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>
75 #include <linux/kthread.h>
77 STATIC struct quotactl_ops linvfs_qops;
78 STATIC struct super_operations linvfs_sops;
79 STATIC kmem_zone_t *xfs_vnode_zone;
80 STATIC kmem_zone_t *xfs_ioend_zone;
81 mempool_t *xfs_ioend_pool;
83 STATIC struct xfs_mount_args *
85 struct super_block *sb)
87 struct xfs_mount_args *args;
89 args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP);
90 args->logbufs = args->logbufsize = -1;
91 strncpy(args->fsname, sb->s_id, MAXNAMELEN);
93 /* Copy the already-parsed mount(2) flags we're interested in */
94 if (sb->s_flags & MS_NOATIME)
95 args->flags |= XFSMNT_NOATIME;
96 if (sb->s_flags & MS_DIRSYNC)
97 args->flags |= XFSMNT_DIRSYNC;
98 if (sb->s_flags & MS_SYNCHRONOUS)
99 args->flags |= XFSMNT_WSYNC;
101 /* Default to 32 bit inodes on Linux all the time */
102 args->flags |= XFSMNT_32BITINODES;
109 unsigned int blockshift)
111 unsigned int pagefactor = 1;
112 unsigned int bitshift = BITS_PER_LONG - 1;
114 /* Figure out maximum filesize, on Linux this can depend on
115 * the filesystem blocksize (on 32 bit platforms).
116 * __block_prepare_write does this in an [unsigned] long...
117 * page->index << (PAGE_CACHE_SHIFT - bbits)
118 * So, for page sized blocks (4K on 32 bit platforms),
119 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
120 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
121 * but for smaller blocksizes it is less (bbits = log2 bsize).
122 * Note1: get_block_t takes a long (implicit cast from above)
123 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
124 * can optionally convert the [unsigned] long from above into
125 * an [unsigned] long long.
128 #if BITS_PER_LONG == 32
129 # if defined(CONFIG_LBD)
130 ASSERT(sizeof(sector_t) == 8);
131 pagefactor = PAGE_CACHE_SIZE;
132 bitshift = BITS_PER_LONG;
134 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
138 return (((__uint64_t)pagefactor) << bitshift) - 1;
141 STATIC __inline__ void
145 switch (inode->i_mode & S_IFMT) {
147 inode->i_op = &linvfs_file_inode_operations;
148 inode->i_fop = &linvfs_file_operations;
149 inode->i_mapping->a_ops = &linvfs_aops;
152 inode->i_op = &linvfs_dir_inode_operations;
153 inode->i_fop = &linvfs_dir_operations;
156 inode->i_op = &linvfs_symlink_inode_operations;
158 inode->i_mapping->a_ops = &linvfs_aops;
161 inode->i_op = &linvfs_file_inode_operations;
162 init_special_inode(inode, inode->i_mode, inode->i_rdev);
167 STATIC __inline__ void
168 xfs_revalidate_inode(
173 struct inode *inode = LINVFS_GET_IP(vp);
175 inode->i_mode = ip->i_d.di_mode;
176 inode->i_nlink = ip->i_d.di_nlink;
177 inode->i_uid = ip->i_d.di_uid;
178 inode->i_gid = ip->i_d.di_gid;
180 switch (inode->i_mode & S_IFMT) {
184 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
185 sysv_minor(ip->i_df.if_u2.if_rdev));
192 inode->i_blksize = PAGE_CACHE_SIZE;
193 inode->i_generation = ip->i_d.di_gen;
194 i_size_write(inode, ip->i_d.di_size);
196 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
197 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec;
198 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
199 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
200 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
201 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
202 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
203 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
204 inode->i_flags |= S_IMMUTABLE;
206 inode->i_flags &= ~S_IMMUTABLE;
207 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
208 inode->i_flags |= S_APPEND;
210 inode->i_flags &= ~S_APPEND;
211 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
212 inode->i_flags |= S_SYNC;
214 inode->i_flags &= ~S_SYNC;
215 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
216 inode->i_flags |= S_NOATIME;
218 inode->i_flags &= ~S_NOATIME;
219 vp->v_flag &= ~VMODIFIED;
223 xfs_initialize_vnode(
226 bhv_desc_t *inode_bhv,
229 xfs_inode_t *ip = XFS_BHVTOI(inode_bhv);
230 struct inode *inode = LINVFS_GET_IP(vp);
232 if (!inode_bhv->bd_vobj) {
233 vp->v_vfsp = bhvtovfs(bdp);
234 bhv_desc_init(inode_bhv, ip, vp, &xfs_vnodeops);
235 bhv_insert(VN_BHV_HEAD(vp), inode_bhv);
239 * We need to set the ops vectors, and unlock the inode, but if
240 * we have been called during the new inode create process, it is
241 * too early to fill in the Linux inode. We will get called a
242 * second time once the inode is properly set up, and then we can
245 if (ip->i_d.di_mode != 0 && unlock && (inode->i_state & I_NEW)) {
246 xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
247 xfs_set_inodeops(inode);
249 ip->i_flags &= ~XFS_INEW;
252 unlock_new_inode(inode);
260 struct block_device **bdevp)
264 *bdevp = open_bdev_excl(name, 0, mp);
265 if (IS_ERR(*bdevp)) {
266 error = PTR_ERR(*bdevp);
267 printk("XFS: Invalid device [%s], error=%d\n", name, error);
275 struct block_device *bdev)
278 close_bdev_excl(bdev);
282 STATIC struct inode *
284 struct super_block *sb)
288 vp = kmem_cache_alloc(xfs_vnode_zone, kmem_flags_convert(KM_SLEEP));
291 return LINVFS_GET_IP(vp);
295 linvfs_destroy_inode(
298 kmem_zone_free(xfs_vnode_zone, LINVFS_GET_VP(inode));
302 linvfs_inode_init_once(
304 kmem_cache_t *cachep,
307 vnode_t *vp = (vnode_t *)data;
309 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
310 SLAB_CTOR_CONSTRUCTOR)
311 inode_init_once(LINVFS_GET_IP(vp));
315 linvfs_init_zones(void)
317 xfs_vnode_zone = kmem_cache_create("xfs_vnode",
318 sizeof(vnode_t), 0, SLAB_RECLAIM_ACCOUNT,
319 linvfs_inode_init_once, NULL);
323 xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
325 goto out_destroy_vnode_zone;
327 xfs_ioend_pool = mempool_create(4 * MAX_BUF_PER_PAGE,
328 mempool_alloc_slab, mempool_free_slab,
331 goto out_free_ioend_zone;
337 kmem_zone_destroy(xfs_ioend_zone);
338 out_destroy_vnode_zone:
339 kmem_zone_destroy(xfs_vnode_zone);
345 linvfs_destroy_zones(void)
347 mempool_destroy(xfs_ioend_pool);
348 kmem_zone_destroy(xfs_vnode_zone);
349 kmem_zone_destroy(xfs_ioend_zone);
353 * Attempt to flush the inode, this will actually fail
354 * if the inode is pinned, but we dirty the inode again
355 * at the point when it is unpinned after a log write,
356 * since this is when the inode itself becomes flushable.
363 vnode_t *vp = LINVFS_GET_VP(inode);
364 int error = 0, flags = FLUSH_INODE;
367 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
370 VOP_IFLUSH(vp, flags, error);
371 if (error == EAGAIN) {
373 VOP_IFLUSH(vp, flags | FLUSH_LOG, error);
386 vnode_t *vp = LINVFS_GET_VP(inode);
389 vn_trace_entry(vp, "clear_inode", (inst_t *)__return_address);
391 XFS_STATS_INC(vn_rele);
392 XFS_STATS_INC(vn_remove);
393 XFS_STATS_INC(vn_reclaim);
394 XFS_STATS_DEC(vn_active);
397 * This can happen because xfs_iget_core calls xfs_idestroy if we
398 * find an inode with di_mode == 0 but without IGET_CREATE set.
401 VOP_INACTIVE(vp, NULL, cache);
404 vp->v_flag &= ~VMODIFIED;
408 VOP_RECLAIM(vp, error);
410 panic("vn_purge: cannot reclaim");
413 ASSERT(vp->v_fbhv == NULL);
415 #ifdef XFS_VNODE_TRACE
416 ktrace_free(vp->v_trace);
421 * Enqueue a work item to be picked up by the vfs xfssyncd thread.
422 * Doing this has two advantages:
423 * - It saves on stack space, which is tight in certain situations
424 * - It can be used (with care) as a mechanism to avoid deadlocks.
425 * Flushing while allocating in a full filesystem requires both.
428 xfs_syncd_queue_work(
431 void (*syncer)(vfs_t *, void *))
433 vfs_sync_work_t *work;
435 work = kmem_alloc(sizeof(struct vfs_sync_work), KM_SLEEP);
436 INIT_LIST_HEAD(&work->w_list);
437 work->w_syncer = syncer;
440 spin_lock(&vfs->vfs_sync_lock);
441 list_add_tail(&work->w_list, &vfs->vfs_sync_list);
442 spin_unlock(&vfs->vfs_sync_lock);
443 wake_up_process(vfs->vfs_sync_task);
447 * Flush delayed allocate data, attempting to free up reserved space
448 * from existing allocations. At this point a new allocation attempt
449 * has failed with ENOSPC and we are in the process of scratching our
450 * heads, looking about for more room...
453 xfs_flush_inode_work(
457 filemap_flush(((struct inode *)inode)->i_mapping);
458 iput((struct inode *)inode);
465 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
466 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
469 xfs_syncd_queue_work(vfs, inode, xfs_flush_inode_work);
474 * This is the "bigger hammer" version of xfs_flush_inode_work...
475 * (IOW, "If at first you don't succeed, use a Bigger Hammer").
478 xfs_flush_device_work(
482 sync_blockdev(vfs->vfs_super->s_bdev);
483 iput((struct inode *)inode);
490 struct inode *inode = LINVFS_GET_IP(XFS_ITOV(ip));
491 struct vfs *vfs = XFS_MTOVFS(ip->i_mount);
494 xfs_syncd_queue_work(vfs, inode, xfs_flush_device_work);
496 xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
499 #define SYNCD_FLAGS (SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR)
507 if (!(vfsp->vfs_flag & VFS_RDONLY))
508 VFS_SYNC(vfsp, SYNCD_FLAGS, NULL, error);
509 vfsp->vfs_sync_seq++;
511 wake_up(&vfsp->vfs_wait_single_sync_task);
519 vfs_t *vfsp = (vfs_t *) arg;
520 struct vfs_sync_work *work, *n;
523 timeleft = (xfs_syncd_centisecs * HZ) / 100;
525 set_current_state(TASK_INTERRUPTIBLE);
526 timeleft = schedule_timeout(timeleft);
529 if (kthread_should_stop())
532 spin_lock(&vfsp->vfs_sync_lock);
534 * We can get woken by laptop mode, to do a sync -
535 * that's the (only!) case where the list would be
536 * empty with time remaining.
538 if (!timeleft || list_empty(&vfsp->vfs_sync_list)) {
540 timeleft = (xfs_syncd_centisecs * HZ) / 100;
541 INIT_LIST_HEAD(&vfsp->vfs_sync_work.w_list);
542 list_add_tail(&vfsp->vfs_sync_work.w_list,
543 &vfsp->vfs_sync_list);
545 list_for_each_entry_safe(work, n, &vfsp->vfs_sync_list, w_list)
546 list_move(&work->w_list, &tmp);
547 spin_unlock(&vfsp->vfs_sync_lock);
549 list_for_each_entry_safe(work, n, &tmp, w_list) {
550 (*work->w_syncer)(vfsp, work->w_data);
551 list_del(&work->w_list);
552 if (work == &vfsp->vfs_sync_work)
554 kmem_free(work, sizeof(struct vfs_sync_work));
565 vfsp->vfs_sync_work.w_syncer = vfs_sync_worker;
566 vfsp->vfs_sync_work.w_vfs = vfsp;
567 vfsp->vfs_sync_task = kthread_run(xfssyncd, vfsp, "xfssyncd");
568 if (IS_ERR(vfsp->vfs_sync_task))
569 return -PTR_ERR(vfsp->vfs_sync_task);
577 kthread_stop(vfsp->vfs_sync_task);
582 struct super_block *sb)
584 vfs_t *vfsp = LINVFS_GET_VFS(sb);
587 linvfs_stop_syncd(vfsp);
588 VFS_SYNC(vfsp, SYNC_ATTR|SYNC_DELWRI, NULL, error);
590 VFS_UNMOUNT(vfsp, 0, NULL, error);
592 printk("XFS unmount got error %d\n", error);
593 printk("%s: vfsp/0x%p left dangling!\n", __FUNCTION__, vfsp);
597 vfs_deallocate(vfsp);
602 struct super_block *sb)
604 vfs_t *vfsp = LINVFS_GET_VFS(sb);
607 if (sb->s_flags & MS_RDONLY) {
608 sb->s_dirt = 0; /* paranoia */
611 /* Push the log and superblock a little */
612 VFS_SYNC(vfsp, SYNC_FSDATA, NULL, error);
618 struct super_block *sb,
621 vfs_t *vfsp = LINVFS_GET_VFS(sb);
623 int flags = SYNC_FSDATA;
625 if (unlikely(sb->s_frozen == SB_FREEZE_WRITE))
626 flags = SYNC_QUIESCE;
628 flags = SYNC_FSDATA | (wait ? SYNC_WAIT : 0);
630 VFS_SYNC(vfsp, flags, NULL, error);
633 if (unlikely(laptop_mode)) {
634 int prev_sync_seq = vfsp->vfs_sync_seq;
637 * The disk must be active because we're syncing.
638 * We schedule xfssyncd now (now that the disk is
639 * active) instead of later (when it might not be).
641 wake_up_process(vfsp->vfs_sync_task);
643 * We have to wait for the sync iteration to complete.
644 * If we don't, the disk activity caused by the sync
645 * will come after the sync is completed, and that
646 * triggers another sync from laptop mode.
648 wait_event(vfsp->vfs_wait_single_sync_task,
649 vfsp->vfs_sync_seq != prev_sync_seq);
657 struct super_block *sb,
658 struct kstatfs *statp)
660 vfs_t *vfsp = LINVFS_GET_VFS(sb);
663 VFS_STATVFS(vfsp, statp, NULL, error);
669 struct super_block *sb,
673 vfs_t *vfsp = LINVFS_GET_VFS(sb);
674 struct xfs_mount_args *args = xfs_args_allocate(sb);
677 VFS_PARSEARGS(vfsp, options, args, 1, error);
679 VFS_MNTUPDATE(vfsp, flags, args, error);
680 kmem_free(args, sizeof(*args));
686 struct super_block *sb)
688 VFS_FREEZE(LINVFS_GET_VFS(sb));
694 struct vfsmount *mnt)
696 struct vfs *vfsp = LINVFS_GET_VFS(mnt->mnt_sb);
699 VFS_SHOWARGS(vfsp, m, error);
705 struct super_block *sb,
706 struct fs_quota_stat *fqs)
708 struct vfs *vfsp = LINVFS_GET_VFS(sb);
711 VFS_QUOTACTL(vfsp, Q_XGETQSTAT, 0, (caddr_t)fqs, error);
717 struct super_block *sb,
721 struct vfs *vfsp = LINVFS_GET_VFS(sb);
724 VFS_QUOTACTL(vfsp, op, 0, (caddr_t)&flags, error);
730 struct super_block *sb,
733 struct fs_disk_quota *fdq)
735 struct vfs *vfsp = LINVFS_GET_VFS(sb);
738 getmode = (type == USRQUOTA) ? Q_XGETQUOTA :
739 ((type == GRPQUOTA) ? Q_XGETGQUOTA : Q_XGETPQUOTA);
740 VFS_QUOTACTL(vfsp, getmode, id, (caddr_t)fdq, error);
746 struct super_block *sb,
749 struct fs_disk_quota *fdq)
751 struct vfs *vfsp = LINVFS_GET_VFS(sb);
754 setmode = (type == USRQUOTA) ? Q_XSETQLIM :
755 ((type == GRPQUOTA) ? Q_XSETGQLIM : Q_XSETPQLIM);
756 VFS_QUOTACTL(vfsp, setmode, id, (caddr_t)fdq, error);
762 struct super_block *sb,
767 struct vfs *vfsp = vfs_allocate();
768 struct xfs_mount_args *args = xfs_args_allocate(sb);
769 struct kstatfs statvfs;
772 vfsp->vfs_super = sb;
773 LINVFS_SET_VFS(sb, vfsp);
774 if (sb->s_flags & MS_RDONLY)
775 vfsp->vfs_flag |= VFS_RDONLY;
776 bhv_insert_all_vfsops(vfsp);
778 VFS_PARSEARGS(vfsp, (char *)data, args, 0, error);
780 bhv_remove_all_vfsops(vfsp, 1);
784 sb_min_blocksize(sb, BBSIZE);
785 #ifdef CONFIG_XFS_EXPORT
786 sb->s_export_op = &linvfs_export_ops;
788 sb->s_qcop = &linvfs_qops;
789 sb->s_op = &linvfs_sops;
791 VFS_MOUNT(vfsp, args, NULL, error);
793 bhv_remove_all_vfsops(vfsp, 1);
797 VFS_STATVFS(vfsp, &statvfs, NULL, error);
802 sb->s_magic = statvfs.f_type;
803 sb->s_blocksize = statvfs.f_bsize;
804 sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1;
805 sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
807 set_posix_acl_flag(sb);
809 VFS_ROOT(vfsp, &rootvp, error);
813 sb->s_root = d_alloc_root(LINVFS_GET_IP(rootvp));
818 if (is_bad_inode(sb->s_root->d_inode)) {
822 if ((error = linvfs_start_syncd(vfsp)))
824 vn_trace_exit(rootvp, __FUNCTION__, (inst_t *)__return_address);
826 kmem_free(args, sizeof(*args));
838 VFS_UNMOUNT(vfsp, 0, NULL, error2);
841 vfs_deallocate(vfsp);
842 kmem_free(args, sizeof(*args));
846 STATIC struct super_block *
848 struct file_system_type *fs_type,
850 const char *dev_name,
853 return get_sb_bdev(fs_type, flags, dev_name, data, linvfs_fill_super);
856 STATIC struct super_operations linvfs_sops = {
857 .alloc_inode = linvfs_alloc_inode,
858 .destroy_inode = linvfs_destroy_inode,
859 .write_inode = linvfs_write_inode,
860 .clear_inode = linvfs_clear_inode,
861 .put_super = linvfs_put_super,
862 .write_super = linvfs_write_super,
863 .sync_fs = linvfs_sync_super,
864 .write_super_lockfs = linvfs_freeze_fs,
865 .statfs = linvfs_statfs,
866 .remount_fs = linvfs_remount,
867 .show_options = linvfs_show_options,
870 STATIC struct quotactl_ops linvfs_qops = {
871 .get_xstate = linvfs_getxstate,
872 .set_xstate = linvfs_setxstate,
873 .get_xquota = linvfs_getxquota,
874 .set_xquota = linvfs_setxquota,
877 STATIC struct file_system_type xfs_fs_type = {
878 .owner = THIS_MODULE,
880 .get_sb = linvfs_get_sb,
881 .kill_sb = kill_block_super,
882 .fs_flags = FS_REQUIRES_DEV,
891 static char message[] __initdata = KERN_INFO \
892 XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n";
897 xfs_physmem = si.totalram;
901 error = linvfs_init_zones();
905 error = pagebuf_init();
914 error = register_filesystem(&xfs_fs_type);
917 XFS_DM_INIT(&xfs_fs_type);
924 linvfs_destroy_zones();
934 XFS_DM_EXIT(&xfs_fs_type);
935 unregister_filesystem(&xfs_fs_type);
938 linvfs_destroy_zones();
942 module_init(init_xfs_fs);
943 module_exit(exit_xfs_fs);
945 MODULE_AUTHOR("Silicon Graphics, Inc.");
946 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
947 MODULE_LICENSE("GPL");