2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/blkdev.h>
18 #include <linux/kthread.h>
19 #include <linux/export.h>
20 #include <linux/namei.h>
21 #include <linux/mount.h>
22 #include <linux/gfs2_ondisk.h>
23 #include <linux/quotaops.h>
24 #include <linux/lockdep.h>
25 #include <linux/module.h>
26 #include <linux/backing-dev.h>
43 #include "trace_gfs2.h"
49 * gfs2_tune_init - Fill a gfs2_tune structure with default values
54 static void gfs2_tune_init(struct gfs2_tune
*gt
)
56 spin_lock_init(>
->gt_spin
);
58 gt
->gt_quota_warn_period
= 10;
59 gt
->gt_quota_scale_num
= 1;
60 gt
->gt_quota_scale_den
= 1;
61 gt
->gt_new_files_jdata
= 0;
62 gt
->gt_max_readahead
= BIT(18);
63 gt
->gt_complain_secs
= 10;
66 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
69 struct address_space
*mapping
;
71 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
77 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
78 if (!sdp
->sd_lkstats
) {
83 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
84 gfs2_tune_init(&sdp
->sd_tune
);
86 init_waitqueue_head(&sdp
->sd_glock_wait
);
87 atomic_set(&sdp
->sd_glock_disposal
, 0);
88 init_completion(&sdp
->sd_locking_init
);
89 init_completion(&sdp
->sd_wdack
);
90 spin_lock_init(&sdp
->sd_statfs_spin
);
92 spin_lock_init(&sdp
->sd_rindex_spin
);
93 sdp
->sd_rindex_tree
.rb_node
= NULL
;
95 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
96 spin_lock_init(&sdp
->sd_jindex_spin
);
97 mutex_init(&sdp
->sd_jindex_mutex
);
98 init_completion(&sdp
->sd_journal_ready
);
100 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
101 mutex_init(&sdp
->sd_quota_mutex
);
102 mutex_init(&sdp
->sd_quota_sync_mutex
);
103 init_waitqueue_head(&sdp
->sd_quota_wait
);
104 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
105 spin_lock_init(&sdp
->sd_trunc_lock
);
106 spin_lock_init(&sdp
->sd_bitmap_lock
);
108 mapping
= &sdp
->sd_aspace
;
110 address_space_init_once(mapping
);
111 mapping
->a_ops
= &gfs2_rgrp_aops
;
112 mapping
->host
= sb
->s_bdev
->bd_inode
;
114 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
115 mapping
->private_data
= NULL
;
116 mapping
->writeback_index
= 0;
118 spin_lock_init(&sdp
->sd_log_lock
);
119 atomic_set(&sdp
->sd_log_pinned
, 0);
120 INIT_LIST_HEAD(&sdp
->sd_log_le_revoke
);
121 INIT_LIST_HEAD(&sdp
->sd_log_le_ordered
);
122 spin_lock_init(&sdp
->sd_ordered_lock
);
124 init_waitqueue_head(&sdp
->sd_log_waitq
);
125 init_waitqueue_head(&sdp
->sd_logd_waitq
);
126 spin_lock_init(&sdp
->sd_ail_lock
);
127 INIT_LIST_HEAD(&sdp
->sd_ail1_list
);
128 INIT_LIST_HEAD(&sdp
->sd_ail2_list
);
130 init_rwsem(&sdp
->sd_log_flush_lock
);
131 atomic_set(&sdp
->sd_log_in_flight
, 0);
132 atomic_set(&sdp
->sd_reserving_log
, 0);
133 init_waitqueue_head(&sdp
->sd_reserving_log_wait
);
134 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
135 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
136 mutex_init(&sdp
->sd_freeze_mutex
);
143 * gfs2_check_sb - Check superblock
144 * @sdp: the filesystem
145 * @sb: The superblock
146 * @silent: Don't print a message if the check fails
148 * Checks the version code of the FS is one that we understand how to
149 * read and that the sizes of the various on-disk structures have not
153 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
155 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
157 if (sb
->sb_magic
!= GFS2_MAGIC
||
158 sb
->sb_type
!= GFS2_METATYPE_SB
) {
160 pr_warn("not a GFS2 filesystem\n");
164 /* If format numbers match exactly, we're done. */
166 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
167 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
170 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
175 static void end_bio_io_page(struct bio
*bio
)
177 struct page
*page
= bio
->bi_private
;
180 SetPageUptodate(page
);
182 pr_warn("error %d reading superblock\n", bio
->bi_status
);
186 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
188 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
189 struct super_block
*s
= sdp
->sd_vfs
;
190 const struct gfs2_sb
*str
= buf
;
192 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
193 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
194 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
195 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
196 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
197 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
198 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
199 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
200 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
201 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
202 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
204 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
205 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
206 memcpy(&s
->s_uuid
, str
->sb_uuid
, 16);
210 * gfs2_read_super - Read the gfs2 super block from disk
211 * @sdp: The GFS2 super block
212 * @sector: The location of the super block
213 * @error: The error code to return
215 * This uses the bio functions to read the super block from disk
216 * because we want to be 100% sure that we never read cached data.
217 * A super block is read twice only during each GFS2 mount and is
218 * never written to by the filesystem. The first time its read no
219 * locks are held, and the only details which are looked at are those
220 * relating to the locking protocol. Once locking is up and working,
221 * the sb is read again under the lock to establish the location of
222 * the master directory (contains pointers to journals etc) and the
225 * Returns: 0 on success or error
228 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
230 struct super_block
*sb
= sdp
->sd_vfs
;
235 page
= alloc_page(GFP_NOFS
);
239 ClearPageUptodate(page
);
240 ClearPageDirty(page
);
243 bio
= bio_alloc(GFP_NOFS
, 1);
244 bio
->bi_iter
.bi_sector
= sector
* (sb
->s_blocksize
>> 9);
245 bio_set_dev(bio
, sb
->s_bdev
);
246 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
248 bio
->bi_end_io
= end_bio_io_page
;
249 bio
->bi_private
= page
;
250 bio_set_op_attrs(bio
, REQ_OP_READ
, REQ_META
);
252 wait_on_page_locked(page
);
254 if (!PageUptodate(page
)) {
262 return gfs2_check_sb(sdp
, silent
);
266 * gfs2_read_sb - Read super block
267 * @sdp: The GFS2 superblock
268 * @silent: Don't print message if mount fails
272 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
274 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
279 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
282 fs_err(sdp
, "can't read superblock\n");
286 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
287 GFS2_BASIC_BLOCK_SHIFT
;
288 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
289 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
290 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
291 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
292 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
293 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
294 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
295 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
296 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
297 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
298 sizeof(struct gfs2_meta_header
)) /
299 sizeof(struct gfs2_quota_change
);
300 sdp
->sd_blocks_per_bitmap
= (sdp
->sd_sb
.sb_bsize
-
301 sizeof(struct gfs2_meta_header
))
302 * GFS2_NBBY
; /* not the rgrp bitmap, subsequent bitmaps only */
304 /* Compute maximum reservation required to add a entry to a directory */
306 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * BIT(GFS2_DIR_MAX_DEPTH
),
310 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
311 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
312 ind_blocks
+= tmp_blocks
;
315 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
317 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
319 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
320 sizeof(struct gfs2_dinode
);
321 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
326 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
328 m
= do_div(d
, sdp
->sd_inptrs
);
330 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
332 sdp
->sd_heightsize
[x
] = space
;
334 sdp
->sd_max_height
= x
;
335 sdp
->sd_heightsize
[x
] = ~0;
336 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
338 sdp
->sd_jheightsize
[0] = sdp
->sd_sb
.sb_bsize
-
339 sizeof(struct gfs2_dinode
);
340 sdp
->sd_jheightsize
[1] = sdp
->sd_jbsize
* sdp
->sd_diptrs
;
345 space
= sdp
->sd_jheightsize
[x
- 1] * sdp
->sd_inptrs
;
347 m
= do_div(d
, sdp
->sd_inptrs
);
349 if (d
!= sdp
->sd_jheightsize
[x
- 1] || m
)
351 sdp
->sd_jheightsize
[x
] = space
;
353 sdp
->sd_max_jheight
= x
;
354 sdp
->sd_jheightsize
[x
] = ~0;
355 gfs2_assert(sdp
, sdp
->sd_max_jheight
<= GFS2_MAX_META_HEIGHT
);
357 sdp
->sd_max_dents_per_leaf
= (sdp
->sd_sb
.sb_bsize
-
358 sizeof(struct gfs2_leaf
)) /
359 GFS2_MIN_DIRENT_SIZE
;
363 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
368 proto
= sdp
->sd_args
.ar_lockproto
;
369 table
= sdp
->sd_args
.ar_locktable
;
371 /* Try to autodetect */
373 if (!proto
[0] || !table
[0]) {
374 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
379 proto
= sdp
->sd_sb
.sb_lockproto
;
381 table
= sdp
->sd_sb
.sb_locktable
;
385 table
= sdp
->sd_vfs
->s_id
;
387 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
388 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
390 table
= sdp
->sd_table_name
;
391 while ((table
= strchr(table
, '/')))
397 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
405 error
= gfs2_glock_nq_num(sdp
,
406 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
407 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
410 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
414 error
= gfs2_glock_nq_num(sdp
,
415 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
417 LM_FLAG_NOEXP
| GL_EXACT
,
420 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
424 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
425 CREATE
, &sdp
->sd_rename_gl
);
427 fs_err(sdp
, "can't create rename glock: %d\n", error
);
431 error
= gfs2_glock_get(sdp
, GFS2_FREEZE_LOCK
, &gfs2_freeze_glops
,
432 CREATE
, &sdp
->sd_freeze_gl
);
434 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
441 gfs2_glock_put(sdp
->sd_freeze_gl
);
443 gfs2_glock_put(sdp
->sd_rename_gl
);
445 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
447 gfs2_glock_dq_uninit(mount_gh
);
452 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
453 u64 no_addr
, const char *name
)
455 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
456 struct dentry
*dentry
;
459 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0,
460 GFS2_BLKST_FREE
/* ignore */);
462 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
463 return PTR_ERR(inode
);
465 dentry
= d_make_root(inode
);
467 fs_err(sdp
, "can't alloc %s dentry\n", name
);
474 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
476 struct super_block
*sb
= sdp
->sd_vfs
;
477 struct gfs2_holder sb_gh
;
481 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
482 LM_ST_SHARED
, 0, &sb_gh
);
484 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
488 ret
= gfs2_read_sb(sdp
, silent
);
490 fs_err(sdp
, "can't read superblock: %d\n", ret
);
494 /* Set up the buffer cache and SB for real */
495 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
497 fs_err(sdp
, "FS block size (%u) is too small for device "
499 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
502 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
504 fs_err(sdp
, "FS block size (%u) is too big for machine "
506 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
509 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
511 /* Get the root inode */
512 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
513 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
517 /* Get the master inode */
518 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
519 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
521 dput(sdp
->sd_root_dir
);
524 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
526 gfs2_glock_dq_uninit(&sb_gh
);
530 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
532 char *message
= "FIRSTMOUNT=Done";
533 char *envp
[] = { message
, NULL
};
535 fs_info(sdp
, "first mount done, others may mount\n");
537 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
538 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
540 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
544 * gfs2_jindex_hold - Grab a lock on the jindex
545 * @sdp: The GFS2 superblock
546 * @ji_gh: the holder for the jindex glock
551 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
553 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
556 struct gfs2_jdesc
*jd
;
561 mutex_lock(&sdp
->sd_jindex_mutex
);
564 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
568 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
569 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
571 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
572 if (error
== -ENOENT
) {
577 gfs2_glock_dq_uninit(ji_gh
);
583 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
587 INIT_LIST_HEAD(&jd
->extent_list
);
588 INIT_LIST_HEAD(&jd
->jd_revoke_list
);
590 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
591 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
592 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
596 error
= PTR_ERR(jd
->jd_inode
);
601 spin_lock(&sdp
->sd_jindex_spin
);
602 jd
->jd_jid
= sdp
->sd_journals
++;
603 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
604 spin_unlock(&sdp
->sd_jindex_spin
);
607 mutex_unlock(&sdp
->sd_jindex_mutex
);
613 * check_journal_clean - Make sure a journal is clean for a spectator mount
614 * @sdp: The GFS2 superblock
615 * @jd: The journal descriptor
617 * Returns: 0 if the journal is clean or locked, else an error
619 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
622 struct gfs2_holder j_gh
;
623 struct gfs2_log_header_host head
;
624 struct gfs2_inode
*ip
;
626 ip
= GFS2_I(jd
->jd_inode
);
627 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
628 GL_EXACT
| GL_NOCACHE
, &j_gh
);
630 fs_err(sdp
, "Error locking journal for spectator mount.\n");
633 error
= gfs2_jdesc_check(jd
);
635 fs_err(sdp
, "Error checking journal for spectator mount.\n");
638 error
= gfs2_find_jhead(jd
, &head
);
640 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
643 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
645 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
646 "must not be a spectator.\n", jd
->jd_jid
);
650 gfs2_glock_dq_uninit(&j_gh
);
654 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
656 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
657 struct gfs2_holder ji_gh
;
658 struct gfs2_inode
*ip
;
667 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
668 if (IS_ERR(sdp
->sd_jindex
)) {
669 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
670 return PTR_ERR(sdp
->sd_jindex
);
673 /* Load in the journal index special file */
675 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
677 fs_err(sdp
, "can't read journal index: %d\n", error
);
682 if (!gfs2_jindex_size(sdp
)) {
683 fs_err(sdp
, "no journals!\n");
687 atomic_set(&sdp
->sd_log_blks_needed
, 0);
688 if (sdp
->sd_args
.ar_spectator
) {
689 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
690 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
691 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
692 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
694 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
695 fs_err(sdp
, "can't mount journal #%u\n",
696 sdp
->sd_lockstruct
.ls_jid
);
697 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
698 gfs2_jindex_size(sdp
),
699 gfs2_jindex_size(sdp
) - 1);
702 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
704 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
706 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
707 &sdp
->sd_journal_gh
);
709 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
713 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
714 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
715 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
718 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
720 goto fail_journal_gh
;
723 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
725 fs_err(sdp
, "my journal (%u) is bad: %d\n",
726 sdp
->sd_jdesc
->jd_jid
, error
);
729 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
730 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
731 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
733 /* Map the extents for this journal's blocks */
734 gfs2_map_journal_extents(sdp
, sdp
->sd_jdesc
);
736 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
738 if (sdp
->sd_lockstruct
.ls_first
) {
740 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
741 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
743 if (sdp
->sd_args
.ar_spectator
) {
744 error
= check_journal_clean(sdp
, jd
);
749 error
= gfs2_recover_journal(jd
, true);
751 fs_err(sdp
, "error recovering journal %u: %d\n",
757 gfs2_others_may_mount(sdp
);
758 } else if (!sdp
->sd_args
.ar_spectator
) {
759 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
761 fs_err(sdp
, "error recovering my journal: %d\n", error
);
766 sdp
->sd_log_idle
= 1;
767 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
768 gfs2_glock_dq_uninit(&ji_gh
);
770 INIT_WORK(&sdp
->sd_freeze_work
, gfs2_freeze_func
);
774 if (!sdp
->sd_args
.ar_spectator
)
775 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
777 if (!sdp
->sd_args
.ar_spectator
)
778 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
780 gfs2_jindex_free(sdp
);
782 gfs2_glock_dq_uninit(&ji_gh
);
784 iput(sdp
->sd_jindex
);
788 static struct lock_class_key gfs2_quota_imutex_key
;
790 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
793 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
798 error
= init_journal(sdp
, undo
);
799 complete_all(&sdp
->sd_journal_ready
);
803 /* Read in the master statfs inode */
804 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
805 if (IS_ERR(sdp
->sd_statfs_inode
)) {
806 error
= PTR_ERR(sdp
->sd_statfs_inode
);
807 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
811 /* Read in the resource index inode */
812 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
813 if (IS_ERR(sdp
->sd_rindex
)) {
814 error
= PTR_ERR(sdp
->sd_rindex
);
815 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
818 sdp
->sd_rindex_uptodate
= 0;
820 /* Read in the quota inode */
821 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
822 if (IS_ERR(sdp
->sd_quota_inode
)) {
823 error
= PTR_ERR(sdp
->sd_quota_inode
);
824 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
828 * i_mutex on quota files is special. Since this inode is hidden system
829 * file, we are safe to define locking ourselves.
831 lockdep_set_class(&sdp
->sd_quota_inode
->i_rwsem
,
832 &gfs2_quota_imutex_key
);
834 error
= gfs2_rindex_update(sdp
);
841 iput(sdp
->sd_quota_inode
);
843 gfs2_clear_rgrpd(sdp
);
844 iput(sdp
->sd_rindex
);
846 iput(sdp
->sd_statfs_inode
);
848 init_journal(sdp
, UNDO
);
853 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
855 struct inode
*pn
= NULL
;
858 struct gfs2_inode
*ip
;
859 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
861 if (sdp
->sd_args
.ar_spectator
)
867 pn
= gfs2_lookup_simple(master
, "per_node");
870 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
874 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
875 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
876 if (IS_ERR(sdp
->sd_sc_inode
)) {
877 error
= PTR_ERR(sdp
->sd_sc_inode
);
878 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
882 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
883 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
884 if (IS_ERR(sdp
->sd_qc_inode
)) {
885 error
= PTR_ERR(sdp
->sd_qc_inode
);
886 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
893 ip
= GFS2_I(sdp
->sd_sc_inode
);
894 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
897 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
901 ip
= GFS2_I(sdp
->sd_qc_inode
);
902 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
905 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
912 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
914 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
916 iput(sdp
->sd_qc_inode
);
918 iput(sdp
->sd_sc_inode
);
924 static const match_table_t nolock_tokens
= {
925 { Opt_jid
, "jid=%d\n", },
929 static const struct lm_lockops nolock_ops
= {
930 .lm_proto_name
= "lock_nolock",
931 .lm_put_lock
= gfs2_glock_free
,
932 .lm_tokens
= &nolock_tokens
,
936 * gfs2_lm_mount - mount a locking protocol
937 * @sdp: the filesystem
938 * @args: mount arguments
939 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
944 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
946 const struct lm_lockops
*lm
;
947 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
948 struct gfs2_args
*args
= &sdp
->sd_args
;
949 const char *proto
= sdp
->sd_proto_name
;
950 const char *table
= sdp
->sd_table_name
;
954 if (!strcmp("lock_nolock", proto
)) {
956 sdp
->sd_args
.ar_localflocks
= 1;
957 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
958 } else if (!strcmp("lock_dlm", proto
)) {
962 pr_info("can't find protocol %s\n", proto
);
966 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
971 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
972 substring_t tmp
[MAX_OPT_ARGS
];
978 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
981 ret
= match_int(&tmp
[0], &option
);
982 if (ret
|| option
< 0)
984 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
989 /* Obsolete, but left for backward compat purposes */
992 ret
= match_int(&tmp
[0], &option
);
993 if (ret
|| (option
!= 0 && option
!= 1))
995 ls
->ls_first
= option
;
1000 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
1005 if (lm
->lm_mount
== NULL
) {
1006 fs_info(sdp
, "Now mounting FS...\n");
1007 complete_all(&sdp
->sd_locking_init
);
1010 ret
= lm
->lm_mount(sdp
, table
);
1012 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
1013 complete_all(&sdp
->sd_locking_init
);
1017 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1019 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1020 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1022 lm
->lm_unmount(sdp
);
1025 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1027 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1030 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, TASK_INTERRUPTIBLE
)
1034 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1036 struct super_block
*sb
= sdp
->sd_vfs
;
1039 char *envp
[] = { ro
, spectator
, NULL
};
1040 sprintf(ro
, "RDONLY=%d", sb_rdonly(sb
));
1041 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1042 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1046 * fill_super - Read in superblock
1047 * @sb: The VFS superblock
1048 * @data: Mount options
1049 * @silent: Don't complain if it's not a GFS2 filesystem
1054 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1056 struct gfs2_sbd
*sdp
;
1057 struct gfs2_holder mount_gh
;
1062 pr_warn("can't alloc struct gfs2_sbd\n");
1065 sdp
->sd_args
= *args
;
1067 if (sdp
->sd_args
.ar_spectator
) {
1068 sb
->s_flags
|= SB_RDONLY
;
1069 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1071 if (sdp
->sd_args
.ar_posix_acl
)
1072 sb
->s_flags
|= SB_POSIXACL
;
1073 if (sdp
->sd_args
.ar_nobarrier
)
1074 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1076 sb
->s_flags
|= SB_NOSEC
;
1077 sb
->s_magic
= GFS2_MAGIC
;
1078 sb
->s_op
= &gfs2_super_ops
;
1079 sb
->s_d_op
= &gfs2_dops
;
1080 sb
->s_export_op
= &gfs2_export_ops
;
1081 sb
->s_xattr
= gfs2_xattr_handlers
;
1082 sb
->s_qcop
= &gfs2_quotactl_ops
;
1083 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
1084 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1085 sb
->s_time_gran
= 1;
1086 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1088 /* Set up the buffer cache and fill in some fake block size values
1089 to allow us to read-in the on-disk superblock. */
1090 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1091 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1092 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1093 GFS2_BASIC_BLOCK_SHIFT
;
1094 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
1096 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1097 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1098 if (sdp
->sd_args
.ar_statfs_quantum
) {
1099 sdp
->sd_tune
.gt_statfs_slow
= 0;
1100 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1102 sdp
->sd_tune
.gt_statfs_slow
= 1;
1103 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1106 error
= init_names(sdp
, silent
);
1108 /* In this case, we haven't initialized sysfs, so we have to
1109 manually free the sdp. */
1110 free_percpu(sdp
->sd_lkstats
);
1112 sb
->s_fs_info
= NULL
;
1116 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s", sdp
->sd_table_name
);
1118 error
= gfs2_sys_fs_add(sdp
);
1120 * If we hit an error here, gfs2_sys_fs_add will have called function
1121 * kobject_put which causes the sysfs usage count to go to zero, which
1122 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1123 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1124 * kobject_put to free sdp.
1129 gfs2_create_debugfs_file(sdp
);
1131 error
= gfs2_lm_mount(sdp
, silent
);
1135 error
= init_locking(sdp
, &mount_gh
, DO
);
1139 error
= init_sb(sdp
, silent
);
1143 error
= wait_on_journal(sdp
);
1148 * If user space has failed to join the cluster or some similar
1149 * failure has occurred, then the journal id will contain a
1150 * negative (error) number. This will then be returned to the
1151 * caller (of the mount syscall). We do this even for spectator
1152 * mounts (which just write a jid of 0 to indicate "ok" even though
1153 * the jid is unused in the spectator case)
1155 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1156 error
= sdp
->sd_lockstruct
.ls_jid
;
1157 sdp
->sd_lockstruct
.ls_jid
= 0;
1161 if (sdp
->sd_args
.ar_spectator
)
1162 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.s",
1163 sdp
->sd_table_name
);
1165 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.%u",
1166 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1168 error
= init_inodes(sdp
, DO
);
1172 error
= init_per_node(sdp
, DO
);
1176 error
= gfs2_statfs_init(sdp
);
1178 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1182 if (!sb_rdonly(sb
)) {
1183 error
= gfs2_make_fs_rw(sdp
);
1185 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1190 gfs2_glock_dq_uninit(&mount_gh
);
1191 gfs2_online_uevent(sdp
);
1195 init_per_node(sdp
, UNDO
);
1197 init_inodes(sdp
, UNDO
);
1199 if (sdp
->sd_root_dir
)
1200 dput(sdp
->sd_root_dir
);
1201 if (sdp
->sd_master_dir
)
1202 dput(sdp
->sd_master_dir
);
1207 init_locking(sdp
, &mount_gh
, UNDO
);
1209 complete_all(&sdp
->sd_journal_ready
);
1210 gfs2_gl_hash_clear(sdp
);
1211 gfs2_lm_unmount(sdp
);
1213 gfs2_delete_debugfs_file(sdp
);
1214 free_percpu(sdp
->sd_lkstats
);
1215 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1216 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1217 gfs2_sys_fs_del(sdp
);
1218 sb
->s_fs_info
= NULL
;
1222 static int set_gfs2_super(struct super_block
*s
, void *data
)
1225 s
->s_dev
= s
->s_bdev
->bd_dev
;
1226 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1230 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1232 struct block_device
*bdev
= ptr
;
1233 return (bdev
== s
->s_bdev
);
1237 * gfs2_mount - Get the GFS2 superblock
1238 * @fs_type: The GFS2 filesystem type
1239 * @flags: Mount flags
1240 * @dev_name: The name of the device
1241 * @data: The mount arguments
1243 * Q. Why not use get_sb_bdev() ?
1244 * A. We need to select one of two root directories to mount, independent
1245 * of whether this is the initial, or subsequent, mount of this sb
1247 * Returns: 0 or -ve on error
1250 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1251 const char *dev_name
, void *data
)
1253 struct block_device
*bdev
;
1254 struct super_block
*s
;
1255 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1257 struct gfs2_args args
;
1258 struct gfs2_sbd
*sdp
;
1260 if (!(flags
& SB_RDONLY
))
1261 mode
|= FMODE_WRITE
;
1263 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1265 return ERR_CAST(bdev
);
1268 * once the super is inserted into the list by sget, s_umount
1269 * will protect the lockfs code from trying to start a snapshot
1270 * while we are mounting
1272 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1273 if (bdev
->bd_fsfreeze_count
> 0) {
1274 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1278 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1279 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1286 * s_umount nests inside bd_mutex during
1287 * __invalidate_device(). blkdev_put() acquires
1288 * bd_mutex and can't be called under s_umount. Drop
1289 * s_umount temporarily. This is safe as we're
1290 * holding an active reference.
1292 up_write(&s
->s_umount
);
1293 blkdev_put(bdev
, mode
);
1294 down_write(&s
->s_umount
);
1296 /* s_mode must be set before deactivate_locked_super calls */
1300 memset(&args
, 0, sizeof(args
));
1301 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1302 args
.ar_data
= GFS2_DATA_DEFAULT
;
1303 args
.ar_commit
= 30;
1304 args
.ar_statfs_quantum
= 30;
1305 args
.ar_quota_quantum
= 60;
1306 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1308 error
= gfs2_mount_args(&args
, data
);
1310 pr_warn("can't parse mount arguments\n");
1316 if ((flags
^ s
->s_flags
) & SB_RDONLY
)
1319 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1320 sb_set_blocksize(s
, block_size(bdev
));
1321 error
= fill_super(s
, &args
, flags
& SB_SILENT
? 1 : 0);
1324 s
->s_flags
|= SB_ACTIVE
;
1330 return dget(sdp
->sd_master_dir
);
1332 return dget(sdp
->sd_root_dir
);
1335 deactivate_locked_super(s
);
1336 return ERR_PTR(error
);
1338 blkdev_put(bdev
, mode
);
1339 return ERR_PTR(error
);
1342 static int set_meta_super(struct super_block
*s
, void *ptr
)
1347 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1348 int flags
, const char *dev_name
, void *data
)
1350 struct super_block
*s
;
1351 struct gfs2_sbd
*sdp
;
1355 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1357 pr_warn("path_lookup on %s returned error %d\n",
1359 return ERR_PTR(error
);
1361 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1362 path
.dentry
->d_sb
->s_bdev
);
1365 pr_warn("gfs2 mount does not exist\n");
1368 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1369 deactivate_locked_super(s
);
1370 return ERR_PTR(-EBUSY
);
1373 return dget(sdp
->sd_master_dir
);
1376 static void gfs2_kill_sb(struct super_block
*sb
)
1378 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1381 kill_block_super(sb
);
1385 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_SYNC
| GFS2_LFC_KILL_SB
);
1386 dput(sdp
->sd_root_dir
);
1387 dput(sdp
->sd_master_dir
);
1388 sdp
->sd_root_dir
= NULL
;
1389 sdp
->sd_master_dir
= NULL
;
1390 shrink_dcache_sb(sb
);
1391 free_percpu(sdp
->sd_lkstats
);
1392 kill_block_super(sb
);
1395 struct file_system_type gfs2_fs_type
= {
1397 .fs_flags
= FS_REQUIRES_DEV
,
1398 .mount
= gfs2_mount
,
1399 .kill_sb
= gfs2_kill_sb
,
1400 .owner
= THIS_MODULE
,
1402 MODULE_ALIAS_FS("gfs2");
1404 struct file_system_type gfs2meta_fs_type
= {
1406 .fs_flags
= FS_REQUIRES_DEV
,
1407 .mount
= gfs2_mount_meta
,
1408 .owner
= THIS_MODULE
,
1410 MODULE_ALIAS_FS("gfs2meta");