsize
>> 10);
191 seq_show_option(m
, "logdev", mp
->m_logname
);
193 seq_show_option(m
, "rtdev", mp
->m_rtname
);
195 if (mp
->m_dalign
> 0)
196 seq_printf(m
, ",sunit=%d",
197 (int)XFS_FSB_TO_BB(mp
, mp
->m_dalign
));
198 if (mp
->m_swidth
> 0)
199 seq_printf(m
, ",swidth=%d",
200 (int)XFS_FSB_TO_BB(mp
, mp
->m_swidth
));
202 if (mp
->m_qflags
& (XFS_UQUOTA_ACCT
|XFS_UQUOTA_ENFD
))
203 seq_puts(m
, ",usrquota");
204 else if (mp
->m_qflags
& XFS_UQUOTA_ACCT
)
205 seq_puts(m
, ",uqnoenforce");
207 if (mp
->m_qflags
& XFS_PQUOTA_ACCT
) {
208 if (mp
->m_qflags
& XFS_PQUOTA_ENFD
)
209 seq_puts(m
, ",prjquota");
211 seq_puts(m
, ",pqnoenforce");
213 if (mp
->m_qflags
& XFS_GQUOTA_ACCT
) {
214 if (mp
->m_qflags
& XFS_GQUOTA_ENFD
)
215 seq_puts(m
, ",grpquota");
217 seq_puts(m
, ",gqnoenforce");
220 if (!(mp
->m_qflags
& XFS_ALL_QUOTA_ACCT
))
221 seq_puts(m
, ",noquota");
227 * Set parameters for inode allocation heuristics, taking into account
228 * filesystem size and inode32/inode64 mount options; i.e. specifically
229 * whether or not XFS_MOUNT_SMALL_INUMS is set.
231 * Inode allocation patterns are altered only if inode32 is requested
232 * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
233 * If altered, XFS_MOUNT_32BITINODES is set as well.
235 * An agcount independent of that in the mount structure is provided
236 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
237 * to the potentially higher ag count.
239 * Returns the maximum AG index which may contain inodes.
243 struct xfs_mount
*mp
,
244 xfs_agnumber_t agcount
)
246 xfs_agnumber_t index
;
247 xfs_agnumber_t maxagi
= 0;
248 xfs_sb_t
*sbp
= &mp
->m_sb
;
249 xfs_agnumber_t max_metadata
;
254 * Calculate how much should be reserved for inodes to meet
255 * the max inode percentage. Used only for inode32.
257 if (M_IGEO(mp
)->maxicount
) {
260 icount
= sbp
->sb_dblocks
* sbp
->sb_imax_pct
;
262 icount
+= sbp
->sb_agblocks
- 1;
263 do_div(icount
, sbp
->sb_agblocks
);
264 max_metadata
= icount
;
266 max_metadata
= agcount
;
269 /* Get the last possible inode in the filesystem */
270 agino
= XFS_AGB_TO_AGINO(mp
, sbp
->sb_agblocks
- 1);
271 ino
= XFS_AGINO_TO_INO(mp
, agcount
- 1, agino
);
274 * If user asked for no more than 32-bit inodes, and the fs is
275 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
276 * the allocator to accommodate the request.
278 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) && ino
> XFS_MAXINUMBER_32
)
279 mp
->m_flags
|= XFS_MOUNT_32BITINODES
;
281 mp
->m_flags
&= ~XFS_MOUNT_32BITINODES
;
283 for (index
= 0; index
< agcount
; index
++) {
284 struct xfs_perag
*pag
;
286 ino
= XFS_AGINO_TO_INO(mp
, index
, agino
);
288 pag
= xfs_perag_get(mp
, index
);
290 if (mp
->m_flags
& XFS_MOUNT_32BITINODES
) {
291 if (ino
> XFS_MAXINUMBER_32
) {
292 pag
->pagi_inodeok
= 0;
293 pag
->pagf_metadata
= 0;
295 pag
->pagi_inodeok
= 1;
297 if (index
< max_metadata
)
298 pag
->pagf_metadata
= 1;
300 pag
->pagf_metadata
= 0;
303 pag
->pagi_inodeok
= 1;
304 pag
->pagf_metadata
= 0;
310 return (mp
->m_flags
& XFS_MOUNT_32BITINODES
) ? maxagi
: agcount
;
317 struct block_device
**bdevp
)
321 *bdevp
= blkdev_get_by_path(name
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
323 if (IS_ERR(*bdevp
)) {
324 error
= PTR_ERR(*bdevp
);
325 xfs_warn(mp
, "Invalid device [%s], error=%d", name
, error
);
333 struct block_device
*bdev
)
336 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
340 xfs_blkdev_issue_flush(
341 xfs_buftarg_t
*buftarg
)
343 blkdev_issue_flush(buftarg
->bt_bdev
, GFP_NOFS
);
348 struct xfs_mount
*mp
)
350 struct dax_device
*dax_ddev
= mp
->m_ddev_targp
->bt_daxdev
;
352 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
353 struct block_device
*logdev
= mp
->m_logdev_targp
->bt_bdev
;
354 struct dax_device
*dax_logdev
= mp
->m_logdev_targp
->bt_daxdev
;
356 xfs_free_buftarg(mp
->m_logdev_targp
);
357 xfs_blkdev_put(logdev
);
358 fs_put_dax(dax_logdev
);
360 if (mp
->m_rtdev_targp
) {
361 struct block_device
*rtdev
= mp
->m_rtdev_targp
->bt_bdev
;
362 struct dax_device
*dax_rtdev
= mp
->m_rtdev_targp
->bt_daxdev
;
364 xfs_free_buftarg(mp
->m_rtdev_targp
);
365 xfs_blkdev_put(rtdev
);
366 fs_put_dax(dax_rtdev
);
368 xfs_free_buftarg(mp
->m_ddev_targp
);
369 fs_put_dax(dax_ddev
);
373 * The file system configurations are:
374 * (1) device (partition) with data and internal log
375 * (2) logical volume with data and log subvolumes.
376 * (3) logical volume with data, log, and realtime subvolumes.
378 * We only have to handle opening the log and realtime volumes here if
379 * they are present. The data subvolume has already been opened by
380 * get_sb_bdev() and is stored in sb->s_bdev.
384 struct xfs_mount
*mp
)
386 struct block_device
*ddev
= mp
->m_super
->s_bdev
;
387 struct dax_device
*dax_ddev
= fs_dax_get_by_bdev(ddev
);
388 struct dax_device
*dax_logdev
= NULL
, *dax_rtdev
= NULL
;
389 struct block_device
*logdev
= NULL
, *rtdev
= NULL
;
393 * Open real time and log devices - order is important.
396 error
= xfs_blkdev_get(mp
, mp
->m_logname
, &logdev
);
399 dax_logdev
= fs_dax_get_by_bdev(logdev
);
403 error
= xfs_blkdev_get(mp
, mp
->m_rtname
, &rtdev
);
405 goto out_close_logdev
;
407 if (rtdev
== ddev
|| rtdev
== logdev
) {
409 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
411 goto out_close_rtdev
;
413 dax_rtdev
= fs_dax_get_by_bdev(rtdev
);
417 * Setup xfs_mount buffer target pointers
420 mp
->m_ddev_targp
= xfs_alloc_buftarg(mp
, ddev
, dax_ddev
);
421 if (!mp
->m_ddev_targp
)
422 goto out_close_rtdev
;
425 mp
->m_rtdev_targp
= xfs_alloc_buftarg(mp
, rtdev
, dax_rtdev
);
426 if (!mp
->m_rtdev_targp
)
427 goto out_free_ddev_targ
;
430 if (logdev
&& logdev
!= ddev
) {
431 mp
->m_logdev_targp
= xfs_alloc_buftarg(mp
, logdev
, dax_logdev
);
432 if (!mp
->m_logdev_targp
)
433 goto out_free_rtdev_targ
;
435 mp
->m_logdev_targp
= mp
->m_ddev_targp
;
441 if (mp
->m_rtdev_targp
)
442 xfs_free_buftarg(mp
->m_rtdev_targp
);
444 xfs_free_buftarg(mp
->m_ddev_targp
);
446 xfs_blkdev_put(rtdev
);
447 fs_put_dax(dax_rtdev
);
449 if (logdev
&& logdev
!= ddev
) {
450 xfs_blkdev_put(logdev
);
451 fs_put_dax(dax_logdev
);
454 fs_put_dax(dax_ddev
);
459 * Setup xfs_mount buffer target pointers based on superblock
463 struct xfs_mount
*mp
)
467 error
= xfs_setsize_buftarg(mp
->m_ddev_targp
, mp
->m_sb
.sb_sectsize
);
471 if (mp
->m_logdev_targp
&& mp
->m_logdev_targp
!= mp
->m_ddev_targp
) {
472 unsigned int log_sector_size
= BBSIZE
;
474 if (xfs_sb_version_hassector(&mp
->m_sb
))
475 log_sector_size
= mp
->m_sb
.sb_logsectsize
;
476 error
= xfs_setsize_buftarg(mp
->m_logdev_targp
,
481 if (mp
->m_rtdev_targp
) {
482 error
= xfs_setsize_buftarg(mp
->m_rtdev_targp
,
483 mp
->m_sb
.sb_sectsize
);
492 xfs_init_mount_workqueues(
493 struct xfs_mount
*mp
)
495 mp
->m_buf_workqueue
= alloc_workqueue("xfs-buf/%s",
496 WQ_MEM_RECLAIM
|WQ_FREEZABLE
, 1, mp
->m_super
->s_id
);
497 if (!mp
->m_buf_workqueue
)
500 mp
->m_unwritten_workqueue
= alloc_workqueue("xfs-conv/%s",
501 WQ_MEM_RECLAIM
|WQ_FREEZABLE
, 0, mp
->m_super
->s_id
);
502 if (!mp
->m_unwritten_workqueue
)
503 goto out_destroy_buf
;
505 mp
->m_cil_workqueue
= alloc_workqueue("xfs-cil/%s",
506 WQ_MEM_RECLAIM
| WQ_FREEZABLE
| WQ_UNBOUND
,
507 0, mp
->m_super
->s_id
);
508 if (!mp
->m_cil_workqueue
)
509 goto out_destroy_unwritten
;
511 mp
->m_reclaim_workqueue
= alloc_workqueue("xfs-reclaim/%s",
512 WQ_MEM_RECLAIM
|WQ_FREEZABLE
, 0, mp
->m_super
->s_id
);
513 if (!mp
->m_reclaim_workqueue
)
514 goto out_destroy_cil
;
516 mp
->m_eofblocks_workqueue
= alloc_workqueue("xfs-eofblocks/%s",
517 WQ_MEM_RECLAIM
|WQ_FREEZABLE
, 0, mp
->m_super
->s_id
);
518 if (!mp
->m_eofblocks_workqueue
)
519 goto out_destroy_reclaim
;
521 mp
->m_sync_workqueue
= alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE
, 0,
523 if (!mp
->m_sync_workqueue
)
524 goto out_destroy_eofb
;
529 destroy_workqueue(mp
->m_eofblocks_workqueue
);
531 destroy_workqueue(mp
->m_reclaim_workqueue
);
533 destroy_workqueue(mp
->m_cil_workqueue
);
534 out_destroy_unwritten
:
535 destroy_workqueue(mp
->m_unwritten_workqueue
);
537 destroy_workqueue(mp
->m_buf_workqueue
);
543 xfs_destroy_mount_workqueues(
544 struct xfs_mount
*mp
)
546 destroy_workqueue(mp
->m_sync_workqueue
);
547 destroy_workqueue(mp
->m_eofblocks_workqueue
);
548 destroy_workqueue(mp
->m_reclaim_workqueue
);
549 destroy_workqueue(mp
->m_cil_workqueue
);
550 destroy_workqueue(mp
->m_unwritten_workqueue
);
551 destroy_workqueue(mp
->m_buf_workqueue
);
555 xfs_flush_inodes_worker(
556 struct work_struct
*work
)
558 struct xfs_mount
*mp
= container_of(work
, struct xfs_mount
,
559 m_flush_inodes_work
);
560 struct super_block
*sb
= mp
->m_super
;
562 if (down_read_trylock(&sb
->s_umount
)) {
564 up_read(&sb
->s_umount
);
569 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
570 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
571 * for IO to complete so that we effectively throttle multiple callers to the
572 * rate at which IO is completing.
576 struct xfs_mount
*mp
)
579 * If flush_work() returns true then that means we waited for a flush
580 * which was already in progress. Don't bother running another scan.
582 if (flush_work(&mp
->m_flush_inodes_work
))
585 queue_work(mp
->m_sync_workqueue
, &mp
->m_flush_inodes_work
);
586 flush_work(&mp
->m_flush_inodes_work
);
589 /* Catch misguided souls that try to use this interface on XFS */
590 STATIC
struct inode
*
592 struct super_block
*sb
)
601 struct xfs_inode
*ip
,
604 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
605 struct xfs_bmbt_irec got
;
606 struct xfs_iext_cursor icur
;
608 if (!ifp
|| !xfs_iext_lookup_extent(ip
, ifp
, 0, &icur
, &got
))
611 if (isnullstartblock(got
.br_startblock
)) {
612 xfs_warn(ip
->i_mount
,
613 "ino %llx %s fork has delalloc extent at [0x%llx:0x%llx]",
615 whichfork
== XFS_DATA_FORK
? "data" : "cow",
616 got
.br_startoff
, got
.br_blockcount
);
618 } while (xfs_iext_next_extent(ifp
, &icur
, &got
));
621 #define xfs_check_delalloc(ip, whichfork) do { } while (0)
625 * Now that the generic code is guaranteed not to be accessing
626 * the linux inode, we can inactivate and reclaim the inode.
629 xfs_fs_destroy_inode(
632 struct xfs_inode
*ip
= XFS_I(inode
);
634 trace_xfs_destroy_inode(ip
);
636 ASSERT(!rwsem_is_locked(&inode
->i_rwsem
));
637 XFS_STATS_INC(ip
->i_mount
, vn_rele
);
638 XFS_STATS_INC(ip
->i_mount
, vn_remove
);
642 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
) && ip
->i_delayed_blks
) {
643 xfs_check_delalloc(ip
, XFS_DATA_FORK
);
644 xfs_check_delalloc(ip
, XFS_COW_FORK
);
648 XFS_STATS_INC(ip
->i_mount
, vn_reclaim
);
651 * We should never get here with one of the reclaim flags already set.
653 ASSERT_ALWAYS(!xfs_iflags_test(ip
, XFS_IRECLAIMABLE
));
654 ASSERT_ALWAYS(!xfs_iflags_test(ip
, XFS_IRECLAIM
));
657 * We always use background reclaim here because even if the
658 * inode is clean, it still may be under IO and hence we have
659 * to take the flush lock. The background reclaim path handles
660 * this more efficiently than we can here, so simply let background
661 * reclaim tear down all inodes.
663 xfs_inode_set_reclaim_tag(ip
);
671 struct xfs_inode
*ip
= XFS_I(inode
);
672 struct xfs_mount
*mp
= ip
->i_mount
;
673 struct xfs_trans
*tp
;
675 if (!(inode
->i_sb
->s_flags
& SB_LAZYTIME
))
677 if (flag
!= I_DIRTY_SYNC
|| !(inode
->i_state
& I_DIRTY_TIME
))
680 if (xfs_trans_alloc(mp
, &M_RES(mp
)->tr_fsyncts
, 0, 0, 0, &tp
))
682 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
683 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
684 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_TIMESTAMP
);
685 xfs_trans_commit(tp
);
689 * Slab object creation initialisation for the XFS inode.
690 * This covers only the idempotent fields in the XFS inode;
691 * all other fields need to be initialised on allocation
692 * from the slab. This avoids the need to repeatedly initialise
693 * fields in the xfs inode that left in the initialise state
694 * when freeing the inode.
697 xfs_fs_inode_init_once(
700 struct xfs_inode
*ip
= inode
;
702 memset(ip
, 0, sizeof(struct xfs_inode
));
705 inode_init_once(VFS_I(ip
));
708 atomic_set(&ip
->i_pincount
, 0);
709 spin_lock_init(&ip
->i_flags_lock
);
711 mrlock_init(&ip
->i_mmaplock
, MRLOCK_ALLOW_EQUAL_PRI
|MRLOCK_BARRIER
,
712 "xfsino", ip
->i_ino
);
713 mrlock_init(&ip
->i_lock
, MRLOCK_ALLOW_EQUAL_PRI
|MRLOCK_BARRIER
,
714 "xfsino", ip
->i_ino
);
718 * We do an unlocked check for XFS_IDONTCACHE here because we are already
719 * serialised against cache hits here via the inode->i_lock and igrab() in
720 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
721 * racing with us, and it avoids needing to grab a spinlock here for every inode
722 * we drop the final reference on.
728 struct xfs_inode
*ip
= XFS_I(inode
);
731 * If this unlinked inode is in the middle of recovery, don't
732 * drop the inode just yet; log recovery will take care of
733 * that. See the comment for this inode flag.
735 if (ip
->i_flags
& XFS_IRECOVERY
) {
736 ASSERT(ip
->i_mount
->m_log
->l_flags
& XLOG_RECOVERY_NEEDED
);
740 return generic_drop_inode(inode
);
745 struct xfs_mount
*mp
)
748 kfree(mp
->m_logname
);
754 struct super_block
*sb
,
757 struct xfs_mount
*mp
= XFS_M(sb
);
760 * Doing anything during the async pass would be counterproductive.
765 xfs_log_force(mp
, XFS_LOG_SYNC
);
768 * The disk must be active because we're syncing.
769 * We schedule log work now (now that the disk is
770 * active) instead of later (when it might not be).
772 flush_delayed_work(&mp
->m_log
->l_work
);
780 struct dentry
*dentry
,
781 struct kstatfs
*statp
)
783 struct xfs_mount
*mp
= XFS_M(dentry
->d_sb
);
784 xfs_sb_t
*sbp
= &mp
->m_sb
;
785 struct xfs_inode
*ip
= XFS_I(d_inode(dentry
));
786 uint64_t fakeinos
, id
;
793 statp
->f_type
= XFS_SUPER_MAGIC
;
794 statp
->f_namelen
= MAXNAMELEN
- 1;
796 id
= huge_encode_dev(mp
->m_ddev_targp
->bt_dev
);
797 statp
->f_fsid
.val
[0] = (u32
)id
;
798 statp
->f_fsid
.val
[1] = (u32
)(id
>> 32);
800 icount
= percpu_counter_sum(&mp
->m_icount
);
801 ifree
= percpu_counter_sum(&mp
->m_ifree
);
802 fdblocks
= percpu_counter_sum(&mp
->m_fdblocks
);
804 spin_lock(&mp
->m_sb_lock
);
805 statp
->f_bsize
= sbp
->sb_blocksize
;
806 lsize
= sbp
->sb_logstart
? sbp
->sb_logblocks
: 0;
807 statp
->f_blocks
= sbp
->sb_dblocks
- lsize
;
808 spin_unlock(&mp
->m_sb_lock
);
810 /* make sure statp->f_bfree does not underflow */
811 statp
->f_bfree
= max_t(int64_t, fdblocks
- mp
->m_alloc_set_aside
, 0);
812 statp
->f_bavail
= statp
->f_bfree
;
814 fakeinos
= XFS_FSB_TO_INO(mp
, statp
->f_bfree
);
815 statp
->f_files
= min(icount
+ fakeinos
, (uint64_t)XFS_MAXINUMBER
);
816 if (M_IGEO(mp
)->maxicount
)
817 statp
->f_files
= min_t(typeof(statp
->f_files
),
819 M_IGEO(mp
)->maxicount
);
821 /* If sb_icount overshot maxicount, report actual allocation */
822 statp
->f_files
= max_t(typeof(statp
->f_files
),
826 /* make sure statp->f_ffree does not underflow */
827 ffree
= statp
->f_files
- (icount
- ifree
);
828 statp
->f_ffree
= max_t(int64_t, ffree
, 0);
831 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
832 ((mp
->m_qflags
& (XFS_PQUOTA_ACCT
|XFS_PQUOTA_ENFD
))) ==
833 (XFS_PQUOTA_ACCT
|XFS_PQUOTA_ENFD
))
834 xfs_qm_statvfs(ip
, statp
);
836 if (XFS_IS_REALTIME_MOUNT(mp
) &&
837 (ip
->i_d
.di_flags
& (XFS_DIFLAG_RTINHERIT
| XFS_DIFLAG_REALTIME
))) {
838 statp
->f_blocks
= sbp
->sb_rblocks
;
839 statp
->f_bavail
= statp
->f_bfree
=
840 sbp
->sb_frextents
* sbp
->sb_rextsize
;
847 xfs_save_resvblks(struct xfs_mount
*mp
)
849 uint64_t resblks
= 0;
851 mp
->m_resblks_save
= mp
->m_resblks
;
852 xfs_reserve_blocks(mp
, &resblks
, NULL
);
856 xfs_restore_resvblks(struct xfs_mount
*mp
)
860 if (mp
->m_resblks_save
) {
861 resblks
= mp
->m_resblks_save
;
862 mp
->m_resblks_save
= 0;
864 resblks
= xfs_default_resblks(mp
);
866 xfs_reserve_blocks(mp
, &resblks
, NULL
);
870 * Trigger writeback of all the dirty metadata in the file system.
872 * This ensures that the metadata is written to their location on disk rather
873 * than just existing in transactions in the log. This means after a quiesce
874 * there is no log replay required to write the inodes to disk - this is the
875 * primary difference between a sync and a quiesce.
877 * We cancel log work early here to ensure all transactions the log worker may
878 * run have finished before we clean up and log the superblock and write an
879 * unmount record. The unfreeze process is responsible for restarting the log
884 struct xfs_mount
*mp
)
888 cancel_delayed_work_sync(&mp
->m_log
->l_work
);
890 /* force the log to unpin objects from the now complete transactions */
891 xfs_log_force(mp
, XFS_LOG_SYNC
);
894 /* Push the superblock and write an unmount record */
895 error
= xfs_log_sbcount(mp
);
897 xfs_warn(mp
, "xfs_attr_quiesce: failed to log sb changes. "
898 "Frozen image may not be consistent.");
903 * Second stage of a freeze. The data is already frozen so we only
904 * need to take care of the metadata. Once that's done sync the superblock
905 * to the log to dirty it in case of a crash while frozen. This ensures that we
906 * will recover the unlinked inode lists on the next mount.
910 struct super_block
*sb
)
912 struct xfs_mount
*mp
= XFS_M(sb
);
917 * The filesystem is now frozen far enough that memory reclaim
918 * cannot safely operate on the filesystem. Hence we need to
919 * set a GFP_NOFS context here to avoid recursion deadlocks.
921 flags
= memalloc_nofs_save();
922 xfs_stop_block_reaping(mp
);
923 xfs_save_resvblks(mp
);
924 xfs_quiesce_attr(mp
);
925 ret
= xfs_sync_sb(mp
, true);
926 memalloc_nofs_restore(flags
);
932 struct super_block
*sb
)
934 struct xfs_mount
*mp
= XFS_M(sb
);
936 xfs_restore_resvblks(mp
);
937 xfs_log_work_queue(mp
);
938 xfs_start_block_reaping(mp
);
943 * This function fills in xfs_mount_t fields based on mount args.
944 * Note: the superblock _has_ now been read in.
948 struct xfs_mount
*mp
)
950 int ronly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
952 /* Fail a mount where the logbuf is smaller than the log stripe */
953 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
954 if (mp
->m_logbsize
<= 0 &&
955 mp
->m_sb
.sb_logsunit
> XLOG_BIG_RECORD_BSIZE
) {
956 mp
->m_logbsize
= mp
->m_sb
.sb_logsunit
;
957 } else if (mp
->m_logbsize
> 0 &&
958 mp
->m_logbsize
< mp
->m_sb
.sb_logsunit
) {
960 "logbuf size must be greater than or equal to log stripe size");
964 /* Fail a mount if the logbuf is larger than 32K */
965 if (mp
->m_logbsize
> XLOG_BIG_RECORD_BSIZE
) {
967 "logbuf size for version 1 logs must be 16K or 32K");
973 * V5 filesystems always use attr2 format for attributes.
975 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
976 (mp
->m_flags
& XFS_MOUNT_NOATTR2
)) {
977 xfs_warn(mp
, "Cannot mount a V5 filesystem as noattr2. "
978 "attr2 is always enabled for V5 filesystems.");
983 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
984 * told by noattr2 to turn it off
986 if (xfs_sb_version_hasattr2(&mp
->m_sb
) &&
987 !(mp
->m_flags
& XFS_MOUNT_NOATTR2
))
988 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
991 * prohibit r/w mounts of read-only filesystems
993 if ((mp
->m_sb
.sb_flags
& XFS_SBF_READONLY
) && !ronly
) {
995 "cannot mount a read-only filesystem as read-write");
999 if ((mp
->m_qflags
& (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
)) &&
1000 (mp
->m_qflags
& (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
)) &&
1001 !xfs_sb_version_has_pquotino(&mp
->m_sb
)) {
1003 "Super block does not support project and group quota together");
1011 xfs_init_percpu_counters(
1012 struct xfs_mount
*mp
)
1016 error
= percpu_counter_init(&mp
->m_icount
, 0, GFP_KERNEL
);
1020 error
= percpu_counter_init(&mp
->m_ifree
, 0, GFP_KERNEL
);
1024 error
= percpu_counter_init(&mp
->m_fdblocks
, 0, GFP_KERNEL
);
1028 error
= percpu_counter_init(&mp
->m_delalloc_blks
, 0, GFP_KERNEL
);
1035 percpu_counter_destroy(&mp
->m_fdblocks
);
1037 percpu_counter_destroy(&mp
->m_ifree
);
1039 percpu_counter_destroy(&mp
->m_icount
);
1044 xfs_reinit_percpu_counters(
1045 struct xfs_mount
*mp
)
1047 percpu_counter_set(&mp
->m_icount
, mp
->m_sb
.sb_icount
);
1048 percpu_counter_set(&mp
->m_ifree
, mp
->m_sb
.sb_ifree
);
1049 percpu_counter_set(&mp
->m_fdblocks
, mp
->m_sb
.sb_fdblocks
);
1053 xfs_destroy_percpu_counters(
1054 struct xfs_mount
*mp
)
1056 percpu_counter_destroy(&mp
->m_icount
);
1057 percpu_counter_destroy(&mp
->m_ifree
);
1058 percpu_counter_destroy(&mp
->m_fdblocks
);
1059 ASSERT(XFS_FORCED_SHUTDOWN(mp
) ||
1060 percpu_counter_sum(&mp
->m_delalloc_blks
) == 0);
1061 percpu_counter_destroy(&mp
->m_delalloc_blks
);
1066 struct super_block
*sb
)
1068 struct xfs_mount
*mp
= XFS_M(sb
);
1070 /* if ->fill_super failed, we have no mount to tear down */
1074 xfs_notice(mp
, "Unmounting Filesystem");
1075 xfs_filestream_unmount(mp
);
1079 free_percpu(mp
->m_stats
.xs_stats
);
1080 xfs_destroy_percpu_counters(mp
);
1081 xfs_destroy_mount_workqueues(mp
);
1082 xfs_close_devices(mp
);
1084 sb
->s_fs_info
= NULL
;
1089 xfs_fs_nr_cached_objects(
1090 struct super_block
*sb
,
1091 struct shrink_control
*sc
)
1093 /* Paranoia: catch incorrect calls during mount setup or teardown */
1094 if (WARN_ON_ONCE(!sb
->s_fs_info
))
1096 return xfs_reclaim_inodes_count(XFS_M(sb
));
1100 xfs_fs_free_cached_objects(
1101 struct super_block
*sb
,
1102 struct shrink_control
*sc
)
1104 return xfs_reclaim_inodes_nr(XFS_M(sb
), sc
->nr_to_scan
);
1107 static const struct super_operations xfs_super_operations
= {
1108 .alloc_inode
= xfs_fs_alloc_inode
,
1109 .destroy_inode
= xfs_fs_destroy_inode
,
1110 .dirty_inode
= xfs_fs_dirty_inode
,
1111 .drop_inode
= xfs_fs_drop_inode
,
1112 .put_super
= xfs_fs_put_super
,
1113 .sync_fs
= xfs_fs_sync_fs
,
1114 .freeze_fs
= xfs_fs_freeze
,
1115 .unfreeze_fs
= xfs_fs_unfreeze
,
1116 .statfs
= xfs_fs_statfs
,
1117 .show_options
= xfs_fs_show_options
,
1118 .nr_cached_objects
= xfs_fs_nr_cached_objects
,
1119 .free_cached_objects
= xfs_fs_free_cached_objects
,
1128 int last
, shift_left_factor
= 0, _res
;
1132 value
= kstrdup(s
, GFP_KERNEL
);
1136 last
= strlen(value
) - 1;
1137 if (value
[last
] == 'K' || value
[last
] == 'k') {
1138 shift_left_factor
= 10;
1141 if (value
[last
] == 'M' || value
[last
] == 'm') {
1142 shift_left_factor
= 20;
1145 if (value
[last
] == 'G' || value
[last
] == 'g') {
1146 shift_left_factor
= 30;
1150 if (kstrtoint(value
, base
, &_res
))
1153 *res
= _res
<< shift_left_factor
;
1158 * Set mount state from a mount option.
1160 * NOTE: mp->m_super is NULL here!
1164 struct fs_context
*fc
,
1165 struct fs_parameter
*param
)
1167 struct xfs_mount
*mp
= fc
->s_fs_info
;
1168 struct fs_parse_result result
;
1172 opt
= fs_parse(fc
, xfs_fs_parameters
, param
, &result
);
1178 mp
->m_logbufs
= result
.uint_32
;
1181 if (suffix_kstrtoint(param
->string
, 10, &mp
->m_logbsize
))
1185 kfree(mp
->m_logname
);
1186 mp
->m_logname
= kstrdup(param
->string
, GFP_KERNEL
);
1191 kfree(mp
->m_rtname
);
1192 mp
->m_rtname
= kstrdup(param
->string
, GFP_KERNEL
);
1197 if (suffix_kstrtoint(param
->string
, 10, &size
))
1199 mp
->m_allocsize_log
= ffs(size
) - 1;
1200 mp
->m_flags
|= XFS_MOUNT_ALLOCSIZE
;
1204 mp
->m_flags
|= XFS_MOUNT_GRPID
;
1207 case Opt_sysvgroups
:
1208 mp
->m_flags
&= ~XFS_MOUNT_GRPID
;
1211 mp
->m_flags
|= XFS_MOUNT_WSYNC
;
1213 case Opt_norecovery
:
1214 mp
->m_flags
|= XFS_MOUNT_NORECOVERY
;
1217 mp
->m_flags
|= XFS_MOUNT_NOALIGN
;
1220 mp
->m_flags
|= XFS_MOUNT_SWALLOC
;
1223 mp
->m_dalign
= result
.uint_32
;
1226 mp
->m_swidth
= result
.uint_32
;
1229 mp
->m_flags
|= XFS_MOUNT_SMALL_INUMS
;
1232 mp
->m_flags
&= ~XFS_MOUNT_SMALL_INUMS
;
1235 mp
->m_flags
|= XFS_MOUNT_NOUUID
;
1238 mp
->m_flags
|= XFS_MOUNT_IKEEP
;
1241 mp
->m_flags
&= ~XFS_MOUNT_IKEEP
;
1244 mp
->m_flags
|= XFS_MOUNT_LARGEIO
;
1247 mp
->m_flags
&= ~XFS_MOUNT_LARGEIO
;
1250 mp
->m_flags
|= XFS_MOUNT_ATTR2
;
1253 mp
->m_flags
&= ~XFS_MOUNT_ATTR2
;
1254 mp
->m_flags
|= XFS_MOUNT_NOATTR2
;
1256 case Opt_filestreams
:
1257 mp
->m_flags
|= XFS_MOUNT_FILESTREAMS
;
1260 mp
->m_qflags
&= ~XFS_ALL_QUOTA_ACCT
;
1261 mp
->m_qflags
&= ~XFS_ALL_QUOTA_ENFD
;
1262 mp
->m_qflags
&= ~XFS_ALL_QUOTA_ACTIVE
;
1267 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
|
1270 case Opt_qnoenforce
:
1271 case Opt_uqnoenforce
:
1272 mp
->m_qflags
|= (XFS_UQUOTA_ACCT
| XFS_UQUOTA_ACTIVE
);
1273 mp
->m_qflags
&= ~XFS_UQUOTA_ENFD
;
1277 mp
->m_qflags
|= (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
|
1280 case Opt_pqnoenforce
:
1281 mp
->m_qflags
|= (XFS_PQUOTA_ACCT
| XFS_PQUOTA_ACTIVE
);
1282 mp
->m_qflags
&= ~XFS_PQUOTA_ENFD
;
1286 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
|
1289 case Opt_gqnoenforce
:
1290 mp
->m_qflags
|= (XFS_GQUOTA_ACCT
| XFS_GQUOTA_ACTIVE
);
1291 mp
->m_qflags
&= ~XFS_GQUOTA_ENFD
;
1294 mp
->m_flags
|= XFS_MOUNT_DISCARD
;
1297 mp
->m_flags
&= ~XFS_MOUNT_DISCARD
;
1299 #ifdef CONFIG_FS_DAX
1301 xfs_mount_set_dax_mode(mp
, XFS_DAX_ALWAYS
);
1304 xfs_mount_set_dax_mode(mp
, result
.uint_32
);
1308 xfs_warn(mp
, "unknown mount option [%s].", param
->key
);
1316 xfs_fc_validate_params(
1317 struct xfs_mount
*mp
)
1320 * no recovery flag requires a read-only mount
1322 if ((mp
->m_flags
& XFS_MOUNT_NORECOVERY
) &&
1323 !(mp
->m_flags
& XFS_MOUNT_RDONLY
)) {
1324 xfs_warn(mp
, "no-recovery mounts must be read-only.");
1328 if ((mp
->m_flags
& XFS_MOUNT_NOALIGN
) &&
1329 (mp
->m_dalign
|| mp
->m_swidth
)) {
1331 "sunit and swidth options incompatible with the noalign option");
1335 if (!IS_ENABLED(CONFIG_XFS_QUOTA
) && mp
->m_qflags
!= 0) {
1336 xfs_warn(mp
, "quota support not available in this kernel.");
1340 if ((mp
->m_dalign
&& !mp
->m_swidth
) ||
1341 (!mp
->m_dalign
&& mp
->m_swidth
)) {
1342 xfs_warn(mp
, "sunit and swidth must be specified together");
1346 if (mp
->m_dalign
&& (mp
->m_swidth
% mp
->m_dalign
!= 0)) {
1348 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1349 mp
->m_swidth
, mp
->m_dalign
);
1353 if (mp
->m_logbufs
!= -1 &&
1354 mp
->m_logbufs
!= 0 &&
1355 (mp
->m_logbufs
< XLOG_MIN_ICLOGS
||
1356 mp
->m_logbufs
> XLOG_MAX_ICLOGS
)) {
1357 xfs_warn(mp
, "invalid logbufs value: %d [not %d-%d]",
1358 mp
->m_logbufs
, XLOG_MIN_ICLOGS
, XLOG_MAX_ICLOGS
);
1362 if (mp
->m_logbsize
!= -1 &&
1363 mp
->m_logbsize
!= 0 &&
1364 (mp
->m_logbsize
< XLOG_MIN_RECORD_BSIZE
||
1365 mp
->m_logbsize
> XLOG_MAX_RECORD_BSIZE
||
1366 !is_power_of_2(mp
->m_logbsize
))) {
1368 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1373 if ((mp
->m_flags
& XFS_MOUNT_ALLOCSIZE
) &&
1374 (mp
->m_allocsize_log
> XFS_MAX_IO_LOG
||
1375 mp
->m_allocsize_log
< XFS_MIN_IO_LOG
)) {
1376 xfs_warn(mp
, "invalid log iosize: %d [not %d-%d]",
1377 mp
->m_allocsize_log
, XFS_MIN_IO_LOG
, XFS_MAX_IO_LOG
);
1386 struct super_block
*sb
,
1387 struct fs_context
*fc
)
1389 struct xfs_mount
*mp
= sb
->s_fs_info
;
1391 int flags
= 0, error
;
1395 error
= xfs_fc_validate_params(mp
);
1397 goto out_free_names
;
1399 sb_min_blocksize(sb
, BBSIZE
);
1400 sb
->s_xattr
= xfs_xattr_handlers
;
1401 sb
->s_export_op
= &xfs_export_operations
;
1402 #ifdef CONFIG_XFS_QUOTA
1403 sb
->s_qcop
= &xfs_quotactl_operations
;
1404 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
| QTYPE_MASK_PRJ
;
1406 sb
->s_op
= &xfs_super_operations
;
1409 * Delay mount work if the debug hook is set. This is debug
1410 * instrumention to coordinate simulation of xfs mount failures with
1411 * VFS superblock operations
1413 if (xfs_globals
.mount_delay
) {
1414 xfs_notice(mp
, "Delaying mount for %d seconds.",
1415 xfs_globals
.mount_delay
);
1416 msleep(xfs_globals
.mount_delay
* 1000);
1419 if (fc
->sb_flags
& SB_SILENT
)
1420 flags
|= XFS_MFSI_QUIET
;
1422 error
= xfs_open_devices(mp
);
1424 goto out_free_names
;
1426 error
= xfs_init_mount_workqueues(mp
);
1428 goto out_close_devices
;
1430 error
= xfs_init_percpu_counters(mp
);
1432 goto out_destroy_workqueues
;
1434 /* Allocate stats memory before we do operations that might use it */
1435 mp
->m_stats
.xs_stats
= alloc_percpu(struct xfsstats
);
1436 if (!mp
->m_stats
.xs_stats
) {
1438 goto out_destroy_counters
;
1441 error
= xfs_readsb(mp
, flags
);
1443 goto out_free_stats
;
1445 error
= xfs_finish_flags(mp
);
1449 error
= xfs_setup_devices(mp
);
1454 * XFS block mappings use 54 bits to store the logical block offset.
1455 * This should suffice to handle the maximum file size that the VFS
1456 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1457 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1458 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1459 * to check this assertion.
1461 * Avoid integer overflow by comparing the maximum bmbt offset to the
1462 * maximum pagecache offset in units of fs blocks.
1464 if (XFS_B_TO_FSBT(mp
, MAX_LFS_FILESIZE
) > XFS_MAX_FILEOFF
) {
1466 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1467 XFS_B_TO_FSBT(mp
, MAX_LFS_FILESIZE
),
1473 error
= xfs_filestream_mount(mp
);
1478 * we must configure the block size in the superblock before we run the
1479 * full mount process as the mount process can lookup and cache inodes.
1481 sb
->s_magic
= XFS_SUPER_MAGIC
;
1482 sb
->s_blocksize
= mp
->m_sb
.sb_blocksize
;
1483 sb
->s_blocksize_bits
= ffs(sb
->s_blocksize
) - 1;
1484 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1485 sb
->s_max_links
= XFS_MAXLINK
;
1486 sb
->s_time_gran
= 1;
1487 sb
->s_time_min
= S32_MIN
;
1488 sb
->s_time_max
= S32_MAX
;
1489 sb
->s_iflags
|= SB_I_CGROUPWB
;
1491 set_posix_acl_flag(sb
);
1493 /* version 5 superblocks support inode version counters. */
1494 if (XFS_SB_VERSION_NUM(&mp
->m_sb
) == XFS_SB_VERSION_5
)
1495 sb
->s_flags
|= SB_I_VERSION
;
1497 if (mp
->m_flags
& XFS_MOUNT_DAX_ALWAYS
) {
1498 bool rtdev_is_dax
= false, datadev_is_dax
;
1501 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1503 datadev_is_dax
= bdev_dax_supported(mp
->m_ddev_targp
->bt_bdev
,
1505 if (mp
->m_rtdev_targp
)
1506 rtdev_is_dax
= bdev_dax_supported(
1507 mp
->m_rtdev_targp
->bt_bdev
, sb
->s_blocksize
);
1508 if (!rtdev_is_dax
&& !datadev_is_dax
) {
1510 "DAX unsupported by block device. Turning off DAX.");
1511 xfs_mount_set_dax_mode(mp
, XFS_DAX_NEVER
);
1513 if (xfs_sb_version_hasreflink(&mp
->m_sb
)) {
1515 "DAX and reflink cannot be used together!");
1517 goto out_filestream_unmount
;
1521 if (mp
->m_flags
& XFS_MOUNT_DISCARD
) {
1522 struct request_queue
*q
= bdev_get_queue(sb
->s_bdev
);
1524 if (!blk_queue_discard(q
)) {
1525 xfs_warn(mp
, "mounting with \"discard\" option, but "
1526 "the device does not support discard");
1527 mp
->m_flags
&= ~XFS_MOUNT_DISCARD
;
1531 if (xfs_sb_version_hasreflink(&mp
->m_sb
)) {
1532 if (mp
->m_sb
.sb_rblocks
) {
1534 "reflink not compatible with realtime device!");
1536 goto out_filestream_unmount
;
1539 if (xfs_globals
.always_cow
) {
1540 xfs_info(mp
, "using DEBUG-only always_cow mode.");
1541 mp
->m_always_cow
= true;
1545 if (xfs_sb_version_hasrmapbt(&mp
->m_sb
) && mp
->m_sb
.sb_rblocks
) {
1547 "reverse mapping btree not compatible with realtime device!");
1549 goto out_filestream_unmount
;
1552 error
= xfs_mountfs(mp
);
1554 goto out_filestream_unmount
;
1556 root
= igrab(VFS_I(mp
->m_rootip
));
1561 sb
->s_root
= d_make_root(root
);
1569 out_filestream_unmount
:
1570 xfs_filestream_unmount(mp
);
1574 free_percpu(mp
->m_stats
.xs_stats
);
1575 out_destroy_counters
:
1576 xfs_destroy_percpu_counters(mp
);
1577 out_destroy_workqueues
:
1578 xfs_destroy_mount_workqueues(mp
);
1580 xfs_close_devices(mp
);
1582 sb
->s_fs_info
= NULL
;
1587 xfs_filestream_unmount(mp
);
1594 struct fs_context
*fc
)
1596 return get_tree_bdev(fc
, xfs_fc_fill_super
);
1601 struct xfs_mount
*mp
)
1603 struct xfs_sb
*sbp
= &mp
->m_sb
;
1606 if (mp
->m_flags
& XFS_MOUNT_NORECOVERY
) {
1608 "ro->rw transition prohibited on norecovery mount");
1612 if (XFS_SB_VERSION_NUM(sbp
) == XFS_SB_VERSION_5
&&
1613 xfs_sb_has_ro_compat_feature(sbp
, XFS_SB_FEAT_RO_COMPAT_UNKNOWN
)) {
1615 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1616 (sbp
->sb_features_ro_compat
&
1617 XFS_SB_FEAT_RO_COMPAT_UNKNOWN
));
1621 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
1624 * If this is the first remount to writeable state we might have some
1625 * superblock changes to update.
1627 if (mp
->m_update_sb
) {
1628 error
= xfs_sync_sb(mp
, false);
1630 xfs_warn(mp
, "failed to write sb changes");
1633 mp
->m_update_sb
= false;
1637 * Fill out the reserve pool if it is empty. Use the stashed value if
1638 * it is non-zero, otherwise go with the default.
1640 xfs_restore_resvblks(mp
);
1641 xfs_log_work_queue(mp
);
1643 /* Recover any CoW blocks that never got remapped. */
1644 error
= xfs_reflink_recover_cow(mp
);
1647 "Error %d recovering leftover CoW allocations.", error
);
1648 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1651 xfs_start_block_reaping(mp
);
1653 /* Create the per-AG metadata reservation pool .*/
1654 error
= xfs_fs_reserve_ag_blocks(mp
);
1655 if (error
&& error
!= -ENOSPC
)
1663 struct xfs_mount
*mp
)
1668 * Cancel background eofb scanning so it cannot race with the final
1669 * log force+buftarg wait and deadlock the remount.
1671 xfs_stop_block_reaping(mp
);
1673 /* Get rid of any leftover CoW reservations... */
1674 error
= xfs_icache_free_cowblocks(mp
, NULL
);
1676 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1680 /* Free the per-AG metadata reservation pool. */
1681 error
= xfs_fs_unreserve_ag_blocks(mp
);
1683 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1688 * Before we sync the metadata, we need to free up the reserve block
1689 * pool so that the used block count in the superblock on disk is
1690 * correct at the end of the remount. Stash the current* reserve pool
1691 * size so that if we get remounted rw, we can return it to the same
1694 xfs_save_resvblks(mp
);
1696 xfs_quiesce_attr(mp
);
1697 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
1703 * Logically we would return an error here to prevent users from believing
1704 * they might have changed mount options using remount which can't be changed.
1706 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1707 * arguments in some cases so we can't blindly reject options, but have to
1708 * check for each specified option if it actually differs from the currently
1709 * set option and only reject it if that's the case.
1711 * Until that is implemented we return success for every remount request, and
1712 * silently ignore all options that we can't actually change.
1716 struct fs_context
*fc
)
1718 struct xfs_mount
*mp
= XFS_M(fc
->root
->d_sb
);
1719 struct xfs_mount
*new_mp
= fc
->s_fs_info
;
1720 xfs_sb_t
*sbp
= &mp
->m_sb
;
1721 int flags
= fc
->sb_flags
;
1724 error
= xfs_fc_validate_params(new_mp
);
1728 sync_filesystem(mp
->m_super
);
1730 /* inode32 -> inode64 */
1731 if ((mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) &&
1732 !(new_mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
)) {
1733 mp
->m_flags
&= ~XFS_MOUNT_SMALL_INUMS
;
1734 mp
->m_maxagi
= xfs_set_inode_alloc(mp
, sbp
->sb_agcount
);
1737 /* inode64 -> inode32 */
1738 if (!(mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
) &&
1739 (new_mp
->m_flags
& XFS_MOUNT_SMALL_INUMS
)) {
1740 mp
->m_flags
|= XFS_MOUNT_SMALL_INUMS
;
1741 mp
->m_maxagi
= xfs_set_inode_alloc(mp
, sbp
->sb_agcount
);
1745 if ((mp
->m_flags
& XFS_MOUNT_RDONLY
) && !(flags
& SB_RDONLY
)) {
1746 error
= xfs_remount_rw(mp
);
1752 if (!(mp
->m_flags
& XFS_MOUNT_RDONLY
) && (flags
& SB_RDONLY
)) {
1753 error
= xfs_remount_ro(mp
);
1761 static void xfs_fc_free(
1762 struct fs_context
*fc
)
1764 struct xfs_mount
*mp
= fc
->s_fs_info
;
1767 * mp is stored in the fs_context when it is initialized.
1768 * mp is transferred to the superblock on a successful mount,
1769 * but if an error occurs before the transfer we have to free
1776 static const struct fs_context_operations xfs_context_ops
= {
1777 .parse_param
= xfs_fc_parse_param
,
1778 .get_tree
= xfs_fc_get_tree
,
1779 .reconfigure
= xfs_fc_reconfigure
,
1780 .free
= xfs_fc_free
,
1783 static int xfs_init_fs_context(
1784 struct fs_context
*fc
)
1786 struct xfs_mount
*mp
;
1788 mp
= kmem_alloc(sizeof(struct xfs_mount
), KM_ZERO
);
1792 spin_lock_init(&mp
->m_sb_lock
);
1793 spin_lock_init(&mp
->m_agirotor_lock
);
1794 INIT_RADIX_TREE(&mp
->m_perag_tree
, GFP_ATOMIC
);
1795 spin_lock_init(&mp
->m_perag_lock
);
1796 mutex_init(&mp
->m_growlock
);
1797 INIT_WORK(&mp
->m_flush_inodes_work
, xfs_flush_inodes_worker
);
1798 INIT_DELAYED_WORK(&mp
->m_reclaim_work
, xfs_reclaim_worker
);
1799 INIT_DELAYED_WORK(&mp
->m_eofblocks_work
, xfs_eofblocks_worker
);
1800 INIT_DELAYED_WORK(&mp
->m_cowblocks_work
, xfs_cowblocks_worker
);
1801 mp
->m_kobj
.kobject
.kset
= xfs_kset
;
1803 * We don't create the finobt per-ag space reservation until after log
1804 * recovery, so we must set this to true so that an ifree transaction
1805 * started during log recovery will not depend on space reservations
1806 * for finobt expansion.
1808 mp
->m_finobt_nores
= true;
1811 * These can be overridden by the mount option parsing.
1814 mp
->m_logbsize
= -1;
1815 mp
->m_allocsize_log
= 16; /* 64k */
1818 * Copy binary VFS mount flags we are interested in.
1820 if (fc
->sb_flags
& SB_RDONLY
)
1821 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
1822 if (fc
->sb_flags
& SB_DIRSYNC
)
1823 mp
->m_flags
|= XFS_MOUNT_DIRSYNC
;
1824 if (fc
->sb_flags
& SB_SYNCHRONOUS
)
1825 mp
->m_flags
|= XFS_MOUNT_WSYNC
;
1828 fc
->ops
= &xfs_context_ops
;
1833 static struct file_system_type xfs_fs_type
= {
1834 .owner
= THIS_MODULE
,
1836 .init_fs_context
= xfs_init_fs_context
,
1837 .parameters
= xfs_fs_parameters
,
1838 .kill_sb
= kill_block_super
,
1839 .fs_flags
= FS_REQUIRES_DEV
,
1841 MODULE_ALIAS_FS("xfs");
1844 xfs_init_zones(void)
1846 xfs_log_ticket_zone
= kmem_cache_create("xfs_log_ticket",
1847 sizeof(struct xlog_ticket
),
1849 if (!xfs_log_ticket_zone
)
1852 xfs_bmap_free_item_zone
= kmem_cache_create("xfs_bmap_free_item",
1853 sizeof(struct xfs_extent_free_item
),
1855 if (!xfs_bmap_free_item_zone
)
1856 goto out_destroy_log_ticket_zone
;
1858 xfs_btree_cur_zone
= kmem_cache_create("xfs_btree_cur",
1859 sizeof(struct xfs_btree_cur
),
1861 if (!xfs_btree_cur_zone
)
1862 goto out_destroy_bmap_free_item_zone
;
1864 xfs_da_state_zone
= kmem_cache_create("xfs_da_state",
1865 sizeof(struct xfs_da_state
),
1867 if (!xfs_da_state_zone
)
1868 goto out_destroy_btree_cur_zone
;
1870 xfs_ifork_zone
= kmem_cache_create("xfs_ifork",
1871 sizeof(struct xfs_ifork
),
1873 if (!xfs_ifork_zone
)
1874 goto out_destroy_da_state_zone
;
1876 xfs_trans_zone
= kmem_cache_create("xf_trans",
1877 sizeof(struct xfs_trans
),
1879 if (!xfs_trans_zone
)
1880 goto out_destroy_ifork_zone
;
1884 * The size of the zone allocated buf log item is the maximum
1885 * size possible under XFS. This wastes a little bit of memory,
1886 * but it is much faster.
1888 xfs_buf_item_zone
= kmem_cache_create("xfs_buf_item",
1889 sizeof(struct xfs_buf_log_item
),
1891 if (!xfs_buf_item_zone
)
1892 goto out_destroy_trans_zone
;
1894 xfs_efd_zone
= kmem_cache_create("xfs_efd_item",
1895 (sizeof(struct xfs_efd_log_item
) +
1896 (XFS_EFD_MAX_FAST_EXTENTS
- 1) *
1897 sizeof(struct xfs_extent
)),
1900 goto out_destroy_buf_item_zone
;
1902 xfs_efi_zone
= kmem_cache_create("xfs_efi_item",
1903 (sizeof(struct xfs_efi_log_item
) +
1904 (XFS_EFI_MAX_FAST_EXTENTS
- 1) *
1905 sizeof(struct xfs_extent
)),
1908 goto out_destroy_efd_zone
;
1910 xfs_inode_zone
= kmem_cache_create("xfs_inode",
1911 sizeof(struct xfs_inode
), 0,
1912 (SLAB_HWCACHE_ALIGN
|
1913 SLAB_RECLAIM_ACCOUNT
|
1914 SLAB_MEM_SPREAD
| SLAB_ACCOUNT
),
1915 xfs_fs_inode_init_once
);
1916 if (!xfs_inode_zone
)
1917 goto out_destroy_efi_zone
;
1919 xfs_ili_zone
= kmem_cache_create("xfs_ili",
1920 sizeof(struct xfs_inode_log_item
), 0,
1921 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
1924 goto out_destroy_inode_zone
;
1926 xfs_icreate_zone
= kmem_cache_create("xfs_icr",
1927 sizeof(struct xfs_icreate_item
),
1929 if (!xfs_icreate_zone
)
1930 goto out_destroy_ili_zone
;
1932 xfs_rud_zone
= kmem_cache_create("xfs_rud_item",
1933 sizeof(struct xfs_rud_log_item
),
1936 goto out_destroy_icreate_zone
;
1938 xfs_rui_zone
= kmem_cache_create("xfs_rui_item",
1939 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS
),
1942 goto out_destroy_rud_zone
;
1944 xfs_cud_zone
= kmem_cache_create("xfs_cud_item",
1945 sizeof(struct xfs_cud_log_item
),
1948 goto out_destroy_rui_zone
;
1950 xfs_cui_zone
= kmem_cache_create("xfs_cui_item",
1951 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS
),
1954 goto out_destroy_cud_zone
;
1956 xfs_bud_zone
= kmem_cache_create("xfs_bud_item",
1957 sizeof(struct xfs_bud_log_item
),
1960 goto out_destroy_cui_zone
;
1962 xfs_bui_zone
= kmem_cache_create("xfs_bui_item",
1963 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS
),
1966 goto out_destroy_bud_zone
;
1970 out_destroy_bud_zone
:
1971 kmem_cache_destroy(xfs_bud_zone
);
1972 out_destroy_cui_zone
:
1973 kmem_cache_destroy(xfs_cui_zone
);
1974 out_destroy_cud_zone
:
1975 kmem_cache_destroy(xfs_cud_zone
);
1976 out_destroy_rui_zone
:
1977 kmem_cache_destroy(xfs_rui_zone
);
1978 out_destroy_rud_zone
:
1979 kmem_cache_destroy(xfs_rud_zone
);
1980 out_destroy_icreate_zone
:
1981 kmem_cache_destroy(xfs_icreate_zone
);
1982 out_destroy_ili_zone
:
1983 kmem_cache_destroy(xfs_ili_zone
);
1984 out_destroy_inode_zone
:
1985 kmem_cache_destroy(xfs_inode_zone
);
1986 out_destroy_efi_zone
:
1987 kmem_cache_destroy(xfs_efi_zone
);
1988 out_destroy_efd_zone
:
1989 kmem_cache_destroy(xfs_efd_zone
);
1990 out_destroy_buf_item_zone
:
1991 kmem_cache_destroy(xfs_buf_item_zone
);
1992 out_destroy_trans_zone
:
1993 kmem_cache_destroy(xfs_trans_zone
);
1994 out_destroy_ifork_zone
:
1995 kmem_cache_destroy(xfs_ifork_zone
);
1996 out_destroy_da_state_zone
:
1997 kmem_cache_destroy(xfs_da_state_zone
);
1998 out_destroy_btree_cur_zone
:
1999 kmem_cache_destroy(xfs_btree_cur_zone
);
2000 out_destroy_bmap_free_item_zone
:
2001 kmem_cache_destroy(xfs_bmap_free_item_zone
);
2002 out_destroy_log_ticket_zone
:
2003 kmem_cache_destroy(xfs_log_ticket_zone
);
2009 xfs_destroy_zones(void)
2012 * Make sure all delayed rcu free are flushed before we
2016 kmem_cache_destroy(xfs_bui_zone
);
2017 kmem_cache_destroy(xfs_bud_zone
);
2018 kmem_cache_destroy(xfs_cui_zone
);
2019 kmem_cache_destroy(xfs_cud_zone
);
2020 kmem_cache_destroy(xfs_rui_zone
);
2021 kmem_cache_destroy(xfs_rud_zone
);
2022 kmem_cache_destroy(xfs_icreate_zone
);
2023 kmem_cache_destroy(xfs_ili_zone
);
2024 kmem_cache_destroy(xfs_inode_zone
);
2025 kmem_cache_destroy(xfs_efi_zone
);
2026 kmem_cache_destroy(xfs_efd_zone
);
2027 kmem_cache_destroy(xfs_buf_item_zone
);
2028 kmem_cache_destroy(xfs_trans_zone
);
2029 kmem_cache_destroy(xfs_ifork_zone
);
2030 kmem_cache_destroy(xfs_da_state_zone
);
2031 kmem_cache_destroy(xfs_btree_cur_zone
);
2032 kmem_cache_destroy(xfs_bmap_free_item_zone
);
2033 kmem_cache_destroy(xfs_log_ticket_zone
);
2037 xfs_init_workqueues(void)
2040 * The allocation workqueue can be used in memory reclaim situations
2041 * (writepage path), and parallelism is only limited by the number of
2042 * AGs in all the filesystems mounted. Hence use the default large
2043 * max_active value for this workqueue.
2045 xfs_alloc_wq
= alloc_workqueue("xfsalloc",
2046 WQ_MEM_RECLAIM
|WQ_FREEZABLE
, 0);
2050 xfs_discard_wq
= alloc_workqueue("xfsdiscard", WQ_UNBOUND
, 0);
2051 if (!xfs_discard_wq
)
2052 goto out_free_alloc_wq
;
2056 destroy_workqueue(xfs_alloc_wq
);
2061 xfs_destroy_workqueues(void)
2063 destroy_workqueue(xfs_discard_wq
);
2064 destroy_workqueue(xfs_alloc_wq
);
2072 xfs_check_ondisk_structs();
2074 printk(KERN_INFO XFS_VERSION_STRING
" with "
2075 XFS_BUILD_OPTIONS
" enabled\n");
2079 error
= xfs_init_zones();
2083 error
= xfs_init_workqueues();
2085 goto out_destroy_zones
;
2087 error
= xfs_mru_cache_init();
2089 goto out_destroy_wq
;
2091 error
= xfs_buf_init();
2093 goto out_mru_cache_uninit
;
2095 error
= xfs_init_procfs();
2097 goto out_buf_terminate
;
2099 error
= xfs_sysctl_register();
2101 goto out_cleanup_procfs
;
2103 xfs_kset
= kset_create_and_add("xfs", NULL
, fs_kobj
);
2106 goto out_sysctl_unregister
;
2109 xfsstats
.xs_kobj
.kobject
.kset
= xfs_kset
;
2111 xfsstats
.xs_stats
= alloc_percpu(struct xfsstats
);
2112 if (!xfsstats
.xs_stats
) {
2114 goto out_kset_unregister
;
2117 error
= xfs_sysfs_init(&xfsstats
.xs_kobj
, &xfs_stats_ktype
, NULL
,
2120 goto out_free_stats
;
2123 xfs_dbg_kobj
.kobject
.kset
= xfs_kset
;
2124 error
= xfs_sysfs_init(&xfs_dbg_kobj
, &xfs_dbg_ktype
, NULL
, "debug");
2126 goto out_remove_stats_kobj
;
2129 error
= xfs_qm_init();
2131 goto out_remove_dbg_kobj
;
2133 error
= register_filesystem(&xfs_fs_type
);
2140 out_remove_dbg_kobj
:
2142 xfs_sysfs_del(&xfs_dbg_kobj
);
2143 out_remove_stats_kobj
:
2145 xfs_sysfs_del(&xfsstats
.xs_kobj
);
2147 free_percpu(xfsstats
.xs_stats
);
2148 out_kset_unregister
:
2149 kset_unregister(xfs_kset
);
2150 out_sysctl_unregister
:
2151 xfs_sysctl_unregister();
2153 xfs_cleanup_procfs();
2155 xfs_buf_terminate();
2156 out_mru_cache_uninit
:
2157 xfs_mru_cache_uninit();
2159 xfs_destroy_workqueues();
2161 xfs_destroy_zones();
2170 unregister_filesystem(&xfs_fs_type
);
2172 xfs_sysfs_del(&xfs_dbg_kobj
);
2174 xfs_sysfs_del(&xfsstats
.xs_kobj
);
2175 free_percpu(xfsstats
.xs_stats
);
2176 kset_unregister(xfs_kset
);
2177 xfs_sysctl_unregister();
2178 xfs_cleanup_procfs();
2179 xfs_buf_terminate();
2180 xfs_mru_cache_uninit();
2181 xfs_destroy_workqueues();
2182 xfs_destroy_zones();
2183 xfs_uuid_table_free();
2186 module_init(init_xfs_fs
);
2187 module_exit(exit_xfs_fs
);
2189 MODULE_AUTHOR("Silicon Graphics, Inc.");
2190 MODULE_DESCRIPTION(XFS_VERSION_STRING
" with " XFS_BUILD_OPTIONS
" enabled");
2191 MODULE_LICENSE("GPL");