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>
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);
145 while (sdp->sd_glockd_num--)
146 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
148 if (!(sb->s_flags & MS_RDONLY)) {
149 error = gfs2_make_fs_ro(sdp);
153 /* At this point, we're through modifying the disk */
157 iput(sdp->sd_jindex);
158 iput(sdp->sd_inum_inode);
159 iput(sdp->sd_statfs_inode);
160 iput(sdp->sd_rindex);
161 iput(sdp->sd_quota_inode);
163 gfs2_glock_put(sdp->sd_rename_gl);
164 gfs2_glock_put(sdp->sd_trans_gl);
166 if (!sdp->sd_args.ar_spectator) {
167 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
168 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
169 gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
170 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
171 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
172 iput(sdp->sd_ir_inode);
173 iput(sdp->sd_sc_inode);
174 iput(sdp->sd_qc_inode);
177 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
178 gfs2_clear_rgrpd(sdp);
179 gfs2_jindex_free(sdp);
180 /* Take apart glock structures and buffer lists */
181 gfs2_gl_hash_clear(sdp);
182 /* Unmount the locking protocol */
183 gfs2_lm_unmount(sdp);
185 /* At this point, we're through participating in the lockspace */
186 gfs2_sys_fs_del(sdp);
192 * @sb: the superblock
196 static void gfs2_write_super(struct super_block *sb)
202 * gfs2_sync_fs - sync the filesystem
203 * @sb: the superblock
205 * Flushes the log to disk.
208 static int gfs2_sync_fs(struct super_block *sb, int wait)
211 if (wait && sb->s_fs_info)
212 gfs2_log_flush(sb->s_fs_info, NULL);
217 * gfs2_write_super_lockfs - prevent further writes to the filesystem
218 * @sb: the VFS structure for the filesystem
222 static void gfs2_write_super_lockfs(struct super_block *sb)
224 struct gfs2_sbd *sdp = sb->s_fs_info;
227 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
231 error = gfs2_freeze_fs(sdp);
237 fs_err(sdp, "waiting for recovery before freeze\n");
241 fs_err(sdp, "error freezing FS: %d\n", error);
245 fs_err(sdp, "retrying...\n");
251 * gfs2_unlockfs - reallow writes to the filesystem
252 * @sb: the VFS structure for the filesystem
256 static void gfs2_unlockfs(struct super_block *sb)
258 gfs2_unfreeze_fs(sb->s_fs_info);
262 * gfs2_statfs - Gather and return stats about the filesystem
263 * @sb: The superblock
264 * @statfsbuf: The buffer
266 * Returns: 0 on success or error code
269 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
271 struct super_block *sb = dentry->d_inode->i_sb;
272 struct gfs2_sbd *sdp = sb->s_fs_info;
273 struct gfs2_statfs_change_host sc;
276 if (gfs2_tune_get(sdp, gt_statfs_slow))
277 error = gfs2_statfs_slow(sdp, &sc);
279 error = gfs2_statfs_i(sdp, &sc);
284 buf->f_type = GFS2_MAGIC;
285 buf->f_bsize = sdp->sd_sb.sb_bsize;
286 buf->f_blocks = sc.sc_total;
287 buf->f_bfree = sc.sc_free;
288 buf->f_bavail = sc.sc_free;
289 buf->f_files = sc.sc_dinodes + sc.sc_free;
290 buf->f_ffree = sc.sc_free;
291 buf->f_namelen = GFS2_FNAMESIZE;
297 * gfs2_remount_fs - called when the FS is remounted
298 * @sb: the filesystem
299 * @flags: the remount flags
300 * @data: extra data passed in (not used right now)
305 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data)
307 struct gfs2_sbd *sdp = sb->s_fs_info;
310 error = gfs2_mount_args(sdp, data, 1);
314 if (sdp->sd_args.ar_spectator)
317 if (*flags & MS_RDONLY) {
318 if (!(sb->s_flags & MS_RDONLY))
319 error = gfs2_make_fs_ro(sdp);
320 } else if (!(*flags & MS_RDONLY) &&
321 (sb->s_flags & MS_RDONLY)) {
322 error = gfs2_make_fs_rw(sdp);
330 * gfs2_drop_inode - Drop an inode (test for remote unlink)
331 * @inode: The inode to drop
333 * If we've received a callback on an iopen lock then its because a
334 * remote node tried to deallocate the inode but failed due to this node
335 * still having the inode open. Here we mark the link count zero
336 * since we know that it must have reached zero if the GLF_DEMOTE flag
337 * is set on the iopen glock. If we didn't do a disk read since the
338 * remote node removed the final link then we might otherwise miss
339 * this event. This check ensures that this node will deallocate the
340 * inode's blocks, or alternatively pass the baton on to another
341 * node for later deallocation.
344 static void gfs2_drop_inode(struct inode *inode)
346 struct gfs2_inode *ip = GFS2_I(inode);
348 if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) {
349 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
350 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags))
353 generic_drop_inode(inode);
357 * gfs2_clear_inode - Deallocate an inode when VFS is done with it
358 * @inode: The VFS inode
362 static void gfs2_clear_inode(struct inode *inode)
364 struct gfs2_inode *ip = GFS2_I(inode);
366 /* This tells us its a "real" inode and not one which only
367 * serves to contain an address space (see rgrp.c, meta_io.c)
368 * which therefore doesn't have its own glocks.
370 if (test_bit(GIF_USER, &ip->i_flags)) {
371 ip->i_gl->gl_object = NULL;
372 gfs2_glock_schedule_for_reclaim(ip->i_gl);
373 gfs2_glock_put(ip->i_gl);
375 if (ip->i_iopen_gh.gh_gl) {
376 ip->i_iopen_gh.gh_gl->gl_object = NULL;
377 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
382 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
388 } while (!IS_ROOT(d1));
393 * gfs2_show_options - Show mount options for /proc/mounts
394 * @s: seq_file structure
397 * Returns: 0 on success or error code
400 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
402 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info;
403 struct gfs2_args *args = &sdp->sd_args;
405 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir))
406 seq_printf(s, ",meta");
407 if (args->ar_lockproto[0])
408 seq_printf(s, ",lockproto=%s", args->ar_lockproto);
409 if (args->ar_locktable[0])
410 seq_printf(s, ",locktable=%s", args->ar_locktable);
411 if (args->ar_hostdata[0])
412 seq_printf(s, ",hostdata=%s", args->ar_hostdata);
413 if (args->ar_spectator)
414 seq_printf(s, ",spectator");
415 if (args->ar_ignore_local_fs)
416 seq_printf(s, ",ignore_local_fs");
417 if (args->ar_localflocks)
418 seq_printf(s, ",localflocks");
419 if (args->ar_localcaching)
420 seq_printf(s, ",localcaching");
422 seq_printf(s, ",debug");
423 if (args->ar_upgrade)
424 seq_printf(s, ",upgrade");
425 if (args->ar_num_glockd != GFS2_GLOCKD_DEFAULT)
426 seq_printf(s, ",num_glockd=%u", args->ar_num_glockd);
427 if (args->ar_posix_acl)
428 seq_printf(s, ",acl");
429 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
431 switch (args->ar_quota) {
435 case GFS2_QUOTA_ACCOUNT:
445 seq_printf(s, ",quota=%s", state);
447 if (args->ar_suiddir)
448 seq_printf(s, ",suiddir");
449 if (args->ar_data != GFS2_DATA_DEFAULT) {
451 switch (args->ar_data) {
452 case GFS2_DATA_WRITEBACK:
455 case GFS2_DATA_ORDERED:
462 seq_printf(s, ",data=%s", state);
469 * We have to (at the moment) hold the inodes main lock to cover
470 * the gap between unlocking the shared lock on the iopen lock and
471 * taking the exclusive lock. I'd rather do a shared -> exclusive
472 * conversion on the iopen lock, but we can change that later. This
473 * is safe, just less efficient.
476 static void gfs2_delete_inode(struct inode *inode)
478 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
479 struct gfs2_inode *ip = GFS2_I(inode);
480 struct gfs2_holder gh;
483 if (!test_bit(GIF_USER, &ip->i_flags))
486 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
487 if (unlikely(error)) {
488 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
492 gfs2_glock_dq_wait(&ip->i_iopen_gh);
493 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh);
494 error = gfs2_glock_nq(&ip->i_iopen_gh);
498 if (S_ISDIR(inode->i_mode) &&
499 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
500 error = gfs2_dir_exhash_dealloc(ip);
506 error = gfs2_ea_dealloc(ip);
511 if (!gfs2_is_stuffed(ip)) {
512 error = gfs2_file_dealloc(ip);
517 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 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags))
531 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,