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 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/blkdev.h>
16 #include <linux/kthread.h>
17 #include <linux/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
37 #include "trace_gfs2.h"
43 * gfs2_tune_init - Fill a gfs2_tune structure with default values
48 static void gfs2_tune_init(struct gfs2_tune *gt)
50 spin_lock_init(>->gt_spin);
52 gt->gt_quota_simul_sync = 64;
53 gt->gt_quota_warn_period = 10;
54 gt->gt_quota_scale_num = 1;
55 gt->gt_quota_scale_den = 1;
56 gt->gt_new_files_jdata = 0;
57 gt->gt_max_readahead = 1 << 18;
58 gt->gt_complain_secs = 10;
61 static struct gfs2_sbd *init_sbd(struct super_block *sb)
65 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
71 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
72 if (!sdp->sd_lkstats) {
77 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
78 gfs2_tune_init(&sdp->sd_tune);
80 init_waitqueue_head(&sdp->sd_glock_wait);
81 atomic_set(&sdp->sd_glock_disposal, 0);
82 init_completion(&sdp->sd_locking_init);
83 spin_lock_init(&sdp->sd_statfs_spin);
85 spin_lock_init(&sdp->sd_rindex_spin);
86 sdp->sd_rindex_tree.rb_node = NULL;
88 INIT_LIST_HEAD(&sdp->sd_jindex_list);
89 spin_lock_init(&sdp->sd_jindex_spin);
90 mutex_init(&sdp->sd_jindex_mutex);
92 INIT_LIST_HEAD(&sdp->sd_quota_list);
93 mutex_init(&sdp->sd_quota_mutex);
94 init_waitqueue_head(&sdp->sd_quota_wait);
95 INIT_LIST_HEAD(&sdp->sd_trunc_list);
96 spin_lock_init(&sdp->sd_trunc_lock);
98 spin_lock_init(&sdp->sd_log_lock);
99 atomic_set(&sdp->sd_log_pinned, 0);
100 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
101 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
102 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
103 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
105 init_waitqueue_head(&sdp->sd_log_waitq);
106 init_waitqueue_head(&sdp->sd_logd_waitq);
107 spin_lock_init(&sdp->sd_ail_lock);
108 INIT_LIST_HEAD(&sdp->sd_ail1_list);
109 INIT_LIST_HEAD(&sdp->sd_ail2_list);
111 init_rwsem(&sdp->sd_log_flush_lock);
112 atomic_set(&sdp->sd_log_in_flight, 0);
113 init_waitqueue_head(&sdp->sd_log_flush_wait);
115 INIT_LIST_HEAD(&sdp->sd_revoke_list);
117 mutex_init(&sdp->sd_freeze_lock);
124 * gfs2_check_sb - Check superblock
125 * @sdp: the filesystem
126 * @sb: The superblock
127 * @silent: Don't print a message if the check fails
129 * Checks the version code of the FS is one that we understand how to
130 * read and that the sizes of the various on-disk structures have not
134 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
136 struct gfs2_sb_host *sb = &sdp->sd_sb;
138 if (sb->sb_magic != GFS2_MAGIC ||
139 sb->sb_type != GFS2_METATYPE_SB) {
141 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
145 /* If format numbers match exactly, we're done. */
147 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
148 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
151 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
156 static void end_bio_io_page(struct bio *bio, int error)
158 struct page *page = bio->bi_private;
161 SetPageUptodate(page);
163 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
167 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
169 struct gfs2_sb_host *sb = &sdp->sd_sb;
170 struct super_block *s = sdp->sd_vfs;
171 const struct gfs2_sb *str = buf;
173 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
174 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
175 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
176 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
177 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
178 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
179 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
180 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
181 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
182 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
183 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
185 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
186 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
187 memcpy(s->s_uuid, str->sb_uuid, 16);
191 * gfs2_read_super - Read the gfs2 super block from disk
192 * @sdp: The GFS2 super block
193 * @sector: The location of the super block
194 * @error: The error code to return
196 * This uses the bio functions to read the super block from disk
197 * because we want to be 100% sure that we never read cached data.
198 * A super block is read twice only during each GFS2 mount and is
199 * never written to by the filesystem. The first time its read no
200 * locks are held, and the only details which are looked at are those
201 * relating to the locking protocol. Once locking is up and working,
202 * the sb is read again under the lock to establish the location of
203 * the master directory (contains pointers to journals etc) and the
206 * Returns: 0 on success or error
209 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
211 struct super_block *sb = sdp->sd_vfs;
216 page = alloc_page(GFP_NOFS);
220 ClearPageUptodate(page);
221 ClearPageDirty(page);
224 bio = bio_alloc(GFP_NOFS, 1);
225 bio->bi_sector = sector * (sb->s_blocksize >> 9);
226 bio->bi_bdev = sb->s_bdev;
227 bio_add_page(bio, page, PAGE_SIZE, 0);
229 bio->bi_end_io = end_bio_io_page;
230 bio->bi_private = page;
231 submit_bio(READ_SYNC | REQ_META, bio);
232 wait_on_page_locked(page);
234 if (!PageUptodate(page)) {
242 return gfs2_check_sb(sdp, silent);
246 * gfs2_read_sb - Read super block
247 * @sdp: The GFS2 superblock
248 * @silent: Don't print message if mount fails
252 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
254 u32 hash_blocks, ind_blocks, leaf_blocks;
259 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
262 fs_err(sdp, "can't read superblock\n");
266 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
267 GFS2_BASIC_BLOCK_SHIFT;
268 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
269 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
270 sizeof(struct gfs2_dinode)) / sizeof(u64);
271 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
272 sizeof(struct gfs2_meta_header)) / sizeof(u64);
273 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
274 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
275 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
276 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
277 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
278 sizeof(struct gfs2_meta_header)) /
279 sizeof(struct gfs2_quota_change);
281 /* Compute maximum reservation required to add a entry to a directory */
283 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
287 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
288 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
289 ind_blocks += tmp_blocks;
292 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
294 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
296 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
297 sizeof(struct gfs2_dinode);
298 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
303 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
305 m = do_div(d, sdp->sd_inptrs);
307 if (d != sdp->sd_heightsize[x - 1] || m)
309 sdp->sd_heightsize[x] = space;
311 sdp->sd_max_height = x;
312 sdp->sd_heightsize[x] = ~0;
313 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
315 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
316 sizeof(struct gfs2_dinode);
317 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
322 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
324 m = do_div(d, sdp->sd_inptrs);
326 if (d != sdp->sd_jheightsize[x - 1] || m)
328 sdp->sd_jheightsize[x] = space;
330 sdp->sd_max_jheight = x;
331 sdp->sd_jheightsize[x] = ~0;
332 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
337 static int init_names(struct gfs2_sbd *sdp, int silent)
342 proto = sdp->sd_args.ar_lockproto;
343 table = sdp->sd_args.ar_locktable;
345 /* Try to autodetect */
347 if (!proto[0] || !table[0]) {
348 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
353 proto = sdp->sd_sb.sb_lockproto;
355 table = sdp->sd_sb.sb_locktable;
359 table = sdp->sd_vfs->s_id;
361 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
362 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
364 table = sdp->sd_table_name;
365 while ((table = strchr(table, '/')))
371 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
379 error = gfs2_glock_nq_num(sdp,
380 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
381 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
384 fs_err(sdp, "can't acquire mount glock: %d\n", error);
388 error = gfs2_glock_nq_num(sdp,
389 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
391 LM_FLAG_NOEXP | GL_EXACT,
394 fs_err(sdp, "can't acquire live glock: %d\n", error);
398 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
399 CREATE, &sdp->sd_rename_gl);
401 fs_err(sdp, "can't create rename glock: %d\n", error);
405 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
406 CREATE, &sdp->sd_trans_gl);
408 fs_err(sdp, "can't create transaction glock: %d\n", error);
415 gfs2_glock_put(sdp->sd_trans_gl);
417 gfs2_glock_put(sdp->sd_rename_gl);
419 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
421 gfs2_glock_dq_uninit(mount_gh);
426 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
427 u64 no_addr, const char *name)
429 struct gfs2_sbd *sdp = sb->s_fs_info;
430 struct dentry *dentry;
433 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
435 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
436 return PTR_ERR(inode);
438 dentry = d_make_root(inode);
440 fs_err(sdp, "can't alloc %s dentry\n", name);
447 static int init_sb(struct gfs2_sbd *sdp, int silent)
449 struct super_block *sb = sdp->sd_vfs;
450 struct gfs2_holder sb_gh;
454 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
455 LM_ST_SHARED, 0, &sb_gh);
457 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
461 ret = gfs2_read_sb(sdp, silent);
463 fs_err(sdp, "can't read superblock: %d\n", ret);
467 /* Set up the buffer cache and SB for real */
468 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
470 fs_err(sdp, "FS block size (%u) is too small for device "
472 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
475 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
477 fs_err(sdp, "FS block size (%u) is too big for machine "
479 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
482 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
484 /* Get the root inode */
485 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
486 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
490 /* Get the master inode */
491 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
492 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
494 dput(sdp->sd_root_dir);
497 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
499 gfs2_glock_dq_uninit(&sb_gh);
504 * map_journal_extents - create a reusable "extent" mapping from all logical
505 * blocks to all physical blocks for the given journal. This will save
506 * us time when writing journal blocks. Most journals will have only one
507 * extent that maps all their logical blocks. That's because gfs2.mkfs
508 * arranges the journal blocks sequentially to maximize performance.
509 * So the extent would map the first block for the entire file length.
510 * However, gfs2_jadd can happen while file activity is happening, so
511 * those journals may not be sequential. Less likely is the case where
512 * the users created their own journals by mounting the metafs and
513 * laying it out. But it's still possible. These journals might have
516 * TODO: This should be done in bigger chunks rather than one block at a time,
517 * but since it's only done at mount time, I'm not worried about the
520 static int map_journal_extents(struct gfs2_sbd *sdp)
522 struct gfs2_jdesc *jd = sdp->sd_jdesc;
524 u64 db, prev_db; /* logical block, disk block, prev disk block */
525 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
526 struct gfs2_journal_extent *jext = NULL;
527 struct buffer_head bh;
532 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
535 bh.b_size = 1 << ip->i_inode.i_blkbits;
536 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
539 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
540 "%u db=%llu\n", rc, lb, (unsigned long long)db);
543 if (!prev_db || db != prev_db + 1) {
544 jext = kzalloc(sizeof(struct gfs2_journal_extent),
547 printk(KERN_INFO "GFS2 error: out of memory "
548 "mapping journal extents.\n");
555 list_add_tail(&jext->extent_list, &jd->extent_list);
564 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
566 char *message = "FIRSTMOUNT=Done";
567 char *envp[] = { message, NULL };
569 fs_info(sdp, "first mount done, others may mount\n");
571 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
572 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
574 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
578 * gfs2_jindex_hold - Grab a lock on the jindex
579 * @sdp: The GFS2 superblock
580 * @ji_gh: the holder for the jindex glock
585 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
587 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
590 struct gfs2_jdesc *jd;
595 mutex_lock(&sdp->sd_jindex_mutex);
598 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
602 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
603 name.hash = gfs2_disk_hash(name.name, name.len);
605 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
606 if (error == -ENOENT) {
611 gfs2_glock_dq_uninit(ji_gh);
617 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
621 INIT_LIST_HEAD(&jd->extent_list);
622 INIT_WORK(&jd->jd_work, gfs2_recover_func);
623 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
624 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
628 error = PTR_ERR(jd->jd_inode);
633 spin_lock(&sdp->sd_jindex_spin);
634 jd->jd_jid = sdp->sd_journals++;
635 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
636 spin_unlock(&sdp->sd_jindex_spin);
639 mutex_unlock(&sdp->sd_jindex_mutex);
644 static int init_journal(struct gfs2_sbd *sdp, int undo)
646 struct inode *master = sdp->sd_master_dir->d_inode;
647 struct gfs2_holder ji_gh;
648 struct gfs2_inode *ip;
657 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
658 if (IS_ERR(sdp->sd_jindex)) {
659 fs_err(sdp, "can't lookup journal index: %d\n", error);
660 return PTR_ERR(sdp->sd_jindex);
663 /* Load in the journal index special file */
665 error = gfs2_jindex_hold(sdp, &ji_gh);
667 fs_err(sdp, "can't read journal index: %d\n", error);
672 if (!gfs2_jindex_size(sdp)) {
673 fs_err(sdp, "no journals!\n");
677 if (sdp->sd_args.ar_spectator) {
678 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
679 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
680 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
681 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
683 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
684 fs_err(sdp, "can't mount journal #%u\n",
685 sdp->sd_lockstruct.ls_jid);
686 fs_err(sdp, "there are only %u journals (0 - %u)\n",
687 gfs2_jindex_size(sdp),
688 gfs2_jindex_size(sdp) - 1);
691 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
693 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
695 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
696 &sdp->sd_journal_gh);
698 fs_err(sdp, "can't acquire journal glock: %d\n", error);
702 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
703 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
704 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
707 fs_err(sdp, "can't acquire journal inode glock: %d\n",
709 goto fail_journal_gh;
712 error = gfs2_jdesc_check(sdp->sd_jdesc);
714 fs_err(sdp, "my journal (%u) is bad: %d\n",
715 sdp->sd_jdesc->jd_jid, error);
718 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
719 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
720 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
722 /* Map the extents for this journal's blocks */
723 map_journal_extents(sdp);
725 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
727 if (sdp->sd_lockstruct.ls_first) {
729 for (x = 0; x < sdp->sd_journals; x++) {
730 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
733 fs_err(sdp, "error recovering journal %u: %d\n",
739 gfs2_others_may_mount(sdp);
740 } else if (!sdp->sd_args.ar_spectator) {
741 error = gfs2_recover_journal(sdp->sd_jdesc, true);
743 fs_err(sdp, "error recovering my journal: %d\n", error);
748 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
749 gfs2_glock_dq_uninit(&ji_gh);
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);
770 static int init_inodes(struct gfs2_sbd *sdp, int undo)
773 struct inode *master = sdp->sd_master_dir->d_inode;
778 error = init_journal(sdp, undo);
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 error = gfs2_rindex_update(sdp);
814 iput(sdp->sd_quota_inode);
816 gfs2_clear_rgrpd(sdp);
817 iput(sdp->sd_rindex);
819 iput(sdp->sd_statfs_inode);
821 init_journal(sdp, UNDO);
826 static int init_per_node(struct gfs2_sbd *sdp, int undo)
828 struct inode *pn = NULL;
831 struct gfs2_inode *ip;
832 struct inode *master = sdp->sd_master_dir->d_inode;
834 if (sdp->sd_args.ar_spectator)
840 pn = gfs2_lookup_simple(master, "per_node");
843 fs_err(sdp, "can't find per_node directory: %d\n", error);
847 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
848 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
849 if (IS_ERR(sdp->sd_sc_inode)) {
850 error = PTR_ERR(sdp->sd_sc_inode);
851 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
855 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
856 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
857 if (IS_ERR(sdp->sd_qc_inode)) {
858 error = PTR_ERR(sdp->sd_qc_inode);
859 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
866 ip = GFS2_I(sdp->sd_sc_inode);
867 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
870 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
874 ip = GFS2_I(sdp->sd_qc_inode);
875 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
878 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
885 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
887 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
889 iput(sdp->sd_qc_inode);
891 iput(sdp->sd_sc_inode);
898 static int init_threads(struct gfs2_sbd *sdp, int undo)
900 struct task_struct *p;
906 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
909 fs_err(sdp, "can't start logd thread: %d\n", error);
912 sdp->sd_logd_process = p;
914 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
917 fs_err(sdp, "can't start quotad thread: %d\n", error);
920 sdp->sd_quotad_process = p;
926 kthread_stop(sdp->sd_quotad_process);
928 kthread_stop(sdp->sd_logd_process);
932 static const match_table_t nolock_tokens = {
933 { Opt_jid, "jid=%d\n", },
937 static const struct lm_lockops nolock_ops = {
938 .lm_proto_name = "lock_nolock",
939 .lm_put_lock = gfs2_glock_free,
940 .lm_tokens = &nolock_tokens,
944 * gfs2_lm_mount - mount a locking protocol
945 * @sdp: the filesystem
946 * @args: mount arguments
947 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
952 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
954 const struct lm_lockops *lm;
955 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
956 struct gfs2_args *args = &sdp->sd_args;
957 const char *proto = sdp->sd_proto_name;
958 const char *table = sdp->sd_table_name;
962 if (!strcmp("lock_nolock", proto)) {
964 sdp->sd_args.ar_localflocks = 1;
965 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
966 } else if (!strcmp("lock_dlm", proto)) {
970 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
974 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
979 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
980 substring_t tmp[MAX_OPT_ARGS];
986 token = match_token(o, *lm->lm_tokens, tmp);
989 ret = match_int(&tmp[0], &option);
990 if (ret || option < 0)
992 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
996 /* Obsolete, but left for backward compat purposes */
999 ret = match_int(&tmp[0], &option);
1000 if (ret || (option != 0 && option != 1))
1001 goto hostdata_error;
1002 ls->ls_first = option;
1005 ret = match_int(&tmp[0], &option);
1006 if (ret || (option != 0 && option != 1))
1007 goto hostdata_error;
1008 ls->ls_nodir = option;
1013 fs_info(sdp, "unknown hostdata (%s)\n", o);
1018 if (lm->lm_mount == NULL) {
1019 fs_info(sdp, "Now mounting FS...\n");
1020 complete_all(&sdp->sd_locking_init);
1023 ret = lm->lm_mount(sdp, table);
1025 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1026 complete_all(&sdp->sd_locking_init);
1030 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1032 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1033 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1035 lm->lm_unmount(sdp);
1038 static int gfs2_journalid_wait(void *word)
1040 if (signal_pending(current))
1046 static int wait_on_journal(struct gfs2_sbd *sdp)
1048 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1051 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1054 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1056 struct super_block *sb = sdp->sd_vfs;
1059 char *envp[] = { ro, spectator, NULL };
1060 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1061 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1062 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1066 * fill_super - Read in superblock
1067 * @sb: The VFS superblock
1068 * @data: Mount options
1069 * @silent: Don't complain if it's not a GFS2 filesystem
1074 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1076 struct gfs2_sbd *sdp;
1077 struct gfs2_holder mount_gh;
1082 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1085 sdp->sd_args = *args;
1087 if (sdp->sd_args.ar_spectator) {
1088 sb->s_flags |= MS_RDONLY;
1089 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1091 if (sdp->sd_args.ar_posix_acl)
1092 sb->s_flags |= MS_POSIXACL;
1093 if (sdp->sd_args.ar_nobarrier)
1094 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1096 sb->s_flags |= MS_NOSEC;
1097 sb->s_magic = GFS2_MAGIC;
1098 sb->s_op = &gfs2_super_ops;
1099 sb->s_d_op = &gfs2_dops;
1100 sb->s_export_op = &gfs2_export_ops;
1101 sb->s_xattr = gfs2_xattr_handlers;
1102 sb->s_qcop = &gfs2_quotactl_ops;
1103 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1104 sb->s_time_gran = 1;
1105 sb->s_maxbytes = MAX_LFS_FILESIZE;
1107 /* Set up the buffer cache and fill in some fake block size values
1108 to allow us to read-in the on-disk superblock. */
1109 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1110 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1111 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1112 GFS2_BASIC_BLOCK_SHIFT;
1113 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1115 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1116 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1117 if (sdp->sd_args.ar_statfs_quantum) {
1118 sdp->sd_tune.gt_statfs_slow = 0;
1119 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1121 sdp->sd_tune.gt_statfs_slow = 1;
1122 sdp->sd_tune.gt_statfs_quantum = 30;
1125 error = init_names(sdp, silent);
1129 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1131 gfs2_create_debugfs_file(sdp);
1133 error = gfs2_sys_fs_add(sdp);
1137 error = gfs2_lm_mount(sdp, silent);
1141 error = init_locking(sdp, &mount_gh, DO);
1145 error = init_sb(sdp, silent);
1149 error = wait_on_journal(sdp);
1154 * If user space has failed to join the cluster or some similar
1155 * failure has occurred, then the journal id will contain a
1156 * negative (error) number. This will then be returned to the
1157 * caller (of the mount syscall). We do this even for spectator
1158 * mounts (which just write a jid of 0 to indicate "ok" even though
1159 * the jid is unused in the spectator case)
1161 if (sdp->sd_lockstruct.ls_jid < 0) {
1162 error = sdp->sd_lockstruct.ls_jid;
1163 sdp->sd_lockstruct.ls_jid = 0;
1167 if (sdp->sd_args.ar_spectator)
1168 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1169 sdp->sd_table_name);
1171 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1172 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1174 error = init_inodes(sdp, DO);
1178 error = init_per_node(sdp, DO);
1182 error = gfs2_statfs_init(sdp);
1184 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1188 error = init_threads(sdp, DO);
1192 if (!(sb->s_flags & MS_RDONLY)) {
1193 error = gfs2_make_fs_rw(sdp);
1195 fs_err(sdp, "can't make FS RW: %d\n", error);
1200 gfs2_glock_dq_uninit(&mount_gh);
1201 gfs2_online_uevent(sdp);
1205 init_threads(sdp, UNDO);
1207 init_per_node(sdp, UNDO);
1209 init_inodes(sdp, UNDO);
1211 if (sdp->sd_root_dir)
1212 dput(sdp->sd_root_dir);
1213 if (sdp->sd_master_dir)
1214 dput(sdp->sd_master_dir);
1219 init_locking(sdp, &mount_gh, UNDO);
1221 gfs2_gl_hash_clear(sdp);
1222 gfs2_lm_unmount(sdp);
1224 gfs2_sys_fs_del(sdp);
1226 gfs2_delete_debugfs_file(sdp);
1227 free_percpu(sdp->sd_lkstats);
1229 sb->s_fs_info = NULL;
1233 static int set_gfs2_super(struct super_block *s, void *data)
1236 s->s_dev = s->s_bdev->bd_dev;
1239 * We set the bdi here to the queue backing, file systems can
1240 * overwrite this in ->fill_super()
1242 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1246 static int test_gfs2_super(struct super_block *s, void *ptr)
1248 struct block_device *bdev = ptr;
1249 return (bdev == s->s_bdev);
1253 * gfs2_mount - Get the GFS2 superblock
1254 * @fs_type: The GFS2 filesystem type
1255 * @flags: Mount flags
1256 * @dev_name: The name of the device
1257 * @data: The mount arguments
1259 * Q. Why not use get_sb_bdev() ?
1260 * A. We need to select one of two root directories to mount, independent
1261 * of whether this is the initial, or subsequent, mount of this sb
1263 * Returns: 0 or -ve on error
1266 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1267 const char *dev_name, void *data)
1269 struct block_device *bdev;
1270 struct super_block *s;
1271 fmode_t mode = FMODE_READ | FMODE_EXCL;
1273 struct gfs2_args args;
1274 struct gfs2_sbd *sdp;
1276 if (!(flags & MS_RDONLY))
1277 mode |= FMODE_WRITE;
1279 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1281 return ERR_CAST(bdev);
1284 * once the super is inserted into the list by sget, s_umount
1285 * will protect the lockfs code from trying to start a snapshot
1286 * while we are mounting
1288 mutex_lock(&bdev->bd_fsfreeze_mutex);
1289 if (bdev->bd_fsfreeze_count > 0) {
1290 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1294 s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1295 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1301 blkdev_put(bdev, mode);
1303 memset(&args, 0, sizeof(args));
1304 args.ar_quota = GFS2_QUOTA_DEFAULT;
1305 args.ar_data = GFS2_DATA_DEFAULT;
1306 args.ar_commit = 30;
1307 args.ar_statfs_quantum = 30;
1308 args.ar_quota_quantum = 60;
1309 args.ar_errors = GFS2_ERRORS_DEFAULT;
1311 error = gfs2_mount_args(&args, data);
1313 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1319 if ((flags ^ s->s_flags) & MS_RDONLY)
1322 char b[BDEVNAME_SIZE];
1326 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1327 sb_set_blocksize(s, block_size(bdev));
1328 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1331 s->s_flags |= MS_ACTIVE;
1337 return dget(sdp->sd_master_dir);
1339 return dget(sdp->sd_root_dir);
1342 deactivate_locked_super(s);
1343 return ERR_PTR(error);
1345 blkdev_put(bdev, mode);
1346 return ERR_PTR(error);
1349 static int set_meta_super(struct super_block *s, void *ptr)
1354 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1355 int flags, const char *dev_name, void *data)
1357 struct super_block *s;
1358 struct gfs2_sbd *sdp;
1362 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1364 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1366 return ERR_PTR(error);
1368 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1369 path.dentry->d_inode->i_sb->s_bdev);
1372 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1375 if ((flags ^ s->s_flags) & MS_RDONLY) {
1376 deactivate_locked_super(s);
1377 return ERR_PTR(-EBUSY);
1380 return dget(sdp->sd_master_dir);
1383 static void gfs2_kill_sb(struct super_block *sb)
1385 struct gfs2_sbd *sdp = sb->s_fs_info;
1388 kill_block_super(sb);
1392 gfs2_meta_syncfs(sdp);
1393 dput(sdp->sd_root_dir);
1394 dput(sdp->sd_master_dir);
1395 sdp->sd_root_dir = NULL;
1396 sdp->sd_master_dir = NULL;
1397 shrink_dcache_sb(sb);
1398 kill_block_super(sb);
1399 gfs2_delete_debugfs_file(sdp);
1400 free_percpu(sdp->sd_lkstats);
1404 struct file_system_type gfs2_fs_type = {
1406 .fs_flags = FS_REQUIRES_DEV,
1407 .mount = gfs2_mount,
1408 .kill_sb = gfs2_kill_sb,
1409 .owner = THIS_MODULE,
1412 struct file_system_type gfs2meta_fs_type = {
1414 .fs_flags = FS_REQUIRES_DEV,
1415 .mount = gfs2_mount_meta,
1416 .owner = THIS_MODULE,