2 * Copyright (c) 2000-2005 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"
70 #include <linux/namei.h>
71 #include <linux/init.h>
72 #include <linux/mount.h>
73 #include <linux/mempool.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 *xfs_vnode_zone;
79 STATIC kmem_zone_t *xfs_ioend_zone;
80 mempool_t *xfs_ioend_pool;
82 STATIC struct xfs_mount_args *
84 struct super_block *sb)
86 struct xfs_mount_args *args;
88 args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
89 args->logbufs = args->logbufsize = -1;
90 strncpy(args->fsname, sb->s_id, MAXNAMELEN);
92 /* Copy the already-parsed mount(2) flags we're interested in */
93 if (sb->s_flags & MS_NOATIME)
94 args->flags |= XFSMNT_NOATIME;
95 if (sb->s_flags & MS_DIRSYNC)
96 args->flags |= XFSMNT_DIRSYNC;
97 if (sb->s_flags & MS_SYNCHRONOUS)
98 args->flags |= XFSMNT_WSYNC;
100 /* Default to 32 bit inodes on Linux all the time */
101 args->flags |= XFSMNT_32BITINODES;
108 unsigned int blockshift)
110 unsigned int pagefactor = 1;
111 unsigned int bitshift = BITS_PER_LONG - 1;
113 /* Figure out maximum filesize, on Linux this can depend on
114 * the filesystem blocksize (on 32 bit platforms).
115 * __block_prepare_write does this in an [unsigned] long...
116 * page->index << (PAGE_CACHE_SHIFT - bbits)
117 * So, for page sized blocks (4K on 32 bit platforms),
118 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
119 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
120 * but for smaller blocksizes it is less (bbits = log2 bsize).
121 * Note1: get_block_t takes a long (implicit cast from above)
122 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
123 * can optionally convert the [unsigned] long from above into
124 * an [unsigned] long long.
127 #if BITS_PER_LONG == 32
128 # if defined(CONFIG_LBD)
129 ASSERT(sizeof(sector_t) == 8);
130 pagefactor = PAGE_CACHE_SIZE;
131 bitshift = BITS_PER_LONG;
133 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
137 return (((__uint64_t)pagefactor) << bitshift) - 1;
140 STATIC __inline__ void
144 switch (inode->i_mode & S_IFMT) {
146 inode->i_op = &linvfs_file_inode_operations;
147 inode->i_fop = &linvfs_file_operations;
148 inode->i_mapping->a_ops = &linvfs_aops;
151 inode->i_op = &linvfs_dir_inode_operations;
152 inode->i_fop = &linvfs_dir_operations;
155 inode->i_op = &linvfs_symlink_inode_operations;
157 inode->i_mapping->a_ops = &linvfs_aops;
160 inode->i_op = &linvfs_file_inode_operations;
161 init_special_inode(inode, inode->i_mode, inode->i_rdev);
166 STATIC __inline__ void
167 xfs_revalidate_inode(
172 struct inode *inode = LINVFS_GET_IP(vp);
174 inode->i_mode = ip->i_d.di_mode;
175 inode->i_nlink = ip->i_d.di_nlink;
176 inode->i_uid = ip->i_d.di_uid;
177 inode->i_gid = ip->i_d.di_gid;
179 switch (inode->i_mode & S_IFMT) {
183 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
184 sysv_minor(ip->i_df.if_u2.if_rdev));
191 inode->i_blksize = PAGE_CACHE_SIZE;
192 inode->i_generation = ip->i_d.di_gen;
193 i_size_write(inode, ip->i_d.di_size);
195 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
196 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
197 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
198 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
199 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
200 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
201 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
202 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
203 inode->i_flags |= S_IMMUTABLE;
205 inode->i_flags &= ~S_IMMUTABLE;
206 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
207 inode->i_flags |= S_APPEND;
209 inode->i_flags &= ~S_APPEND;
210 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
211 inode->i_flags |= S_SYNC;
213 inode->i_flags &= ~S_SYNC;
214 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
215 inode->i_flags |= S_NOATIME;
217 inode->i_flags &= ~S_NOATIME;
218 vp->v_flag &= ~VMODIFIED;
222 xfs_initialize_vnode(
225 bhv_desc_t *inode_bhv,
228 xfs_inode_t *ip = XFS_BHVTOI(inode_bhv);
229 struct inode *inode = LINVFS_GET_IP(vp);
231 if (!inode_bhv->bd_vobj) {
232 vp->v_vfsp = bhvtovfs(bdp);
233 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
234 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
238 * We need to set the ops vectors, and unlock the inode, but if
239 * we have been called during the new inode create process, it is
240 * too early to fill in the Linux inode. We will get called a
241 * second time once the inode is properly set up, and then we can
244 if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
245 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
246 xfs_set_inodeops(inode);
248 ip->i_flags &= ~XFS_INEW;
251 unlock_new_inode(inode);
259 struct block_device **bdevp)
263 *bdevp = open_bdev_excl(name, 0, mp);
264 if (IS_ERR(*bdevp)) {
265 error = PTR_ERR(*bdevp);
266 printk("XFS: Invalid device [%s], error=%d\n", name, error);
274 struct block_device *bdev)
277 close_bdev_excl(bdev);
281 STATIC struct inode *
283 struct super_block *sb)
287 vp = kmem_cache_alloc(xfs_vnode_zone, kmem_flags_convert(KM_SLEEP));
290 return LINVFS_GET_IP(vp);
294 linvfs_destroy_inode(
297 kmem_zone_free(xfs_vnode_zone, LINVFS_GET_VP(inode));
301 linvfs_inode_init_once(
303 kmem_cache_t *cachep,
306 vnode_t *vp = (vnode_t *)data;
308 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
309 SLAB_CTOR_CONSTRUCTOR)
310 inode_init_once(LINVFS_GET_IP(vp));
314 linvfs_init_zones(void)
316 xfs_vnode_zone = kmem_cache_create("xfs_vnode",
317 sizeof(vnode_t), 0, SLAB_RECLAIM_ACCOUNT,
318 linvfs_inode_init_once, NULL);
322 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
324 goto out_destroy_vnode_zone;
326 xfs_ioend_pool = mempool_create(4 * MAX_BUF_PER_PAGE,
327 mempool_alloc_slab, mempool_free_slab,
330 goto out_free_ioend_zone;
336 kmem_zone_destroy(xfs_ioend_zone);
337 out_destroy_vnode_zone:
338 kmem_zone_destroy(xfs_vnode_zone);
344 linvfs_destroy_zones(void)
346 mempool_destroy(xfs_ioend_pool);
347 kmem_zone_destroy(xfs_vnode_zone);
348 kmem_zone_destroy(xfs_ioend_zone);
352 * Attempt to flush the inode, this will actually fail
353 * if the inode is pinned, but we dirty the inode again
354 * at the point when it is unpinned after a log write,
355 * since this is when the inode itself becomes flushable.
362 vnode_t *vp = LINVFS_GET_VP(inode);
363 int error = 0, flags = FLUSH_INODE;
366 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
369 VOP_IFLUSH(vp, flags, error);
370 if (error == EAGAIN) {
372 VOP_IFLUSH(vp, flags | FLUSH_LOG, error);
385 vnode_t *vp = LINVFS_GET_VP(inode);
389 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
391 * Do all our cleanup, and remove this vnode.
399 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
400 * Doing this has two advantages:
401 * - It saves on stack space, which is tight in certain situations
402 * - It can be used (with care) as a mechanism to avoid deadlocks.
403 * Flushing while allocating in a full filesystem requires both.
406 xfs_syncd_queue_work(
409 void (*syncer)(vfs_t *, void *))
411 vfs_sync_work_t *work;
413 work = kmem_alloc(sizeof(struct vfs_sync_work), KM_SLEEP);
414 INIT_LIST_HEAD(&work->w_list);
415 work->w_syncer = syncer;
418 spin_lock(&vfs->vfs_sync_lock);
419 list_add_tail(&work->w_list, &vfs->vfs_sync_list);
420 spin_unlock(&vfs->vfs_sync_lock);
421 wake_up_process(vfs->vfs_sync_task);
425 * Flush delayed allocate data, attempting to free up reserved space
426 * from existing allocations. At this point a new allocation attempt
427 * has failed with ENOSPC and we are in the process of scratching our
428 * heads, looking about for more room...
431 xfs_flush_inode_work(
435 filemap_flush(((struct inode *)inode)->i_mapping);
436 iput((struct inode *)inode);
443 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
444 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
447 xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
452 * This is the "bigger hammer" version of xfs_flush_inode_work...
453 * (IOW, "If at first you don't succeed, use a Bigger Hammer").
456 xfs_flush_device_work(
460 sync_blockdev(vfs->vfs_super->s_bdev);
461 iput((struct inode *)inode);
468 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
469 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
472 xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
474 xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
477 #define SYNCD_FLAGS (SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
485 if (!(vfsp->vfs_flag & VFS_RDONLY))
486 VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
487 vfsp->vfs_sync_seq++;
489 wake_up(&vfsp->vfs_wait_single_sync_task);
497 vfs_t *vfsp = (vfs_t *) arg;
498 struct list_head tmp;
499 struct vfs_sync_work *work, *n;
501 daemonize("xfssyncd");
503 vfsp->vfs_sync_work.w_vfs = vfsp;
504 vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
505 vfsp->vfs_sync_task = current;
507 wake_up(&vfsp->vfs_wait_sync_task);
509 INIT_LIST_HEAD(&tmp);
510 timeleft = (xfs_syncd_centisecs * HZ) / 100;
512 set_current_state(TASK_INTERRUPTIBLE);
513 timeleft = schedule_timeout(timeleft);
516 if (vfsp->vfs_flag & VFS_UMOUNT)
519 spin_lock(&vfsp->vfs_sync_lock);
521 * We can get woken by laptop mode, to do a sync -
522 * that's the (only!) case where the list would be
523 * empty with time remaining.
525 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
527 timeleft = (xfs_syncd_centisecs * HZ) / 100;
528 INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
529 list_add_tail(&vfsp->vfs_sync_work.w_list,
530 &vfsp->vfs_sync_list);
532 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
533 list_move(&work->w_list, &tmp);
534 spin_unlock(&vfsp->vfs_sync_lock);
536 list_for_each_entry_safe(work, n, &tmp, w_list) {
537 (*work->w_syncer)(vfsp, work->w_data);
538 list_del(&work->w_list);
539 if (work == &vfsp->vfs_sync_work)
541 kmem_free(work, sizeof(struct vfs_sync_work));
545 vfsp->vfs_sync_task = NULL;
547 wake_up(&vfsp->vfs_wait_sync_task);
558 pid = kernel_thread(xfssyncd, (void *) vfsp,
559 CLONE_VM | CLONE_FS | CLONE_FILES);
562 wait_event(vfsp->vfs_wait_sync_task, vfsp->vfs_sync_task);
570 vfsp->vfs_flag |= VFS_UMOUNT;
573 wake_up_process(vfsp->vfs_sync_task);
574 wait_event(vfsp->vfs_wait_sync_task, !vfsp->vfs_sync_task);
579 struct super_block *sb)
581 vfs_t *vfsp = LINVFS_GET_VFS(sb);
584 linvfs_stop_syncd(vfsp);
585 VFS_SYNC(vfsp, SYNC_ATTR|SYNC_DELWRI, NULL, error);
587 VFS_UNMOUNT(vfsp, 0, NULL, error);
589 printk("XFS unmount got error %d\n", error);
590 printk("%s: vfsp/0x%p left dangling!\n", __FUNCTION__, vfsp);
594 vfs_deallocate(vfsp);
599 struct super_block *sb)
601 vfs_t *vfsp = LINVFS_GET_VFS(sb);
604 if (sb->s_flags & MS_RDONLY) {
605 sb->s_dirt = 0; /* paranoia */
608 /* Push the log and superblock a little */
609 VFS_SYNC(vfsp, SYNC_FSDATA, NULL, error);
615 struct super_block *sb,
618 vfs_t *vfsp = LINVFS_GET_VFS(sb);
620 int flags = SYNC_FSDATA;
622 if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
623 flags = SYNC_QUIESCE;
625 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
627 VFS_SYNC(vfsp, flags, NULL, error);
630 if (unlikely(laptop_mode)) {
631 int prev_sync_seq = vfsp->vfs_sync_seq;
634 * The disk must be active because we're syncing.
635 * We schedule xfssyncd now (now that the disk is
636 * active) instead of later (when it might not be).
638 wake_up_process(vfsp->vfs_sync_task);
640 * We have to wait for the sync iteration to complete.
641 * If we don't, the disk activity caused by the sync
642 * will come after the sync is completed, and that
643 * triggers another sync from laptop mode.
645 wait_event(vfsp->vfs_wait_single_sync_task,
646 vfsp->vfs_sync_seq != prev_sync_seq);
654 struct super_block *sb,
655 struct kstatfs *statp)
657 vfs_t *vfsp = LINVFS_GET_VFS(sb);
660 VFS_STATVFS(vfsp, statp, NULL, error);
666 struct super_block *sb,
670 vfs_t *vfsp = LINVFS_GET_VFS(sb);
671 struct xfs_mount_args *args = xfs_args_allocate(sb);
674 VFS_PARSEARGS(vfsp, options, args, 1, error);
676 VFS_MNTUPDATE(vfsp, flags, args, error);
677 kmem_free(args, sizeof(*args));
683 struct super_block *sb)
685 VFS_FREEZE(LINVFS_GET_VFS(sb));
691 struct vfsmount *mnt)
693 struct vfs *vfsp = LINVFS_GET_VFS(mnt->mnt_sb);
696 VFS_SHOWARGS(vfsp, m, error);
702 struct super_block *sb,
703 struct fs_quota_stat *fqs)
705 struct vfs *vfsp = LINVFS_GET_VFS(sb);
708 VFS_QUOTACTL(vfsp, Q_XGETQSTAT, 0, (caddr_t)fqs, error);
714 struct super_block *sb,
718 struct vfs *vfsp = LINVFS_GET_VFS(sb);
721 VFS_QUOTACTL(vfsp, op, 0, (caddr_t)&flags, error);
727 struct super_block *sb,
730 struct fs_disk_quota *fdq)
732 struct vfs *vfsp = LINVFS_GET_VFS(sb);
735 getmode = (type == USRQUOTA) ? Q_XGETQUOTA :
736 ((type == GRPQUOTA) ? Q_XGETGQUOTA : Q_XGETPQUOTA);
737 VFS_QUOTACTL(vfsp, getmode, id, (caddr_t)fdq, error);
743 struct super_block *sb,
746 struct fs_disk_quota *fdq)
748 struct vfs *vfsp = LINVFS_GET_VFS(sb);
751 setmode = (type == USRQUOTA) ? Q_XSETQLIM :
752 ((type == GRPQUOTA) ? Q_XSETGQLIM : Q_XSETPQLIM);
753 VFS_QUOTACTL(vfsp, setmode, id, (caddr_t)fdq, error);
759 struct super_block *sb,
764 struct vfs *vfsp = vfs_allocate();
765 struct xfs_mount_args *args = xfs_args_allocate(sb);
766 struct kstatfs statvfs;
769 vfsp->vfs_super = sb;
770 LINVFS_SET_VFS(sb, vfsp);
771 if (sb->s_flags & MS_RDONLY)
772 vfsp->vfs_flag |= VFS_RDONLY;
773 bhv_insert_all_vfsops(vfsp);
775 VFS_PARSEARGS(vfsp, (char *)data, args, 0, error);
777 bhv_remove_all_vfsops(vfsp, 1);
781 sb_min_blocksize(sb, BBSIZE);
782 #ifdef CONFIG_XFS_EXPORT
783 sb->s_export_op = &linvfs_export_ops;
785 sb->s_qcop = &linvfs_qops;
786 sb->s_op = &linvfs_sops;
788 VFS_MOUNT(vfsp, args, NULL, error);
790 bhv_remove_all_vfsops(vfsp, 1);
794 VFS_STATVFS(vfsp, &statvfs, NULL, error);
799 sb->s_magic = statvfs.f_type;
800 sb->s_blocksize = statvfs.f_bsize;
801 sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
802 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
804 set_posix_acl_flag(sb);
806 VFS_ROOT(vfsp, &rootvp, error);
810 sb->s_root = d_alloc_root(LINVFS_GET_IP(rootvp));
815 if (is_bad_inode(sb->s_root->d_inode)) {
819 if ((error = linvfs_start_syncd(vfsp)))
821 vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
823 kmem_free(args, sizeof(*args));
835 VFS_UNMOUNT(vfsp, 0, NULL, error2);
838 vfs_deallocate(vfsp);
839 kmem_free(args, sizeof(*args));
843 STATIC struct super_block *
845 struct file_system_type *fs_type,
847 const char *dev_name,
850 return get_sb_bdev(fs_type, flags, dev_name, data, linvfs_fill_super);
853 STATIC struct super_operations linvfs_sops = {
854 .alloc_inode = linvfs_alloc_inode,
855 .destroy_inode = linvfs_destroy_inode,
856 .write_inode = linvfs_write_inode,
857 .clear_inode = linvfs_clear_inode,
858 .put_super = linvfs_put_super,
859 .write_super = linvfs_write_super,
860 .sync_fs = linvfs_sync_super,
861 .write_super_lockfs = linvfs_freeze_fs,
862 .statfs = linvfs_statfs,
863 .remount_fs = linvfs_remount,
864 .show_options = linvfs_show_options,
867 STATIC struct quotactl_ops linvfs_qops = {
868 .get_xstate = linvfs_getxstate,
869 .set_xstate = linvfs_setxstate,
870 .get_xquota = linvfs_getxquota,
871 .set_xquota = linvfs_setxquota,
874 STATIC struct file_system_type xfs_fs_type = {
875 .owner = THIS_MODULE,
877 .get_sb = linvfs_get_sb,
878 .kill_sb = kill_block_super,
879 .fs_flags = FS_REQUIRES_DEV,
888 static char message[] __initdata = KERN_INFO \
889 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
894 xfs_physmem = si.totalram;
898 error = linvfs_init_zones();
902 error = pagebuf_init();
911 error = register_filesystem(&xfs_fs_type);
914 XFS_DM_INIT(&xfs_fs_type);
921 linvfs_destroy_zones();
931 XFS_DM_EXIT(&xfs_fs_type);
932 unregister_filesystem(&xfs_fs_type);
935 linvfs_destroy_zones();
939 module_init(init_xfs_fs);
940 module_exit(exit_xfs_fs);
942 MODULE_AUTHOR("Silicon Graphics, Inc.");
943 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
944 MODULE_LICENSE("GPL");