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>
23 #include <linux/module.h>
39 #include "trace_gfs2.h"
45 * gfs2_tune_init - Fill a gfs2_tune structure with default values
50 static void gfs2_tune_init(struct gfs2_tune *gt)
52 spin_lock_init(>->gt_spin);
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 init_completion(&sdp->sd_wdack);
85 spin_lock_init(&sdp->sd_statfs_spin);
87 spin_lock_init(&sdp->sd_rindex_spin);
88 sdp->sd_rindex_tree.rb_node = NULL;
90 INIT_LIST_HEAD(&sdp->sd_jindex_list);
91 spin_lock_init(&sdp->sd_jindex_spin);
92 mutex_init(&sdp->sd_jindex_mutex);
94 INIT_LIST_HEAD(&sdp->sd_quota_list);
95 mutex_init(&sdp->sd_quota_mutex);
96 mutex_init(&sdp->sd_quota_sync_mutex);
97 init_waitqueue_head(&sdp->sd_quota_wait);
98 INIT_LIST_HEAD(&sdp->sd_trunc_list);
99 spin_lock_init(&sdp->sd_trunc_lock);
101 spin_lock_init(&sdp->sd_log_lock);
102 atomic_set(&sdp->sd_log_pinned, 0);
103 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
104 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
105 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
106 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
107 spin_lock_init(&sdp->sd_ordered_lock);
109 init_waitqueue_head(&sdp->sd_log_waitq);
110 init_waitqueue_head(&sdp->sd_logd_waitq);
111 spin_lock_init(&sdp->sd_ail_lock);
112 INIT_LIST_HEAD(&sdp->sd_ail1_list);
113 INIT_LIST_HEAD(&sdp->sd_ail2_list);
115 init_rwsem(&sdp->sd_log_flush_lock);
116 atomic_set(&sdp->sd_log_in_flight, 0);
117 init_waitqueue_head(&sdp->sd_log_flush_wait);
119 INIT_LIST_HEAD(&sdp->sd_revoke_list);
126 * gfs2_check_sb - Check superblock
127 * @sdp: the filesystem
128 * @sb: The superblock
129 * @silent: Don't print a message if the check fails
131 * Checks the version code of the FS is one that we understand how to
132 * read and that the sizes of the various on-disk structures have not
136 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
138 struct gfs2_sb_host *sb = &sdp->sd_sb;
140 if (sb->sb_magic != GFS2_MAGIC ||
141 sb->sb_type != GFS2_METATYPE_SB) {
143 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
147 /* If format numbers match exactly, we're done. */
149 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
150 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
153 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
158 static void end_bio_io_page(struct bio *bio, int error)
160 struct page *page = bio->bi_private;
163 SetPageUptodate(page);
165 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
169 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
171 struct gfs2_sb_host *sb = &sdp->sd_sb;
172 struct super_block *s = sdp->sd_vfs;
173 const struct gfs2_sb *str = buf;
175 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
176 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
177 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
178 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
179 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
180 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
181 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
182 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
183 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
184 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
185 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
187 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
188 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
189 memcpy(s->s_uuid, str->sb_uuid, 16);
193 * gfs2_read_super - Read the gfs2 super block from disk
194 * @sdp: The GFS2 super block
195 * @sector: The location of the super block
196 * @error: The error code to return
198 * This uses the bio functions to read the super block from disk
199 * because we want to be 100% sure that we never read cached data.
200 * A super block is read twice only during each GFS2 mount and is
201 * never written to by the filesystem. The first time its read no
202 * locks are held, and the only details which are looked at are those
203 * relating to the locking protocol. Once locking is up and working,
204 * the sb is read again under the lock to establish the location of
205 * the master directory (contains pointers to journals etc) and the
208 * Returns: 0 on success or error
211 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
213 struct super_block *sb = sdp->sd_vfs;
218 page = alloc_page(GFP_NOFS);
222 ClearPageUptodate(page);
223 ClearPageDirty(page);
226 bio = bio_alloc(GFP_NOFS, 1);
227 bio->bi_sector = sector * (sb->s_blocksize >> 9);
228 bio->bi_bdev = sb->s_bdev;
229 bio_add_page(bio, page, PAGE_SIZE, 0);
231 bio->bi_end_io = end_bio_io_page;
232 bio->bi_private = page;
233 submit_bio(READ_SYNC | REQ_META, bio);
234 wait_on_page_locked(page);
236 if (!PageUptodate(page)) {
244 return gfs2_check_sb(sdp, silent);
248 * gfs2_read_sb - Read super block
249 * @sdp: The GFS2 superblock
250 * @silent: Don't print message if mount fails
254 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
256 u32 hash_blocks, ind_blocks, leaf_blocks;
261 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
264 fs_err(sdp, "can't read superblock\n");
268 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
269 GFS2_BASIC_BLOCK_SHIFT;
270 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
271 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
272 sizeof(struct gfs2_dinode)) / sizeof(u64);
273 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
274 sizeof(struct gfs2_meta_header)) / sizeof(u64);
275 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
276 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
277 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
278 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
279 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
280 sizeof(struct gfs2_meta_header)) /
281 sizeof(struct gfs2_quota_change);
282 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
283 sizeof(struct gfs2_meta_header))
284 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
286 /* Compute maximum reservation required to add a entry to a directory */
288 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
292 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
293 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
294 ind_blocks += tmp_blocks;
297 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
299 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
301 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
302 sizeof(struct gfs2_dinode);
303 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
308 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
310 m = do_div(d, sdp->sd_inptrs);
312 if (d != sdp->sd_heightsize[x - 1] || m)
314 sdp->sd_heightsize[x] = space;
316 sdp->sd_max_height = x;
317 sdp->sd_heightsize[x] = ~0;
318 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
320 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
321 sizeof(struct gfs2_dinode);
322 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
327 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
329 m = do_div(d, sdp->sd_inptrs);
331 if (d != sdp->sd_jheightsize[x - 1] || m)
333 sdp->sd_jheightsize[x] = space;
335 sdp->sd_max_jheight = x;
336 sdp->sd_jheightsize[x] = ~0;
337 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
342 static int init_names(struct gfs2_sbd *sdp, int silent)
347 proto = sdp->sd_args.ar_lockproto;
348 table = sdp->sd_args.ar_locktable;
350 /* Try to autodetect */
352 if (!proto[0] || !table[0]) {
353 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
358 proto = sdp->sd_sb.sb_lockproto;
360 table = sdp->sd_sb.sb_locktable;
364 table = sdp->sd_vfs->s_id;
366 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
367 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
369 table = sdp->sd_table_name;
370 while ((table = strchr(table, '/')))
376 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
384 error = gfs2_glock_nq_num(sdp,
385 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
386 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
389 fs_err(sdp, "can't acquire mount glock: %d\n", error);
393 error = gfs2_glock_nq_num(sdp,
394 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
396 LM_FLAG_NOEXP | GL_EXACT,
399 fs_err(sdp, "can't acquire live glock: %d\n", error);
403 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
404 CREATE, &sdp->sd_rename_gl);
406 fs_err(sdp, "can't create rename glock: %d\n", error);
410 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
411 CREATE, &sdp->sd_trans_gl);
413 fs_err(sdp, "can't create transaction glock: %d\n", error);
420 gfs2_glock_put(sdp->sd_trans_gl);
422 gfs2_glock_put(sdp->sd_rename_gl);
424 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
426 gfs2_glock_dq_uninit(mount_gh);
431 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
432 u64 no_addr, const char *name)
434 struct gfs2_sbd *sdp = sb->s_fs_info;
435 struct dentry *dentry;
438 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
440 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
441 return PTR_ERR(inode);
443 dentry = d_make_root(inode);
445 fs_err(sdp, "can't alloc %s dentry\n", name);
452 static int init_sb(struct gfs2_sbd *sdp, int silent)
454 struct super_block *sb = sdp->sd_vfs;
455 struct gfs2_holder sb_gh;
459 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
460 LM_ST_SHARED, 0, &sb_gh);
462 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
466 ret = gfs2_read_sb(sdp, silent);
468 fs_err(sdp, "can't read superblock: %d\n", ret);
472 /* Set up the buffer cache and SB for real */
473 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
475 fs_err(sdp, "FS block size (%u) is too small for device "
477 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
480 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
482 fs_err(sdp, "FS block size (%u) is too big for machine "
484 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
487 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
489 /* Get the root inode */
490 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
491 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
495 /* Get the master inode */
496 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
497 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
499 dput(sdp->sd_root_dir);
502 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
504 gfs2_glock_dq_uninit(&sb_gh);
509 * map_journal_extents - create a reusable "extent" mapping from all logical
510 * blocks to all physical blocks for the given journal. This will save
511 * us time when writing journal blocks. Most journals will have only one
512 * extent that maps all their logical blocks. That's because gfs2.mkfs
513 * arranges the journal blocks sequentially to maximize performance.
514 * So the extent would map the first block for the entire file length.
515 * However, gfs2_jadd can happen while file activity is happening, so
516 * those journals may not be sequential. Less likely is the case where
517 * the users created their own journals by mounting the metafs and
518 * laying it out. But it's still possible. These journals might have
521 * TODO: This should be done in bigger chunks rather than one block at a time,
522 * but since it's only done at mount time, I'm not worried about the
525 static int map_journal_extents(struct gfs2_sbd *sdp)
527 struct gfs2_jdesc *jd = sdp->sd_jdesc;
529 u64 db, prev_db; /* logical block, disk block, prev disk block */
530 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
531 struct gfs2_journal_extent *jext = NULL;
532 struct buffer_head bh;
537 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
540 bh.b_size = 1 << ip->i_inode.i_blkbits;
541 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
544 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
545 "%u db=%llu\n", rc, lb, (unsigned long long)db);
548 if (!prev_db || db != prev_db + 1) {
549 jext = kzalloc(sizeof(struct gfs2_journal_extent),
552 printk(KERN_INFO "GFS2 error: out of memory "
553 "mapping journal extents.\n");
560 list_add_tail(&jext->extent_list, &jd->extent_list);
569 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
571 char *message = "FIRSTMOUNT=Done";
572 char *envp[] = { message, NULL };
574 fs_info(sdp, "first mount done, others may mount\n");
576 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
577 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
579 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
583 * gfs2_jindex_hold - Grab a lock on the jindex
584 * @sdp: The GFS2 superblock
585 * @ji_gh: the holder for the jindex glock
590 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
592 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
595 struct gfs2_jdesc *jd;
600 mutex_lock(&sdp->sd_jindex_mutex);
603 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
607 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
608 name.hash = gfs2_disk_hash(name.name, name.len);
610 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
611 if (error == -ENOENT) {
616 gfs2_glock_dq_uninit(ji_gh);
622 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
626 INIT_LIST_HEAD(&jd->extent_list);
627 INIT_WORK(&jd->jd_work, gfs2_recover_func);
628 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
629 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
633 error = PTR_ERR(jd->jd_inode);
638 spin_lock(&sdp->sd_jindex_spin);
639 jd->jd_jid = sdp->sd_journals++;
640 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
641 spin_unlock(&sdp->sd_jindex_spin);
644 mutex_unlock(&sdp->sd_jindex_mutex);
650 * check_journal_clean - Make sure a journal is clean for a spectator mount
651 * @sdp: The GFS2 superblock
652 * @jd: The journal descriptor
654 * Returns: 0 if the journal is clean or locked, else an error
656 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
659 struct gfs2_holder j_gh;
660 struct gfs2_log_header_host head;
661 struct gfs2_inode *ip;
663 ip = GFS2_I(jd->jd_inode);
664 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
665 GL_EXACT | GL_NOCACHE, &j_gh);
667 fs_err(sdp, "Error locking journal for spectator mount.\n");
670 error = gfs2_jdesc_check(jd);
672 fs_err(sdp, "Error checking journal for spectator mount.\n");
675 error = gfs2_find_jhead(jd, &head);
677 fs_err(sdp, "Error parsing journal for spectator mount.\n");
680 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
682 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
683 "must not be a spectator.\n", jd->jd_jid);
687 gfs2_glock_dq_uninit(&j_gh);
691 static int init_journal(struct gfs2_sbd *sdp, int undo)
693 struct inode *master = sdp->sd_master_dir->d_inode;
694 struct gfs2_holder ji_gh;
695 struct gfs2_inode *ip;
704 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
705 if (IS_ERR(sdp->sd_jindex)) {
706 fs_err(sdp, "can't lookup journal index: %d\n", error);
707 return PTR_ERR(sdp->sd_jindex);
710 /* Load in the journal index special file */
712 error = gfs2_jindex_hold(sdp, &ji_gh);
714 fs_err(sdp, "can't read journal index: %d\n", error);
719 if (!gfs2_jindex_size(sdp)) {
720 fs_err(sdp, "no journals!\n");
724 if (sdp->sd_args.ar_spectator) {
725 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
726 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
727 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
728 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
730 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
731 fs_err(sdp, "can't mount journal #%u\n",
732 sdp->sd_lockstruct.ls_jid);
733 fs_err(sdp, "there are only %u journals (0 - %u)\n",
734 gfs2_jindex_size(sdp),
735 gfs2_jindex_size(sdp) - 1);
738 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
740 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
742 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
743 &sdp->sd_journal_gh);
745 fs_err(sdp, "can't acquire journal glock: %d\n", error);
749 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
750 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
751 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
754 fs_err(sdp, "can't acquire journal inode glock: %d\n",
756 goto fail_journal_gh;
759 error = gfs2_jdesc_check(sdp->sd_jdesc);
761 fs_err(sdp, "my journal (%u) is bad: %d\n",
762 sdp->sd_jdesc->jd_jid, error);
765 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
766 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
767 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
769 /* Map the extents for this journal's blocks */
770 map_journal_extents(sdp);
772 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
774 if (sdp->sd_lockstruct.ls_first) {
776 for (x = 0; x < sdp->sd_journals; x++) {
777 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
779 if (sdp->sd_args.ar_spectator) {
780 error = check_journal_clean(sdp, jd);
785 error = gfs2_recover_journal(jd, true);
787 fs_err(sdp, "error recovering journal %u: %d\n",
793 gfs2_others_may_mount(sdp);
794 } else if (!sdp->sd_args.ar_spectator) {
795 error = gfs2_recover_journal(sdp->sd_jdesc, true);
797 fs_err(sdp, "error recovering my journal: %d\n", error);
802 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
803 gfs2_glock_dq_uninit(&ji_gh);
809 if (!sdp->sd_args.ar_spectator)
810 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
812 if (!sdp->sd_args.ar_spectator)
813 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
815 gfs2_jindex_free(sdp);
817 gfs2_glock_dq_uninit(&ji_gh);
819 iput(sdp->sd_jindex);
823 static struct lock_class_key gfs2_quota_imutex_key;
825 static int init_inodes(struct gfs2_sbd *sdp, int undo)
828 struct inode *master = sdp->sd_master_dir->d_inode;
833 error = init_journal(sdp, undo);
837 /* Read in the master statfs inode */
838 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
839 if (IS_ERR(sdp->sd_statfs_inode)) {
840 error = PTR_ERR(sdp->sd_statfs_inode);
841 fs_err(sdp, "can't read in statfs inode: %d\n", error);
845 /* Read in the resource index inode */
846 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
847 if (IS_ERR(sdp->sd_rindex)) {
848 error = PTR_ERR(sdp->sd_rindex);
849 fs_err(sdp, "can't get resource index inode: %d\n", error);
852 sdp->sd_rindex_uptodate = 0;
854 /* Read in the quota inode */
855 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
856 if (IS_ERR(sdp->sd_quota_inode)) {
857 error = PTR_ERR(sdp->sd_quota_inode);
858 fs_err(sdp, "can't get quota file inode: %d\n", error);
862 * i_mutex on quota files is special. Since this inode is hidden system
863 * file, we are safe to define locking ourselves.
865 lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
866 &gfs2_quota_imutex_key);
868 error = gfs2_rindex_update(sdp);
875 iput(sdp->sd_quota_inode);
877 gfs2_clear_rgrpd(sdp);
878 iput(sdp->sd_rindex);
880 iput(sdp->sd_statfs_inode);
882 init_journal(sdp, UNDO);
887 static int init_per_node(struct gfs2_sbd *sdp, int undo)
889 struct inode *pn = NULL;
892 struct gfs2_inode *ip;
893 struct inode *master = sdp->sd_master_dir->d_inode;
895 if (sdp->sd_args.ar_spectator)
901 pn = gfs2_lookup_simple(master, "per_node");
904 fs_err(sdp, "can't find per_node directory: %d\n", error);
908 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
909 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
910 if (IS_ERR(sdp->sd_sc_inode)) {
911 error = PTR_ERR(sdp->sd_sc_inode);
912 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
916 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
917 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
918 if (IS_ERR(sdp->sd_qc_inode)) {
919 error = PTR_ERR(sdp->sd_qc_inode);
920 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
927 ip = GFS2_I(sdp->sd_sc_inode);
928 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
931 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
935 ip = GFS2_I(sdp->sd_qc_inode);
936 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
939 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
946 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
948 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
950 iput(sdp->sd_qc_inode);
952 iput(sdp->sd_sc_inode);
959 static int init_threads(struct gfs2_sbd *sdp, int undo)
961 struct task_struct *p;
967 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
970 fs_err(sdp, "can't start logd thread: %d\n", error);
973 sdp->sd_logd_process = p;
975 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
978 fs_err(sdp, "can't start quotad thread: %d\n", error);
981 sdp->sd_quotad_process = p;
987 kthread_stop(sdp->sd_quotad_process);
989 kthread_stop(sdp->sd_logd_process);
993 static const match_table_t nolock_tokens = {
994 { Opt_jid, "jid=%d\n", },
998 static const struct lm_lockops nolock_ops = {
999 .lm_proto_name = "lock_nolock",
1000 .lm_put_lock = gfs2_glock_free,
1001 .lm_tokens = &nolock_tokens,
1005 * gfs2_lm_mount - mount a locking protocol
1006 * @sdp: the filesystem
1007 * @args: mount arguments
1008 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
1013 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1015 const struct lm_lockops *lm;
1016 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1017 struct gfs2_args *args = &sdp->sd_args;
1018 const char *proto = sdp->sd_proto_name;
1019 const char *table = sdp->sd_table_name;
1023 if (!strcmp("lock_nolock", proto)) {
1025 sdp->sd_args.ar_localflocks = 1;
1026 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1027 } else if (!strcmp("lock_dlm", proto)) {
1031 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
1035 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1040 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1041 substring_t tmp[MAX_OPT_ARGS];
1047 token = match_token(o, *lm->lm_tokens, tmp);
1050 ret = match_int(&tmp[0], &option);
1051 if (ret || option < 0)
1052 goto hostdata_error;
1053 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1054 ls->ls_jid = option;
1058 /* Obsolete, but left for backward compat purposes */
1061 ret = match_int(&tmp[0], &option);
1062 if (ret || (option != 0 && option != 1))
1063 goto hostdata_error;
1064 ls->ls_first = option;
1069 fs_info(sdp, "unknown hostdata (%s)\n", o);
1074 if (lm->lm_mount == NULL) {
1075 fs_info(sdp, "Now mounting FS...\n");
1076 complete_all(&sdp->sd_locking_init);
1079 ret = lm->lm_mount(sdp, table);
1081 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1082 complete_all(&sdp->sd_locking_init);
1086 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1088 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1089 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1091 lm->lm_unmount(sdp);
1094 static int gfs2_journalid_wait(void *word)
1096 if (signal_pending(current))
1102 static int wait_on_journal(struct gfs2_sbd *sdp)
1104 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1107 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1110 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1112 struct super_block *sb = sdp->sd_vfs;
1115 char *envp[] = { ro, spectator, NULL };
1116 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1117 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1118 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1122 * fill_super - Read in superblock
1123 * @sb: The VFS superblock
1124 * @data: Mount options
1125 * @silent: Don't complain if it's not a GFS2 filesystem
1130 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1132 struct gfs2_sbd *sdp;
1133 struct gfs2_holder mount_gh;
1138 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1141 sdp->sd_args = *args;
1143 if (sdp->sd_args.ar_spectator) {
1144 sb->s_flags |= MS_RDONLY;
1145 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1147 if (sdp->sd_args.ar_posix_acl)
1148 sb->s_flags |= MS_POSIXACL;
1149 if (sdp->sd_args.ar_nobarrier)
1150 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1152 sb->s_flags |= MS_NOSEC;
1153 sb->s_magic = GFS2_MAGIC;
1154 sb->s_op = &gfs2_super_ops;
1155 sb->s_d_op = &gfs2_dops;
1156 sb->s_export_op = &gfs2_export_ops;
1157 sb->s_xattr = gfs2_xattr_handlers;
1158 sb->s_qcop = &gfs2_quotactl_ops;
1159 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1160 sb->s_time_gran = 1;
1161 sb->s_maxbytes = MAX_LFS_FILESIZE;
1163 /* Set up the buffer cache and fill in some fake block size values
1164 to allow us to read-in the on-disk superblock. */
1165 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1166 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1167 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1168 GFS2_BASIC_BLOCK_SHIFT;
1169 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1171 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1172 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1173 if (sdp->sd_args.ar_statfs_quantum) {
1174 sdp->sd_tune.gt_statfs_slow = 0;
1175 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1177 sdp->sd_tune.gt_statfs_slow = 1;
1178 sdp->sd_tune.gt_statfs_quantum = 30;
1181 error = init_names(sdp, silent);
1183 /* In this case, we haven't initialized sysfs, so we have to
1184 manually free the sdp. */
1185 free_percpu(sdp->sd_lkstats);
1187 sb->s_fs_info = NULL;
1191 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1193 error = gfs2_sys_fs_add(sdp);
1195 * If we hit an error here, gfs2_sys_fs_add will have called function
1196 * kobject_put which causes the sysfs usage count to go to zero, which
1197 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1198 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1199 * kobject_put to free sdp.
1204 gfs2_create_debugfs_file(sdp);
1206 error = gfs2_lm_mount(sdp, silent);
1210 error = init_locking(sdp, &mount_gh, DO);
1214 error = init_sb(sdp, silent);
1218 error = wait_on_journal(sdp);
1223 * If user space has failed to join the cluster or some similar
1224 * failure has occurred, then the journal id will contain a
1225 * negative (error) number. This will then be returned to the
1226 * caller (of the mount syscall). We do this even for spectator
1227 * mounts (which just write a jid of 0 to indicate "ok" even though
1228 * the jid is unused in the spectator case)
1230 if (sdp->sd_lockstruct.ls_jid < 0) {
1231 error = sdp->sd_lockstruct.ls_jid;
1232 sdp->sd_lockstruct.ls_jid = 0;
1236 if (sdp->sd_args.ar_spectator)
1237 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1238 sdp->sd_table_name);
1240 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1241 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1243 error = init_inodes(sdp, DO);
1247 error = init_per_node(sdp, DO);
1251 error = gfs2_statfs_init(sdp);
1253 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1257 error = init_threads(sdp, DO);
1261 if (!(sb->s_flags & MS_RDONLY)) {
1262 error = gfs2_make_fs_rw(sdp);
1264 fs_err(sdp, "can't make FS RW: %d\n", error);
1269 gfs2_glock_dq_uninit(&mount_gh);
1270 gfs2_online_uevent(sdp);
1274 init_threads(sdp, UNDO);
1276 init_per_node(sdp, UNDO);
1278 init_inodes(sdp, UNDO);
1280 if (sdp->sd_root_dir)
1281 dput(sdp->sd_root_dir);
1282 if (sdp->sd_master_dir)
1283 dput(sdp->sd_master_dir);
1288 init_locking(sdp, &mount_gh, UNDO);
1290 gfs2_gl_hash_clear(sdp);
1291 gfs2_lm_unmount(sdp);
1293 gfs2_delete_debugfs_file(sdp);
1294 free_percpu(sdp->sd_lkstats);
1295 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1296 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1297 gfs2_sys_fs_del(sdp);
1298 sb->s_fs_info = NULL;
1302 static int set_gfs2_super(struct super_block *s, void *data)
1305 s->s_dev = s->s_bdev->bd_dev;
1308 * We set the bdi here to the queue backing, file systems can
1309 * overwrite this in ->fill_super()
1311 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1315 static int test_gfs2_super(struct super_block *s, void *ptr)
1317 struct block_device *bdev = ptr;
1318 return (bdev == s->s_bdev);
1322 * gfs2_mount - Get the GFS2 superblock
1323 * @fs_type: The GFS2 filesystem type
1324 * @flags: Mount flags
1325 * @dev_name: The name of the device
1326 * @data: The mount arguments
1328 * Q. Why not use get_sb_bdev() ?
1329 * A. We need to select one of two root directories to mount, independent
1330 * of whether this is the initial, or subsequent, mount of this sb
1332 * Returns: 0 or -ve on error
1335 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1336 const char *dev_name, void *data)
1338 struct block_device *bdev;
1339 struct super_block *s;
1340 fmode_t mode = FMODE_READ | FMODE_EXCL;
1342 struct gfs2_args args;
1343 struct gfs2_sbd *sdp;
1345 if (!(flags & MS_RDONLY))
1346 mode |= FMODE_WRITE;
1348 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1350 return ERR_CAST(bdev);
1353 * once the super is inserted into the list by sget, s_umount
1354 * will protect the lockfs code from trying to start a snapshot
1355 * while we are mounting
1357 mutex_lock(&bdev->bd_fsfreeze_mutex);
1358 if (bdev->bd_fsfreeze_count > 0) {
1359 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1363 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1364 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1371 * s_umount nests inside bd_mutex during
1372 * __invalidate_device(). blkdev_put() acquires
1373 * bd_mutex and can't be called under s_umount. Drop
1374 * s_umount temporarily. This is safe as we're
1375 * holding an active reference.
1377 up_write(&s->s_umount);
1378 blkdev_put(bdev, mode);
1379 down_write(&s->s_umount);
1382 memset(&args, 0, sizeof(args));
1383 args.ar_quota = GFS2_QUOTA_DEFAULT;
1384 args.ar_data = GFS2_DATA_DEFAULT;
1385 args.ar_commit = 30;
1386 args.ar_statfs_quantum = 30;
1387 args.ar_quota_quantum = 60;
1388 args.ar_errors = GFS2_ERRORS_DEFAULT;
1390 error = gfs2_mount_args(&args, data);
1392 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1398 if ((flags ^ s->s_flags) & MS_RDONLY)
1401 char b[BDEVNAME_SIZE];
1404 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1405 sb_set_blocksize(s, block_size(bdev));
1406 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1409 s->s_flags |= MS_ACTIVE;
1415 return dget(sdp->sd_master_dir);
1417 return dget(sdp->sd_root_dir);
1420 deactivate_locked_super(s);
1421 return ERR_PTR(error);
1423 blkdev_put(bdev, mode);
1424 return ERR_PTR(error);
1427 static int set_meta_super(struct super_block *s, void *ptr)
1432 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1433 int flags, const char *dev_name, void *data)
1435 struct super_block *s;
1436 struct gfs2_sbd *sdp;
1440 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1442 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1444 return ERR_PTR(error);
1446 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1447 path.dentry->d_inode->i_sb->s_bdev);
1450 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1453 if ((flags ^ s->s_flags) & MS_RDONLY) {
1454 deactivate_locked_super(s);
1455 return ERR_PTR(-EBUSY);
1458 return dget(sdp->sd_master_dir);
1461 static void gfs2_kill_sb(struct super_block *sb)
1463 struct gfs2_sbd *sdp = sb->s_fs_info;
1466 kill_block_super(sb);
1470 gfs2_meta_syncfs(sdp);
1471 dput(sdp->sd_root_dir);
1472 dput(sdp->sd_master_dir);
1473 sdp->sd_root_dir = NULL;
1474 sdp->sd_master_dir = NULL;
1475 shrink_dcache_sb(sb);
1476 gfs2_delete_debugfs_file(sdp);
1477 free_percpu(sdp->sd_lkstats);
1478 kill_block_super(sb);
1481 struct file_system_type gfs2_fs_type = {
1483 .fs_flags = FS_REQUIRES_DEV,
1484 .mount = gfs2_mount,
1485 .kill_sb = gfs2_kill_sb,
1486 .owner = THIS_MODULE,
1488 MODULE_ALIAS_FS("gfs2");
1490 struct file_system_type gfs2meta_fs_type = {
1492 .fs_flags = FS_REQUIRES_DEV,
1493 .mount = gfs2_mount_meta,
1494 .owner = THIS_MODULE,
1496 MODULE_ALIAS_FS("gfs2meta");