2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "xfs_trans.h"
27 #include "xfs_alloc.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_quota.h"
30 #include "xfs_mount.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_btree.h"
39 #include "xfs_ialloc.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_itable.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_utils.h"
49 #include "xfs_vnodeops.h"
50 #include "xfs_version.h"
52 #include <linux/namei.h>
53 #include <linux/init.h>
54 #include <linux/mount.h>
55 #include <linux/mempool.h>
56 #include <linux/writeback.h>
57 #include <linux/kthread.h>
58 #include <linux/freezer.h>
60 static struct quotactl_ops xfs_quotactl_operations;
61 static struct super_operations xfs_super_operations;
62 static kmem_zone_t *xfs_vnode_zone;
63 static kmem_zone_t *xfs_ioend_zone;
64 mempool_t *xfs_ioend_pool;
66 STATIC struct xfs_mount_args *
68 struct super_block *sb,
71 struct xfs_mount_args *args;
73 args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
74 args->logbufs = args->logbufsize = -1;
75 strncpy(args->fsname, sb->s_id, MAXNAMELEN);
77 /* Copy the already-parsed mount(2) flags we're interested in */
78 if (sb->s_flags & MS_DIRSYNC)
79 args->flags |= XFSMNT_DIRSYNC;
80 if (sb->s_flags & MS_SYNCHRONOUS)
81 args->flags |= XFSMNT_WSYNC;
83 args->flags |= XFSMNT_QUIET;
84 args->flags |= XFSMNT_32BITINODES;
91 unsigned int blockshift)
93 unsigned int pagefactor = 1;
94 unsigned int bitshift = BITS_PER_LONG - 1;
96 /* Figure out maximum filesize, on Linux this can depend on
97 * the filesystem blocksize (on 32 bit platforms).
98 * __block_prepare_write does this in an [unsigned] long...
99 * page->index << (PAGE_CACHE_SHIFT - bbits)
100 * So, for page sized blocks (4K on 32 bit platforms),
101 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
102 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
103 * but for smaller blocksizes it is less (bbits = log2 bsize).
104 * Note1: get_block_t takes a long (implicit cast from above)
105 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
106 * can optionally convert the [unsigned] long from above into
107 * an [unsigned] long long.
110 #if BITS_PER_LONG == 32
111 # if defined(CONFIG_LBD)
112 ASSERT(sizeof(sector_t) == 8);
113 pagefactor = PAGE_CACHE_SIZE;
114 bitshift = BITS_PER_LONG;
116 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
120 return (((__uint64_t)pagefactor) << bitshift) - 1;
127 switch (inode->i_mode & S_IFMT) {
129 inode->i_op = &xfs_inode_operations;
130 inode->i_fop = &xfs_file_operations;
131 inode->i_mapping->a_ops = &xfs_address_space_operations;
134 inode->i_op = &xfs_dir_inode_operations;
135 inode->i_fop = &xfs_dir_file_operations;
138 inode->i_op = &xfs_symlink_inode_operations;
140 inode->i_mapping->a_ops = &xfs_address_space_operations;
143 inode->i_op = &xfs_inode_operations;
144 init_special_inode(inode, inode->i_mode, inode->i_rdev);
150 xfs_revalidate_inode(
155 struct inode *inode = vn_to_inode(vp);
157 inode->i_mode = ip->i_d.di_mode;
158 inode->i_nlink = ip->i_d.di_nlink;
159 inode->i_uid = ip->i_d.di_uid;
160 inode->i_gid = ip->i_d.di_gid;
162 switch (inode->i_mode & S_IFMT) {
166 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
167 sysv_minor(ip->i_df.if_u2.if_rdev));
174 inode->i_generation = ip->i_d.di_gen;
175 i_size_write(inode, ip->i_d.di_size);
177 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
178 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
179 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
180 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
181 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
182 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
183 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
184 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
185 inode->i_flags |= S_IMMUTABLE;
187 inode->i_flags &= ~S_IMMUTABLE;
188 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
189 inode->i_flags |= S_APPEND;
191 inode->i_flags &= ~S_APPEND;
192 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
193 inode->i_flags |= S_SYNC;
195 inode->i_flags &= ~S_SYNC;
196 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
197 inode->i_flags |= S_NOATIME;
199 inode->i_flags &= ~S_NOATIME;
200 xfs_iflags_clear(ip, XFS_IMODIFIED);
204 xfs_initialize_vnode(
207 struct xfs_inode *ip,
210 struct inode *inode = vn_to_inode(vp);
214 inode->i_private = ip;
218 * We need to set the ops vectors, and unlock the inode, but if
219 * we have been called during the new inode create process, it is
220 * too early to fill in the Linux inode. We will get called a
221 * second time once the inode is properly set up, and then we can
224 if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
225 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
226 xfs_set_inodeops(inode);
228 xfs_iflags_clear(ip, XFS_INEW);
231 unlock_new_inode(inode);
239 struct block_device **bdevp)
243 *bdevp = open_bdev_excl(name, 0, mp);
244 if (IS_ERR(*bdevp)) {
245 error = PTR_ERR(*bdevp);
246 printk("XFS: Invalid device [%s], error=%d\n", name, error);
254 struct block_device *bdev)
257 close_bdev_excl(bdev);
261 * Try to write out the superblock using barriers.
267 xfs_buf_t *sbp = xfs_getsb(mp, 0);
272 XFS_BUF_UNDELAYWRITE(sbp);
274 XFS_BUF_UNASYNC(sbp);
275 XFS_BUF_ORDERED(sbp);
278 error = xfs_iowait(sbp);
281 * Clear all the flags we set and possible error state in the
282 * buffer. We only did the write to try out whether barriers
283 * worked and shouldn't leave any traces in the superblock
287 XFS_BUF_ERROR(sbp, 0);
288 XFS_BUF_UNORDERED(sbp);
295 xfs_mountfs_check_barriers(xfs_mount_t *mp)
299 if (mp->m_logdev_targp != mp->m_ddev_targp) {
300 xfs_fs_cmn_err(CE_NOTE, mp,
301 "Disabling barriers, not supported with external log device");
302 mp->m_flags &= ~XFS_MOUNT_BARRIER;
306 if (xfs_readonly_buftarg(mp->m_ddev_targp)) {
307 xfs_fs_cmn_err(CE_NOTE, mp,
308 "Disabling barriers, underlying device is readonly");
309 mp->m_flags &= ~XFS_MOUNT_BARRIER;
313 error = xfs_barrier_test(mp);
315 xfs_fs_cmn_err(CE_NOTE, mp,
316 "Disabling barriers, trial barrier write failed");
317 mp->m_flags &= ~XFS_MOUNT_BARRIER;
323 xfs_blkdev_issue_flush(
324 xfs_buftarg_t *buftarg)
326 blkdev_issue_flush(buftarg->bt_bdev, NULL);
329 STATIC struct inode *
331 struct super_block *sb)
335 vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP);
338 return vn_to_inode(vp);
342 xfs_fs_destroy_inode(
345 kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode));
349 xfs_fs_inode_init_once(
354 inode_init_once(vn_to_inode((bhv_vnode_t *)vnode));
360 xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode",
361 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
363 xfs_fs_inode_init_once);
367 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
369 goto out_destroy_vnode_zone;
371 xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
374 goto out_free_ioend_zone;
378 kmem_zone_destroy(xfs_ioend_zone);
379 out_destroy_vnode_zone:
380 kmem_zone_destroy(xfs_vnode_zone);
386 xfs_destroy_zones(void)
388 mempool_destroy(xfs_ioend_pool);
389 kmem_zone_destroy(xfs_vnode_zone);
390 kmem_zone_destroy(xfs_ioend_zone);
394 * Attempt to flush the inode, this will actually fail
395 * if the inode is pinned, but we dirty the inode again
396 * at the point when it is unpinned after a log write,
397 * since this is when the inode itself becomes flushable.
404 int error = 0, flags = FLUSH_INODE;
406 vn_trace_entry(XFS_I(inode), __FUNCTION__,
407 (inst_t *)__return_address);
409 filemap_fdatawait(inode->i_mapping);
412 error = xfs_inode_flush(XFS_I(inode), flags);
413 if (error == EAGAIN) {
415 error = xfs_inode_flush(XFS_I(inode),
428 xfs_inode_t *ip = XFS_I(inode);
431 * ip can be null when xfs_iget_core calls xfs_idestroy if we
432 * find an inode with di_mode == 0 but without IGET_CREATE set.
435 vn_trace_entry(ip, __FUNCTION__, (inst_t *)__return_address);
437 XFS_STATS_INC(vn_rele);
438 XFS_STATS_INC(vn_remove);
439 XFS_STATS_INC(vn_reclaim);
440 XFS_STATS_DEC(vn_active);
443 xfs_iflags_clear(ip, XFS_IMODIFIED);
445 panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, inode);
448 ASSERT(XFS_I(inode) == NULL);
452 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
453 * Doing this has two advantages:
454 * - It saves on stack space, which is tight in certain situations
455 * - It can be used (with care) as a mechanism to avoid deadlocks.
456 * Flushing while allocating in a full filesystem requires both.
459 xfs_syncd_queue_work(
462 void (*syncer)(bhv_vfs_t *, void *))
464 struct bhv_vfs_sync_work *work;
466 work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP);
467 INIT_LIST_HEAD(&work->w_list);
468 work->w_syncer = syncer;
471 spin_lock(&vfs->vfs_sync_lock);
472 list_add_tail(&work->w_list, &vfs->vfs_sync_list);
473 spin_unlock(&vfs->vfs_sync_lock);
474 wake_up_process(vfs->vfs_sync_task);
478 * Flush delayed allocate data, attempting to free up reserved space
479 * from existing allocations. At this point a new allocation attempt
480 * has failed with ENOSPC and we are in the process of scratching our
481 * heads, looking about for more room...
484 xfs_flush_inode_work(
488 filemap_flush(((struct inode *)inode)->i_mapping);
489 iput((struct inode *)inode);
496 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
497 struct bhv_vfs *vfs = XFS_MTOVFS(ip->i_mount);
500 xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
501 delay(msecs_to_jiffies(500));
505 * This is the "bigger hammer" version of xfs_flush_inode_work...
506 * (IOW, "If at first you don't succeed, use a Bigger Hammer").
509 xfs_flush_device_work(
513 sync_blockdev(vfs->vfs_super->s_bdev);
514 iput((struct inode *)inode);
521 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
522 struct bhv_vfs *vfs = XFS_MTOVFS(ip->i_mount);
525 xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
526 delay(msecs_to_jiffies(500));
527 xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
537 if (!(vfsp->vfs_flag & VFS_RDONLY))
538 error = bhv_vfs_sync(vfsp, SYNC_FSDATA | SYNC_BDFLUSH | \
539 SYNC_ATTR | SYNC_REFCACHE | SYNC_SUPER,
541 vfsp->vfs_sync_seq++;
542 wake_up(&vfsp->vfs_wait_single_sync_task);
550 bhv_vfs_t *vfsp = (bhv_vfs_t *) arg;
551 bhv_vfs_sync_work_t *work, *n;
555 timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
557 timeleft = schedule_timeout_interruptible(timeleft);
560 if (kthread_should_stop() && list_empty(&vfsp->vfs_sync_list))
563 spin_lock(&vfsp->vfs_sync_lock);
565 * We can get woken by laptop mode, to do a sync -
566 * that's the (only!) case where the list would be
567 * empty with time remaining.
569 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
571 timeleft = xfs_syncd_centisecs *
572 msecs_to_jiffies(10);
573 INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
574 list_add_tail(&vfsp->vfs_sync_work.w_list,
575 &vfsp->vfs_sync_list);
577 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
578 list_move(&work->w_list, &tmp);
579 spin_unlock(&vfsp->vfs_sync_lock);
581 list_for_each_entry_safe(work, n, &tmp, w_list) {
582 (*work->w_syncer)(vfsp, work->w_data);
583 list_del(&work->w_list);
584 if (work == &vfsp->vfs_sync_work)
586 kmem_free(work, sizeof(struct bhv_vfs_sync_work));
597 vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
598 vfsp->vfs_sync_work.w_vfs = vfsp;
599 vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
600 if (IS_ERR(vfsp->vfs_sync_task))
601 return -PTR_ERR(vfsp->vfs_sync_task);
609 kthread_stop(vfsp->vfs_sync_task);
614 struct super_block *sb)
616 bhv_vfs_t *vfsp = vfs_from_sb(sb);
619 xfs_fs_stop_syncd(vfsp);
620 bhv_vfs_sync(vfsp, SYNC_ATTR | SYNC_DELWRI, NULL);
621 error = bhv_vfs_unmount(vfsp, 0, NULL);
623 printk("XFS: unmount got error=%d\n", error);
624 printk("%s: vfs=0x%p left dangling!\n", __FUNCTION__, vfsp);
626 vfs_deallocate(vfsp);
632 struct super_block *sb)
634 if (!(sb->s_flags & MS_RDONLY))
635 bhv_vfs_sync(vfs_from_sb(sb), SYNC_FSDATA, NULL);
641 struct super_block *sb,
644 bhv_vfs_t *vfsp = vfs_from_sb(sb);
648 if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
650 * First stage of freeze - no more writers will make progress
651 * now we are here, so we flush delwri and delalloc buffers
652 * here, then wait for all I/O to complete. Data is frozen at
653 * that point. Metadata is not frozen, transactions can still
654 * occur here so don't bother flushing the buftarg (i.e
655 * SYNC_QUIESCE) because it'll just get dirty again.
657 flags = SYNC_DATA_QUIESCE;
659 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
661 error = bhv_vfs_sync(vfsp, flags, NULL);
664 if (unlikely(laptop_mode)) {
665 int prev_sync_seq = vfsp->vfs_sync_seq;
668 * The disk must be active because we're syncing.
669 * We schedule xfssyncd now (now that the disk is
670 * active) instead of later (when it might not be).
672 wake_up_process(vfsp->vfs_sync_task);
674 * We have to wait for the sync iteration to complete.
675 * If we don't, the disk activity caused by the sync
676 * will come after the sync is completed, and that
677 * triggers another sync from laptop mode.
679 wait_event(vfsp->vfs_wait_single_sync_task,
680 vfsp->vfs_sync_seq != prev_sync_seq);
688 struct dentry *dentry,
689 struct kstatfs *statp)
691 return -bhv_vfs_statvfs(vfs_from_sb(dentry->d_sb), statp,
692 vn_from_inode(dentry->d_inode));
697 struct super_block *sb,
701 bhv_vfs_t *vfsp = vfs_from_sb(sb);
702 struct xfs_mount_args *args = xfs_args_allocate(sb, 0);
705 error = bhv_vfs_parseargs(vfsp, options, args, 1);
707 error = bhv_vfs_mntupdate(vfsp, flags, args);
708 kmem_free(args, sizeof(*args));
714 struct super_block *sb)
716 bhv_vfs_freeze(vfs_from_sb(sb));
722 struct vfsmount *mnt)
724 return -bhv_vfs_showargs(vfs_from_sb(mnt->mnt_sb), m);
729 struct super_block *sb,
732 return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XQUOTASYNC, 0, NULL);
737 struct super_block *sb,
738 struct fs_quota_stat *fqs)
740 return -bhv_vfs_quotactl(vfs_from_sb(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
745 struct super_block *sb,
749 return -bhv_vfs_quotactl(vfs_from_sb(sb), op, 0, (caddr_t)&flags);
754 struct super_block *sb,
757 struct fs_disk_quota *fdq)
759 return -bhv_vfs_quotactl(vfs_from_sb(sb),
760 (type == USRQUOTA) ? Q_XGETQUOTA :
761 ((type == GRPQUOTA) ? Q_XGETGQUOTA :
762 Q_XGETPQUOTA), id, (caddr_t)fdq);
767 struct super_block *sb,
770 struct fs_disk_quota *fdq)
772 return -bhv_vfs_quotactl(vfs_from_sb(sb),
773 (type == USRQUOTA) ? Q_XSETQLIM :
774 ((type == GRPQUOTA) ? Q_XSETGQLIM :
775 Q_XSETPQLIM), id, (caddr_t)fdq);
780 struct super_block *sb,
784 struct bhv_vnode *rootvp;
785 struct bhv_vfs *vfsp = vfs_allocate(sb);
786 struct xfs_mount_args *args = xfs_args_allocate(sb, silent);
787 struct kstatfs statvfs;
790 bhv_insert_all_vfsops(vfsp);
792 error = bhv_vfs_parseargs(vfsp, (char *)data, args, 0);
794 bhv_remove_all_vfsops(vfsp, 1);
798 sb_min_blocksize(sb, BBSIZE);
799 sb->s_export_op = &xfs_export_operations;
800 sb->s_qcop = &xfs_quotactl_operations;
801 sb->s_op = &xfs_super_operations;
803 error = bhv_vfs_mount(vfsp, args, NULL);
805 bhv_remove_all_vfsops(vfsp, 1);
809 error = bhv_vfs_statvfs(vfsp, &statvfs, NULL);
814 sb->s_magic = statvfs.f_type;
815 sb->s_blocksize = statvfs.f_bsize;
816 sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
817 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
819 set_posix_acl_flag(sb);
821 error = bhv_vfs_root(vfsp, &rootvp);
825 sb->s_root = d_alloc_root(vn_to_inode(rootvp));
830 if (is_bad_inode(sb->s_root->d_inode)) {
834 if ((error = xfs_fs_start_syncd(vfsp)))
836 vn_trace_exit(XFS_I(sb->s_root->d_inode), __FUNCTION__,
837 (inst_t *)__return_address);
839 kmem_free(args, sizeof(*args));
851 bhv_vfs_unmount(vfsp, 0, NULL);
854 vfs_deallocate(vfsp);
855 kmem_free(args, sizeof(*args));
861 struct file_system_type *fs_type,
863 const char *dev_name,
865 struct vfsmount *mnt)
867 return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
871 static struct super_operations xfs_super_operations = {
872 .alloc_inode = xfs_fs_alloc_inode,
873 .destroy_inode = xfs_fs_destroy_inode,
874 .write_inode = xfs_fs_write_inode,
875 .clear_inode = xfs_fs_clear_inode,
876 .put_super = xfs_fs_put_super,
877 .write_super = xfs_fs_write_super,
878 .sync_fs = xfs_fs_sync_super,
879 .write_super_lockfs = xfs_fs_lockfs,
880 .statfs = xfs_fs_statfs,
881 .remount_fs = xfs_fs_remount,
882 .show_options = xfs_fs_show_options,
885 static struct quotactl_ops xfs_quotactl_operations = {
886 .quota_sync = xfs_fs_quotasync,
887 .get_xstate = xfs_fs_getxstate,
888 .set_xstate = xfs_fs_setxstate,
889 .get_xquota = xfs_fs_getxquota,
890 .set_xquota = xfs_fs_setxquota,
893 static struct file_system_type xfs_fs_type = {
894 .owner = THIS_MODULE,
896 .get_sb = xfs_fs_get_sb,
897 .kill_sb = kill_block_super,
898 .fs_flags = FS_REQUIRES_DEV,
906 static char message[] __initdata = KERN_INFO \
907 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
913 error = xfs_init_zones();
917 error = xfs_buf_init();
926 error = register_filesystem(&xfs_fs_type);
945 unregister_filesystem(&xfs_fs_type);
952 module_init(init_xfs_fs);
953 module_exit(exit_xfs_fs);
955 MODULE_AUTHOR("Silicon Graphics, Inc.");
956 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
957 MODULE_LICENSE("GPL");