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_vfsops.h"
51 #include "xfs_version.h"
53 #include <linux/namei.h>
54 #include <linux/init.h>
55 #include <linux/mount.h>
56 #include <linux/mempool.h>
57 #include <linux/writeback.h>
58 #include <linux/kthread.h>
59 #include <linux/freezer.h>
61 static struct quotactl_ops xfs_quotactl_operations;
62 static struct super_operations xfs_super_operations;
63 static kmem_zone_t *xfs_vnode_zone;
64 static kmem_zone_t *xfs_ioend_zone;
65 mempool_t *xfs_ioend_pool;
67 STATIC struct xfs_mount_args *
69 struct super_block *sb,
72 struct xfs_mount_args *args;
74 args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
75 args->logbufs = args->logbufsize = -1;
76 strncpy(args->fsname, sb->s_id, MAXNAMELEN);
78 /* Copy the already-parsed mount(2) flags we're interested in */
79 if (sb->s_flags & MS_DIRSYNC)
80 args->flags |= XFSMNT_DIRSYNC;
81 if (sb->s_flags & MS_SYNCHRONOUS)
82 args->flags |= XFSMNT_WSYNC;
84 args->flags |= XFSMNT_QUIET;
85 args->flags |= XFSMNT_32BITINODES;
92 unsigned int blockshift)
94 unsigned int pagefactor = 1;
95 unsigned int bitshift = BITS_PER_LONG - 1;
97 /* Figure out maximum filesize, on Linux this can depend on
98 * the filesystem blocksize (on 32 bit platforms).
99 * __block_prepare_write does this in an [unsigned] long...
100 * page->index << (PAGE_CACHE_SHIFT - bbits)
101 * So, for page sized blocks (4K on 32 bit platforms),
102 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
103 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
104 * but for smaller blocksizes it is less (bbits = log2 bsize).
105 * Note1: get_block_t takes a long (implicit cast from above)
106 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
107 * can optionally convert the [unsigned] long from above into
108 * an [unsigned] long long.
111 #if BITS_PER_LONG == 32
112 # if defined(CONFIG_LBD)
113 ASSERT(sizeof(sector_t) == 8);
114 pagefactor = PAGE_CACHE_SIZE;
115 bitshift = BITS_PER_LONG;
117 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
121 return (((__uint64_t)pagefactor) << bitshift) - 1;
128 switch (inode->i_mode & S_IFMT) {
130 inode->i_op = &xfs_inode_operations;
131 inode->i_fop = &xfs_file_operations;
132 inode->i_mapping->a_ops = &xfs_address_space_operations;
135 inode->i_op = &xfs_dir_inode_operations;
136 inode->i_fop = &xfs_dir_file_operations;
139 inode->i_op = &xfs_symlink_inode_operations;
141 inode->i_mapping->a_ops = &xfs_address_space_operations;
144 inode->i_op = &xfs_inode_operations;
145 init_special_inode(inode, inode->i_mode, inode->i_rdev);
151 xfs_revalidate_inode(
156 struct inode *inode = vn_to_inode(vp);
158 inode->i_mode = ip->i_d.di_mode;
159 inode->i_nlink = ip->i_d.di_nlink;
160 inode->i_uid = ip->i_d.di_uid;
161 inode->i_gid = ip->i_d.di_gid;
163 switch (inode->i_mode & S_IFMT) {
167 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
168 sysv_minor(ip->i_df.if_u2.if_rdev));
175 inode->i_generation = ip->i_d.di_gen;
176 i_size_write(inode, ip->i_d.di_size);
178 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
179 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
180 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
181 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
182 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
183 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
184 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
185 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
186 inode->i_flags |= S_IMMUTABLE;
188 inode->i_flags &= ~S_IMMUTABLE;
189 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
190 inode->i_flags |= S_APPEND;
192 inode->i_flags &= ~S_APPEND;
193 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
194 inode->i_flags |= S_SYNC;
196 inode->i_flags &= ~S_SYNC;
197 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
198 inode->i_flags |= S_NOATIME;
200 inode->i_flags &= ~S_NOATIME;
201 xfs_iflags_clear(ip, XFS_IMODIFIED);
205 xfs_initialize_vnode(
206 struct xfs_mount *mp,
208 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 && (inode->i_state & I_NEW)) {
225 xfs_revalidate_inode(mp, 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(
460 struct xfs_mount *mp,
462 void (*syncer)(struct xfs_mount *, 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(&mp->m_sync_lock);
472 list_add_tail(&work->w_list, &mp->m_sync_list);
473 spin_unlock(&mp->m_sync_lock);
474 wake_up_process(mp->m_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(
485 struct xfs_mount *mp,
488 struct inode *inode = arg;
489 filemap_flush(inode->i_mapping);
497 struct inode *inode = ip->i_vnode;
500 xfs_syncd_queue_work(ip->i_mount, 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(
510 struct xfs_mount *mp,
513 struct inode *inode = arg;
514 sync_blockdev(mp->m_super->s_bdev);
522 struct inode *inode = vn_to_inode(XFS_ITOV(ip));
525 xfs_syncd_queue_work(ip->i_mount, 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);
532 struct xfs_mount *mp,
537 if (!(mp->m_flags & XFS_MOUNT_RDONLY))
538 error = xfs_sync(mp, SYNC_FSDATA | SYNC_BDFLUSH | SYNC_ATTR |
539 SYNC_REFCACHE | SYNC_SUPER);
541 wake_up(&mp->m_wait_single_sync_task);
548 struct xfs_mount *mp = arg;
550 bhv_vfs_sync_work_t *work, *n;
554 timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10);
556 timeleft = schedule_timeout_interruptible(timeleft);
559 if (kthread_should_stop() && list_empty(&mp->m_sync_list))
562 spin_lock(&mp->m_sync_lock);
564 * We can get woken by laptop mode, to do a sync -
565 * that's the (only!) case where the list would be
566 * empty with time remaining.
568 if (!timeleft || list_empty(&mp->m_sync_list)) {
570 timeleft = xfs_syncd_centisecs *
571 msecs_to_jiffies(10);
572 INIT_LIST_HEAD(&mp->m_sync_work.w_list);
573 list_add_tail(&mp->m_sync_work.w_list,
576 list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list)
577 list_move(&work->w_list, &tmp);
578 spin_unlock(&mp->m_sync_lock);
580 list_for_each_entry_safe(work, n, &tmp, w_list) {
581 (*work->w_syncer)(mp, work->w_data);
582 list_del(&work->w_list);
583 if (work == &mp->m_sync_work)
585 kmem_free(work, sizeof(struct bhv_vfs_sync_work));
594 struct super_block *sb)
596 struct xfs_mount *mp = XFS_M(sb);
599 kthread_stop(mp->m_sync_task);
601 xfs_sync(mp, SYNC_ATTR | SYNC_DELWRI);
602 error = xfs_unmount(mp, 0, NULL);
604 printk("XFS: unmount got error=%d\n", error);
609 struct super_block *sb)
611 if (!(sb->s_flags & MS_RDONLY))
612 xfs_sync(XFS_M(sb), SYNC_FSDATA);
618 struct super_block *sb,
621 struct xfs_mount *mp = XFS_M(sb);
625 if (unlikely(sb->s_frozen == SB_FREEZE_WRITE)) {
627 * First stage of freeze - no more writers will make progress
628 * now we are here, so we flush delwri and delalloc buffers
629 * here, then wait for all I/O to complete. Data is frozen at
630 * that point. Metadata is not frozen, transactions can still
631 * occur here so don't bother flushing the buftarg (i.e
632 * SYNC_QUIESCE) because it'll just get dirty again.
634 flags = SYNC_DATA_QUIESCE;
636 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
638 error = xfs_sync(mp, flags);
641 if (unlikely(laptop_mode)) {
642 int prev_sync_seq = mp->m_sync_seq;
645 * The disk must be active because we're syncing.
646 * We schedule xfssyncd now (now that the disk is
647 * active) instead of later (when it might not be).
649 wake_up_process(mp->m_sync_task);
651 * We have to wait for the sync iteration to complete.
652 * If we don't, the disk activity caused by the sync
653 * will come after the sync is completed, and that
654 * triggers another sync from laptop mode.
656 wait_event(mp->m_wait_single_sync_task,
657 mp->m_sync_seq != prev_sync_seq);
665 struct dentry *dentry,
666 struct kstatfs *statp)
668 return -xfs_statvfs(XFS_M(dentry->d_sb), statp,
669 vn_from_inode(dentry->d_inode));
674 struct super_block *sb,
678 struct xfs_mount *mp = XFS_M(sb);
679 struct xfs_mount_args *args = xfs_args_allocate(sb, 0);
682 error = xfs_parseargs(mp, options, args, 1);
684 error = xfs_mntupdate(mp, flags, args);
685 kmem_free(args, sizeof(*args));
691 struct super_block *sb)
693 xfs_freeze(XFS_M(sb));
699 struct vfsmount *mnt)
701 return -xfs_showargs(XFS_M(mnt->mnt_sb), m);
706 struct super_block *sb,
709 return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XQUOTASYNC, 0, NULL);
714 struct super_block *sb,
715 struct fs_quota_stat *fqs)
717 return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XGETQSTAT, 0, (caddr_t)fqs);
722 struct super_block *sb,
726 return -XFS_QM_QUOTACTL(XFS_M(sb), op, 0, (caddr_t)&flags);
731 struct super_block *sb,
734 struct fs_disk_quota *fdq)
736 return -XFS_QM_QUOTACTL(XFS_M(sb),
737 (type == USRQUOTA) ? Q_XGETQUOTA :
738 ((type == GRPQUOTA) ? Q_XGETGQUOTA :
739 Q_XGETPQUOTA), id, (caddr_t)fdq);
744 struct super_block *sb,
747 struct fs_disk_quota *fdq)
749 return -XFS_QM_QUOTACTL(XFS_M(sb),
750 (type == USRQUOTA) ? Q_XSETQLIM :
751 ((type == GRPQUOTA) ? Q_XSETGQLIM :
752 Q_XSETPQLIM), id, (caddr_t)fdq);
757 struct super_block *sb,
761 struct inode *rootvp;
762 struct xfs_mount *mp = NULL;
763 struct xfs_mount_args *args = xfs_args_allocate(sb, silent);
764 struct kstatfs statvfs;
767 mp = xfs_mount_init();
769 INIT_LIST_HEAD(&mp->m_sync_list);
770 spin_lock_init(&mp->m_sync_lock);
771 init_waitqueue_head(&mp->m_wait_single_sync_task);
776 if (sb->s_flags & MS_RDONLY)
777 mp->m_flags |= XFS_MOUNT_RDONLY;
779 error = xfs_parseargs(mp, (char *)data, args, 0);
783 sb_min_blocksize(sb, BBSIZE);
784 sb->s_export_op = &xfs_export_operations;
785 sb->s_qcop = &xfs_quotactl_operations;
786 sb->s_op = &xfs_super_operations;
788 error = xfs_mount(mp, args, NULL);
792 error = xfs_statvfs(mp, &statvfs, NULL);
797 sb->s_magic = statvfs.f_type;
798 sb->s_blocksize = statvfs.f_bsize;
799 sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
800 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
802 set_posix_acl_flag(sb);
804 error = xfs_root(mp, &rootvp);
808 sb->s_root = d_alloc_root(vn_to_inode(rootvp));
813 if (is_bad_inode(sb->s_root->d_inode)) {
818 mp->m_sync_work.w_syncer = xfs_sync_worker;
819 mp->m_sync_work.w_mount = mp;
820 mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd");
821 if (IS_ERR(mp->m_sync_task)) {
822 error = -PTR_ERR(mp->m_sync_task);
826 vn_trace_exit(XFS_I(sb->s_root->d_inode), __FUNCTION__,
827 (inst_t *)__return_address);
829 kmem_free(args, sizeof(*args));
841 xfs_unmount(mp, 0, NULL);
844 kmem_free(args, sizeof(*args));
850 struct file_system_type *fs_type,
852 const char *dev_name,
854 struct vfsmount *mnt)
856 return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super,
860 static struct super_operations xfs_super_operations = {
861 .alloc_inode = xfs_fs_alloc_inode,
862 .destroy_inode = xfs_fs_destroy_inode,
863 .write_inode = xfs_fs_write_inode,
864 .clear_inode = xfs_fs_clear_inode,
865 .put_super = xfs_fs_put_super,
866 .write_super = xfs_fs_write_super,
867 .sync_fs = xfs_fs_sync_super,
868 .write_super_lockfs = xfs_fs_lockfs,
869 .statfs = xfs_fs_statfs,
870 .remount_fs = xfs_fs_remount,
871 .show_options = xfs_fs_show_options,
874 static struct quotactl_ops xfs_quotactl_operations = {
875 .quota_sync = xfs_fs_quotasync,
876 .get_xstate = xfs_fs_getxstate,
877 .set_xstate = xfs_fs_setxstate,
878 .get_xquota = xfs_fs_getxquota,
879 .set_xquota = xfs_fs_setxquota,
882 static struct file_system_type xfs_fs_type = {
883 .owner = THIS_MODULE,
885 .get_sb = xfs_fs_get_sb,
886 .kill_sb = kill_block_super,
887 .fs_flags = FS_REQUIRES_DEV,
895 static char message[] __initdata = KERN_INFO \
896 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
902 error = xfs_init_zones();
906 error = xfs_buf_init();
915 error = register_filesystem(&xfs_fs_type);
934 unregister_filesystem(&xfs_fs_type);
941 module_init(init_xfs_fs);
942 module_exit(exit_xfs_fs);
944 MODULE_AUTHOR("Silicon Graphics, Inc.");
945 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
946 MODULE_LICENSE("GPL");