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>
22 #include <linux/lockdep.h>
38 #include "trace_gfs2.h"
44 * gfs2_tune_init - Fill a gfs2_tune structure with default values
49 static void gfs2_tune_init(struct gfs2_tune *gt)
51 spin_lock_init(>->gt_spin);
53 gt->gt_quota_simul_sync = 64;
54 gt->gt_quota_warn_period = 10;
55 gt->gt_quota_scale_num = 1;
56 gt->gt_quota_scale_den = 1;
57 gt->gt_new_files_jdata = 0;
58 gt->gt_max_readahead = 1 << 18;
59 gt->gt_complain_secs = 10;
62 static struct gfs2_sbd *init_sbd(struct super_block *sb)
66 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
72 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
73 if (!sdp->sd_lkstats) {
78 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
79 gfs2_tune_init(&sdp->sd_tune);
81 init_waitqueue_head(&sdp->sd_glock_wait);
82 atomic_set(&sdp->sd_glock_disposal, 0);
83 init_completion(&sdp->sd_locking_init);
84 spin_lock_init(&sdp->sd_statfs_spin);
86 spin_lock_init(&sdp->sd_rindex_spin);
87 sdp->sd_rindex_tree.rb_node = NULL;
89 INIT_LIST_HEAD(&sdp->sd_jindex_list);
90 spin_lock_init(&sdp->sd_jindex_spin);
91 mutex_init(&sdp->sd_jindex_mutex);
93 INIT_LIST_HEAD(&sdp->sd_quota_list);
94 mutex_init(&sdp->sd_quota_mutex);
95 init_waitqueue_head(&sdp->sd_quota_wait);
96 INIT_LIST_HEAD(&sdp->sd_trunc_list);
97 spin_lock_init(&sdp->sd_trunc_lock);
99 spin_lock_init(&sdp->sd_log_lock);
100 atomic_set(&sdp->sd_log_pinned, 0);
101 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
102 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
103 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
104 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
106 init_waitqueue_head(&sdp->sd_log_waitq);
107 init_waitqueue_head(&sdp->sd_logd_waitq);
108 spin_lock_init(&sdp->sd_ail_lock);
109 INIT_LIST_HEAD(&sdp->sd_ail1_list);
110 INIT_LIST_HEAD(&sdp->sd_ail2_list);
112 init_rwsem(&sdp->sd_log_flush_lock);
113 atomic_set(&sdp->sd_log_in_flight, 0);
114 init_waitqueue_head(&sdp->sd_log_flush_wait);
116 INIT_LIST_HEAD(&sdp->sd_revoke_list);
123 * gfs2_check_sb - Check superblock
124 * @sdp: the filesystem
125 * @sb: The superblock
126 * @silent: Don't print a message if the check fails
128 * Checks the version code of the FS is one that we understand how to
129 * read and that the sizes of the various on-disk structures have not
133 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
135 struct gfs2_sb_host *sb = &sdp->sd_sb;
137 if (sb->sb_magic != GFS2_MAGIC ||
138 sb->sb_type != GFS2_METATYPE_SB) {
140 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
144 /* If format numbers match exactly, we're done. */
146 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
147 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
150 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
155 static void end_bio_io_page(struct bio *bio, int error)
157 struct page *page = bio->bi_private;
160 SetPageUptodate(page);
162 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
166 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
168 struct gfs2_sb_host *sb = &sdp->sd_sb;
169 struct super_block *s = sdp->sd_vfs;
170 const struct gfs2_sb *str = buf;
172 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
173 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
174 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
175 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
176 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
177 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
178 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
179 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
180 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
181 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
182 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
184 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
185 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
186 memcpy(s->s_uuid, str->sb_uuid, 16);
190 * gfs2_read_super - Read the gfs2 super block from disk
191 * @sdp: The GFS2 super block
192 * @sector: The location of the super block
193 * @error: The error code to return
195 * This uses the bio functions to read the super block from disk
196 * because we want to be 100% sure that we never read cached data.
197 * A super block is read twice only during each GFS2 mount and is
198 * never written to by the filesystem. The first time its read no
199 * locks are held, and the only details which are looked at are those
200 * relating to the locking protocol. Once locking is up and working,
201 * the sb is read again under the lock to establish the location of
202 * the master directory (contains pointers to journals etc) and the
205 * Returns: 0 on success or error
208 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
210 struct super_block *sb = sdp->sd_vfs;
215 page = alloc_page(GFP_NOFS);
219 ClearPageUptodate(page);
220 ClearPageDirty(page);
223 bio = bio_alloc(GFP_NOFS, 1);
224 bio->bi_sector = sector * (sb->s_blocksize >> 9);
225 bio->bi_bdev = sb->s_bdev;
226 bio_add_page(bio, page, PAGE_SIZE, 0);
228 bio->bi_end_io = end_bio_io_page;
229 bio->bi_private = page;
230 submit_bio(READ_SYNC | REQ_META, bio);
231 wait_on_page_locked(page);
233 if (!PageUptodate(page)) {
241 return gfs2_check_sb(sdp, silent);
245 * gfs2_read_sb - Read super block
246 * @sdp: The GFS2 superblock
247 * @silent: Don't print message if mount fails
251 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
253 u32 hash_blocks, ind_blocks, leaf_blocks;
258 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
261 fs_err(sdp, "can't read superblock\n");
265 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
266 GFS2_BASIC_BLOCK_SHIFT;
267 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
268 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
269 sizeof(struct gfs2_dinode)) / sizeof(u64);
270 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
271 sizeof(struct gfs2_meta_header)) / sizeof(u64);
272 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
273 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
274 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
275 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
276 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
277 sizeof(struct gfs2_meta_header)) /
278 sizeof(struct gfs2_quota_change);
279 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
280 sizeof(struct gfs2_meta_header))
281 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
283 /* Compute maximum reservation required to add a entry to a directory */
285 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
289 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
290 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
291 ind_blocks += tmp_blocks;
294 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
296 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
298 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
299 sizeof(struct gfs2_dinode);
300 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
305 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
307 m = do_div(d, sdp->sd_inptrs);
309 if (d != sdp->sd_heightsize[x - 1] || m)
311 sdp->sd_heightsize[x] = space;
313 sdp->sd_max_height = x;
314 sdp->sd_heightsize[x] = ~0;
315 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
317 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
318 sizeof(struct gfs2_dinode);
319 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
324 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
326 m = do_div(d, sdp->sd_inptrs);
328 if (d != sdp->sd_jheightsize[x - 1] || m)
330 sdp->sd_jheightsize[x] = space;
332 sdp->sd_max_jheight = x;
333 sdp->sd_jheightsize[x] = ~0;
334 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
339 static int init_names(struct gfs2_sbd *sdp, int silent)
344 proto = sdp->sd_args.ar_lockproto;
345 table = sdp->sd_args.ar_locktable;
347 /* Try to autodetect */
349 if (!proto[0] || !table[0]) {
350 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
355 proto = sdp->sd_sb.sb_lockproto;
357 table = sdp->sd_sb.sb_locktable;
361 table = sdp->sd_vfs->s_id;
363 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
364 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
366 table = sdp->sd_table_name;
367 while ((table = strchr(table, '/')))
373 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
381 error = gfs2_glock_nq_num(sdp,
382 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
383 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
386 fs_err(sdp, "can't acquire mount glock: %d\n", error);
390 error = gfs2_glock_nq_num(sdp,
391 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
393 LM_FLAG_NOEXP | GL_EXACT,
396 fs_err(sdp, "can't acquire live glock: %d\n", error);
400 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
401 CREATE, &sdp->sd_rename_gl);
403 fs_err(sdp, "can't create rename glock: %d\n", error);
407 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
408 CREATE, &sdp->sd_trans_gl);
410 fs_err(sdp, "can't create transaction glock: %d\n", error);
417 gfs2_glock_put(sdp->sd_trans_gl);
419 gfs2_glock_put(sdp->sd_rename_gl);
421 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
423 gfs2_glock_dq_uninit(mount_gh);
428 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
429 u64 no_addr, const char *name)
431 struct gfs2_sbd *sdp = sb->s_fs_info;
432 struct dentry *dentry;
435 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
437 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
438 return PTR_ERR(inode);
440 dentry = d_make_root(inode);
442 fs_err(sdp, "can't alloc %s dentry\n", name);
449 static int init_sb(struct gfs2_sbd *sdp, int silent)
451 struct super_block *sb = sdp->sd_vfs;
452 struct gfs2_holder sb_gh;
456 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
457 LM_ST_SHARED, 0, &sb_gh);
459 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
463 ret = gfs2_read_sb(sdp, silent);
465 fs_err(sdp, "can't read superblock: %d\n", ret);
469 /* Set up the buffer cache and SB for real */
470 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
472 fs_err(sdp, "FS block size (%u) is too small for device "
474 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
477 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
479 fs_err(sdp, "FS block size (%u) is too big for machine "
481 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
484 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
486 /* Get the root inode */
487 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
488 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
492 /* Get the master inode */
493 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
494 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
496 dput(sdp->sd_root_dir);
499 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
501 gfs2_glock_dq_uninit(&sb_gh);
506 * map_journal_extents - create a reusable "extent" mapping from all logical
507 * blocks to all physical blocks for the given journal. This will save
508 * us time when writing journal blocks. Most journals will have only one
509 * extent that maps all their logical blocks. That's because gfs2.mkfs
510 * arranges the journal blocks sequentially to maximize performance.
511 * So the extent would map the first block for the entire file length.
512 * However, gfs2_jadd can happen while file activity is happening, so
513 * those journals may not be sequential. Less likely is the case where
514 * the users created their own journals by mounting the metafs and
515 * laying it out. But it's still possible. These journals might have
518 * TODO: This should be done in bigger chunks rather than one block at a time,
519 * but since it's only done at mount time, I'm not worried about the
522 static int map_journal_extents(struct gfs2_sbd *sdp)
524 struct gfs2_jdesc *jd = sdp->sd_jdesc;
526 u64 db, prev_db; /* logical block, disk block, prev disk block */
527 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
528 struct gfs2_journal_extent *jext = NULL;
529 struct buffer_head bh;
534 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
537 bh.b_size = 1 << ip->i_inode.i_blkbits;
538 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
541 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
542 "%u db=%llu\n", rc, lb, (unsigned long long)db);
545 if (!prev_db || db != prev_db + 1) {
546 jext = kzalloc(sizeof(struct gfs2_journal_extent),
549 printk(KERN_INFO "GFS2 error: out of memory "
550 "mapping journal extents.\n");
557 list_add_tail(&jext->extent_list, &jd->extent_list);
566 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
568 char *message = "FIRSTMOUNT=Done";
569 char *envp[] = { message, NULL };
571 fs_info(sdp, "first mount done, others may mount\n");
573 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
574 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
576 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
580 * gfs2_jindex_hold - Grab a lock on the jindex
581 * @sdp: The GFS2 superblock
582 * @ji_gh: the holder for the jindex glock
587 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
589 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
592 struct gfs2_jdesc *jd;
597 mutex_lock(&sdp->sd_jindex_mutex);
600 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
604 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
605 name.hash = gfs2_disk_hash(name.name, name.len);
607 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
608 if (error == -ENOENT) {
613 gfs2_glock_dq_uninit(ji_gh);
619 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
623 INIT_LIST_HEAD(&jd->extent_list);
624 INIT_WORK(&jd->jd_work, gfs2_recover_func);
625 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
626 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
630 error = PTR_ERR(jd->jd_inode);
635 spin_lock(&sdp->sd_jindex_spin);
636 jd->jd_jid = sdp->sd_journals++;
637 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
638 spin_unlock(&sdp->sd_jindex_spin);
641 mutex_unlock(&sdp->sd_jindex_mutex);
646 static int init_journal(struct gfs2_sbd *sdp, int undo)
648 struct inode *master = sdp->sd_master_dir->d_inode;
649 struct gfs2_holder ji_gh;
650 struct gfs2_inode *ip;
659 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
660 if (IS_ERR(sdp->sd_jindex)) {
661 fs_err(sdp, "can't lookup journal index: %d\n", error);
662 return PTR_ERR(sdp->sd_jindex);
665 /* Load in the journal index special file */
667 error = gfs2_jindex_hold(sdp, &ji_gh);
669 fs_err(sdp, "can't read journal index: %d\n", error);
674 if (!gfs2_jindex_size(sdp)) {
675 fs_err(sdp, "no journals!\n");
679 if (sdp->sd_args.ar_spectator) {
680 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
681 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
682 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
683 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
685 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
686 fs_err(sdp, "can't mount journal #%u\n",
687 sdp->sd_lockstruct.ls_jid);
688 fs_err(sdp, "there are only %u journals (0 - %u)\n",
689 gfs2_jindex_size(sdp),
690 gfs2_jindex_size(sdp) - 1);
693 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
695 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
697 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
698 &sdp->sd_journal_gh);
700 fs_err(sdp, "can't acquire journal glock: %d\n", error);
704 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
705 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
706 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
709 fs_err(sdp, "can't acquire journal inode glock: %d\n",
711 goto fail_journal_gh;
714 error = gfs2_jdesc_check(sdp->sd_jdesc);
716 fs_err(sdp, "my journal (%u) is bad: %d\n",
717 sdp->sd_jdesc->jd_jid, error);
720 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
721 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
722 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
724 /* Map the extents for this journal's blocks */
725 map_journal_extents(sdp);
727 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
729 if (sdp->sd_lockstruct.ls_first) {
731 for (x = 0; x < sdp->sd_journals; x++) {
732 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
735 fs_err(sdp, "error recovering journal %u: %d\n",
741 gfs2_others_may_mount(sdp);
742 } else if (!sdp->sd_args.ar_spectator) {
743 error = gfs2_recover_journal(sdp->sd_jdesc, true);
745 fs_err(sdp, "error recovering my journal: %d\n", error);
750 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
751 gfs2_glock_dq_uninit(&ji_gh);
757 if (!sdp->sd_args.ar_spectator)
758 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
760 if (!sdp->sd_args.ar_spectator)
761 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
763 gfs2_jindex_free(sdp);
765 gfs2_glock_dq_uninit(&ji_gh);
767 iput(sdp->sd_jindex);
771 static struct lock_class_key gfs2_quota_imutex_key;
773 static int init_inodes(struct gfs2_sbd *sdp, int undo)
776 struct inode *master = sdp->sd_master_dir->d_inode;
781 error = init_journal(sdp, undo);
785 /* Read in the master statfs inode */
786 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
787 if (IS_ERR(sdp->sd_statfs_inode)) {
788 error = PTR_ERR(sdp->sd_statfs_inode);
789 fs_err(sdp, "can't read in statfs inode: %d\n", error);
793 /* Read in the resource index inode */
794 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
795 if (IS_ERR(sdp->sd_rindex)) {
796 error = PTR_ERR(sdp->sd_rindex);
797 fs_err(sdp, "can't get resource index inode: %d\n", error);
800 sdp->sd_rindex_uptodate = 0;
802 /* Read in the quota inode */
803 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
804 if (IS_ERR(sdp->sd_quota_inode)) {
805 error = PTR_ERR(sdp->sd_quota_inode);
806 fs_err(sdp, "can't get quota file inode: %d\n", error);
810 * i_mutex on quota files is special. Since this inode is hidden system
811 * file, we are safe to define locking ourselves.
813 lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
814 &gfs2_quota_imutex_key);
816 error = gfs2_rindex_update(sdp);
823 iput(sdp->sd_quota_inode);
825 gfs2_clear_rgrpd(sdp);
826 iput(sdp->sd_rindex);
828 iput(sdp->sd_statfs_inode);
830 init_journal(sdp, UNDO);
835 static int init_per_node(struct gfs2_sbd *sdp, int undo)
837 struct inode *pn = NULL;
840 struct gfs2_inode *ip;
841 struct inode *master = sdp->sd_master_dir->d_inode;
843 if (sdp->sd_args.ar_spectator)
849 pn = gfs2_lookup_simple(master, "per_node");
852 fs_err(sdp, "can't find per_node directory: %d\n", error);
856 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
857 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
858 if (IS_ERR(sdp->sd_sc_inode)) {
859 error = PTR_ERR(sdp->sd_sc_inode);
860 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
864 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
865 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
866 if (IS_ERR(sdp->sd_qc_inode)) {
867 error = PTR_ERR(sdp->sd_qc_inode);
868 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
875 ip = GFS2_I(sdp->sd_sc_inode);
876 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
879 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
883 ip = GFS2_I(sdp->sd_qc_inode);
884 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
887 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
894 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
896 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
898 iput(sdp->sd_qc_inode);
900 iput(sdp->sd_sc_inode);
907 static int init_threads(struct gfs2_sbd *sdp, int undo)
909 struct task_struct *p;
915 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
918 fs_err(sdp, "can't start logd thread: %d\n", error);
921 sdp->sd_logd_process = p;
923 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
926 fs_err(sdp, "can't start quotad thread: %d\n", error);
929 sdp->sd_quotad_process = p;
935 kthread_stop(sdp->sd_quotad_process);
937 kthread_stop(sdp->sd_logd_process);
941 static const match_table_t nolock_tokens = {
942 { Opt_jid, "jid=%d\n", },
946 static const struct lm_lockops nolock_ops = {
947 .lm_proto_name = "lock_nolock",
948 .lm_put_lock = gfs2_glock_free,
949 .lm_tokens = &nolock_tokens,
953 * gfs2_lm_mount - mount a locking protocol
954 * @sdp: the filesystem
955 * @args: mount arguments
956 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
961 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
963 const struct lm_lockops *lm;
964 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
965 struct gfs2_args *args = &sdp->sd_args;
966 const char *proto = sdp->sd_proto_name;
967 const char *table = sdp->sd_table_name;
971 if (!strcmp("lock_nolock", proto)) {
973 sdp->sd_args.ar_localflocks = 1;
974 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
975 } else if (!strcmp("lock_dlm", proto)) {
979 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
983 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
988 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
989 substring_t tmp[MAX_OPT_ARGS];
995 token = match_token(o, *lm->lm_tokens, tmp);
998 ret = match_int(&tmp[0], &option);
999 if (ret || option < 0)
1000 goto hostdata_error;
1001 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1002 ls->ls_jid = option;
1006 /* Obsolete, but left for backward compat purposes */
1009 ret = match_int(&tmp[0], &option);
1010 if (ret || (option != 0 && option != 1))
1011 goto hostdata_error;
1012 ls->ls_first = option;
1017 fs_info(sdp, "unknown hostdata (%s)\n", o);
1022 if (lm->lm_mount == NULL) {
1023 fs_info(sdp, "Now mounting FS...\n");
1024 complete_all(&sdp->sd_locking_init);
1027 ret = lm->lm_mount(sdp, table);
1029 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1030 complete_all(&sdp->sd_locking_init);
1034 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1036 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1037 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1039 lm->lm_unmount(sdp);
1042 static int gfs2_journalid_wait(void *word)
1044 if (signal_pending(current))
1050 static int wait_on_journal(struct gfs2_sbd *sdp)
1052 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1055 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1058 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1060 struct super_block *sb = sdp->sd_vfs;
1063 char *envp[] = { ro, spectator, NULL };
1064 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1065 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1066 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1070 * fill_super - Read in superblock
1071 * @sb: The VFS superblock
1072 * @data: Mount options
1073 * @silent: Don't complain if it's not a GFS2 filesystem
1078 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1080 struct gfs2_sbd *sdp;
1081 struct gfs2_holder mount_gh;
1086 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1089 sdp->sd_args = *args;
1091 if (sdp->sd_args.ar_spectator) {
1092 sb->s_flags |= MS_RDONLY;
1093 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1095 if (sdp->sd_args.ar_posix_acl)
1096 sb->s_flags |= MS_POSIXACL;
1097 if (sdp->sd_args.ar_nobarrier)
1098 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1100 sb->s_flags |= MS_NOSEC;
1101 sb->s_magic = GFS2_MAGIC;
1102 sb->s_op = &gfs2_super_ops;
1103 sb->s_d_op = &gfs2_dops;
1104 sb->s_export_op = &gfs2_export_ops;
1105 sb->s_xattr = gfs2_xattr_handlers;
1106 sb->s_qcop = &gfs2_quotactl_ops;
1107 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1108 sb->s_time_gran = 1;
1109 sb->s_maxbytes = MAX_LFS_FILESIZE;
1111 /* Set up the buffer cache and fill in some fake block size values
1112 to allow us to read-in the on-disk superblock. */
1113 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1114 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1115 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1116 GFS2_BASIC_BLOCK_SHIFT;
1117 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1119 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1120 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1121 if (sdp->sd_args.ar_statfs_quantum) {
1122 sdp->sd_tune.gt_statfs_slow = 0;
1123 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1125 sdp->sd_tune.gt_statfs_slow = 1;
1126 sdp->sd_tune.gt_statfs_quantum = 30;
1129 error = init_names(sdp, silent);
1131 /* In this case, we haven't initialized sysfs, so we have to
1132 manually free the sdp. */
1133 free_percpu(sdp->sd_lkstats);
1135 sb->s_fs_info = NULL;
1139 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1141 error = gfs2_sys_fs_add(sdp);
1143 * If we hit an error here, gfs2_sys_fs_add will have called function
1144 * kobject_put which causes the sysfs usage count to go to zero, which
1145 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1146 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1147 * kobject_put to free sdp.
1152 gfs2_create_debugfs_file(sdp);
1154 error = gfs2_lm_mount(sdp, silent);
1158 error = init_locking(sdp, &mount_gh, DO);
1162 error = init_sb(sdp, silent);
1166 error = wait_on_journal(sdp);
1171 * If user space has failed to join the cluster or some similar
1172 * failure has occurred, then the journal id will contain a
1173 * negative (error) number. This will then be returned to the
1174 * caller (of the mount syscall). We do this even for spectator
1175 * mounts (which just write a jid of 0 to indicate "ok" even though
1176 * the jid is unused in the spectator case)
1178 if (sdp->sd_lockstruct.ls_jid < 0) {
1179 error = sdp->sd_lockstruct.ls_jid;
1180 sdp->sd_lockstruct.ls_jid = 0;
1184 if (sdp->sd_args.ar_spectator)
1185 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1186 sdp->sd_table_name);
1188 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1189 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1191 error = init_inodes(sdp, DO);
1195 error = init_per_node(sdp, DO);
1199 error = gfs2_statfs_init(sdp);
1201 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1205 error = init_threads(sdp, DO);
1209 if (!(sb->s_flags & MS_RDONLY)) {
1210 error = gfs2_make_fs_rw(sdp);
1212 fs_err(sdp, "can't make FS RW: %d\n", error);
1217 gfs2_glock_dq_uninit(&mount_gh);
1218 gfs2_online_uevent(sdp);
1222 init_threads(sdp, UNDO);
1224 init_per_node(sdp, UNDO);
1226 init_inodes(sdp, UNDO);
1228 if (sdp->sd_root_dir)
1229 dput(sdp->sd_root_dir);
1230 if (sdp->sd_master_dir)
1231 dput(sdp->sd_master_dir);
1236 init_locking(sdp, &mount_gh, UNDO);
1238 gfs2_gl_hash_clear(sdp);
1239 gfs2_lm_unmount(sdp);
1241 gfs2_delete_debugfs_file(sdp);
1242 free_percpu(sdp->sd_lkstats);
1243 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1244 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1245 gfs2_sys_fs_del(sdp);
1246 sb->s_fs_info = NULL;
1250 static int set_gfs2_super(struct super_block *s, void *data)
1253 s->s_dev = s->s_bdev->bd_dev;
1256 * We set the bdi here to the queue backing, file systems can
1257 * overwrite this in ->fill_super()
1259 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1263 static int test_gfs2_super(struct super_block *s, void *ptr)
1265 struct block_device *bdev = ptr;
1266 return (bdev == s->s_bdev);
1270 * gfs2_mount - Get the GFS2 superblock
1271 * @fs_type: The GFS2 filesystem type
1272 * @flags: Mount flags
1273 * @dev_name: The name of the device
1274 * @data: The mount arguments
1276 * Q. Why not use get_sb_bdev() ?
1277 * A. We need to select one of two root directories to mount, independent
1278 * of whether this is the initial, or subsequent, mount of this sb
1280 * Returns: 0 or -ve on error
1283 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1284 const char *dev_name, void *data)
1286 struct block_device *bdev;
1287 struct super_block *s;
1288 fmode_t mode = FMODE_READ | FMODE_EXCL;
1290 struct gfs2_args args;
1291 struct gfs2_sbd *sdp;
1293 if (!(flags & MS_RDONLY))
1294 mode |= FMODE_WRITE;
1296 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1298 return ERR_CAST(bdev);
1301 * once the super is inserted into the list by sget, s_umount
1302 * will protect the lockfs code from trying to start a snapshot
1303 * while we are mounting
1305 mutex_lock(&bdev->bd_fsfreeze_mutex);
1306 if (bdev->bd_fsfreeze_count > 0) {
1307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1311 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1312 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1318 blkdev_put(bdev, mode);
1320 memset(&args, 0, sizeof(args));
1321 args.ar_quota = GFS2_QUOTA_DEFAULT;
1322 args.ar_data = GFS2_DATA_DEFAULT;
1323 args.ar_commit = 30;
1324 args.ar_statfs_quantum = 30;
1325 args.ar_quota_quantum = 60;
1326 args.ar_errors = GFS2_ERRORS_DEFAULT;
1328 error = gfs2_mount_args(&args, data);
1330 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1336 if ((flags ^ s->s_flags) & MS_RDONLY)
1339 char b[BDEVNAME_SIZE];
1342 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1343 sb_set_blocksize(s, block_size(bdev));
1344 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1347 s->s_flags |= MS_ACTIVE;
1353 return dget(sdp->sd_master_dir);
1355 return dget(sdp->sd_root_dir);
1358 deactivate_locked_super(s);
1359 return ERR_PTR(error);
1361 blkdev_put(bdev, mode);
1362 return ERR_PTR(error);
1365 static int set_meta_super(struct super_block *s, void *ptr)
1370 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1371 int flags, const char *dev_name, void *data)
1373 struct super_block *s;
1374 struct gfs2_sbd *sdp;
1378 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1380 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1382 return ERR_PTR(error);
1384 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1385 path.dentry->d_inode->i_sb->s_bdev);
1388 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1391 if ((flags ^ s->s_flags) & MS_RDONLY) {
1392 deactivate_locked_super(s);
1393 return ERR_PTR(-EBUSY);
1396 return dget(sdp->sd_master_dir);
1399 static void gfs2_kill_sb(struct super_block *sb)
1401 struct gfs2_sbd *sdp = sb->s_fs_info;
1404 kill_block_super(sb);
1408 gfs2_meta_syncfs(sdp);
1409 dput(sdp->sd_root_dir);
1410 dput(sdp->sd_master_dir);
1411 sdp->sd_root_dir = NULL;
1412 sdp->sd_master_dir = NULL;
1413 shrink_dcache_sb(sb);
1414 gfs2_delete_debugfs_file(sdp);
1415 free_percpu(sdp->sd_lkstats);
1416 kill_block_super(sb);
1419 struct file_system_type gfs2_fs_type = {
1421 .fs_flags = FS_REQUIRES_DEV,
1422 .mount = gfs2_mount,
1423 .kill_sb = gfs2_kill_sb,
1424 .owner = THIS_MODULE,
1427 struct file_system_type gfs2meta_fs_type = {
1429 .fs_flags = FS_REQUIRES_DEV,
1430 .mount = gfs2_mount_meta,
1431 .owner = THIS_MODULE,