1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
40 #include "trace_gfs2.h"
47 * gfs2_tune_init - Fill a gfs2_tune structure with default values
52 static void gfs2_tune_init(struct gfs2_tune
*gt
)
54 spin_lock_init(>
->gt_spin
);
56 gt
->gt_quota_warn_period
= 10;
57 gt
->gt_quota_scale_num
= 1;
58 gt
->gt_quota_scale_den
= 1;
59 gt
->gt_new_files_jdata
= 0;
60 gt
->gt_max_readahead
= BIT(18);
61 gt
->gt_complain_secs
= 10;
64 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
67 struct address_space
*mapping
;
69 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
74 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
75 if (!sdp
->sd_lkstats
) {
81 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
82 gfs2_tune_init(&sdp
->sd_tune
);
84 init_waitqueue_head(&sdp
->sd_glock_wait
);
85 atomic_set(&sdp
->sd_glock_disposal
, 0);
86 init_completion(&sdp
->sd_locking_init
);
87 init_completion(&sdp
->sd_wdack
);
88 spin_lock_init(&sdp
->sd_statfs_spin
);
90 spin_lock_init(&sdp
->sd_rindex_spin
);
91 sdp
->sd_rindex_tree
.rb_node
= NULL
;
93 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
94 spin_lock_init(&sdp
->sd_jindex_spin
);
95 mutex_init(&sdp
->sd_jindex_mutex
);
96 init_completion(&sdp
->sd_journal_ready
);
98 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
99 mutex_init(&sdp
->sd_quota_mutex
);
100 mutex_init(&sdp
->sd_quota_sync_mutex
);
101 init_waitqueue_head(&sdp
->sd_quota_wait
);
102 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
103 spin_lock_init(&sdp
->sd_trunc_lock
);
104 spin_lock_init(&sdp
->sd_bitmap_lock
);
106 mapping
= &sdp
->sd_aspace
;
108 address_space_init_once(mapping
);
109 mapping
->a_ops
= &gfs2_rgrp_aops
;
110 mapping
->host
= sb
->s_bdev
->bd_inode
;
112 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
113 mapping
->private_data
= NULL
;
114 mapping
->writeback_index
= 0;
116 spin_lock_init(&sdp
->sd_log_lock
);
117 atomic_set(&sdp
->sd_log_pinned
, 0);
118 INIT_LIST_HEAD(&sdp
->sd_log_revokes
);
119 INIT_LIST_HEAD(&sdp
->sd_log_ordered
);
120 spin_lock_init(&sdp
->sd_ordered_lock
);
122 init_waitqueue_head(&sdp
->sd_log_waitq
);
123 init_waitqueue_head(&sdp
->sd_logd_waitq
);
124 spin_lock_init(&sdp
->sd_ail_lock
);
125 INIT_LIST_HEAD(&sdp
->sd_ail1_list
);
126 INIT_LIST_HEAD(&sdp
->sd_ail2_list
);
128 init_rwsem(&sdp
->sd_log_flush_lock
);
129 atomic_set(&sdp
->sd_log_in_flight
, 0);
130 atomic_set(&sdp
->sd_reserving_log
, 0);
131 init_waitqueue_head(&sdp
->sd_reserving_log_wait
);
132 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
133 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
134 mutex_init(&sdp
->sd_freeze_mutex
);
141 * gfs2_check_sb - Check superblock
142 * @sdp: the filesystem
143 * @sb: The superblock
144 * @silent: Don't print a message if the check fails
146 * Checks the version code of the FS is one that we understand how to
147 * read and that the sizes of the various on-disk structures have not
151 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
153 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
155 if (sb
->sb_magic
!= GFS2_MAGIC
||
156 sb
->sb_type
!= GFS2_METATYPE_SB
) {
158 pr_warn("not a GFS2 filesystem\n");
162 /* If format numbers match exactly, we're done. */
164 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
165 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
168 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
173 static void end_bio_io_page(struct bio
*bio
)
175 struct page
*page
= bio
->bi_private
;
178 SetPageUptodate(page
);
180 pr_warn("error %d reading superblock\n", bio
->bi_status
);
184 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
186 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
187 struct super_block
*s
= sdp
->sd_vfs
;
188 const struct gfs2_sb
*str
= buf
;
190 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
191 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
192 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
193 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
194 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
195 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
196 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
197 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
198 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
199 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
200 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
202 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
203 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
204 memcpy(&s
->s_uuid
, str
->sb_uuid
, 16);
208 * gfs2_read_super - Read the gfs2 super block from disk
209 * @sdp: The GFS2 super block
210 * @sector: The location of the super block
211 * @error: The error code to return
213 * This uses the bio functions to read the super block from disk
214 * because we want to be 100% sure that we never read cached data.
215 * A super block is read twice only during each GFS2 mount and is
216 * never written to by the filesystem. The first time its read no
217 * locks are held, and the only details which are looked at are those
218 * relating to the locking protocol. Once locking is up and working,
219 * the sb is read again under the lock to establish the location of
220 * the master directory (contains pointers to journals etc) and the
223 * Returns: 0 on success or error
226 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
228 struct super_block
*sb
= sdp
->sd_vfs
;
233 page
= alloc_page(GFP_NOFS
);
237 ClearPageUptodate(page
);
238 ClearPageDirty(page
);
241 bio
= bio_alloc(GFP_NOFS
, 1);
242 bio
->bi_iter
.bi_sector
= sector
* (sb
->s_blocksize
>> 9);
243 bio_set_dev(bio
, sb
->s_bdev
);
244 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
246 bio
->bi_end_io
= end_bio_io_page
;
247 bio
->bi_private
= page
;
248 bio_set_op_attrs(bio
, REQ_OP_READ
, REQ_META
);
250 wait_on_page_locked(page
);
252 if (!PageUptodate(page
)) {
260 return gfs2_check_sb(sdp
, silent
);
264 * gfs2_read_sb - Read super block
265 * @sdp: The GFS2 superblock
266 * @silent: Don't print message if mount fails
270 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
272 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
277 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
280 fs_err(sdp
, "can't read superblock\n");
284 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
285 GFS2_BASIC_BLOCK_SHIFT
;
286 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
287 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
288 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
289 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
290 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
291 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
292 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
293 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
294 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
295 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
296 sizeof(struct gfs2_meta_header
)) /
297 sizeof(struct gfs2_quota_change
);
298 sdp
->sd_blocks_per_bitmap
= (sdp
->sd_sb
.sb_bsize
-
299 sizeof(struct gfs2_meta_header
))
300 * GFS2_NBBY
; /* not the rgrp bitmap, subsequent bitmaps only */
302 /* Compute maximum reservation required to add a entry to a directory */
304 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * BIT(GFS2_DIR_MAX_DEPTH
),
308 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
309 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
310 ind_blocks
+= tmp_blocks
;
313 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
315 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
317 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
318 sizeof(struct gfs2_dinode
);
319 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
324 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
326 m
= do_div(d
, sdp
->sd_inptrs
);
328 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
330 sdp
->sd_heightsize
[x
] = space
;
332 sdp
->sd_max_height
= x
;
333 sdp
->sd_heightsize
[x
] = ~0;
334 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
336 sdp
->sd_max_dents_per_leaf
= (sdp
->sd_sb
.sb_bsize
-
337 sizeof(struct gfs2_leaf
)) /
338 GFS2_MIN_DIRENT_SIZE
;
342 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
347 proto
= sdp
->sd_args
.ar_lockproto
;
348 table
= sdp
->sd_args
.ar_locktable
;
350 /* Try to autodetect */
352 if (!proto
[0] || !table
[0]) {
353 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
358 proto
= sdp
->sd_sb
.sb_lockproto
;
360 table
= sdp
->sd_sb
.sb_locktable
;
364 table
= sdp
->sd_vfs
->s_id
;
366 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
367 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
369 table
= sdp
->sd_table_name
;
370 while ((table
= strchr(table
, '/')))
376 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
384 error
= gfs2_glock_nq_num(sdp
,
385 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
386 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
389 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
393 error
= gfs2_glock_nq_num(sdp
,
394 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
396 LM_FLAG_NOEXP
| GL_EXACT
,
399 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
403 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
404 CREATE
, &sdp
->sd_rename_gl
);
406 fs_err(sdp
, "can't create rename glock: %d\n", error
);
410 error
= gfs2_glock_get(sdp
, GFS2_FREEZE_LOCK
, &gfs2_freeze_glops
,
411 CREATE
, &sdp
->sd_freeze_gl
);
413 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
420 gfs2_glock_put(sdp
->sd_freeze_gl
);
422 gfs2_glock_put(sdp
->sd_rename_gl
);
424 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
426 gfs2_glock_dq_uninit(mount_gh
);
431 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
432 u64 no_addr
, const char *name
)
434 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
435 struct dentry
*dentry
;
438 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0,
439 GFS2_BLKST_FREE
/* ignore */);
441 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
442 return PTR_ERR(inode
);
444 dentry
= d_make_root(inode
);
446 fs_err(sdp
, "can't alloc %s dentry\n", name
);
453 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
455 struct super_block
*sb
= sdp
->sd_vfs
;
456 struct gfs2_holder sb_gh
;
460 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
461 LM_ST_SHARED
, 0, &sb_gh
);
463 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
467 ret
= gfs2_read_sb(sdp
, silent
);
469 fs_err(sdp
, "can't read superblock: %d\n", ret
);
473 /* Set up the buffer cache and SB for real */
474 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
476 fs_err(sdp
, "FS block size (%u) is too small for device "
478 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
481 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
483 fs_err(sdp
, "FS block size (%u) is too big for machine "
485 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
488 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
490 /* Get the root inode */
491 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
492 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
496 /* Get the master inode */
497 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
498 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
500 dput(sdp
->sd_root_dir
);
503 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
505 gfs2_glock_dq_uninit(&sb_gh
);
509 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
511 char *message
= "FIRSTMOUNT=Done";
512 char *envp
[] = { message
, NULL
};
514 fs_info(sdp
, "first mount done, others may mount\n");
516 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
517 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
519 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
523 * gfs2_jindex_hold - Grab a lock on the jindex
524 * @sdp: The GFS2 superblock
525 * @ji_gh: the holder for the jindex glock
530 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
532 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
535 struct gfs2_jdesc
*jd
;
540 mutex_lock(&sdp
->sd_jindex_mutex
);
543 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
547 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
548 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
550 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
551 if (error
== -ENOENT
) {
556 gfs2_glock_dq_uninit(ji_gh
);
562 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
566 INIT_LIST_HEAD(&jd
->extent_list
);
567 INIT_LIST_HEAD(&jd
->jd_revoke_list
);
569 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
570 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
571 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
575 error
= PTR_ERR(jd
->jd_inode
);
580 spin_lock(&sdp
->sd_jindex_spin
);
581 jd
->jd_jid
= sdp
->sd_journals
++;
582 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
583 spin_unlock(&sdp
->sd_jindex_spin
);
586 mutex_unlock(&sdp
->sd_jindex_mutex
);
592 * check_journal_clean - Make sure a journal is clean for a spectator mount
593 * @sdp: The GFS2 superblock
594 * @jd: The journal descriptor
596 * Returns: 0 if the journal is clean or locked, else an error
598 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
601 struct gfs2_holder j_gh
;
602 struct gfs2_log_header_host head
;
603 struct gfs2_inode
*ip
;
605 ip
= GFS2_I(jd
->jd_inode
);
606 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
607 GL_EXACT
| GL_NOCACHE
, &j_gh
);
609 fs_err(sdp
, "Error locking journal for spectator mount.\n");
612 error
= gfs2_jdesc_check(jd
);
614 fs_err(sdp
, "Error checking journal for spectator mount.\n");
617 error
= gfs2_find_jhead(jd
, &head
, false);
619 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
622 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
624 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
625 "must not be a spectator.\n", jd
->jd_jid
);
629 gfs2_glock_dq_uninit(&j_gh
);
633 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
635 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
636 struct gfs2_holder ji_gh
;
637 struct gfs2_inode
*ip
;
646 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
647 if (IS_ERR(sdp
->sd_jindex
)) {
648 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
649 return PTR_ERR(sdp
->sd_jindex
);
652 /* Load in the journal index special file */
654 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
656 fs_err(sdp
, "can't read journal index: %d\n", error
);
661 if (!gfs2_jindex_size(sdp
)) {
662 fs_err(sdp
, "no journals!\n");
666 atomic_set(&sdp
->sd_log_blks_needed
, 0);
667 if (sdp
->sd_args
.ar_spectator
) {
668 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
669 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
670 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
671 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
673 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
674 fs_err(sdp
, "can't mount journal #%u\n",
675 sdp
->sd_lockstruct
.ls_jid
);
676 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
677 gfs2_jindex_size(sdp
),
678 gfs2_jindex_size(sdp
) - 1);
681 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
683 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
685 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
686 &sdp
->sd_journal_gh
);
688 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
692 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
693 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
694 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
697 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
699 goto fail_journal_gh
;
702 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
704 fs_err(sdp
, "my journal (%u) is bad: %d\n",
705 sdp
->sd_jdesc
->jd_jid
, error
);
708 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
709 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
710 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
712 /* Map the extents for this journal's blocks */
713 gfs2_map_journal_extents(sdp
, sdp
->sd_jdesc
);
715 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
717 if (sdp
->sd_lockstruct
.ls_first
) {
719 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
720 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
722 if (sdp
->sd_args
.ar_spectator
) {
723 error
= check_journal_clean(sdp
, jd
);
728 error
= gfs2_recover_journal(jd
, true);
730 fs_err(sdp
, "error recovering journal %u: %d\n",
736 gfs2_others_may_mount(sdp
);
737 } else if (!sdp
->sd_args
.ar_spectator
) {
738 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
740 fs_err(sdp
, "error recovering my journal: %d\n", error
);
745 sdp
->sd_log_idle
= 1;
746 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
747 gfs2_glock_dq_uninit(&ji_gh
);
749 INIT_WORK(&sdp
->sd_freeze_work
, gfs2_freeze_func
);
753 if (!sdp
->sd_args
.ar_spectator
)
754 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
756 if (!sdp
->sd_args
.ar_spectator
)
757 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
759 gfs2_jindex_free(sdp
);
761 gfs2_glock_dq_uninit(&ji_gh
);
763 iput(sdp
->sd_jindex
);
767 static struct lock_class_key gfs2_quota_imutex_key
;
769 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
772 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
777 error
= init_journal(sdp
, undo
);
778 complete_all(&sdp
->sd_journal_ready
);
782 /* Read in the master statfs inode */
783 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
784 if (IS_ERR(sdp
->sd_statfs_inode
)) {
785 error
= PTR_ERR(sdp
->sd_statfs_inode
);
786 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
790 /* Read in the resource index inode */
791 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
792 if (IS_ERR(sdp
->sd_rindex
)) {
793 error
= PTR_ERR(sdp
->sd_rindex
);
794 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
797 sdp
->sd_rindex_uptodate
= 0;
799 /* Read in the quota inode */
800 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
801 if (IS_ERR(sdp
->sd_quota_inode
)) {
802 error
= PTR_ERR(sdp
->sd_quota_inode
);
803 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
807 * i_rwsem on quota files is special. Since this inode is hidden system
808 * file, we are safe to define locking ourselves.
810 lockdep_set_class(&sdp
->sd_quota_inode
->i_rwsem
,
811 &gfs2_quota_imutex_key
);
813 error
= gfs2_rindex_update(sdp
);
820 iput(sdp
->sd_quota_inode
);
822 gfs2_clear_rgrpd(sdp
);
823 iput(sdp
->sd_rindex
);
825 iput(sdp
->sd_statfs_inode
);
827 init_journal(sdp
, UNDO
);
832 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
834 struct inode
*pn
= NULL
;
837 struct gfs2_inode
*ip
;
838 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
840 if (sdp
->sd_args
.ar_spectator
)
846 pn
= gfs2_lookup_simple(master
, "per_node");
849 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
853 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
854 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
855 if (IS_ERR(sdp
->sd_sc_inode
)) {
856 error
= PTR_ERR(sdp
->sd_sc_inode
);
857 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
861 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
862 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
863 if (IS_ERR(sdp
->sd_qc_inode
)) {
864 error
= PTR_ERR(sdp
->sd_qc_inode
);
865 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
872 ip
= GFS2_I(sdp
->sd_sc_inode
);
873 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
876 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
880 ip
= GFS2_I(sdp
->sd_qc_inode
);
881 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
884 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
891 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
893 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
895 iput(sdp
->sd_qc_inode
);
897 iput(sdp
->sd_sc_inode
);
903 static const match_table_t nolock_tokens
= {
904 { Opt_jid
, "jid=%d\n", },
908 static const struct lm_lockops nolock_ops
= {
909 .lm_proto_name
= "lock_nolock",
910 .lm_put_lock
= gfs2_glock_free
,
911 .lm_tokens
= &nolock_tokens
,
915 * gfs2_lm_mount - mount a locking protocol
916 * @sdp: the filesystem
917 * @args: mount arguments
918 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
923 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
925 const struct lm_lockops
*lm
;
926 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
927 struct gfs2_args
*args
= &sdp
->sd_args
;
928 const char *proto
= sdp
->sd_proto_name
;
929 const char *table
= sdp
->sd_table_name
;
933 if (!strcmp("lock_nolock", proto
)) {
935 sdp
->sd_args
.ar_localflocks
= 1;
936 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
937 } else if (!strcmp("lock_dlm", proto
)) {
941 pr_info("can't find protocol %s\n", proto
);
945 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
950 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
951 substring_t tmp
[MAX_OPT_ARGS
];
957 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
960 ret
= match_int(&tmp
[0], &option
);
961 if (ret
|| option
< 0)
963 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
968 /* Obsolete, but left for backward compat purposes */
971 ret
= match_int(&tmp
[0], &option
);
972 if (ret
|| (option
!= 0 && option
!= 1))
974 ls
->ls_first
= option
;
979 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
984 if (lm
->lm_mount
== NULL
) {
985 fs_info(sdp
, "Now mounting FS...\n");
986 complete_all(&sdp
->sd_locking_init
);
989 ret
= lm
->lm_mount(sdp
, table
);
991 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
992 complete_all(&sdp
->sd_locking_init
);
996 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
998 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
999 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1001 lm
->lm_unmount(sdp
);
1004 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1006 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1009 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, TASK_INTERRUPTIBLE
)
1013 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1015 struct super_block
*sb
= sdp
->sd_vfs
;
1018 char *envp
[] = { ro
, spectator
, NULL
};
1019 sprintf(ro
, "RDONLY=%d", sb_rdonly(sb
));
1020 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1021 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1025 * fill_super - Read in superblock
1026 * @sb: The VFS superblock
1027 * @data: Mount options
1028 * @silent: Don't complain if it's not a GFS2 filesystem
1033 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1035 struct gfs2_sbd
*sdp
;
1036 struct gfs2_holder mount_gh
;
1041 pr_warn("can't alloc struct gfs2_sbd\n");
1044 sdp
->sd_args
= *args
;
1046 if (sdp
->sd_args
.ar_spectator
) {
1047 sb
->s_flags
|= SB_RDONLY
;
1048 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1050 if (sdp
->sd_args
.ar_posix_acl
)
1051 sb
->s_flags
|= SB_POSIXACL
;
1052 if (sdp
->sd_args
.ar_nobarrier
)
1053 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1055 sb
->s_flags
|= SB_NOSEC
;
1056 sb
->s_magic
= GFS2_MAGIC
;
1057 sb
->s_op
= &gfs2_super_ops
;
1058 sb
->s_d_op
= &gfs2_dops
;
1059 sb
->s_export_op
= &gfs2_export_ops
;
1060 sb
->s_xattr
= gfs2_xattr_handlers
;
1061 sb
->s_qcop
= &gfs2_quotactl_ops
;
1062 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
1063 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1064 sb
->s_time_gran
= 1;
1065 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1067 /* Set up the buffer cache and fill in some fake block size values
1068 to allow us to read-in the on-disk superblock. */
1069 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1070 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1071 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1072 GFS2_BASIC_BLOCK_SHIFT
;
1073 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
1075 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1076 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1077 if (sdp
->sd_args
.ar_statfs_quantum
) {
1078 sdp
->sd_tune
.gt_statfs_slow
= 0;
1079 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1081 sdp
->sd_tune
.gt_statfs_slow
= 1;
1082 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1085 error
= init_names(sdp
, silent
);
1087 /* In this case, we haven't initialized sysfs, so we have to
1088 manually free the sdp. */
1089 free_percpu(sdp
->sd_lkstats
);
1091 sb
->s_fs_info
= NULL
;
1095 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s", sdp
->sd_table_name
);
1097 error
= gfs2_sys_fs_add(sdp
);
1099 * If we hit an error here, gfs2_sys_fs_add will have called function
1100 * kobject_put which causes the sysfs usage count to go to zero, which
1101 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1102 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1103 * kobject_put to free sdp.
1108 gfs2_create_debugfs_file(sdp
);
1110 error
= gfs2_lm_mount(sdp
, silent
);
1114 error
= init_locking(sdp
, &mount_gh
, DO
);
1118 error
= init_sb(sdp
, silent
);
1122 error
= wait_on_journal(sdp
);
1127 * If user space has failed to join the cluster or some similar
1128 * failure has occurred, then the journal id will contain a
1129 * negative (error) number. This will then be returned to the
1130 * caller (of the mount syscall). We do this even for spectator
1131 * mounts (which just write a jid of 0 to indicate "ok" even though
1132 * the jid is unused in the spectator case)
1134 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1135 error
= sdp
->sd_lockstruct
.ls_jid
;
1136 sdp
->sd_lockstruct
.ls_jid
= 0;
1140 if (sdp
->sd_args
.ar_spectator
)
1141 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.s",
1142 sdp
->sd_table_name
);
1144 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.%u",
1145 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1147 error
= init_inodes(sdp
, DO
);
1151 error
= init_per_node(sdp
, DO
);
1155 error
= gfs2_statfs_init(sdp
);
1157 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1161 if (!sb_rdonly(sb
)) {
1162 error
= gfs2_make_fs_rw(sdp
);
1164 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1169 gfs2_glock_dq_uninit(&mount_gh
);
1170 gfs2_online_uevent(sdp
);
1174 init_per_node(sdp
, UNDO
);
1176 init_inodes(sdp
, UNDO
);
1178 if (sdp
->sd_root_dir
)
1179 dput(sdp
->sd_root_dir
);
1180 if (sdp
->sd_master_dir
)
1181 dput(sdp
->sd_master_dir
);
1186 init_locking(sdp
, &mount_gh
, UNDO
);
1188 complete_all(&sdp
->sd_journal_ready
);
1189 gfs2_gl_hash_clear(sdp
);
1190 gfs2_lm_unmount(sdp
);
1192 gfs2_delete_debugfs_file(sdp
);
1193 free_percpu(sdp
->sd_lkstats
);
1194 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1195 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1196 gfs2_sys_fs_del(sdp
);
1197 sb
->s_fs_info
= NULL
;
1201 static int set_gfs2_super(struct super_block
*s
, void *data
)
1204 s
->s_dev
= s
->s_bdev
->bd_dev
;
1205 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1209 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1211 struct block_device
*bdev
= ptr
;
1212 return (bdev
== s
->s_bdev
);
1216 * gfs2_mount - Get the GFS2 superblock
1217 * @fs_type: The GFS2 filesystem type
1218 * @flags: Mount flags
1219 * @dev_name: The name of the device
1220 * @data: The mount arguments
1222 * Q. Why not use get_sb_bdev() ?
1223 * A. We need to select one of two root directories to mount, independent
1224 * of whether this is the initial, or subsequent, mount of this sb
1226 * Returns: 0 or -ve on error
1229 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1230 const char *dev_name
, void *data
)
1232 struct block_device
*bdev
;
1233 struct super_block
*s
;
1234 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1236 struct gfs2_args args
;
1237 struct gfs2_sbd
*sdp
;
1239 if (!(flags
& SB_RDONLY
))
1240 mode
|= FMODE_WRITE
;
1242 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1244 return ERR_CAST(bdev
);
1247 * once the super is inserted into the list by sget, s_umount
1248 * will protect the lockfs code from trying to start a snapshot
1249 * while we are mounting
1251 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1252 if (bdev
->bd_fsfreeze_count
> 0) {
1253 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1257 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1258 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1265 * s_umount nests inside bd_mutex during
1266 * __invalidate_device(). blkdev_put() acquires
1267 * bd_mutex and can't be called under s_umount. Drop
1268 * s_umount temporarily. This is safe as we're
1269 * holding an active reference.
1271 up_write(&s
->s_umount
);
1272 blkdev_put(bdev
, mode
);
1273 down_write(&s
->s_umount
);
1275 /* s_mode must be set before deactivate_locked_super calls */
1279 memset(&args
, 0, sizeof(args
));
1280 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1281 args
.ar_data
= GFS2_DATA_DEFAULT
;
1282 args
.ar_commit
= 30;
1283 args
.ar_statfs_quantum
= 30;
1284 args
.ar_quota_quantum
= 60;
1285 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1287 error
= gfs2_mount_args(&args
, data
);
1289 pr_warn("can't parse mount arguments\n");
1295 if ((flags
^ s
->s_flags
) & SB_RDONLY
)
1298 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1299 sb_set_blocksize(s
, block_size(bdev
));
1300 error
= fill_super(s
, &args
, flags
& SB_SILENT
? 1 : 0);
1303 s
->s_flags
|= SB_ACTIVE
;
1309 return dget(sdp
->sd_master_dir
);
1311 return dget(sdp
->sd_root_dir
);
1314 deactivate_locked_super(s
);
1315 return ERR_PTR(error
);
1317 blkdev_put(bdev
, mode
);
1318 return ERR_PTR(error
);
1321 static int set_meta_super(struct super_block
*s
, void *ptr
)
1326 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1327 int flags
, const char *dev_name
, void *data
)
1329 struct super_block
*s
;
1330 struct gfs2_sbd
*sdp
;
1334 if (!dev_name
|| !*dev_name
)
1335 return ERR_PTR(-EINVAL
);
1337 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1339 pr_warn("path_lookup on %s returned error %d\n",
1341 return ERR_PTR(error
);
1343 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1344 path
.dentry
->d_sb
->s_bdev
);
1347 pr_warn("gfs2 mount does not exist\n");
1350 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1351 deactivate_locked_super(s
);
1352 return ERR_PTR(-EBUSY
);
1355 return dget(sdp
->sd_master_dir
);
1358 static void gfs2_kill_sb(struct super_block
*sb
)
1360 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1363 kill_block_super(sb
);
1367 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_SYNC
| GFS2_LFC_KILL_SB
);
1368 dput(sdp
->sd_root_dir
);
1369 dput(sdp
->sd_master_dir
);
1370 sdp
->sd_root_dir
= NULL
;
1371 sdp
->sd_master_dir
= NULL
;
1372 shrink_dcache_sb(sb
);
1373 free_percpu(sdp
->sd_lkstats
);
1374 kill_block_super(sb
);
1377 struct file_system_type gfs2_fs_type
= {
1379 .fs_flags
= FS_REQUIRES_DEV
,
1380 .mount
= gfs2_mount
,
1381 .kill_sb
= gfs2_kill_sb
,
1382 .owner
= THIS_MODULE
,
1384 MODULE_ALIAS_FS("gfs2");
1386 struct file_system_type gfs2meta_fs_type
= {
1388 .fs_flags
= FS_REQUIRES_DEV
,
1389 .mount
= gfs2_mount_meta
,
1390 .owner
= THIS_MODULE
,
1392 MODULE_ALIAS_FS("gfs2meta");