linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / fs / xfs / linux-2.6 / xfs_lrw.c
index 88e2b7c..e0ab45f 100644 (file)
@@ -1,44 +1,25 @@
 /*
- * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
+ * Copyright (c) 2000-2003,2005 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
  */
-/*
- *  fs/xfs/linux/xfs_lrw.c (Linux Read Write stuff)
- *
- */
-
 #include "xfs.h"
-
 #include "xfs_fs.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_ag.h"
 #include "xfs_dmapi.h"
 #include "xfs_quota.h"
 #include "xfs_mount.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_ialloc.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.h"
 #include "xfs_bmap.h"
-#include "xfs_bit.h"
+#include "xfs_btree.h"
+#include "xfs_ialloc.h"
 #include "xfs_rtalloc.h"
 #include "xfs_error.h"
 #include "xfs_itable.h"
@@ -209,30 +189,6 @@ unlock:
        return (-status);
 }
 
-/*
- * xfs_inval_cached_pages
- * 
- * This routine is responsible for keeping direct I/O and buffered I/O
- * somewhat coherent.  From here we make sure that we're at least
- * temporarily holding the inode I/O lock exclusively and then call
- * the page cache to flush and invalidate any cached pages.  If there
- * are no cached pages this routine will be very quick.
- */
-void
-xfs_inval_cached_pages(
-       vnode_t         *vp,
-       xfs_iocore_t    *io,
-       xfs_off_t       offset,
-       int             write,
-       int             relock)
-{
-       if (VN_CACHED(vp)) {
-               xfs_inval_cached_trace(io, offset, -1, ctooff(offtoct(offset)), -1);
-               VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), -1, FI_REMAPF_LOCKED);
-       }
-
-}
-
 ssize_t                        /* bytes read, or (-)  error */
 xfs_read(
        bhv_desc_t              *bdp,
@@ -277,8 +233,8 @@ xfs_read(
                xfs_buftarg_t   *target =
                        (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
                                mp->m_rtdev_targp : mp->m_ddev_targp;
-               if ((*offset & target->pbr_smask) ||
-                   (size & target->pbr_smask)) {
+               if ((*offset & target->bt_smask) ||
+                   (size & target->bt_smask)) {
                        if (*offset == ip->i_d.di_size) {
                                return (0);
                        }
@@ -298,16 +254,17 @@ xfs_read(
        }
 
        if (unlikely(ioflags & IO_ISDIRECT))
-               down(&inode->i_sem);
+               mutex_lock(&inode->i_mutex);
        xfs_ilock(ip, XFS_IOLOCK_SHARED);
 
        if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
            !(ioflags & IO_INVIS)) {
                vrwlock_t locktype = VRWLOCK_READ;
+               int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
 
                ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
                                        BHV_TO_VNODE(bdp), *offset, size,
-                                       FILP_DELAY_FLAG(file), &locktype);
+                                       dmflags, &locktype);
                if (ret) {
                        xfs_iunlock(ip, XFS_IOLOCK_SHARED);
                        goto unlock_isem;
@@ -317,19 +274,16 @@ xfs_read(
        xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
                                (void *)iovp, segs, *offset, ioflags);
        ret = __generic_file_aio_read(iocb, iovp, segs, offset);
-       if (ret == -EIOCBQUEUED)
+       if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
                ret = wait_on_sync_kiocb(iocb);
        if (ret > 0)
                XFS_STATS_ADD(xs_read_bytes, ret);
 
        xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 
-       if (likely(!(ioflags & IO_INVIS)))
-               xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
-
 unlock_isem:
        if (unlikely(ioflags & IO_ISDIRECT))
-               up(&inode->i_sem);
+               mutex_unlock(&inode->i_mutex);
        return ret;
 }
 
@@ -389,9 +343,6 @@ xfs_sendfile(
        if (ret > 0)
                XFS_STATS_ADD(xs_read_bytes, ret);
 
-       if (likely(!(ioflags & IO_INVIS)))
-               xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
-
        return ret;
 }
 
@@ -405,7 +356,6 @@ STATIC int                          /* error (positive) */
 xfs_zero_last_block(
        struct inode    *ip,
        xfs_iocore_t    *io,
-       xfs_off_t       offset,
        xfs_fsize_t     isize,
        xfs_fsize_t     end_size)
 {
@@ -414,19 +364,16 @@ xfs_zero_last_block(
        int             nimaps;
        int             zero_offset;
        int             zero_len;
-       int             isize_fsb_offset;
        int             error = 0;
        xfs_bmbt_irec_t imap;
        loff_t          loff;
-       size_t          lsize;
 
        ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
-       ASSERT(offset > isize);
 
        mp = io->io_mount;
 
-       isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
-       if (isize_fsb_offset == 0) {
+       zero_offset = XFS_B_FSB_OFFSET(mp, isize);
+       if (zero_offset == 0) {
                /*
                 * There are no extra bytes in the last block on disk to
                 * zero, so return.
@@ -456,10 +403,8 @@ xfs_zero_last_block(
         */
        XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
        loff = XFS_FSB_TO_B(mp, last_fsb);
-       lsize = XFS_FSB_TO_B(mp, 1);
 
-       zero_offset = isize_fsb_offset;
-       zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
+       zero_len = mp->m_sb.sb_blocksize - zero_offset;
 
        error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
 
@@ -490,20 +435,17 @@ xfs_zero_eof(
        struct inode    *ip = LINVFS_GET_IP(vp);
        xfs_fileoff_t   start_zero_fsb;
        xfs_fileoff_t   end_zero_fsb;
-       xfs_fileoff_t   prev_zero_fsb;
        xfs_fileoff_t   zero_count_fsb;
        xfs_fileoff_t   last_fsb;
        xfs_extlen_t    buf_len_fsb;
-       xfs_extlen_t    prev_zero_count;
        xfs_mount_t     *mp;
        int             nimaps;
        int             error = 0;
        xfs_bmbt_irec_t imap;
-       loff_t          loff;
-       size_t          lsize;
 
        ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
        ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
+       ASSERT(offset > isize);
 
        mp = io->io_mount;
 
@@ -511,7 +453,7 @@ xfs_zero_eof(
         * First handle zeroing the block on which isize resides.
         * We only zero a part of that block so it is handled specially.
         */
-       error = xfs_zero_last_block(ip, io, offset, isize, end_size);
+       error = xfs_zero_last_block(ip, io, isize, end_size);
        if (error) {
                ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
                ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
@@ -539,8 +481,6 @@ xfs_zero_eof(
        }
 
        ASSERT(start_zero_fsb <= end_zero_fsb);
-       prev_zero_fsb = NULLFILEOFF;
-       prev_zero_count = 0;
        while (start_zero_fsb <= end_zero_fsb) {
                nimaps = 1;
                zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
@@ -562,10 +502,7 @@ xfs_zero_eof(
                         * that sits on a hole and sets the page as P_HOLE
                         * and calls remapf if it is a mapped file.
                         */
-                       prev_zero_fsb = NULLFILEOFF;
-                       prev_zero_count = 0;
-                       start_zero_fsb = imap.br_startoff +
-                                        imap.br_blockcount;
+                       start_zero_fsb = imap.br_startoff + imap.br_blockcount;
                        ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
                        continue;
                }
@@ -586,17 +523,15 @@ xfs_zero_eof(
                 */
                XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 
-               loff = XFS_FSB_TO_B(mp, start_zero_fsb);
-               lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
-
-               error = xfs_iozero(ip, loff, lsize, end_size);
+               error = xfs_iozero(ip,
+                                  XFS_FSB_TO_B(mp, start_zero_fsb),
+                                  XFS_FSB_TO_B(mp, buf_len_fsb),
+                                  end_size);
 
                if (error) {
                        goto out_lock;
                }
 
-               prev_zero_fsb = start_zero_fsb;
-               prev_zero_count = buf_len_fsb;
                start_zero_fsb = imap.br_startoff + buf_len_fsb;
                ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 
@@ -676,12 +611,14 @@ xfs_write(
        if (XFS_FORCED_SHUTDOWN(mp))
                return -EIO;
 
+       fs_check_frozen(vp->v_vfsp, SB_FREEZE_WRITE);
+
        if (ioflags & IO_ISDIRECT) {
                xfs_buftarg_t   *target =
                        (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
                                mp->m_rtdev_targp : mp->m_ddev_targp;
 
-               if ((pos & target->pbr_smask) || (count & target->pbr_smask))
+               if ((pos & target->bt_smask) || (count & target->bt_smask))
                        return XFS_ERROR(-EINVAL);
 
                if (!VN_CACHED(vp) && pos < i_size_read(inode))
@@ -696,7 +633,7 @@ relock:
                iolock = XFS_IOLOCK_EXCL;
                locktype = VRWLOCK_WRITE;
 
-               down(&inode->i_sem);
+               mutex_lock(&inode->i_mutex);
        } else {
                iolock = XFS_IOLOCK_SHARED;
                locktype = VRWLOCK_WRITE_DIRECT;
@@ -727,7 +664,7 @@ start:
                int             dmflags = FILP_DELAY_FLAG(file);
 
                if (need_isem)
-                       dmflags |= DM_FLAGS_ISEM;
+                       dmflags |= DM_FLAGS_IMUX;
 
                xfs_iunlock(xip, XFS_ILOCK_EXCL);
                error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
@@ -753,15 +690,10 @@ start:
                }
        }
 
-       /*
-        * On Linux, generic_file_write updates the times even if
-        * no data is copied in so long as the write had a size.
-        *
-        * We must update xfs' times since revalidate will overcopy xfs.
-        */
-       if (!(ioflags & IO_INVIS)) {
-               xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
-               inode_update_time(inode, 1);
+       if (likely(!(ioflags & IO_INVIS))) {
+               file_update_time(file);
+               xfs_ichgtime_fast(xip, inode,
+                                 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
        }
 
        /*
@@ -818,7 +750,7 @@ retry:
                if (need_isem) {
                        /* demote the lock now the cached pages are gone */
                        XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
-                       up(&inode->i_sem);
+                       mutex_unlock(&inode->i_mutex);
 
                        iolock = XFS_IOLOCK_SHARED;
                        locktype = VRWLOCK_WRITE_DIRECT;
@@ -854,7 +786,7 @@ retry:
 
        current->backing_dev_info = NULL;
 
-       if (ret == -EIOCBQUEUED)
+       if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
                ret = wait_on_sync_kiocb(iocb);
 
        if ((ret == -ENOSPC) &&
@@ -862,16 +794,25 @@ retry:
            !(ioflags & IO_INVIS)) {
 
                xfs_rwunlock(bdp, locktype);
+               if (need_isem)
+                       mutex_unlock(&inode->i_mutex);
                error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
                                DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
                                0, 0, 0); /* Delay flag intentionally  unused */
                if (error)
-                       goto out_unlock_isem;
+                       goto out_nounlocks;
+               if (need_isem)
+                       mutex_lock(&inode->i_mutex);
                xfs_rwlock(bdp, locktype);
                pos = xip->i_d.di_size;
+               ret = 0;
                goto retry;
        }
 
+       isize = i_size_read(inode);
+       if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
+               *offset = isize;
+
        if (*offset > xip->i_d.di_size) {
                xfs_ilock(xip, XFS_ILOCK_EXCL);
                if (*offset > xip->i_d.di_size) {
@@ -960,14 +901,14 @@ retry:
                                xfs_trans_set_sync(tp);
                                error = xfs_trans_commit(tp, 0, NULL);
                                xfs_iunlock(xip, XFS_ILOCK_EXCL);
-                               if (error)
-                                       goto out_unlock_internal;
                        }
+                       if (error)
+                               goto out_unlock_internal;
                }
        
                xfs_rwunlock(bdp, locktype);
                if (need_isem)
-                       up(&inode->i_sem);
+                       mutex_unlock(&inode->i_mutex);
 
                error = sync_page_range(inode, mapping, pos, ret);
                if (!error)
@@ -979,7 +920,8 @@ retry:
        xfs_rwunlock(bdp, locktype);
  out_unlock_isem:
        if (need_isem)
-               up(&inode->i_sem);
+               mutex_unlock(&inode->i_mutex);
+ out_nounlocks:
        return -error;
 }
 
@@ -996,7 +938,7 @@ xfs_bdstrat_cb(struct xfs_buf *bp)
 
        mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
        if (!XFS_FORCED_SHUTDOWN(mp)) {
-               pagebuf_iorequest(bp);
+               xfs_buf_iorequest(bp);
                return 0;
        } else {
                xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
@@ -1049,7 +991,7 @@ xfsbdstrat(
                 * if (XFS_BUF_IS_GRIO(bp)) {
                 */
 
-               pagebuf_iorequest(bp);
+               xfs_buf_iorequest(bp);
                return 0;
        }