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>
40 #include "trace_gfs2.h"
46 * gfs2_tune_init - Fill a gfs2_tune structure with default values
51 static void gfs2_tune_init(struct gfs2_tune *gt)
53 spin_lock_init(>->gt_spin);
55 gt->gt_quota_warn_period = 10;
56 gt->gt_quota_scale_num = 1;
57 gt->gt_quota_scale_den = 1;
58 gt->gt_new_files_jdata = 0;
59 gt->gt_max_readahead = 1 << 18;
60 gt->gt_complain_secs = 10;
63 static struct gfs2_sbd *init_sbd(struct super_block *sb)
66 struct address_space *mapping;
68 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
74 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
75 if (!sdp->sd_lkstats) {
80 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
81 gfs2_tune_init(&sdp->sd_tune);
83 init_waitqueue_head(&sdp->sd_glock_wait);
84 atomic_set(&sdp->sd_glock_disposal, 0);
85 init_completion(&sdp->sd_locking_init);
86 init_completion(&sdp->sd_wdack);
87 spin_lock_init(&sdp->sd_statfs_spin);
89 spin_lock_init(&sdp->sd_rindex_spin);
90 sdp->sd_rindex_tree.rb_node = NULL;
92 INIT_LIST_HEAD(&sdp->sd_jindex_list);
93 spin_lock_init(&sdp->sd_jindex_spin);
94 mutex_init(&sdp->sd_jindex_mutex);
96 INIT_LIST_HEAD(&sdp->sd_quota_list);
97 mutex_init(&sdp->sd_quota_mutex);
98 mutex_init(&sdp->sd_quota_sync_mutex);
99 init_waitqueue_head(&sdp->sd_quota_wait);
100 INIT_LIST_HEAD(&sdp->sd_trunc_list);
101 spin_lock_init(&sdp->sd_trunc_lock);
103 mapping = &sdp->sd_aspace;
105 address_space_init_once(mapping);
106 mapping->a_ops = &gfs2_meta_aops;
107 mapping->host = sb->s_bdev->bd_inode;
109 mapping_set_gfp_mask(mapping, GFP_NOFS);
110 mapping->private_data = NULL;
111 mapping->backing_dev_info = sb->s_bdi;
112 mapping->writeback_index = 0;
114 spin_lock_init(&sdp->sd_log_lock);
115 atomic_set(&sdp->sd_log_pinned, 0);
116 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
117 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
118 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
119 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
120 spin_lock_init(&sdp->sd_ordered_lock);
122 init_waitqueue_head(&sdp->sd_log_waitq);
123 init_waitqueue_head(&sdp->sd_logd_waitq);
124 spin_lock_init(&sdp->sd_ail_lock);
125 INIT_LIST_HEAD(&sdp->sd_ail1_list);
126 INIT_LIST_HEAD(&sdp->sd_ail2_list);
128 init_rwsem(&sdp->sd_log_flush_lock);
129 atomic_set(&sdp->sd_log_in_flight, 0);
130 init_waitqueue_head(&sdp->sd_log_flush_wait);
132 INIT_LIST_HEAD(&sdp->sd_revoke_list);
139 * gfs2_check_sb - Check superblock
140 * @sdp: the filesystem
141 * @sb: The superblock
142 * @silent: Don't print a message if the check fails
144 * Checks the version code of the FS is one that we understand how to
145 * read and that the sizes of the various on-disk structures have not
149 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
151 struct gfs2_sb_host *sb = &sdp->sd_sb;
153 if (sb->sb_magic != GFS2_MAGIC ||
154 sb->sb_type != GFS2_METATYPE_SB) {
156 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
160 /* If format numbers match exactly, we're done. */
162 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
163 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
166 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
171 static void end_bio_io_page(struct bio *bio, int error)
173 struct page *page = bio->bi_private;
176 SetPageUptodate(page);
178 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
182 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
184 struct gfs2_sb_host *sb = &sdp->sd_sb;
185 struct super_block *s = sdp->sd_vfs;
186 const struct gfs2_sb *str = buf;
188 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
189 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
190 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
191 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
192 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
193 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
194 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
195 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
196 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
197 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
198 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
200 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
201 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
202 memcpy(s->s_uuid, str->sb_uuid, 16);
206 * gfs2_read_super - Read the gfs2 super block from disk
207 * @sdp: The GFS2 super block
208 * @sector: The location of the super block
209 * @error: The error code to return
211 * This uses the bio functions to read the super block from disk
212 * because we want to be 100% sure that we never read cached data.
213 * A super block is read twice only during each GFS2 mount and is
214 * never written to by the filesystem. The first time its read no
215 * locks are held, and the only details which are looked at are those
216 * relating to the locking protocol. Once locking is up and working,
217 * the sb is read again under the lock to establish the location of
218 * the master directory (contains pointers to journals etc) and the
221 * Returns: 0 on success or error
224 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
226 struct super_block *sb = sdp->sd_vfs;
231 page = alloc_page(GFP_NOFS);
235 ClearPageUptodate(page);
236 ClearPageDirty(page);
239 bio = bio_alloc(GFP_NOFS, 1);
240 bio->bi_sector = sector * (sb->s_blocksize >> 9);
241 bio->bi_bdev = sb->s_bdev;
242 bio_add_page(bio, page, PAGE_SIZE, 0);
244 bio->bi_end_io = end_bio_io_page;
245 bio->bi_private = page;
246 submit_bio(READ_SYNC | REQ_META, bio);
247 wait_on_page_locked(page);
249 if (!PageUptodate(page)) {
257 return gfs2_check_sb(sdp, silent);
261 * gfs2_read_sb - Read super block
262 * @sdp: The GFS2 superblock
263 * @silent: Don't print message if mount fails
267 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
269 u32 hash_blocks, ind_blocks, leaf_blocks;
274 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
277 fs_err(sdp, "can't read superblock\n");
281 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
282 GFS2_BASIC_BLOCK_SHIFT;
283 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
284 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
285 sizeof(struct gfs2_dinode)) / sizeof(u64);
286 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
287 sizeof(struct gfs2_meta_header)) / sizeof(u64);
288 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
289 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
290 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
291 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
292 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
293 sizeof(struct gfs2_meta_header)) /
294 sizeof(struct gfs2_quota_change);
295 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
296 sizeof(struct gfs2_meta_header))
297 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
299 /* Compute maximum reservation required to add a entry to a directory */
301 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
305 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
306 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
307 ind_blocks += tmp_blocks;
310 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
312 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
314 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
315 sizeof(struct gfs2_dinode);
316 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
321 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
323 m = do_div(d, sdp->sd_inptrs);
325 if (d != sdp->sd_heightsize[x - 1] || m)
327 sdp->sd_heightsize[x] = space;
329 sdp->sd_max_height = x;
330 sdp->sd_heightsize[x] = ~0;
331 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
333 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
334 sizeof(struct gfs2_dinode);
335 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
340 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
342 m = do_div(d, sdp->sd_inptrs);
344 if (d != sdp->sd_jheightsize[x - 1] || m)
346 sdp->sd_jheightsize[x] = space;
348 sdp->sd_max_jheight = x;
349 sdp->sd_jheightsize[x] = ~0;
350 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
355 static int init_names(struct gfs2_sbd *sdp, int silent)
360 proto = sdp->sd_args.ar_lockproto;
361 table = sdp->sd_args.ar_locktable;
363 /* Try to autodetect */
365 if (!proto[0] || !table[0]) {
366 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
371 proto = sdp->sd_sb.sb_lockproto;
373 table = sdp->sd_sb.sb_locktable;
377 table = sdp->sd_vfs->s_id;
379 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
380 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
382 table = sdp->sd_table_name;
383 while ((table = strchr(table, '/')))
389 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
397 error = gfs2_glock_nq_num(sdp,
398 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
399 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
402 fs_err(sdp, "can't acquire mount glock: %d\n", error);
406 error = gfs2_glock_nq_num(sdp,
407 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
409 LM_FLAG_NOEXP | GL_EXACT,
412 fs_err(sdp, "can't acquire live glock: %d\n", error);
416 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
417 CREATE, &sdp->sd_rename_gl);
419 fs_err(sdp, "can't create rename glock: %d\n", error);
423 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
424 CREATE, &sdp->sd_trans_gl);
426 fs_err(sdp, "can't create transaction glock: %d\n", error);
433 gfs2_glock_put(sdp->sd_trans_gl);
435 gfs2_glock_put(sdp->sd_rename_gl);
437 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
439 gfs2_glock_dq_uninit(mount_gh);
444 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
445 u64 no_addr, const char *name)
447 struct gfs2_sbd *sdp = sb->s_fs_info;
448 struct dentry *dentry;
451 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
453 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
454 return PTR_ERR(inode);
456 dentry = d_make_root(inode);
458 fs_err(sdp, "can't alloc %s dentry\n", name);
465 static int init_sb(struct gfs2_sbd *sdp, int silent)
467 struct super_block *sb = sdp->sd_vfs;
468 struct gfs2_holder sb_gh;
472 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
473 LM_ST_SHARED, 0, &sb_gh);
475 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
479 ret = gfs2_read_sb(sdp, silent);
481 fs_err(sdp, "can't read superblock: %d\n", ret);
485 /* Set up the buffer cache and SB for real */
486 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
488 fs_err(sdp, "FS block size (%u) is too small for device "
490 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
493 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
495 fs_err(sdp, "FS block size (%u) is too big for machine "
497 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
500 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
502 /* Get the root inode */
503 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
504 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
508 /* Get the master inode */
509 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
510 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
512 dput(sdp->sd_root_dir);
515 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
517 gfs2_glock_dq_uninit(&sb_gh);
522 * map_journal_extents - create a reusable "extent" mapping from all logical
523 * blocks to all physical blocks for the given journal. This will save
524 * us time when writing journal blocks. Most journals will have only one
525 * extent that maps all their logical blocks. That's because gfs2.mkfs
526 * arranges the journal blocks sequentially to maximize performance.
527 * So the extent would map the first block for the entire file length.
528 * However, gfs2_jadd can happen while file activity is happening, so
529 * those journals may not be sequential. Less likely is the case where
530 * the users created their own journals by mounting the metafs and
531 * laying it out. But it's still possible. These journals might have
534 * TODO: This should be done in bigger chunks rather than one block at a time,
535 * but since it's only done at mount time, I'm not worried about the
538 static int map_journal_extents(struct gfs2_sbd *sdp)
540 struct gfs2_jdesc *jd = sdp->sd_jdesc;
542 u64 db, prev_db; /* logical block, disk block, prev disk block */
543 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
544 struct gfs2_journal_extent *jext = NULL;
545 struct buffer_head bh;
550 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
553 bh.b_size = 1 << ip->i_inode.i_blkbits;
554 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
557 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
558 "%u db=%llu\n", rc, lb, (unsigned long long)db);
561 if (!prev_db || db != prev_db + 1) {
562 jext = kzalloc(sizeof(struct gfs2_journal_extent),
565 printk(KERN_INFO "GFS2 error: out of memory "
566 "mapping journal extents.\n");
573 list_add_tail(&jext->extent_list, &jd->extent_list);
582 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
584 char *message = "FIRSTMOUNT=Done";
585 char *envp[] = { message, NULL };
587 fs_info(sdp, "first mount done, others may mount\n");
589 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
590 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
592 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
596 * gfs2_jindex_hold - Grab a lock on the jindex
597 * @sdp: The GFS2 superblock
598 * @ji_gh: the holder for the jindex glock
603 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
605 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
608 struct gfs2_jdesc *jd;
613 mutex_lock(&sdp->sd_jindex_mutex);
616 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
620 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
621 name.hash = gfs2_disk_hash(name.name, name.len);
623 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
624 if (error == -ENOENT) {
629 gfs2_glock_dq_uninit(ji_gh);
635 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
639 INIT_LIST_HEAD(&jd->extent_list);
640 INIT_WORK(&jd->jd_work, gfs2_recover_func);
641 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
642 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
646 error = PTR_ERR(jd->jd_inode);
651 spin_lock(&sdp->sd_jindex_spin);
652 jd->jd_jid = sdp->sd_journals++;
653 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
654 spin_unlock(&sdp->sd_jindex_spin);
657 mutex_unlock(&sdp->sd_jindex_mutex);
663 * check_journal_clean - Make sure a journal is clean for a spectator mount
664 * @sdp: The GFS2 superblock
665 * @jd: The journal descriptor
667 * Returns: 0 if the journal is clean or locked, else an error
669 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
672 struct gfs2_holder j_gh;
673 struct gfs2_log_header_host head;
674 struct gfs2_inode *ip;
676 ip = GFS2_I(jd->jd_inode);
677 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
678 GL_EXACT | GL_NOCACHE, &j_gh);
680 fs_err(sdp, "Error locking journal for spectator mount.\n");
683 error = gfs2_jdesc_check(jd);
685 fs_err(sdp, "Error checking journal for spectator mount.\n");
688 error = gfs2_find_jhead(jd, &head);
690 fs_err(sdp, "Error parsing journal for spectator mount.\n");
693 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
695 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
696 "must not be a spectator.\n", jd->jd_jid);
700 gfs2_glock_dq_uninit(&j_gh);
704 static int init_journal(struct gfs2_sbd *sdp, int undo)
706 struct inode *master = sdp->sd_master_dir->d_inode;
707 struct gfs2_holder ji_gh;
708 struct gfs2_inode *ip;
717 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
718 if (IS_ERR(sdp->sd_jindex)) {
719 fs_err(sdp, "can't lookup journal index: %d\n", error);
720 return PTR_ERR(sdp->sd_jindex);
723 /* Load in the journal index special file */
725 error = gfs2_jindex_hold(sdp, &ji_gh);
727 fs_err(sdp, "can't read journal index: %d\n", error);
732 if (!gfs2_jindex_size(sdp)) {
733 fs_err(sdp, "no journals!\n");
737 if (sdp->sd_args.ar_spectator) {
738 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
739 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
740 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
741 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
743 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
744 fs_err(sdp, "can't mount journal #%u\n",
745 sdp->sd_lockstruct.ls_jid);
746 fs_err(sdp, "there are only %u journals (0 - %u)\n",
747 gfs2_jindex_size(sdp),
748 gfs2_jindex_size(sdp) - 1);
751 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
753 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
755 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
756 &sdp->sd_journal_gh);
758 fs_err(sdp, "can't acquire journal glock: %d\n", error);
762 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
763 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
764 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
767 fs_err(sdp, "can't acquire journal inode glock: %d\n",
769 goto fail_journal_gh;
772 error = gfs2_jdesc_check(sdp->sd_jdesc);
774 fs_err(sdp, "my journal (%u) is bad: %d\n",
775 sdp->sd_jdesc->jd_jid, error);
778 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
779 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
780 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
782 /* Map the extents for this journal's blocks */
783 map_journal_extents(sdp);
785 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
787 if (sdp->sd_lockstruct.ls_first) {
789 for (x = 0; x < sdp->sd_journals; x++) {
790 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
792 if (sdp->sd_args.ar_spectator) {
793 error = check_journal_clean(sdp, jd);
798 error = gfs2_recover_journal(jd, true);
800 fs_err(sdp, "error recovering journal %u: %d\n",
806 gfs2_others_may_mount(sdp);
807 } else if (!sdp->sd_args.ar_spectator) {
808 error = gfs2_recover_journal(sdp->sd_jdesc, true);
810 fs_err(sdp, "error recovering my journal: %d\n", error);
815 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
816 gfs2_glock_dq_uninit(&ji_gh);
822 if (!sdp->sd_args.ar_spectator)
823 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
825 if (!sdp->sd_args.ar_spectator)
826 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
828 gfs2_jindex_free(sdp);
830 gfs2_glock_dq_uninit(&ji_gh);
832 iput(sdp->sd_jindex);
836 static struct lock_class_key gfs2_quota_imutex_key;
838 static int init_inodes(struct gfs2_sbd *sdp, int undo)
841 struct inode *master = sdp->sd_master_dir->d_inode;
846 error = init_journal(sdp, undo);
850 /* Read in the master statfs inode */
851 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
852 if (IS_ERR(sdp->sd_statfs_inode)) {
853 error = PTR_ERR(sdp->sd_statfs_inode);
854 fs_err(sdp, "can't read in statfs inode: %d\n", error);
858 /* Read in the resource index inode */
859 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
860 if (IS_ERR(sdp->sd_rindex)) {
861 error = PTR_ERR(sdp->sd_rindex);
862 fs_err(sdp, "can't get resource index inode: %d\n", error);
865 sdp->sd_rindex_uptodate = 0;
867 /* Read in the quota inode */
868 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
869 if (IS_ERR(sdp->sd_quota_inode)) {
870 error = PTR_ERR(sdp->sd_quota_inode);
871 fs_err(sdp, "can't get quota file inode: %d\n", error);
875 * i_mutex on quota files is special. Since this inode is hidden system
876 * file, we are safe to define locking ourselves.
878 lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
879 &gfs2_quota_imutex_key);
881 error = gfs2_rindex_update(sdp);
888 iput(sdp->sd_quota_inode);
890 gfs2_clear_rgrpd(sdp);
891 iput(sdp->sd_rindex);
893 iput(sdp->sd_statfs_inode);
895 init_journal(sdp, UNDO);
900 static int init_per_node(struct gfs2_sbd *sdp, int undo)
902 struct inode *pn = NULL;
905 struct gfs2_inode *ip;
906 struct inode *master = sdp->sd_master_dir->d_inode;
908 if (sdp->sd_args.ar_spectator)
914 pn = gfs2_lookup_simple(master, "per_node");
917 fs_err(sdp, "can't find per_node directory: %d\n", error);
921 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
922 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
923 if (IS_ERR(sdp->sd_sc_inode)) {
924 error = PTR_ERR(sdp->sd_sc_inode);
925 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
929 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
930 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
931 if (IS_ERR(sdp->sd_qc_inode)) {
932 error = PTR_ERR(sdp->sd_qc_inode);
933 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
940 ip = GFS2_I(sdp->sd_sc_inode);
941 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
944 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
948 ip = GFS2_I(sdp->sd_qc_inode);
949 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
952 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
959 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
961 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
963 iput(sdp->sd_qc_inode);
965 iput(sdp->sd_sc_inode);
972 static const match_table_t nolock_tokens = {
973 { Opt_jid, "jid=%d\n", },
977 static const struct lm_lockops nolock_ops = {
978 .lm_proto_name = "lock_nolock",
979 .lm_put_lock = gfs2_glock_free,
980 .lm_tokens = &nolock_tokens,
984 * gfs2_lm_mount - mount a locking protocol
985 * @sdp: the filesystem
986 * @args: mount arguments
987 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
992 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
994 const struct lm_lockops *lm;
995 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
996 struct gfs2_args *args = &sdp->sd_args;
997 const char *proto = sdp->sd_proto_name;
998 const char *table = sdp->sd_table_name;
1002 if (!strcmp("lock_nolock", proto)) {
1004 sdp->sd_args.ar_localflocks = 1;
1005 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1006 } else if (!strcmp("lock_dlm", proto)) {
1010 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
1014 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1019 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1020 substring_t tmp[MAX_OPT_ARGS];
1026 token = match_token(o, *lm->lm_tokens, tmp);
1029 ret = match_int(&tmp[0], &option);
1030 if (ret || option < 0)
1031 goto hostdata_error;
1032 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1033 ls->ls_jid = option;
1037 /* Obsolete, but left for backward compat purposes */
1040 ret = match_int(&tmp[0], &option);
1041 if (ret || (option != 0 && option != 1))
1042 goto hostdata_error;
1043 ls->ls_first = option;
1048 fs_info(sdp, "unknown hostdata (%s)\n", o);
1053 if (lm->lm_mount == NULL) {
1054 fs_info(sdp, "Now mounting FS...\n");
1055 complete_all(&sdp->sd_locking_init);
1058 ret = lm->lm_mount(sdp, table);
1060 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1061 complete_all(&sdp->sd_locking_init);
1065 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1067 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1068 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1070 lm->lm_unmount(sdp);
1073 static int gfs2_journalid_wait(void *word)
1075 if (signal_pending(current))
1081 static int wait_on_journal(struct gfs2_sbd *sdp)
1083 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1086 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1089 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1091 struct super_block *sb = sdp->sd_vfs;
1094 char *envp[] = { ro, spectator, NULL };
1095 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1096 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1097 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1101 * fill_super - Read in superblock
1102 * @sb: The VFS superblock
1103 * @data: Mount options
1104 * @silent: Don't complain if it's not a GFS2 filesystem
1109 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1111 struct gfs2_sbd *sdp;
1112 struct gfs2_holder mount_gh;
1117 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1120 sdp->sd_args = *args;
1122 if (sdp->sd_args.ar_spectator) {
1123 sb->s_flags |= MS_RDONLY;
1124 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1126 if (sdp->sd_args.ar_posix_acl)
1127 sb->s_flags |= MS_POSIXACL;
1128 if (sdp->sd_args.ar_nobarrier)
1129 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1131 sb->s_flags |= MS_NOSEC;
1132 sb->s_magic = GFS2_MAGIC;
1133 sb->s_op = &gfs2_super_ops;
1134 sb->s_d_op = &gfs2_dops;
1135 sb->s_export_op = &gfs2_export_ops;
1136 sb->s_xattr = gfs2_xattr_handlers;
1137 sb->s_qcop = &gfs2_quotactl_ops;
1138 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1139 sb->s_time_gran = 1;
1140 sb->s_maxbytes = MAX_LFS_FILESIZE;
1142 /* Set up the buffer cache and fill in some fake block size values
1143 to allow us to read-in the on-disk superblock. */
1144 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1145 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1146 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1147 GFS2_BASIC_BLOCK_SHIFT;
1148 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1150 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1151 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1152 if (sdp->sd_args.ar_statfs_quantum) {
1153 sdp->sd_tune.gt_statfs_slow = 0;
1154 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1156 sdp->sd_tune.gt_statfs_slow = 1;
1157 sdp->sd_tune.gt_statfs_quantum = 30;
1160 error = init_names(sdp, silent);
1162 /* In this case, we haven't initialized sysfs, so we have to
1163 manually free the sdp. */
1164 free_percpu(sdp->sd_lkstats);
1166 sb->s_fs_info = NULL;
1170 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1172 error = gfs2_sys_fs_add(sdp);
1174 * If we hit an error here, gfs2_sys_fs_add will have called function
1175 * kobject_put which causes the sysfs usage count to go to zero, which
1176 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1177 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1178 * kobject_put to free sdp.
1183 gfs2_create_debugfs_file(sdp);
1185 error = gfs2_lm_mount(sdp, silent);
1189 error = init_locking(sdp, &mount_gh, DO);
1193 error = init_sb(sdp, silent);
1197 error = wait_on_journal(sdp);
1202 * If user space has failed to join the cluster or some similar
1203 * failure has occurred, then the journal id will contain a
1204 * negative (error) number. This will then be returned to the
1205 * caller (of the mount syscall). We do this even for spectator
1206 * mounts (which just write a jid of 0 to indicate "ok" even though
1207 * the jid is unused in the spectator case)
1209 if (sdp->sd_lockstruct.ls_jid < 0) {
1210 error = sdp->sd_lockstruct.ls_jid;
1211 sdp->sd_lockstruct.ls_jid = 0;
1215 if (sdp->sd_args.ar_spectator)
1216 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1217 sdp->sd_table_name);
1219 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1220 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1222 error = init_inodes(sdp, DO);
1226 error = init_per_node(sdp, DO);
1230 error = gfs2_statfs_init(sdp);
1232 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1236 if (!(sb->s_flags & MS_RDONLY)) {
1237 error = gfs2_make_fs_rw(sdp);
1239 fs_err(sdp, "can't make FS RW: %d\n", error);
1244 gfs2_glock_dq_uninit(&mount_gh);
1245 gfs2_online_uevent(sdp);
1249 init_per_node(sdp, UNDO);
1251 init_inodes(sdp, UNDO);
1253 if (sdp->sd_root_dir)
1254 dput(sdp->sd_root_dir);
1255 if (sdp->sd_master_dir)
1256 dput(sdp->sd_master_dir);
1261 init_locking(sdp, &mount_gh, UNDO);
1263 gfs2_gl_hash_clear(sdp);
1264 gfs2_lm_unmount(sdp);
1266 gfs2_delete_debugfs_file(sdp);
1267 free_percpu(sdp->sd_lkstats);
1268 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1269 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1270 gfs2_sys_fs_del(sdp);
1271 sb->s_fs_info = NULL;
1275 static int set_gfs2_super(struct super_block *s, void *data)
1278 s->s_dev = s->s_bdev->bd_dev;
1281 * We set the bdi here to the queue backing, file systems can
1282 * overwrite this in ->fill_super()
1284 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1288 static int test_gfs2_super(struct super_block *s, void *ptr)
1290 struct block_device *bdev = ptr;
1291 return (bdev == s->s_bdev);
1295 * gfs2_mount - Get the GFS2 superblock
1296 * @fs_type: The GFS2 filesystem type
1297 * @flags: Mount flags
1298 * @dev_name: The name of the device
1299 * @data: The mount arguments
1301 * Q. Why not use get_sb_bdev() ?
1302 * A. We need to select one of two root directories to mount, independent
1303 * of whether this is the initial, or subsequent, mount of this sb
1305 * Returns: 0 or -ve on error
1308 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1309 const char *dev_name, void *data)
1311 struct block_device *bdev;
1312 struct super_block *s;
1313 fmode_t mode = FMODE_READ | FMODE_EXCL;
1315 struct gfs2_args args;
1316 struct gfs2_sbd *sdp;
1318 if (!(flags & MS_RDONLY))
1319 mode |= FMODE_WRITE;
1321 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1323 return ERR_CAST(bdev);
1326 * once the super is inserted into the list by sget, s_umount
1327 * will protect the lockfs code from trying to start a snapshot
1328 * while we are mounting
1330 mutex_lock(&bdev->bd_fsfreeze_mutex);
1331 if (bdev->bd_fsfreeze_count > 0) {
1332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1336 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1337 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1344 * s_umount nests inside bd_mutex during
1345 * __invalidate_device(). blkdev_put() acquires
1346 * bd_mutex and can't be called under s_umount. Drop
1347 * s_umount temporarily. This is safe as we're
1348 * holding an active reference.
1350 up_write(&s->s_umount);
1351 blkdev_put(bdev, mode);
1352 down_write(&s->s_umount);
1355 memset(&args, 0, sizeof(args));
1356 args.ar_quota = GFS2_QUOTA_DEFAULT;
1357 args.ar_data = GFS2_DATA_DEFAULT;
1358 args.ar_commit = 30;
1359 args.ar_statfs_quantum = 30;
1360 args.ar_quota_quantum = 60;
1361 args.ar_errors = GFS2_ERRORS_DEFAULT;
1363 error = gfs2_mount_args(&args, data);
1365 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1371 if ((flags ^ s->s_flags) & MS_RDONLY)
1374 char b[BDEVNAME_SIZE];
1377 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1378 sb_set_blocksize(s, block_size(bdev));
1379 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1382 s->s_flags |= MS_ACTIVE;
1388 return dget(sdp->sd_master_dir);
1390 return dget(sdp->sd_root_dir);
1393 deactivate_locked_super(s);
1394 return ERR_PTR(error);
1396 blkdev_put(bdev, mode);
1397 return ERR_PTR(error);
1400 static int set_meta_super(struct super_block *s, void *ptr)
1405 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1406 int flags, const char *dev_name, void *data)
1408 struct super_block *s;
1409 struct gfs2_sbd *sdp;
1413 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1415 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1417 return ERR_PTR(error);
1419 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1420 path.dentry->d_inode->i_sb->s_bdev);
1423 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1426 if ((flags ^ s->s_flags) & MS_RDONLY) {
1427 deactivate_locked_super(s);
1428 return ERR_PTR(-EBUSY);
1431 return dget(sdp->sd_master_dir);
1434 static void gfs2_kill_sb(struct super_block *sb)
1436 struct gfs2_sbd *sdp = sb->s_fs_info;
1439 kill_block_super(sb);
1443 gfs2_meta_syncfs(sdp);
1444 dput(sdp->sd_root_dir);
1445 dput(sdp->sd_master_dir);
1446 sdp->sd_root_dir = NULL;
1447 sdp->sd_master_dir = NULL;
1448 shrink_dcache_sb(sb);
1449 gfs2_delete_debugfs_file(sdp);
1450 free_percpu(sdp->sd_lkstats);
1451 kill_block_super(sb);
1454 struct file_system_type gfs2_fs_type = {
1456 .fs_flags = FS_REQUIRES_DEV,
1457 .mount = gfs2_mount,
1458 .kill_sb = gfs2_kill_sb,
1459 .owner = THIS_MODULE,
1461 MODULE_ALIAS_FS("gfs2");
1463 struct file_system_type gfs2meta_fs_type = {
1465 .fs_flags = FS_REQUIRES_DEV,
1466 .mount = gfs2_mount_meta,
1467 .owner = THIS_MODULE,
1469 MODULE_ALIAS_FS("gfs2meta");