xfs: reset the i_iolock lock class in the reclaim path
[safe/jmp/linux-2.6] / fs / xfs / xfs_dfrag.c
index 63abdc2..ab89a7e 100644 (file)
 /*
- * Copyright (c) 2000-2002 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2000-2006 Silicon Graphics, Inc.
+ * All Rights Reserved.
  *
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of version 2 of the GNU General Public License as
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
  * published by the Free Software Foundation.
  *
- * This program is distributed in the hope that it would be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
  *
- * Further, this software is distributed without any warranty that it is
- * free of the rightful claim of any third person regarding infringement
- * or the like.  Any license provided herein, whether implied or
- * otherwise, applies only to this software file.  Patent licenses, if
- * any, provided herein do not apply to combinations of this program with
- * other software, or any other product whatsoever.
- *
- * You should have received a copy of the GNU General Public License along
- * with this program; if not, write the Free Software Foundation, Inc., 59
- * Temple Place - Suite 330, Boston MA 02111-1307, USA.
- *
- * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
- * Mountain View, CA  94043, or:
- *
- * http://www.sgi.com
- *
- * For further information regarding this notice, see:
- *
- * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  */
-
 #include "xfs.h"
-#include "xfs_macros.h"
+#include "xfs_fs.h"
 #include "xfs_types.h"
-#include "xfs_inum.h"
+#include "xfs_bit.h"
 #include "xfs_log.h"
+#include "xfs_inum.h"
 #include "xfs_trans.h"
 #include "xfs_sb.h"
-#include "xfs_dir.h"
+#include "xfs_ag.h"
 #include "xfs_dir2.h"
 #include "xfs_dmapi.h"
 #include "xfs_mount.h"
-#include "xfs_ag.h"
-#include "xfs_alloc_btree.h"
 #include "xfs_bmap_btree.h"
+#include "xfs_alloc_btree.h"
 #include "xfs_ialloc_btree.h"
-#include "xfs_btree.h"
-#include "xfs_attr_sf.h"
-#include "xfs_dir_sf.h"
 #include "xfs_dir2_sf.h"
+#include "xfs_attr_sf.h"
 #include "xfs_dinode.h"
-#include "xfs_inode_item.h"
 #include "xfs_inode.h"
+#include "xfs_inode_item.h"
 #include "xfs_bmap.h"
+#include "xfs_btree.h"
 #include "xfs_ialloc.h"
 #include "xfs_itable.h"
 #include "xfs_dfrag.h"
 #include "xfs_error.h"
-#include "xfs_mac.h"
 #include "xfs_rw.h"
+#include "xfs_vnodeops.h"
 
 /*
  * Syssgi interface for swapext
  */
 int
 xfs_swapext(
-       xfs_swapext_t   __user *sxp)
+       xfs_swapext_t   *sxp)
 {
-       xfs_swapext_t   sx;
-       xfs_inode_t     *ip=NULL, *tip=NULL, *ips[2];
-       xfs_trans_t     *tp;
-       xfs_mount_t     *mp;
-       xfs_bstat_t     *sbp;
-       struct file     *fp = NULL, *tfp = NULL;
-       vnode_t         *vp, *tvp;
-       bhv_desc_t      *bdp, *tbdp;
-       vn_bhv_head_t   *bhp, *tbhp;
-       uint            lock_flags=0;
-       int             ilf_fields, tilf_fields;
+       xfs_inode_t     *ip, *tip;
+       struct file     *file, *target_file;
        int             error = 0;
-       xfs_ifork_t     tempif, *ifp, *tifp;
-       __uint64_t      tmp;
-       int             aforkblks = 0;
-       int             taforkblks = 0;
-       int             locked = 0;
-
-       if (copy_from_user(&sx, sxp, sizeof(sx)))
-               return XFS_ERROR(EFAULT);
 
        /* Pull information for the target fd */
-       if (((fp = fget((int)sx.sx_fdtarget)) == NULL) ||
-           ((vp = LINVFS_GET_VP(fp->f_dentry->d_inode)) == NULL))  {
+       file = fget((int)sxp->sx_fdtarget);
+       if (!file) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out;
        }
 
-       bhp = VN_BHV_HEAD(vp);
-       bdp = vn_bhv_lookup(bhp, &xfs_vnodeops);
-       if (bdp == NULL) {
+       if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) {
                error = XFS_ERROR(EBADF);
-               goto error0;
-       } else {
-               ip = XFS_BHVTOI(bdp);
+               goto out_put_file;
        }
 
-       if (((tfp = fget((int)sx.sx_fdtmp)) == NULL) ||
-           ((tvp = LINVFS_GET_VP(tfp->f_dentry->d_inode)) == NULL)) {
+       target_file = fget((int)sxp->sx_fdtmp);
+       if (!target_file) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_put_file;
        }
 
-       tbhp = VN_BHV_HEAD(tvp);
-       tbdp = vn_bhv_lookup(tbhp, &xfs_vnodeops);
-       if (tbdp == NULL) {
+       if (!(target_file->f_mode & FMODE_WRITE) ||
+           (target_file->f_flags & O_APPEND)) {
                error = XFS_ERROR(EBADF);
-               goto error0;
-       } else {
-               tip = XFS_BHVTOI(tbdp);
+               goto out_put_target_file;
        }
 
+       if (IS_SWAPFILE(file->f_path.dentry->d_inode) ||
+           IS_SWAPFILE(target_file->f_path.dentry->d_inode)) {
+               error = XFS_ERROR(EINVAL);
+               goto out_put_target_file;
+       }
+
+       ip = XFS_I(file->f_path.dentry->d_inode);
+       tip = XFS_I(target_file->f_path.dentry->d_inode);
+
        if (ip->i_mount != tip->i_mount) {
-               error =  XFS_ERROR(EINVAL);
-               goto error0;
+               error = XFS_ERROR(EINVAL);
+               goto out_put_target_file;
        }
 
        if (ip->i_ino == tip->i_ino) {
-               error =  XFS_ERROR(EINVAL);
-               goto error0;
+               error = XFS_ERROR(EINVAL);
+               goto out_put_target_file;
        }
 
-       mp = ip->i_mount;
+       if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
+               error = XFS_ERROR(EIO);
+               goto out_put_target_file;
+       }
 
-       sbp = &sx.sx_stat;
+       error = xfs_swap_extents(ip, tip, sxp);
 
-       if (XFS_FORCED_SHUTDOWN(mp)) {
-               error =  XFS_ERROR(EIO);
-               goto error0;
-       }
+ out_put_target_file:
+       fput(target_file);
+ out_put_file:
+       fput(file);
+ out:
+       return error;
+}
+
+int
+xfs_swap_extents(
+       xfs_inode_t     *ip,
+       xfs_inode_t     *tip,
+       xfs_swapext_t   *sxp)
+{
+       xfs_mount_t     *mp;
+       xfs_trans_t     *tp;
+       xfs_bstat_t     *sbp = &sxp->sx_stat;
+       xfs_ifork_t     *tempifp, *ifp, *tifp;
+       int             ilf_fields, tilf_fields;
+       int             error = 0;
+       int             aforkblks = 0;
+       int             taforkblks = 0;
+       __uint64_t      tmp;
 
-       locked = 1;
+       mp = ip->i_mount;
 
-       /* Lock in i_ino order */
-       if (ip->i_ino < tip->i_ino) {
-               ips[0] = ip;
-               ips[1] = tip;
-       } else {
-               ips[0] = tip;
-               ips[1] = ip;
+       tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
+       if (!tempifp) {
+               error = XFS_ERROR(ENOMEM);
+               goto out;
        }
-       lock_flags = XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL;
-       xfs_lock_inodes(ips, 2, 0, lock_flags);
 
-       /* Check permissions */
-       error = xfs_iaccess(ip, S_IWUSR, NULL);
-       if (error)
-               goto error0;
+       sbp = &sxp->sx_stat;
 
-       error = xfs_iaccess(tip, S_IWUSR, NULL);
-       if (error)
-               goto error0;
+       /*
+        * we have to do two separate lock calls here to keep lockdep
+        * happy. If we try to get all the locks in one call, lock will
+        * report false positives when we drop the ILOCK and regain them
+        * below.
+        */
+       xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
+       xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 
        /* Verify that both files have the same format */
        if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_unlock;
        }
 
        /* Verify both files are either real-time or non-realtime */
-       if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
-           (tip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
+       if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_unlock;
        }
 
        /* Should never get a local format */
        if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
            tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_unlock;
        }
 
-       if (VN_CACHED(tvp) != 0)
-               xfs_inval_cached_pages(XFS_ITOV(tip), &(tip->i_iocore),
-                                               (xfs_off_t)0, 0, 0);
+       if (VN_CACHED(VFS_I(tip)) != 0) {
+               xfs_inval_cached_trace(tip, 0, -1, 0, -1);
+               error = xfs_flushinval_pages(tip, 0, -1,
+                               FI_REMAPF_LOCKED);
+               if (error)
+                       goto out_unlock;
+       }
 
        /* Verify O_DIRECT for ftmp */
-       if (VN_CACHED(tvp) != 0) {
+       if (VN_CACHED(VFS_I(tip)) != 0) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_unlock;
        }
 
        /* Verify all data are being swapped */
-       if (sx.sx_offset != 0 ||
-           sx.sx_length != ip->i_d.di_size ||
-           sx.sx_length != tip->i_d.di_size) {
+       if (sxp->sx_offset != 0 ||
+           sxp->sx_length != ip->i_d.di_size ||
+           sxp->sx_length != tip->i_d.di_size) {
                error = XFS_ERROR(EFAULT);
-               goto error0;
+               goto out_unlock;
        }
 
        /*
@@ -205,7 +196,7 @@ xfs_swapext(
         */
        if ( XFS_IFORK_Q(ip) != XFS_IFORK_Q(tip) ) {
                error = XFS_ERROR(EINVAL);
-               goto error0;
+               goto out_unlock;
        }
 
        /*
@@ -215,23 +206,23 @@ xfs_swapext(
         * process that the file was not changed out from
         * under it.
         */
-       if ((sbp->bs_ctime.tv_sec != ip->i_d.di_ctime.t_sec) ||
-           (sbp->bs_ctime.tv_nsec != ip->i_d.di_ctime.t_nsec) ||
-           (sbp->bs_mtime.tv_sec != ip->i_d.di_mtime.t_sec) ||
-           (sbp->bs_mtime.tv_nsec != ip->i_d.di_mtime.t_nsec)) {
+       if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
+           (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
+           (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
+           (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
                error = XFS_ERROR(EBUSY);
-               goto error0;
+               goto out_unlock;
        }
 
        /* We need to fail if the file is memory mapped.  Once we have tossed
         * all existing pages, the page fault will have no option
         * but to go to the filesystem for pages. By making the page fault call
-        * VOP_READ (or write in the case of autogrow) they block on the iolock
+        * vop_read (or write in the case of autogrow) they block on the iolock
         * until we have switched the extents.
         */
-       if (VN_MAPPED(vp)) {
+       if (VN_MAPPED(VFS_I(ip))) {
                error = XFS_ERROR(EBUSY);
-               goto error0;
+               goto out_unlock;
        }
 
        xfs_iunlock(ip, XFS_ILOCK_EXCL);
@@ -245,7 +236,7 @@ xfs_swapext(
         * fields change.
         */
 
-       VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
+       xfs_tosspages(ip, 0, -1, FI_REMAPF);
 
        tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
        if ((error = xfs_trans_reserve(tp, 0,
@@ -254,9 +245,9 @@ xfs_swapext(
                xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
                xfs_iunlock(tip, XFS_IOLOCK_EXCL);
                xfs_trans_cancel(tp, 0);
-               return error;
+               goto out;
        }
-       xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
+       xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 
        /*
         * Count the number of extended attribute blocks
@@ -264,23 +255,15 @@ xfs_swapext(
        if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
             (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
                error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
-               if (error) {
-                       xfs_iunlock(ip,  lock_flags);
-                       xfs_iunlock(tip, lock_flags);
-                       xfs_trans_cancel(tp, 0);
-                       return error;
-               }
+               if (error)
+                       goto out_trans_cancel;
        }
        if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
             (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
                error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
                        &taforkblks);
-               if (error) {
-                       xfs_iunlock(ip,  lock_flags);
-                       xfs_iunlock(tip, lock_flags);
-                       xfs_trans_cancel(tp, 0);
-                       return error;
-               }
+               if (error)
+                       goto out_trans_cancel;
        }
 
        /*
@@ -288,9 +271,9 @@ xfs_swapext(
         */
        ifp = &ip->i_df;
        tifp = &tip->i_df;
-       tempif = *ifp;  /* struct copy */
-       *ifp = *tifp;   /* struct copy */
-       *tifp = tempif; /* struct copy */
+       *tempifp = *ifp;        /* struct copy */
+       *ifp = *tifp;           /* struct copy */
+       *tifp = *tempifp;       /* struct copy */
 
        /*
         * Fix the on-disk inode values
@@ -345,16 +328,12 @@ xfs_swapext(
                break;
        }
 
-       /*
-        * Increment vnode ref counts since xfs_trans_commit &
-        * xfs_trans_cancel will both unlock the inodes and
-        * decrement the associated ref counts.
-        */
-       VN_HOLD(vp);
-       VN_HOLD(tvp);
 
-       xfs_trans_ijoin(tp, ip, lock_flags);
-       xfs_trans_ijoin(tp, tip, lock_flags);
+       IHOLD(ip);
+       xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
+
+       IHOLD(tip);
+       xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 
        xfs_trans_log_inode(tp, ip,  ilf_fields);
        xfs_trans_log_inode(tp, tip, tilf_fields);
@@ -363,25 +342,21 @@ xfs_swapext(
         * If this is a synchronous mount, make sure that the
         * transaction goes to disk before returning to the user.
         */
-       if (mp->m_flags & XFS_MOUNT_WSYNC) {
+       if (mp->m_flags & XFS_MOUNT_WSYNC)
                xfs_trans_set_sync(tp);
-       }
-
-       error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT, NULL);
 
-       fput(fp);
-       fput(tfp);
+       error = xfs_trans_commit(tp, XFS_TRANS_SWAPEXT);
 
+out:
+       kmem_free(tempifp);
        return error;
 
- error0:
-       if (locked) {
-               xfs_iunlock(ip,  lock_flags);
-               xfs_iunlock(tip, lock_flags);
-       }
-
-       if (fp != NULL) fput(fp);
-       if (tfp != NULL) fput(tfp);
+out_unlock:
+       xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
+       xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
+       goto out;
 
-       return error;
+out_trans_cancel:
+       xfs_trans_cancel(tp, 0);
+       goto out_unlock;
 }