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/lm_interface.h>
23 #include <linux/time.h>
31 #include "ops_super.h"
45 * gfs2_write_inode - Make sure the inode is stable on the disk
47 * @sync: synchronous write flag
52 static int gfs2_write_inode(struct inode *inode, int sync)
54 struct gfs2_inode *ip = GFS2_I(inode);
55 struct gfs2_sbd *sdp = GFS2_SB(inode);
56 struct gfs2_holder gh;
57 struct buffer_head *bh;
58 struct timespec atime;
59 struct gfs2_dinode *di;
62 /* Check this is a "normal" inode, etc */
63 if (!test_bit(GIF_USER, &ip->i_flags) ||
64 (current->flags & PF_MEMALLOC))
66 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
69 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
72 ret = gfs2_meta_inode_buffer(ip, &bh);
74 di = (struct gfs2_dinode *)bh->b_data;
75 atime.tv_sec = be64_to_cpu(di->di_atime);
76 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec);
77 if (timespec_compare(&inode->i_atime, &atime) > 0) {
78 gfs2_trans_add_bh(ip->i_gl, bh, 1);
79 gfs2_dinode_out(ip, bh->b_data);
85 gfs2_glock_dq_uninit(&gh);
88 gfs2_log_flush(GFS2_SB(inode), ip->i_gl);
93 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
94 * @sdp: the filesystem
99 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
101 struct gfs2_holder t_gh;
104 gfs2_quota_sync(sdp);
105 gfs2_statfs_sync(sdp);
107 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
109 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
112 gfs2_meta_syncfs(sdp);
113 gfs2_log_shutdown(sdp);
115 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
118 gfs2_glock_dq_uninit(&t_gh);
120 gfs2_quota_cleanup(sdp);
126 * gfs2_put_super - Unmount the filesystem
127 * @sb: The VFS superblock
131 static void gfs2_put_super(struct super_block *sb)
133 struct gfs2_sbd *sdp = sb->s_fs_info;
136 /* Unfreeze the filesystem, if we need to */
138 mutex_lock(&sdp->sd_freeze_lock);
139 if (sdp->sd_freeze_count)
140 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
141 mutex_unlock(&sdp->sd_freeze_lock);
143 kthread_stop(sdp->sd_quotad_process);
144 kthread_stop(sdp->sd_logd_process);
145 kthread_stop(sdp->sd_recoverd_process);
146 while (sdp->sd_glockd_num--)
147 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
149 if (!(sb->s_flags & MS_RDONLY)) {
150 error = gfs2_make_fs_ro(sdp);
154 /* At this point, we're through modifying the disk */
158 iput(sdp->sd_jindex);
159 iput(sdp->sd_inum_inode);
160 iput(sdp->sd_statfs_inode);
161 iput(sdp->sd_rindex);
162 iput(sdp->sd_quota_inode);
164 gfs2_glock_put(sdp->sd_rename_gl);
165 gfs2_glock_put(sdp->sd_trans_gl);
167 if (!sdp->sd_args.ar_spectator) {
168 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
169 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
170 gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
171 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
172 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
173 iput(sdp->sd_ir_inode);
174 iput(sdp->sd_sc_inode);
175 iput(sdp->sd_qc_inode);
178 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
179 gfs2_clear_rgrpd(sdp);
180 gfs2_jindex_free(sdp);
181 /* Take apart glock structures and buffer lists */
182 gfs2_gl_hash_clear(sdp);
183 /* Unmount the locking protocol */
184 gfs2_lm_unmount(sdp);
186 /* At this point, we're through participating in the lockspace */
187 gfs2_sys_fs_del(sdp);
193 * @sb: the superblock
197 static void gfs2_write_super(struct super_block *sb)
203 * gfs2_sync_fs - sync the filesystem
204 * @sb: the superblock
206 * Flushes the log to disk.
209 static int gfs2_sync_fs(struct super_block *sb, int wait)
212 if (wait && sb->s_fs_info)
213 gfs2_log_flush(sb->s_fs_info, NULL);
218 * gfs2_write_super_lockfs - prevent further writes to the filesystem
219 * @sb: the VFS structure for the filesystem
223 static void gfs2_write_super_lockfs(struct super_block *sb)
225 struct gfs2_sbd *sdp = sb->s_fs_info;
228 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
232 error = gfs2_freeze_fs(sdp);
238 fs_err(sdp, "waiting for recovery before freeze\n");
242 fs_err(sdp, "error freezing FS: %d\n", error);
246 fs_err(sdp, "retrying...\n");
252 * gfs2_unlockfs - reallow writes to the filesystem
253 * @sb: the VFS structure for the filesystem
257 static void gfs2_unlockfs(struct super_block *sb)
259 gfs2_unfreeze_fs(sb->s_fs_info);
263 * gfs2_statfs - Gather and return stats about the filesystem
264 * @sb: The superblock
265 * @statfsbuf: The buffer
267 * Returns: 0 on success or error code
270 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
272 struct super_block *sb = dentry->d_inode->i_sb;
273 struct gfs2_sbd *sdp = sb->s_fs_info;
274 struct gfs2_statfs_change_host sc;
277 if (gfs2_tune_get(sdp, gt_statfs_slow))
278 error = gfs2_statfs_slow(sdp, &sc);
280 error = gfs2_statfs_i(sdp, &sc);
285 buf->f_type = GFS2_MAGIC;
286 buf->f_bsize = sdp->sd_sb.sb_bsize;
287 buf->f_blocks = sc.sc_total;
288 buf->f_bfree = sc.sc_free;
289 buf->f_bavail = sc.sc_free;
290 buf->f_files = sc.sc_dinodes + sc.sc_free;
291 buf->f_ffree = sc.sc_free;
292 buf->f_namelen = GFS2_FNAMESIZE;
298 * gfs2_remount_fs - called when the FS is remounted
299 * @sb: the filesystem
300 * @flags: the remount flags
301 * @data: extra data passed in (not used right now)
306 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
308 struct gfs2_sbd *sdp = sb->s_fs_info;
311 error = gfs2_mount_args(sdp, data, 1);
315 if (sdp->sd_args.ar_spectator)
318 if (*flags & MS_RDONLY) {
319 if (!(sb->s_flags & MS_RDONLY))
320 error = gfs2_make_fs_ro(sdp);
321 } else if (!(*flags & MS_RDONLY) &&
322 (sb->s_flags & MS_RDONLY)) {
323 error = gfs2_make_fs_rw(sdp);
331 * gfs2_drop_inode - Drop an inode (test for remote unlink)
332 * @inode: The inode to drop
334 * If we've received a callback on an iopen lock then its because a
335 * remote node tried to deallocate the inode but failed due to this node
336 * still having the inode open. Here we mark the link count zero
337 * since we know that it must have reached zero if the GLF_DEMOTE flag
338 * is set on the iopen glock. If we didn't do a disk read since the
339 * remote node removed the final link then we might otherwise miss
340 * this event. This check ensures that this node will deallocate the
341 * inode's blocks, or alternatively pass the baton on to another
342 * node for later deallocation.
345 static void gfs2_drop_inode(struct inode *inode)
347 struct gfs2_inode *ip = GFS2_I(inode);
349 if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
350 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
351 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
354 generic_drop_inode(inode);
358 * gfs2_clear_inode - Deallocate an inode when VFS is done with it
359 * @inode: The VFS inode
363 static void gfs2_clear_inode(struct inode *inode)
365 struct gfs2_inode *ip = GFS2_I(inode);
367 /* This tells us its a "real" inode and not one which only
368 * serves to contain an address space (see rgrp.c, meta_io.c)
369 * which therefore doesn't have its own glocks.
371 if (test_bit(GIF_USER, &ip->i_flags)) {
372 ip->i_gl->gl_object = NULL;
373 gfs2_glock_schedule_for_reclaim(ip->i_gl);
374 gfs2_glock_put(ip->i_gl);
376 if (ip->i_iopen_gh.gh_gl) {
377 ip->i_iopen_gh.gh_gl->gl_object = NULL;
378 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
383 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
389 } while (!IS_ROOT(d1));
394 * gfs2_show_options - Show mount options for /proc/mounts
395 * @s: seq_file structure
398 * Returns: 0 on success or error code
401 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
403 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
404 struct gfs2_args *args = &sdp->sd_args;
406 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
407 seq_printf(s, ",meta");
408 if (args->ar_lockproto[0])
409 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
410 if (args->ar_locktable[0])
411 seq_printf(s, ",locktable=%s", args->ar_locktable);
412 if (args->ar_hostdata[0])
413 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
414 if (args->ar_spectator)
415 seq_printf(s, ",spectator");
416 if (args->ar_ignore_local_fs)
417 seq_printf(s, ",ignore_local_fs");
418 if (args->ar_localflocks)
419 seq_printf(s, ",localflocks");
420 if (args->ar_localcaching)
421 seq_printf(s, ",localcaching");
423 seq_printf(s, ",debug");
424 if (args->ar_upgrade)
425 seq_printf(s, ",upgrade");
426 if (args->ar_num_glockd != GFS2_GLOCKD_DEFAULT)
427 seq_printf(s, ",num_glockd=%u", args->ar_num_glockd);
428 if (args->ar_posix_acl)
429 seq_printf(s, ",acl");
430 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
432 switch (args->ar_quota) {
436 case GFS2_QUOTA_ACCOUNT:
446 seq_printf(s, ",quota=%s", state);
448 if (args->ar_suiddir)
449 seq_printf(s, ",suiddir");
450 if (args->ar_data != GFS2_DATA_DEFAULT) {
452 switch (args->ar_data) {
453 case GFS2_DATA_WRITEBACK:
456 case GFS2_DATA_ORDERED:
463 seq_printf(s, ",data=%s", state);
470 * We have to (at the moment) hold the inodes main lock to cover
471 * the gap between unlocking the shared lock on the iopen lock and
472 * taking the exclusive lock. I'd rather do a shared -> exclusive
473 * conversion on the iopen lock, but we can change that later. This
474 * is safe, just less efficient.
477 static void gfs2_delete_inode(struct inode *inode)
479 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
480 struct gfs2_inode *ip = GFS2_I(inode);
481 struct gfs2_holder gh;
484 if (!test_bit(GIF_USER, &ip->i_flags))
487 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
488 if (unlikely(error)) {
489 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
493 gfs2_glock_dq_wait(&ip->i_iopen_gh);
494 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
495 error = gfs2_glock_nq(&ip->i_iopen_gh);
499 if (S_ISDIR(inode->i_mode) &&
500 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
501 error = gfs2_dir_exhash_dealloc(ip);
506 if (ip->i_di.di_eattr) {
507 error = gfs2_ea_dealloc(ip);
512 if (!gfs2_is_stuffed(ip)) {
513 error = gfs2_file_dealloc(ip);
518 error = gfs2_dinode_dealloc(ip);
522 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
525 /* Needs to be done before glock release & also in a transaction */
526 truncate_inode_pages(&inode->i_data, 0);
530 gfs2_glock_dq(&ip->i_iopen_gh);
532 gfs2_holder_uninit(&ip->i_iopen_gh);
533 gfs2_glock_dq_uninit(&gh);
534 if (error && error != GLR_TRYFAILED)
535 fs_warn(sdp, "gfs2_delete_inode: %d\n", error);
537 truncate_inode_pages(&inode->i_data, 0);
541 static struct inode *gfs2_alloc_inode(struct super_block *sb)
543 struct gfs2_inode *ip;
545 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
553 static void gfs2_destroy_inode(struct inode *inode)
555 kmem_cache_free(gfs2_inode_cachep, inode);
558 const struct super_operations gfs2_super_ops = {
559 .alloc_inode = gfs2_alloc_inode,
560 .destroy_inode = gfs2_destroy_inode,
561 .write_inode = gfs2_write_inode,
562 .delete_inode = gfs2_delete_inode,
563 .put_super = gfs2_put_super,
564 .write_super = gfs2_write_super,
565 .sync_fs = gfs2_sync_fs,
566 .write_super_lockfs = gfs2_write_super_lockfs,
567 .unlockfs = gfs2_unlockfs,
568 .statfs = gfs2_statfs,
569 .remount_fs = gfs2_remount_fs,
570 .clear_inode = gfs2_clear_inode,
571 .drop_inode = gfs2_drop_inode,
572 .show_options = gfs2_show_options,