2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
38 #include "xfs_trans.h"
42 #include "xfs_alloc.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_quota.h"
45 #include "xfs_mount.h"
46 #include "xfs_alloc_btree.h"
47 #include "xfs_bmap_btree.h"
48 #include "xfs_ialloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_ialloc.h"
51 #include "xfs_attr_sf.h"
52 #include "xfs_dir_sf.h"
53 #include "xfs_dir2_sf.h"
54 #include "xfs_dinode.h"
55 #include "xfs_inode.h"
58 #include "xfs_rtalloc.h"
59 #include "xfs_error.h"
60 #include "xfs_itable.h"
66 #include "xfs_buf_item.h"
67 #include "xfs_utils.h"
68 #include "xfs_version.h"
69 #include "xfs_ioctl32.h"
71 #include <linux/namei.h>
72 #include <linux/init.h>
73 #include <linux/mount.h>
74 #include <linux/writeback.h>
76 STATIC struct quotactl_ops linvfs_qops;
77 STATIC struct super_operations linvfs_sops;
78 STATIC kmem_zone_t *linvfs_inode_zone;
79 STATIC kmem_shaker_t xfs_inode_shaker;
81 STATIC struct xfs_mount_args *
83 struct super_block *sb)
85 struct xfs_mount_args *args;
87 args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
88 args->logbufs = args->logbufsize = -1;
89 strncpy(args->fsname, sb->s_id, MAXNAMELEN);
91 /* Copy the already-parsed mount(2) flags we're interested in */
92 if (sb->s_flags & MS_NOATIME)
93 args->flags |= XFSMNT_NOATIME;
95 /* Default to 32 bit inodes on Linux all the time */
96 args->flags |= XFSMNT_32BITINODES;
103 unsigned int blockshift)
105 unsigned int pagefactor = 1;
106 unsigned int bitshift = BITS_PER_LONG - 1;
108 /* Figure out maximum filesize, on Linux this can depend on
109 * the filesystem blocksize (on 32 bit platforms).
110 * __block_prepare_write does this in an [unsigned] long...
111 * page->index << (PAGE_CACHE_SHIFT - bbits)
112 * So, for page sized blocks (4K on 32 bit platforms),
113 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
114 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
115 * but for smaller blocksizes it is less (bbits = log2 bsize).
116 * Note1: get_block_t takes a long (implicit cast from above)
117 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
118 * can optionally convert the [unsigned] long from above into
119 * an [unsigned] long long.
122 #if BITS_PER_LONG == 32
123 # if defined(CONFIG_LBD)
124 ASSERT(sizeof(sector_t) == 8);
125 pagefactor = PAGE_CACHE_SIZE;
126 bitshift = BITS_PER_LONG;
128 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
132 return (((__uint64_t)pagefactor) << bitshift) - 1;
135 STATIC __inline__ void
139 vnode_t *vp = LINVFS_GET_VP(inode);
141 if (vp->v_type == VNON) {
143 } else if (S_ISREG(inode->i_mode)) {
144 inode->i_op = &linvfs_file_inode_operations;
145 inode->i_fop = &linvfs_file_operations;
146 inode->i_mapping->a_ops = &linvfs_aops;
147 } else if (S_ISDIR(inode->i_mode)) {
148 inode->i_op = &linvfs_dir_inode_operations;
149 inode->i_fop = &linvfs_dir_operations;
150 } else if (S_ISLNK(inode->i_mode)) {
151 inode->i_op = &linvfs_symlink_inode_operations;
153 inode->i_mapping->a_ops = &linvfs_aops;
155 inode->i_op = &linvfs_file_inode_operations;
156 init_special_inode(inode, inode->i_mode, inode->i_rdev);
160 STATIC __inline__ void
161 xfs_revalidate_inode(
166 struct inode *inode = LINVFS_GET_IP(vp);
168 inode->i_mode = (ip->i_d.di_mode & MODEMASK) | VTTOIF(vp->v_type);
169 inode->i_nlink = ip->i_d.di_nlink;
170 inode->i_uid = ip->i_d.di_uid;
171 inode->i_gid = ip->i_d.di_gid;
172 if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
175 xfs_dev_t dev = ip->i_df.if_u2.if_rdev;
176 inode->i_rdev = MKDEV(sysv_major(dev) & 0x1ff, sysv_minor(dev));
178 inode->i_blksize = PAGE_CACHE_SIZE;
179 inode->i_generation = ip->i_d.di_gen;
180 i_size_write(inode, ip->i_d.di_size);
182 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
183 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
184 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
185 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
186 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
187 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
188 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
189 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
190 inode->i_flags |= S_IMMUTABLE;
192 inode->i_flags &= ~S_IMMUTABLE;
193 if (ip->i_d.di_flags & XFS_DIFLAG_IUNLINK)
194 inode->i_flags |= S_IUNLINK;
196 inode->i_flags &= ~S_IUNLINK;
197 if (ip->i_d.di_flags & XFS_DIFLAG_BARRIER)
198 inode->i_flags |= S_BARRIER;
200 inode->i_flags &= ~S_BARRIER;
201 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
202 inode->i_flags |= S_APPEND;
204 inode->i_flags &= ~S_APPEND;
205 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
206 inode->i_flags |= S_SYNC;
208 inode->i_flags &= ~S_SYNC;
209 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
210 inode->i_flags |= S_NOATIME;
212 inode->i_flags &= ~S_NOATIME;
213 vp->v_flag &= ~VMODIFIED;
217 xfs_initialize_vnode(
220 bhv_desc_t *inode_bhv,
223 xfs_inode_t *ip = XFS_BHVTOI(inode_bhv);
224 struct inode *inode = LINVFS_GET_IP(vp);
226 if (!inode_bhv->bd_vobj) {
227 vp->v_vfsp = bhvtovfs(bdp);
228 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
229 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
233 * We need to set the ops vectors, and unlock the inode, but if
234 * we have been called during the new inode create process, it is
235 * too early to fill in the Linux inode. We will get called a
236 * second time once the inode is properly set up, and then we can
239 if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
240 vp->v_type = IFTOVT(ip->i_d.di_mode);
241 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
242 xfs_set_inodeops(inode);
244 ip->i_flags &= ~XFS_INEW;
247 unlock_new_inode(inode);
255 struct block_device **bdevp)
259 *bdevp = open_bdev_excl(name, 0, mp);
260 if (IS_ERR(*bdevp)) {
261 error = PTR_ERR(*bdevp);
262 printk("XFS: Invalid device [%s], error=%d\n", name, error);
270 struct block_device *bdev)
273 close_bdev_excl(bdev);
277 STATIC struct inode *
279 struct super_block *sb)
283 vp = (vnode_t *)kmem_cache_alloc(linvfs_inode_zone,
284 kmem_flags_convert(KM_SLEEP));
287 return LINVFS_GET_IP(vp);
291 linvfs_destroy_inode(
294 kmem_cache_free(linvfs_inode_zone, LINVFS_GET_VP(inode));
300 unsigned int gfp_mask)
304 pages = kmem_zone_shrink(linvfs_inode_zone);
305 pages += kmem_zone_shrink(xfs_inode_zone);
312 kmem_cache_t *cachep,
315 vnode_t *vp = (vnode_t *)data;
317 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
318 SLAB_CTOR_CONSTRUCTOR)
319 inode_init_once(LINVFS_GET_IP(vp));
323 init_inodecache( void )
325 linvfs_inode_zone = kmem_cache_create("linvfs_icache",
326 sizeof(vnode_t), 0, SLAB_RECLAIM_ACCOUNT,
328 if (linvfs_inode_zone == NULL)
334 destroy_inodecache( void )
336 if (kmem_cache_destroy(linvfs_inode_zone))
337 printk(KERN_WARNING "%s: cache still in use!\n", __FUNCTION__);
341 * Attempt to flush the inode, this will actually fail
342 * if the inode is pinned, but we dirty the inode again
343 * at the point when it is unpinned after a log write,
344 * since this is when the inode itself becomes flushable.
351 vnode_t *vp = LINVFS_GET_VP(inode);
352 int error = 0, flags = FLUSH_INODE;
355 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
358 VOP_IFLUSH(vp, flags, error);
359 if (error == EAGAIN) {
361 VOP_IFLUSH(vp, flags | FLUSH_LOG, error);
374 vnode_t *vp = LINVFS_GET_VP(inode);
378 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
380 * Do all our cleanup, and remove this vnode.
388 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
389 * Doing this has two advantages:
390 * - It saves on stack space, which is tight in certain situations
391 * - It can be used (with care) as a mechanism to avoid deadlocks.
392 * Flushing while allocating in a full filesystem requires both.
395 xfs_syncd_queue_work(
398 void (*syncer)(vfs_t *, void *))
400 vfs_sync_work_t *work;
402 work = kmem_alloc(sizeof(struct vfs_sync_work), KM_SLEEP);
403 INIT_LIST_HEAD(&work->w_list);
404 work->w_syncer = syncer;
407 spin_lock(&vfs->vfs_sync_lock);
408 list_add_tail(&work->w_list, &vfs->vfs_sync_list);
409 spin_unlock(&vfs->vfs_sync_lock);
410 wake_up_process(vfs->vfs_sync_task);
414 * Flush delayed allocate data, attempting to free up reserved space
415 * from existing allocations. At this point a new allocation attempt
416 * has failed with ENOSPC and we are in the process of scratching our
417 * heads, looking about for more room...
420 xfs_flush_inode_work(
424 filemap_flush(((struct inode *)inode)->i_mapping);
425 iput((struct inode *)inode);
432 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
433 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
436 xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
441 * This is the "bigger hammer" version of xfs_flush_inode_work...
442 * (IOW, "If at first you don't succeed, use a Bigger Hammer").
445 xfs_flush_device_work(
449 sync_blockdev(vfs->vfs_super->s_bdev);
450 iput((struct inode *)inode);
457 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
458 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
461 xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
463 xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
466 #define SYNCD_FLAGS (SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
474 if (!(vfsp->vfs_flag & VFS_RDONLY))
475 VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
476 vfsp->vfs_sync_seq++;
478 wake_up(&vfsp->vfs_wait_single_sync_task);
486 vfs_t *vfsp = (vfs_t *) arg;
487 struct list_head tmp;
488 struct vfs_sync_work *work, *n;
490 daemonize("xfssyncd");
492 vfsp->vfs_sync_work.w_vfs = vfsp;
493 vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
494 vfsp->vfs_sync_task = current;
496 wake_up(&vfsp->vfs_wait_sync_task);
498 INIT_LIST_HEAD(&tmp);
499 timeleft = (xfs_syncd_centisecs * HZ) / 100;
501 set_current_state(TASK_INTERRUPTIBLE);
502 timeleft = schedule_timeout(timeleft);
504 try_to_freeze(PF_FREEZE);
505 if (vfsp->vfs_flag & VFS_UMOUNT)
508 spin_lock(&vfsp->vfs_sync_lock);
510 * We can get woken by laptop mode, to do a sync -
511 * that's the (only!) case where the list would be
512 * empty with time remaining.
514 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
516 timeleft = (xfs_syncd_centisecs * HZ) / 100;
517 INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
518 list_add_tail(&vfsp->vfs_sync_work.w_list,
519 &vfsp->vfs_sync_list);
521 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
522 list_move(&work->w_list, &tmp);
523 spin_unlock(&vfsp->vfs_sync_lock);
525 list_for_each_entry_safe(work, n, &tmp, w_list) {
526 (*work->w_syncer)(vfsp, work->w_data);
527 list_del(&work->w_list);
528 if (work == &vfsp->vfs_sync_work)
530 kmem_free(work, sizeof(struct vfs_sync_work));
534 vfsp->vfs_sync_task = NULL;
536 wake_up(&vfsp->vfs_wait_sync_task);
547 pid = kernel_thread(xfssyncd, (void *) vfsp,
548 CLONE_VM | CLONE_FS | CLONE_FILES);
551 wait_event(vfsp->vfs_wait_sync_task, vfsp->vfs_sync_task);
559 vfsp->vfs_flag |= VFS_UMOUNT;
562 wake_up_process(vfsp->vfs_sync_task);
563 wait_event(vfsp->vfs_wait_sync_task, !vfsp->vfs_sync_task);
568 struct super_block *sb)
570 vfs_t *vfsp = LINVFS_GET_VFS(sb);
573 linvfs_stop_syncd(vfsp);
574 VFS_SYNC(vfsp, SYNC_ATTR|SYNC_DELWRI, NULL, error);
576 VFS_UNMOUNT(vfsp, 0, NULL, error);
578 printk("XFS unmount got error %d\n", error);
579 printk("%s: vfsp/0x%p left dangling!\n", __FUNCTION__, vfsp);
583 vfs_deallocate(vfsp);
588 struct super_block *sb)
590 vfs_t *vfsp = LINVFS_GET_VFS(sb);
593 if (sb->s_flags & MS_RDONLY) {
594 sb->s_dirt = 0; /* paranoia */
597 /* Push the log and superblock a little */
598 VFS_SYNC(vfsp, SYNC_FSDATA, NULL, error);
604 struct super_block *sb,
607 vfs_t *vfsp = LINVFS_GET_VFS(sb);
609 int flags = SYNC_FSDATA;
614 VFS_SYNC(vfsp, flags, NULL, error);
617 if (unlikely(laptop_mode)) {
618 int prev_sync_seq = vfsp->vfs_sync_seq;
621 * The disk must be active because we're syncing.
622 * We schedule xfssyncd now (now that the disk is
623 * active) instead of later (when it might not be).
625 wake_up_process(vfsp->vfs_sync_task);
627 * We have to wait for the sync iteration to complete.
628 * If we don't, the disk activity caused by the sync
629 * will come after the sync is completed, and that
630 * triggers another sync from laptop mode.
632 wait_event(vfsp->vfs_wait_single_sync_task,
633 vfsp->vfs_sync_seq != prev_sync_seq);
641 struct super_block *sb,
642 struct kstatfs *statp)
644 vfs_t *vfsp = LINVFS_GET_VFS(sb);
647 VFS_STATVFS(vfsp, statp, NULL, error);
653 struct super_block *sb,
657 vfs_t *vfsp = LINVFS_GET_VFS(sb);
658 struct xfs_mount_args *args = xfs_args_allocate(sb);
661 VFS_PARSEARGS(vfsp, options, args, 1, error);
663 VFS_MNTUPDATE(vfsp, flags, args, error);
664 kmem_free(args, sizeof(*args));
670 struct super_block *sb)
672 VFS_FREEZE(LINVFS_GET_VFS(sb));
678 struct vfsmount *mnt)
680 struct vfs *vfsp = LINVFS_GET_VFS(mnt->mnt_sb);
683 VFS_SHOWARGS(vfsp, m, error);
689 struct super_block *sb,
690 struct fs_quota_stat *fqs)
692 struct vfs *vfsp = LINVFS_GET_VFS(sb);
695 VFS_QUOTACTL(vfsp, Q_XGETQSTAT, 0, (caddr_t)fqs, error);
701 struct super_block *sb,
705 struct vfs *vfsp = LINVFS_GET_VFS(sb);
708 VFS_QUOTACTL(vfsp, op, 0, (caddr_t)&flags, error);
714 struct super_block *sb,
717 struct fs_disk_quota *fdq)
719 struct vfs *vfsp = LINVFS_GET_VFS(sb);
722 getmode = (type == GRPQUOTA) ? Q_XGETGQUOTA : Q_XGETQUOTA;
723 VFS_QUOTACTL(vfsp, getmode, id, (caddr_t)fdq, error);
729 struct super_block *sb,
732 struct fs_disk_quota *fdq)
734 struct vfs *vfsp = LINVFS_GET_VFS(sb);
737 setmode = (type == GRPQUOTA) ? Q_XSETGQLIM : Q_XSETQLIM;
738 VFS_QUOTACTL(vfsp, setmode, id, (caddr_t)fdq, error);
744 struct super_block *sb,
749 struct vfs *vfsp = vfs_allocate();
750 struct xfs_mount_args *args = xfs_args_allocate(sb);
751 struct kstatfs statvfs;
754 vfsp->vfs_super = sb;
755 LINVFS_SET_VFS(sb, vfsp);
756 if (sb->s_flags & MS_RDONLY)
757 vfsp->vfs_flag |= VFS_RDONLY;
758 bhv_insert_all_vfsops(vfsp);
760 VFS_PARSEARGS(vfsp, (char *)data, args, 0, error);
762 bhv_remove_all_vfsops(vfsp, 1);
766 sb_min_blocksize(sb, BBSIZE);
767 #ifdef CONFIG_XFS_EXPORT
768 sb->s_export_op = &linvfs_export_ops;
770 sb->s_qcop = &linvfs_qops;
771 sb->s_op = &linvfs_sops;
773 VFS_MOUNT(vfsp, args, NULL, error);
775 bhv_remove_all_vfsops(vfsp, 1);
779 VFS_STATVFS(vfsp, &statvfs, NULL, error);
784 sb->s_magic = statvfs.f_type;
785 sb->s_blocksize = statvfs.f_bsize;
786 sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
787 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
789 set_posix_acl_flag(sb);
791 VFS_ROOT(vfsp, &rootvp, error);
795 sb->s_root = d_alloc_root(LINVFS_GET_IP(rootvp));
800 if (is_bad_inode(sb->s_root->d_inode)) {
804 if ((error = linvfs_start_syncd(vfsp)))
806 vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
808 kmem_free(args, sizeof(*args));
820 VFS_UNMOUNT(vfsp, 0, NULL, error2);
823 vfs_deallocate(vfsp);
824 kmem_free(args, sizeof(*args));
828 STATIC struct super_block *
830 struct file_system_type *fs_type,
832 const char *dev_name,
835 return get_sb_bdev(fs_type, flags, dev_name, data, linvfs_fill_super);
838 STATIC struct super_operations linvfs_sops = {
839 .alloc_inode = linvfs_alloc_inode,
840 .destroy_inode = linvfs_destroy_inode,
841 .write_inode = linvfs_write_inode,
842 .clear_inode = linvfs_clear_inode,
843 .put_super = linvfs_put_super,
844 .write_super = linvfs_write_super,
845 .sync_fs = linvfs_sync_super,
846 .write_super_lockfs = linvfs_freeze_fs,
847 .statfs = linvfs_statfs,
848 .remount_fs = linvfs_remount,
849 .show_options = linvfs_show_options,
852 STATIC struct quotactl_ops linvfs_qops = {
853 .get_xstate = linvfs_getxstate,
854 .set_xstate = linvfs_setxstate,
855 .get_xquota = linvfs_getxquota,
856 .set_xquota = linvfs_setxquota,
859 STATIC struct file_system_type xfs_fs_type = {
860 .owner = THIS_MODULE,
862 .get_sb = linvfs_get_sb,
863 .kill_sb = kill_block_super,
864 .fs_flags = FS_REQUIRES_DEV,
873 static char message[] __initdata = KERN_INFO \
874 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
879 xfs_physmem = si.totalram;
883 error = init_inodecache();
885 goto undo_inodecache;
887 error = pagebuf_init();
896 xfs_inode_shaker = kmem_shake_register(xfs_inode_shake);
897 if (!xfs_inode_shaker) {
902 error = register_filesystem(&xfs_fs_type);
905 XFS_DM_INIT(&xfs_fs_type);
909 kmem_shake_deregister(xfs_inode_shaker);
915 destroy_inodecache();
925 XFS_DM_EXIT(&xfs_fs_type);
926 unregister_filesystem(&xfs_fs_type);
927 kmem_shake_deregister(xfs_inode_shaker);
930 destroy_inodecache();
934 module_init(init_xfs_fs);
935 module_exit(exit_xfs_fs);
937 MODULE_AUTHOR("Silicon Graphics, Inc.");
938 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
939 MODULE_LICENSE("GPL");