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/statfs.h>
16 #include <linux/seq_file.h>
17 #include <linux/mount.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/crc32.h>
22 #include <linux/time.h>
41 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x)
44 * gfs2_write_inode - Make sure the inode is stable on the disk
46 * @sync: synchronous write flag
51 static int gfs2_write_inode(struct inode *inode, int sync)
53 struct gfs2_inode *ip = GFS2_I(inode);
54 struct gfs2_sbd *sdp = GFS2_SB(inode);
55 struct gfs2_holder gh;
56 struct buffer_head *bh;
57 struct timespec atime;
58 struct gfs2_dinode *di;
61 /* Check this is a "normal" inode, etc */
62 if (!test_bit(GIF_USER, &ip->i_flags) ||
63 (current->flags & PF_MEMALLOC))
65 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
68 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
71 ret = gfs2_meta_inode_buffer(ip, &bh);
73 di = (struct gfs2_dinode *)bh->b_data;
74 atime.tv_sec = be64_to_cpu(di->di_atime);
75 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
76 if (timespec_compare(&inode->i_atime, &atime) > 0) {
77 gfs2_trans_add_bh(ip->i_gl, bh, 1);
78 gfs2_dinode_out(ip, bh->b_data);
84 gfs2_glock_dq_uninit(&gh);
87 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
92 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
93 * @sdp: the filesystem
98 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
100 struct gfs2_holder t_gh;
103 gfs2_quota_sync(sdp);
104 gfs2_statfs_sync(sdp);
106 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
108 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
111 gfs2_meta_syncfs(sdp);
112 gfs2_log_shutdown(sdp);
114 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
117 gfs2_glock_dq_uninit(&t_gh);
119 gfs2_quota_cleanup(sdp);
125 * gfs2_put_super - Unmount the filesystem
126 * @sb: The VFS superblock
130 static void gfs2_put_super(struct super_block *sb)
132 struct gfs2_sbd *sdp = sb->s_fs_info;
135 /* Unfreeze the filesystem, if we need to */
137 mutex_lock(&sdp->sd_freeze_lock);
138 if (sdp->sd_freeze_count)
139 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
140 mutex_unlock(&sdp->sd_freeze_lock);
142 kthread_stop(sdp->sd_quotad_process);
143 kthread_stop(sdp->sd_logd_process);
144 kthread_stop(sdp->sd_recoverd_process);
146 if (!(sb->s_flags & MS_RDONLY)) {
147 error = gfs2_make_fs_ro(sdp);
151 /* At this point, we're through modifying the disk */
155 iput(sdp->sd_jindex);
156 iput(sdp->sd_inum_inode);
157 iput(sdp->sd_statfs_inode);
158 iput(sdp->sd_rindex);
159 iput(sdp->sd_quota_inode);
161 gfs2_glock_put(sdp->sd_rename_gl);
162 gfs2_glock_put(sdp->sd_trans_gl);
164 if (!sdp->sd_args.ar_spectator) {
165 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
166 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
167 gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
168 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
169 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
170 iput(sdp->sd_ir_inode);
171 iput(sdp->sd_sc_inode);
172 iput(sdp->sd_qc_inode);
175 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
176 gfs2_clear_rgrpd(sdp);
177 gfs2_jindex_free(sdp);
178 /* Take apart glock structures and buffer lists */
179 gfs2_gl_hash_clear(sdp);
180 /* Unmount the locking protocol */
181 gfs2_lm_unmount(sdp);
183 /* At this point, we're through participating in the lockspace */
184 gfs2_sys_fs_del(sdp);
189 * @sb: the superblock
193 static void gfs2_write_super(struct super_block *sb)
199 * gfs2_sync_fs - sync the filesystem
200 * @sb: the superblock
202 * Flushes the log to disk.
205 static int gfs2_sync_fs(struct super_block *sb, int wait)
208 if (wait && sb->s_fs_info)
209 gfs2_log_flush(sb->s_fs_info, NULL);
214 * gfs2_freeze - prevent further writes to the filesystem
215 * @sb: the VFS structure for the filesystem
219 static int gfs2_freeze(struct super_block *sb)
221 struct gfs2_sbd *sdp = sb->s_fs_info;
224 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
228 error = gfs2_freeze_fs(sdp);
234 fs_err(sdp, "waiting for recovery before freeze\n");
238 fs_err(sdp, "error freezing FS: %d\n", error);
242 fs_err(sdp, "retrying...\n");
249 * gfs2_unfreeze - reallow writes to the filesystem
250 * @sb: the VFS structure for the filesystem
254 static int gfs2_unfreeze(struct super_block *sb)
256 gfs2_unfreeze_fs(sb->s_fs_info);
261 * statfs_fill - fill in the sg for a given RG
263 * @sc: the sc structure
265 * Returns: 0 on success, -ESTALE if the LVB is invalid
268 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
269 struct gfs2_statfs_change_host *sc)
271 gfs2_rgrp_verify(rgd);
272 sc->sc_total += rgd->rd_data;
273 sc->sc_free += rgd->rd_free;
274 sc->sc_dinodes += rgd->rd_dinodes;
279 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
280 * @sdp: the filesystem
281 * @sc: the sc info that will be returned
283 * Any error (other than a signal) will cause this routine to fall back
284 * to the synchronous version.
286 * FIXME: This really shouldn't busy wait like this.
291 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
293 struct gfs2_holder ri_gh;
294 struct gfs2_rgrpd *rgd_next;
295 struct gfs2_holder *gha, *gh;
296 unsigned int slots = 64;
301 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
302 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
306 error = gfs2_rindex_hold(sdp, &ri_gh);
310 rgd_next = gfs2_rgrpd_get_first(sdp);
315 for (x = 0; x < slots; x++) {
318 if (gh->gh_gl && gfs2_glock_poll(gh)) {
319 err = gfs2_glock_wait(gh);
321 gfs2_holder_uninit(gh);
325 error = statfs_slow_fill(
326 gh->gh_gl->gl_object, sc);
327 gfs2_glock_dq_uninit(gh);
333 else if (rgd_next && !error) {
334 error = gfs2_glock_nq_init(rgd_next->rd_gl,
338 rgd_next = gfs2_rgrpd_get_next(rgd_next);
342 if (signal_pending(current))
343 error = -ERESTARTSYS;
352 gfs2_glock_dq_uninit(&ri_gh);
360 * gfs2_statfs_i - Do a statfs
361 * @sdp: the filesystem
362 * @sg: the sg structure
367 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
369 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
370 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
372 spin_lock(&sdp->sd_statfs_spin);
375 sc->sc_total += l_sc->sc_total;
376 sc->sc_free += l_sc->sc_free;
377 sc->sc_dinodes += l_sc->sc_dinodes;
379 spin_unlock(&sdp->sd_statfs_spin);
383 if (sc->sc_free > sc->sc_total)
384 sc->sc_free = sc->sc_total;
385 if (sc->sc_dinodes < 0)
392 * gfs2_statfs - Gather and return stats about the filesystem
393 * @sb: The superblock
394 * @statfsbuf: The buffer
396 * Returns: 0 on success or error code
399 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
401 struct super_block *sb = dentry->d_inode->i_sb;
402 struct gfs2_sbd *sdp = sb->s_fs_info;
403 struct gfs2_statfs_change_host sc;
406 if (gfs2_tune_get(sdp, gt_statfs_slow))
407 error = gfs2_statfs_slow(sdp, &sc);
409 error = gfs2_statfs_i(sdp, &sc);
414 buf->f_type = GFS2_MAGIC;
415 buf->f_bsize = sdp->sd_sb.sb_bsize;
416 buf->f_blocks = sc.sc_total;
417 buf->f_bfree = sc.sc_free;
418 buf->f_bavail = sc.sc_free;
419 buf->f_files = sc.sc_dinodes + sc.sc_free;
420 buf->f_ffree = sc.sc_free;
421 buf->f_namelen = GFS2_FNAMESIZE;
427 * gfs2_remount_fs - called when the FS is remounted
428 * @sb: the filesystem
429 * @flags: the remount flags
430 * @data: extra data passed in (not used right now)
435 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
437 struct gfs2_sbd *sdp = sb->s_fs_info;
438 struct gfs2_args args = sdp->sd_args; /* Default to current settings */
441 error = gfs2_mount_args(sdp, &args, data);
445 /* Not allowed to change locking details */
446 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) ||
447 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) ||
448 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata))
451 /* Some flags must not be changed */
452 if (args_neq(&args, &sdp->sd_args, spectator) ||
453 args_neq(&args, &sdp->sd_args, ignore_local_fs) ||
454 args_neq(&args, &sdp->sd_args, localflocks) ||
455 args_neq(&args, &sdp->sd_args, localcaching) ||
456 args_neq(&args, &sdp->sd_args, meta))
459 if (sdp->sd_args.ar_spectator)
462 if ((sb->s_flags ^ *flags) & MS_RDONLY) {
463 if (*flags & MS_RDONLY)
464 error = gfs2_make_fs_ro(sdp);
466 error = gfs2_make_fs_rw(sdp);
472 if (sdp->sd_args.ar_posix_acl)
473 sb->s_flags |= MS_POSIXACL;
475 sb->s_flags &= ~MS_POSIXACL;
480 * gfs2_drop_inode - Drop an inode (test for remote unlink)
481 * @inode: The inode to drop
483 * If we've received a callback on an iopen lock then its because a
484 * remote node tried to deallocate the inode but failed due to this node
485 * still having the inode open. Here we mark the link count zero
486 * since we know that it must have reached zero if the GLF_DEMOTE flag
487 * is set on the iopen glock. If we didn't do a disk read since the
488 * remote node removed the final link then we might otherwise miss
489 * this event. This check ensures that this node will deallocate the
490 * inode's blocks, or alternatively pass the baton on to another
491 * node for later deallocation.
494 static void gfs2_drop_inode(struct inode *inode)
496 struct gfs2_inode *ip = GFS2_I(inode);
498 if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
499 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
500 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
503 generic_drop_inode(inode);
507 * gfs2_clear_inode - Deallocate an inode when VFS is done with it
508 * @inode: The VFS inode
512 static void gfs2_clear_inode(struct inode *inode)
514 struct gfs2_inode *ip = GFS2_I(inode);
516 /* This tells us its a "real" inode and not one which only
517 * serves to contain an address space (see rgrp.c, meta_io.c)
518 * which therefore doesn't have its own glocks.
520 if (test_bit(GIF_USER, &ip->i_flags)) {
521 ip->i_gl->gl_object = NULL;
522 gfs2_glock_put(ip->i_gl);
524 if (ip->i_iopen_gh.gh_gl) {
525 ip->i_iopen_gh.gh_gl->gl_object = NULL;
526 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
531 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
537 } while (!IS_ROOT(d1));
542 * gfs2_show_options - Show mount options for /proc/mounts
543 * @s: seq_file structure
546 * Returns: 0 on success or error code
549 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
551 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
552 struct gfs2_args *args = &sdp->sd_args;
554 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
555 seq_printf(s, ",meta");
556 if (args->ar_lockproto[0])
557 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
558 if (args->ar_locktable[0])
559 seq_printf(s, ",locktable=%s", args->ar_locktable);
560 if (args->ar_hostdata[0])
561 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
562 if (args->ar_spectator)
563 seq_printf(s, ",spectator");
564 if (args->ar_ignore_local_fs)
565 seq_printf(s, ",ignore_local_fs");
566 if (args->ar_localflocks)
567 seq_printf(s, ",localflocks");
568 if (args->ar_localcaching)
569 seq_printf(s, ",localcaching");
571 seq_printf(s, ",debug");
572 if (args->ar_upgrade)
573 seq_printf(s, ",upgrade");
574 if (args->ar_posix_acl)
575 seq_printf(s, ",acl");
576 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
578 switch (args->ar_quota) {
582 case GFS2_QUOTA_ACCOUNT:
592 seq_printf(s, ",quota=%s", state);
594 if (args->ar_suiddir)
595 seq_printf(s, ",suiddir");
596 if (args->ar_data != GFS2_DATA_DEFAULT) {
598 switch (args->ar_data) {
599 case GFS2_DATA_WRITEBACK:
602 case GFS2_DATA_ORDERED:
609 seq_printf(s, ",data=%s", state);
611 if (args->ar_discard)
612 seq_printf(s, ",discard");
618 * We have to (at the moment) hold the inodes main lock to cover
619 * the gap between unlocking the shared lock on the iopen lock and
620 * taking the exclusive lock. I'd rather do a shared -> exclusive
621 * conversion on the iopen lock, but we can change that later. This
622 * is safe, just less efficient.
625 static void gfs2_delete_inode(struct inode *inode)
627 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
628 struct gfs2_inode *ip = GFS2_I(inode);
629 struct gfs2_holder gh;
632 if (!test_bit(GIF_USER, &ip->i_flags))
635 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
636 if (unlikely(error)) {
637 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
641 gfs2_glock_dq_wait(&ip->i_iopen_gh);
642 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
643 error = gfs2_glock_nq(&ip->i_iopen_gh);
647 if (S_ISDIR(inode->i_mode) &&
648 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
649 error = gfs2_dir_exhash_dealloc(ip);
655 error = gfs2_ea_dealloc(ip);
660 if (!gfs2_is_stuffed(ip)) {
661 error = gfs2_file_dealloc(ip);
666 error = gfs2_dinode_dealloc(ip);
671 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
674 /* Needs to be done before glock release & also in a transaction */
675 truncate_inode_pages(&inode->i_data, 0);
679 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
680 gfs2_glock_dq(&ip->i_iopen_gh);
681 gfs2_holder_uninit(&ip->i_iopen_gh);
682 gfs2_glock_dq_uninit(&gh);
683 if (error && error != GLR_TRYFAILED)
684 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
686 truncate_inode_pages(&inode->i_data, 0);
690 static struct inode *gfs2_alloc_inode(struct super_block *sb)
692 struct gfs2_inode *ip;
694 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
702 static void gfs2_destroy_inode(struct inode *inode)
704 kmem_cache_free(gfs2_inode_cachep, inode);
707 const struct super_operations gfs2_super_ops = {
708 .alloc_inode = gfs2_alloc_inode,
709 .destroy_inode = gfs2_destroy_inode,
710 .write_inode = gfs2_write_inode,
711 .delete_inode = gfs2_delete_inode,
712 .put_super = gfs2_put_super,
713 .write_super = gfs2_write_super,
714 .sync_fs = gfs2_sync_fs,
715 .freeze_fs = gfs2_freeze,
716 .unfreeze_fs = gfs2_unfreeze,
717 .statfs = gfs2_statfs,
718 .remount_fs = gfs2_remount_fs,
719 .clear_inode = gfs2_clear_inode,
720 .drop_inode = gfs2_drop_inode,
721 .show_options = gfs2_show_options,