Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / fs / xfs / linux-2.6 / xfs_lrw.c
index 67efe33..ee788b1 100644 (file)
@@ -23,7 +23,6 @@
 #include "xfs_trans.h"
 #include "xfs_sb.h"
 #include "xfs_ag.h"
-#include "xfs_dir.h"
 #include "xfs_dir2.h"
 #include "xfs_alloc.h"
 #include "xfs_dmapi.h"
@@ -32,7 +31,6 @@
 #include "xfs_bmap_btree.h"
 #include "xfs_alloc_btree.h"
 #include "xfs_ialloc_btree.h"
-#include "xfs_dir_sf.h"
 #include "xfs_dir2_sf.h"
 #include "xfs_attr_sf.h"
 #include "xfs_dinode.h"
@@ -206,7 +204,7 @@ xfs_read(
        xfs_fsize_t             n;
        xfs_inode_t             *ip;
        xfs_mount_t             *mp;
-       vnode_t                 *vp;
+       bhv_vnode_t             *vp;
        unsigned long           seg;
 
        ip = XFS_BHVTOI(bdp);
@@ -258,7 +256,7 @@ xfs_read(
 
        if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
            !(ioflags & IO_INVIS)) {
-               vrwlock_t locktype = VRWLOCK_READ;
+               bhv_vrwlock_t locktype = VRWLOCK_READ;
                int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
 
                ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
@@ -266,14 +264,19 @@ xfs_read(
                                        dmflags, &locktype);
                if (ret) {
                        xfs_iunlock(ip, XFS_IOLOCK_SHARED);
-                       goto unlock_mutex;
+                       if (unlikely(ioflags & IO_ISDIRECT))
+                               mutex_unlock(&inode->i_mutex);
+                       return ret;
                }
        }
 
        if (unlikely((ioflags & IO_ISDIRECT) && VN_CACHED(vp)))
-               VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(*offset)),
+               bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
                                                -1, FI_REMAPF_LOCKED);
 
+       if (unlikely(ioflags & IO_ISDIRECT))
+               mutex_unlock(&inode->i_mutex);
+
        xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
                                (void *)iovp, segs, *offset, ioflags);
        ret = __generic_file_aio_read(iocb, iovp, segs, offset);
@@ -283,10 +286,6 @@ xfs_read(
                XFS_STATS_ADD(xs_read_bytes, ret);
 
        xfs_iunlock(ip, XFS_IOLOCK_SHARED);
-
-unlock_mutex:
-       if (unlikely(ioflags & IO_ISDIRECT))
-               mutex_unlock(&inode->i_mutex);
        return ret;
 }
 
@@ -313,7 +312,7 @@ xfs_sendfile(
 
        if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
            (!(ioflags & IO_INVIS))) {
-               vrwlock_t locktype = VRWLOCK_READ;
+               bhv_vrwlock_t locktype = VRWLOCK_READ;
                int error;
 
                error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
@@ -357,7 +356,7 @@ xfs_splice_read(
 
        if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
            (!(ioflags & IO_INVIS))) {
-               vrwlock_t locktype = VRWLOCK_READ;
+               bhv_vrwlock_t locktype = VRWLOCK_READ;
                int error;
 
                error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
@@ -392,6 +391,8 @@ xfs_splice_write(
        xfs_inode_t             *ip = XFS_BHVTOI(bdp);
        xfs_mount_t             *mp = ip->i_mount;
        ssize_t                 ret;
+       struct inode            *inode = outfilp->f_mapping->host;
+       xfs_fsize_t             isize;
 
        XFS_STATS_INC(xs_write_calls);
        if (XFS_FORCED_SHUTDOWN(ip->i_mount))
@@ -401,7 +402,7 @@ xfs_splice_write(
 
        if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_WRITE) &&
            (!(ioflags & IO_INVIS))) {
-               vrwlock_t locktype = VRWLOCK_WRITE;
+               bhv_vrwlock_t locktype = VRWLOCK_WRITE;
                int error;
 
                error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, BHV_TO_VNODE(bdp),
@@ -418,6 +419,20 @@ xfs_splice_write(
        if (ret > 0)
                XFS_STATS_ADD(xs_write_bytes, ret);
 
+       isize = i_size_read(inode);
+       if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
+               *ppos = isize;
+
+       if (*ppos > ip->i_d.di_size) {
+               xfs_ilock(ip, XFS_ILOCK_EXCL);
+               if (*ppos > ip->i_d.di_size) {
+                       ip->i_d.di_size = *ppos;
+                       i_size_write(inode, *ppos);
+                       ip->i_update_core = 1;
+                       ip->i_update_size = 1;
+               }
+               xfs_iunlock(ip, XFS_ILOCK_EXCL);
+       }
        xfs_iunlock(ip, XFS_IOLOCK_EXCL);
        return ret;
 }
@@ -458,7 +473,7 @@ xfs_zero_last_block(
        last_fsb = XFS_B_TO_FSBT(mp, isize);
        nimaps = 1;
        error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
-                         &nimaps, NULL);
+                         &nimaps, NULL, NULL);
        if (error) {
                return error;
        }
@@ -499,7 +514,7 @@ xfs_zero_last_block(
 
 int                                    /* error (positive) */
 xfs_zero_eof(
-       vnode_t         *vp,
+       bhv_vnode_t     *vp,
        xfs_iocore_t    *io,
        xfs_off_t       offset,         /* starting I/O offset */
        xfs_fsize_t     isize,          /* current inode size */
@@ -510,7 +525,6 @@ xfs_zero_eof(
        xfs_fileoff_t   end_zero_fsb;
        xfs_fileoff_t   zero_count_fsb;
        xfs_fileoff_t   last_fsb;
-       xfs_extlen_t    buf_len_fsb;
        xfs_mount_t     *mp = io->io_mount;
        int             nimaps;
        int             error = 0;
@@ -556,7 +570,7 @@ xfs_zero_eof(
                nimaps = 1;
                zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
                error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
-                                 0, NULL, 0, &imap, &nimaps, NULL);
+                                 0, NULL, 0, &imap, &nimaps, NULL, NULL);
                if (error) {
                        ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
                        ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
@@ -579,16 +593,7 @@ xfs_zero_eof(
                }
 
                /*
-                * There are blocks in the range requested.
-                * Zero them a single write at a time.  We actually
-                * don't zero the entire range returned if it is
-                * too big and simply loop around to get the rest.
-                * That is not the most efficient thing to do, but it
-                * is simple and this path should not be exercised often.
-                */
-               buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
-                                             mp->m_writeio_blocks << 8);
-               /*
+                * There are blocks we need to zero.
                 * Drop the inode lock while we're doing the I/O.
                 * We'll still have the iolock to protect us.
                 */
@@ -596,14 +601,13 @@ xfs_zero_eof(
 
                error = xfs_iozero(ip,
                                   XFS_FSB_TO_B(mp, start_zero_fsb),
-                                  XFS_FSB_TO_B(mp, buf_len_fsb),
+                                  XFS_FSB_TO_B(mp, imap.br_blockcount),
                                   end_size);
-
                if (error) {
                        goto out_lock;
                }
 
-               start_zero_fsb = imap.br_startoff + buf_len_fsb;
+               start_zero_fsb = imap.br_startoff + imap.br_blockcount;
                ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 
                XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
@@ -637,11 +641,11 @@ xfs_write(
        ssize_t                 ret = 0, error = 0;
        xfs_fsize_t             isize, new_size;
        xfs_iocore_t            *io;
-       vnode_t                 *vp;
+       bhv_vnode_t             *vp;
        unsigned long           seg;
        int                     iolock;
        int                     eventsent = 0;
-       vrwlock_t               locktype;
+       bhv_vrwlock_t           locktype;
        size_t                  ocount = 0, count;
        loff_t                  pos;
        int                     need_i_mutex = 1, need_flush = 0;
@@ -679,11 +683,11 @@ xfs_write(
        io = &xip->i_iocore;
        mp = io->io_mount;
 
+       vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_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) ?
@@ -814,7 +818,7 @@ retry:
                if (need_flush) {
                        xfs_inval_cached_trace(io, pos, -1,
                                        ctooff(offtoct(pos)), -1);
-                       VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(pos)),
+                       bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
                                        -1, FI_REMAPF_LOCKED);
                }
 
@@ -903,79 +907,9 @@ retry:
 
        /* Handle various SYNC-type writes */
        if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
-               /*
-                * If we're treating this as O_DSYNC and we have not updated the
-                * size, force the log.
-                */
-               if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC) &&
-                   !(xip->i_update_size)) {
-                       xfs_inode_log_item_t    *iip = xip->i_itemp;
-
-                       /*
-                        * If an allocation transaction occurred
-                        * without extending the size, then we have to force
-                        * the log up the proper point to ensure that the
-                        * allocation is permanent.  We can't count on
-                        * the fact that buffered writes lock out direct I/O
-                        * writes - the direct I/O write could have extended
-                        * the size nontransactionally, then finished before
-                        * we started.  xfs_write_file will think that the file
-                        * didn't grow but the update isn't safe unless the
-                        * size change is logged.
-                        *
-                        * Force the log if we've committed a transaction
-                        * against the inode or if someone else has and
-                        * the commit record hasn't gone to disk (e.g.
-                        * the inode is pinned).  This guarantees that
-                        * all changes affecting the inode are permanent
-                        * when we return.
-                        */
-                       if (iip && iip->ili_last_lsn) {
-                               xfs_log_force(mp, iip->ili_last_lsn,
-                                               XFS_LOG_FORCE | XFS_LOG_SYNC);
-                       } else if (xfs_ipincount(xip) > 0) {
-                               xfs_log_force(mp, (xfs_lsn_t)0,
-                                               XFS_LOG_FORCE | XFS_LOG_SYNC);
-                       }
-
-               } else {
-                       xfs_trans_t     *tp;
-
-                       /*
-                        * O_SYNC or O_DSYNC _with_ a size update are handled
-                        * the same way.
-                        *
-                        * If the write was synchronous then we need to make
-                        * sure that the inode modification time is permanent.
-                        * We'll have updated the timestamp above, so here
-                        * we use a synchronous transaction to log the inode.
-                        * It's not fast, but it's necessary.
-                        *
-                        * If this a dsync write and the size got changed
-                        * non-transactionally, then we need to ensure that
-                        * the size change gets logged in a synchronous
-                        * transaction.
-                        */
-
-                       tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
-                       if ((error = xfs_trans_reserve(tp, 0,
-                                                     XFS_SWRITE_LOG_RES(mp),
-                                                     0, 0, 0))) {
-                               /* Transaction reserve failed */
-                               xfs_trans_cancel(tp, 0);
-                       } else {
-                               /* Transaction reserve successful */
-                               xfs_ilock(xip, XFS_ILOCK_EXCL);
-                               xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
-                               xfs_trans_ihold(tp, xip);
-                               xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
-                               xfs_trans_set_sync(tp);
-                               error = xfs_trans_commit(tp, 0, NULL);
-                               xfs_iunlock(xip, XFS_ILOCK_EXCL);
-                       }
-                       if (error)
-                               goto out_unlock_internal;
-               }
+               error = xfs_write_sync_logforce(mp, xip);
+               if (error)
+                       goto out_unlock_internal;
 
                xfs_rwunlock(bdp, locktype);
                if (need_i_mutex)