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
);
76 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
77 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_max_dents_per_leaf
= (sdp
->sd_sb
.sb_bsize
-
339 sizeof(struct gfs2_leaf
)) /
340 GFS2_MIN_DIRENT_SIZE
;
344 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
349 proto
= sdp
->sd_args
.ar_lockproto
;
350 table
= sdp
->sd_args
.ar_locktable
;
352 /* Try to autodetect */
354 if (!proto
[0] || !table
[0]) {
355 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
360 proto
= sdp
->sd_sb
.sb_lockproto
;
362 table
= sdp
->sd_sb
.sb_locktable
;
366 table
= sdp
->sd_vfs
->s_id
;
368 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
369 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
371 table
= sdp
->sd_table_name
;
372 while ((table
= strchr(table
, '/')))
378 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
386 error
= gfs2_glock_nq_num(sdp
,
387 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
388 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
391 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
395 error
= gfs2_glock_nq_num(sdp
,
396 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
398 LM_FLAG_NOEXP
| GL_EXACT
,
401 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
405 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
406 CREATE
, &sdp
->sd_rename_gl
);
408 fs_err(sdp
, "can't create rename glock: %d\n", error
);
412 error
= gfs2_glock_get(sdp
, GFS2_FREEZE_LOCK
, &gfs2_freeze_glops
,
413 CREATE
, &sdp
->sd_freeze_gl
);
415 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
422 gfs2_glock_put(sdp
->sd_freeze_gl
);
424 gfs2_glock_put(sdp
->sd_rename_gl
);
426 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
428 gfs2_glock_dq_uninit(mount_gh
);
433 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
434 u64 no_addr
, const char *name
)
436 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
437 struct dentry
*dentry
;
440 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0,
441 GFS2_BLKST_FREE
/* ignore */);
443 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
444 return PTR_ERR(inode
);
446 dentry
= d_make_root(inode
);
448 fs_err(sdp
, "can't alloc %s dentry\n", name
);
455 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
457 struct super_block
*sb
= sdp
->sd_vfs
;
458 struct gfs2_holder sb_gh
;
462 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
463 LM_ST_SHARED
, 0, &sb_gh
);
465 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
469 ret
= gfs2_read_sb(sdp
, silent
);
471 fs_err(sdp
, "can't read superblock: %d\n", ret
);
475 /* Set up the buffer cache and SB for real */
476 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
478 fs_err(sdp
, "FS block size (%u) is too small for device "
480 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
483 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
485 fs_err(sdp
, "FS block size (%u) is too big for machine "
487 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
490 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
492 /* Get the root inode */
493 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
494 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
498 /* Get the master inode */
499 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
500 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
502 dput(sdp
->sd_root_dir
);
505 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
507 gfs2_glock_dq_uninit(&sb_gh
);
511 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
513 char *message
= "FIRSTMOUNT=Done";
514 char *envp
[] = { message
, NULL
};
516 fs_info(sdp
, "first mount done, others may mount\n");
518 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
519 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
521 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
525 * gfs2_jindex_hold - Grab a lock on the jindex
526 * @sdp: The GFS2 superblock
527 * @ji_gh: the holder for the jindex glock
532 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
534 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
537 struct gfs2_jdesc
*jd
;
542 mutex_lock(&sdp
->sd_jindex_mutex
);
545 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
549 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
550 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
552 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
553 if (error
== -ENOENT
) {
558 gfs2_glock_dq_uninit(ji_gh
);
564 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
568 INIT_LIST_HEAD(&jd
->extent_list
);
569 INIT_LIST_HEAD(&jd
->jd_revoke_list
);
571 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
572 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
573 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
577 error
= PTR_ERR(jd
->jd_inode
);
582 spin_lock(&sdp
->sd_jindex_spin
);
583 jd
->jd_jid
= sdp
->sd_journals
++;
584 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
585 spin_unlock(&sdp
->sd_jindex_spin
);
588 mutex_unlock(&sdp
->sd_jindex_mutex
);
594 * check_journal_clean - Make sure a journal is clean for a spectator mount
595 * @sdp: The GFS2 superblock
596 * @jd: The journal descriptor
598 * Returns: 0 if the journal is clean or locked, else an error
600 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
603 struct gfs2_holder j_gh
;
604 struct gfs2_log_header_host head
;
605 struct gfs2_inode
*ip
;
607 ip
= GFS2_I(jd
->jd_inode
);
608 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
609 GL_EXACT
| GL_NOCACHE
, &j_gh
);
611 fs_err(sdp
, "Error locking journal for spectator mount.\n");
614 error
= gfs2_jdesc_check(jd
);
616 fs_err(sdp
, "Error checking journal for spectator mount.\n");
619 error
= gfs2_find_jhead(jd
, &head
);
621 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
624 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
626 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
627 "must not be a spectator.\n", jd
->jd_jid
);
631 gfs2_glock_dq_uninit(&j_gh
);
635 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
637 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
638 struct gfs2_holder ji_gh
;
639 struct gfs2_inode
*ip
;
648 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
649 if (IS_ERR(sdp
->sd_jindex
)) {
650 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
651 return PTR_ERR(sdp
->sd_jindex
);
654 /* Load in the journal index special file */
656 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
658 fs_err(sdp
, "can't read journal index: %d\n", error
);
663 if (!gfs2_jindex_size(sdp
)) {
664 fs_err(sdp
, "no journals!\n");
668 atomic_set(&sdp
->sd_log_blks_needed
, 0);
669 if (sdp
->sd_args
.ar_spectator
) {
670 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
671 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
672 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
673 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
675 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
676 fs_err(sdp
, "can't mount journal #%u\n",
677 sdp
->sd_lockstruct
.ls_jid
);
678 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
679 gfs2_jindex_size(sdp
),
680 gfs2_jindex_size(sdp
) - 1);
683 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
685 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
687 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
688 &sdp
->sd_journal_gh
);
690 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
694 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
695 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
696 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
699 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
701 goto fail_journal_gh
;
704 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
706 fs_err(sdp
, "my journal (%u) is bad: %d\n",
707 sdp
->sd_jdesc
->jd_jid
, error
);
710 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
711 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
712 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
714 /* Map the extents for this journal's blocks */
715 gfs2_map_journal_extents(sdp
, sdp
->sd_jdesc
);
717 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
719 if (sdp
->sd_lockstruct
.ls_first
) {
721 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
722 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
724 if (sdp
->sd_args
.ar_spectator
) {
725 error
= check_journal_clean(sdp
, jd
);
730 error
= gfs2_recover_journal(jd
, true);
732 fs_err(sdp
, "error recovering journal %u: %d\n",
738 gfs2_others_may_mount(sdp
);
739 } else if (!sdp
->sd_args
.ar_spectator
) {
740 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
742 fs_err(sdp
, "error recovering my journal: %d\n", error
);
747 sdp
->sd_log_idle
= 1;
748 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
749 gfs2_glock_dq_uninit(&ji_gh
);
751 INIT_WORK(&sdp
->sd_freeze_work
, gfs2_freeze_func
);
755 if (!sdp
->sd_args
.ar_spectator
)
756 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
758 if (!sdp
->sd_args
.ar_spectator
)
759 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
761 gfs2_jindex_free(sdp
);
763 gfs2_glock_dq_uninit(&ji_gh
);
765 iput(sdp
->sd_jindex
);
769 static struct lock_class_key gfs2_quota_imutex_key
;
771 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
774 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
779 error
= init_journal(sdp
, undo
);
780 complete_all(&sdp
->sd_journal_ready
);
784 /* Read in the master statfs inode */
785 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
786 if (IS_ERR(sdp
->sd_statfs_inode
)) {
787 error
= PTR_ERR(sdp
->sd_statfs_inode
);
788 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
792 /* Read in the resource index inode */
793 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
794 if (IS_ERR(sdp
->sd_rindex
)) {
795 error
= PTR_ERR(sdp
->sd_rindex
);
796 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
799 sdp
->sd_rindex_uptodate
= 0;
801 /* Read in the quota inode */
802 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
803 if (IS_ERR(sdp
->sd_quota_inode
)) {
804 error
= PTR_ERR(sdp
->sd_quota_inode
);
805 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
809 * i_rwsem on quota files is special. Since this inode is hidden system
810 * file, we are safe to define locking ourselves.
812 lockdep_set_class(&sdp
->sd_quota_inode
->i_rwsem
,
813 &gfs2_quota_imutex_key
);
815 error
= gfs2_rindex_update(sdp
);
822 iput(sdp
->sd_quota_inode
);
824 gfs2_clear_rgrpd(sdp
);
825 iput(sdp
->sd_rindex
);
827 iput(sdp
->sd_statfs_inode
);
829 init_journal(sdp
, UNDO
);
834 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
836 struct inode
*pn
= NULL
;
839 struct gfs2_inode
*ip
;
840 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
842 if (sdp
->sd_args
.ar_spectator
)
848 pn
= gfs2_lookup_simple(master
, "per_node");
851 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
855 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
856 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
857 if (IS_ERR(sdp
->sd_sc_inode
)) {
858 error
= PTR_ERR(sdp
->sd_sc_inode
);
859 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
863 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
864 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
865 if (IS_ERR(sdp
->sd_qc_inode
)) {
866 error
= PTR_ERR(sdp
->sd_qc_inode
);
867 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
874 ip
= GFS2_I(sdp
->sd_sc_inode
);
875 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
878 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
882 ip
= GFS2_I(sdp
->sd_qc_inode
);
883 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
886 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
893 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
895 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
897 iput(sdp
->sd_qc_inode
);
899 iput(sdp
->sd_sc_inode
);
905 static const match_table_t nolock_tokens
= {
906 { Opt_jid
, "jid=%d", },
910 static const struct lm_lockops nolock_ops
= {
911 .lm_proto_name
= "lock_nolock",
912 .lm_put_lock
= gfs2_glock_free
,
913 .lm_tokens
= &nolock_tokens
,
917 * gfs2_lm_mount - mount a locking protocol
918 * @sdp: the filesystem
919 * @args: mount arguments
920 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
925 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
927 const struct lm_lockops
*lm
;
928 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
929 struct gfs2_args
*args
= &sdp
->sd_args
;
930 const char *proto
= sdp
->sd_proto_name
;
931 const char *table
= sdp
->sd_table_name
;
935 if (!strcmp("lock_nolock", proto
)) {
937 sdp
->sd_args
.ar_localflocks
= 1;
938 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
939 } else if (!strcmp("lock_dlm", proto
)) {
943 pr_info("can't find protocol %s\n", proto
);
947 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
952 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
953 substring_t tmp
[MAX_OPT_ARGS
];
959 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
962 ret
= match_int(&tmp
[0], &option
);
963 if (ret
|| option
< 0)
965 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
970 /* Obsolete, but left for backward compat purposes */
973 ret
= match_int(&tmp
[0], &option
);
974 if (ret
|| (option
!= 0 && option
!= 1))
976 ls
->ls_first
= option
;
981 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
986 if (lm
->lm_mount
== NULL
) {
987 fs_info(sdp
, "Now mounting FS...\n");
988 complete_all(&sdp
->sd_locking_init
);
991 ret
= lm
->lm_mount(sdp
, table
);
993 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
994 complete_all(&sdp
->sd_locking_init
);
998 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1000 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1001 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1003 lm
->lm_unmount(sdp
);
1006 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1008 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1011 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, TASK_INTERRUPTIBLE
)
1015 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1017 struct super_block
*sb
= sdp
->sd_vfs
;
1020 char *envp
[] = { ro
, spectator
, NULL
};
1021 sprintf(ro
, "RDONLY=%d", sb_rdonly(sb
));
1022 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1023 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1027 * fill_super - Read in superblock
1028 * @sb: The VFS superblock
1029 * @data: Mount options
1030 * @silent: Don't complain if it's not a GFS2 filesystem
1035 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1037 struct gfs2_sbd
*sdp
;
1038 struct gfs2_holder mount_gh
;
1043 pr_warn("can't alloc struct gfs2_sbd\n");
1046 sdp
->sd_args
= *args
;
1048 if (sdp
->sd_args
.ar_spectator
) {
1049 sb
->s_flags
|= SB_RDONLY
;
1050 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1052 if (sdp
->sd_args
.ar_posix_acl
)
1053 sb
->s_flags
|= SB_POSIXACL
;
1054 if (sdp
->sd_args
.ar_nobarrier
)
1055 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1057 sb
->s_flags
|= SB_NOSEC
;
1058 sb
->s_magic
= GFS2_MAGIC
;
1059 sb
->s_op
= &gfs2_super_ops
;
1060 sb
->s_d_op
= &gfs2_dops
;
1061 sb
->s_export_op
= &gfs2_export_ops
;
1062 sb
->s_xattr
= gfs2_xattr_handlers
;
1063 sb
->s_qcop
= &gfs2_quotactl_ops
;
1064 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
1065 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1066 sb
->s_time_gran
= 1;
1067 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1069 /* Set up the buffer cache and fill in some fake block size values
1070 to allow us to read-in the on-disk superblock. */
1071 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1072 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1073 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1074 GFS2_BASIC_BLOCK_SHIFT
;
1075 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
1077 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1078 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1079 if (sdp
->sd_args
.ar_statfs_quantum
) {
1080 sdp
->sd_tune
.gt_statfs_slow
= 0;
1081 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1083 sdp
->sd_tune
.gt_statfs_slow
= 1;
1084 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1087 error
= init_names(sdp
, silent
);
1089 /* In this case, we haven't initialized sysfs, so we have to
1090 manually free the sdp. */
1091 free_percpu(sdp
->sd_lkstats
);
1093 sb
->s_fs_info
= NULL
;
1097 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s", sdp
->sd_table_name
);
1099 error
= gfs2_sys_fs_add(sdp
);
1101 * If we hit an error here, gfs2_sys_fs_add will have called function
1102 * kobject_put which causes the sysfs usage count to go to zero, which
1103 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1104 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1105 * kobject_put to free sdp.
1110 gfs2_create_debugfs_file(sdp
);
1112 error
= gfs2_lm_mount(sdp
, silent
);
1116 error
= init_locking(sdp
, &mount_gh
, DO
);
1120 error
= init_sb(sdp
, silent
);
1124 error
= wait_on_journal(sdp
);
1129 * If user space has failed to join the cluster or some similar
1130 * failure has occurred, then the journal id will contain a
1131 * negative (error) number. This will then be returned to the
1132 * caller (of the mount syscall). We do this even for spectator
1133 * mounts (which just write a jid of 0 to indicate "ok" even though
1134 * the jid is unused in the spectator case)
1136 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1137 error
= sdp
->sd_lockstruct
.ls_jid
;
1138 sdp
->sd_lockstruct
.ls_jid
= 0;
1142 if (sdp
->sd_args
.ar_spectator
)
1143 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.s",
1144 sdp
->sd_table_name
);
1146 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.%u",
1147 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1149 error
= init_inodes(sdp
, DO
);
1153 error
= init_per_node(sdp
, DO
);
1157 error
= gfs2_statfs_init(sdp
);
1159 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1163 if (!sb_rdonly(sb
)) {
1164 error
= gfs2_make_fs_rw(sdp
);
1166 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1171 gfs2_glock_dq_uninit(&mount_gh
);
1172 gfs2_online_uevent(sdp
);
1176 init_per_node(sdp
, UNDO
);
1178 init_inodes(sdp
, UNDO
);
1180 if (sdp
->sd_root_dir
)
1181 dput(sdp
->sd_root_dir
);
1182 if (sdp
->sd_master_dir
)
1183 dput(sdp
->sd_master_dir
);
1188 init_locking(sdp
, &mount_gh
, UNDO
);
1190 complete_all(&sdp
->sd_journal_ready
);
1191 gfs2_gl_hash_clear(sdp
);
1192 gfs2_lm_unmount(sdp
);
1194 gfs2_delete_debugfs_file(sdp
);
1195 free_percpu(sdp
->sd_lkstats
);
1196 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1197 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1198 gfs2_sys_fs_del(sdp
);
1199 sb
->s_fs_info
= NULL
;
1203 static int set_gfs2_super(struct super_block
*s
, void *data
)
1206 s
->s_dev
= s
->s_bdev
->bd_dev
;
1207 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1211 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1213 struct block_device
*bdev
= ptr
;
1214 return (bdev
== s
->s_bdev
);
1218 * gfs2_mount - Get the GFS2 superblock
1219 * @fs_type: The GFS2 filesystem type
1220 * @flags: Mount flags
1221 * @dev_name: The name of the device
1222 * @data: The mount arguments
1224 * Q. Why not use get_sb_bdev() ?
1225 * A. We need to select one of two root directories to mount, independent
1226 * of whether this is the initial, or subsequent, mount of this sb
1228 * Returns: 0 or -ve on error
1231 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1232 const char *dev_name
, void *data
)
1234 struct block_device
*bdev
;
1235 struct super_block
*s
;
1236 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1238 struct gfs2_args args
;
1239 struct gfs2_sbd
*sdp
;
1241 if (!(flags
& SB_RDONLY
))
1242 mode
|= FMODE_WRITE
;
1244 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1246 return ERR_CAST(bdev
);
1249 * once the super is inserted into the list by sget, s_umount
1250 * will protect the lockfs code from trying to start a snapshot
1251 * while we are mounting
1253 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1254 if (bdev
->bd_fsfreeze_count
> 0) {
1255 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1259 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1260 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1267 * s_umount nests inside bd_mutex during
1268 * __invalidate_device(). blkdev_put() acquires
1269 * bd_mutex and can't be called under s_umount. Drop
1270 * s_umount temporarily. This is safe as we're
1271 * holding an active reference.
1273 up_write(&s
->s_umount
);
1274 blkdev_put(bdev
, mode
);
1275 down_write(&s
->s_umount
);
1277 /* s_mode must be set before deactivate_locked_super calls */
1281 memset(&args
, 0, sizeof(args
));
1282 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1283 args
.ar_data
= GFS2_DATA_DEFAULT
;
1284 args
.ar_commit
= 30;
1285 args
.ar_statfs_quantum
= 30;
1286 args
.ar_quota_quantum
= 60;
1287 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1289 error
= gfs2_mount_args(&args
, data
);
1291 pr_warn("can't parse mount arguments\n");
1297 if ((flags
^ s
->s_flags
) & SB_RDONLY
)
1300 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1301 sb_set_blocksize(s
, block_size(bdev
));
1302 error
= fill_super(s
, &args
, flags
& SB_SILENT
? 1 : 0);
1305 s
->s_flags
|= SB_ACTIVE
;
1311 return dget(sdp
->sd_master_dir
);
1313 return dget(sdp
->sd_root_dir
);
1316 deactivate_locked_super(s
);
1317 return ERR_PTR(error
);
1319 blkdev_put(bdev
, mode
);
1320 return ERR_PTR(error
);
1323 static int set_meta_super(struct super_block
*s
, void *ptr
)
1328 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1329 int flags
, const char *dev_name
, void *data
)
1331 struct super_block
*s
;
1332 struct gfs2_sbd
*sdp
;
1336 if (!dev_name
|| !*dev_name
)
1337 return ERR_PTR(-EINVAL
);
1339 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1341 pr_warn("path_lookup on %s returned error %d\n",
1343 return ERR_PTR(error
);
1345 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1346 path
.dentry
->d_sb
->s_bdev
);
1349 pr_warn("gfs2 mount does not exist\n");
1352 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1353 deactivate_locked_super(s
);
1354 return ERR_PTR(-EBUSY
);
1357 return dget(sdp
->sd_master_dir
);
1360 static void gfs2_kill_sb(struct super_block
*sb
)
1362 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1365 kill_block_super(sb
);
1369 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_SYNC
| GFS2_LFC_KILL_SB
);
1370 dput(sdp
->sd_root_dir
);
1371 dput(sdp
->sd_master_dir
);
1372 sdp
->sd_root_dir
= NULL
;
1373 sdp
->sd_master_dir
= NULL
;
1374 shrink_dcache_sb(sb
);
1375 free_percpu(sdp
->sd_lkstats
);
1376 kill_block_super(sb
);
1379 struct file_system_type gfs2_fs_type
= {
1381 .fs_flags
= FS_REQUIRES_DEV
,
1382 .mount
= gfs2_mount
,
1383 .kill_sb
= gfs2_kill_sb
,
1384 .owner
= THIS_MODULE
,
1386 MODULE_ALIAS_FS("gfs2");
1388 struct file_system_type gfs2meta_fs_type
= {
1390 .fs_flags
= FS_REQUIRES_DEV
,
1391 .mount
= gfs2_mount_meta
,
1392 .owner
= THIS_MODULE
,
1394 MODULE_ALIAS_FS("gfs2meta");