fix readahead calculations in xfs_dir2_leaf_getdents()
[safe/jmp/linux-2.6] / fs / xfs / xfs_inode.c
index 663ff9e..b92a4fa 100644 (file)
@@ -23,7 +23,6 @@
 #include "xfs_bit.h"
 #include "xfs_log.h"
 #include "xfs_inum.h"
-#include "xfs_imap.h"
 #include "xfs_trans.h"
 #include "xfs_trans_priv.h"
 #include "xfs_sb.h"
@@ -50,7 +49,6 @@
 #include "xfs_utils.h"
 #include "xfs_dir2_trace.h"
 #include "xfs_quota.h"
-#include "xfs_acl.h"
 #include "xfs_filestream.h"
 #include "xfs_vnodeops.h"
 
@@ -134,10 +132,10 @@ STATIC int
 xfs_imap_to_bp(
        xfs_mount_t     *mp,
        xfs_trans_t     *tp,
-       xfs_imap_t      *imap,
+       struct xfs_imap *imap,
        xfs_buf_t       **bpp,
        uint            buf_flags,
-       uint            imap_flags)
+       uint            iget_flags)
 {
        int             error;
        int             i;
@@ -179,7 +177,7 @@ xfs_imap_to_bp(
                if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
                                                XFS_ERRTAG_ITOBP_INOTOBP,
                                                XFS_RANDOM_ITOBP_INOTOBP))) {
-                       if (imap_flags & XFS_IMAP_BULKSTAT) {
+                       if (iget_flags & XFS_IGET_BULKSTAT) {
                                xfs_trans_brelse(tp, bp);
                                return XFS_ERROR(EINVAL);
                        }
@@ -232,7 +230,7 @@ xfs_inotobp(
        int             *offset,
        uint            imap_flags)
 {
-       xfs_imap_t      imap;
+       struct xfs_imap imap;
        xfs_buf_t       *bp;
        int             error;
 
@@ -266,7 +264,7 @@ xfs_inotobp(
  * in once, thus we can use the mapping information stored in the inode
  * rather than calling xfs_imap().  This allows us to avoid the overhead
  * of looking at the inode btree for small block file systems
- * (see xfs_dilocate()).
+ * (see xfs_imap()).
  */
 int
 xfs_itobp(
@@ -277,17 +275,12 @@ xfs_itobp(
        xfs_buf_t       **bpp,
        uint            buf_flags)
 {
-       xfs_imap_t      imap;
        xfs_buf_t       *bp;
        int             error;
 
-       ASSERT(ip->i_blkno != 0);
-
-       imap.im_blkno = ip->i_blkno;
-       imap.im_len = ip->i_len;
-       imap.im_boffset = ip->i_boffset;
+       ASSERT(ip->i_imap.im_blkno != 0);
 
-       error = xfs_imap_to_bp(mp, tp, &imap, &bp, buf_flags, 0);
+       error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp, buf_flags, 0);
        if (error)
                return error;
 
@@ -298,7 +291,7 @@ xfs_itobp(
                return EAGAIN;
        }
 
-       *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
+       *dipp = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
        *bpp = bp;
        return 0;
 }
@@ -350,6 +343,16 @@ xfs_iformat(
                return XFS_ERROR(EFSCORRUPTED);
        }
 
+       if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
+                    !ip->i_mount->m_rtdev_targp)) {
+               xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
+                       "corrupt dinode %Lu, has realtime flag set.",
+                       ip->i_ino);
+               XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
+                                    XFS_ERRLEVEL_LOW, ip->i_mount, dip);
+               return XFS_ERROR(EFSCORRUPTED);
+       }
+
        switch (ip->i_d.di_mode & S_IFMT) {
        case S_IFIFO:
        case S_IFCHR:
@@ -430,6 +433,19 @@ xfs_iformat(
        case XFS_DINODE_FMT_LOCAL:
                atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
                size = be16_to_cpu(atp->hdr.totsize);
+
+               if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
+                       xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
+                               "corrupt inode %Lu "
+                               "(bad attr fork size %Ld).",
+                               (unsigned long long) ip->i_ino,
+                               (long long) size);
+                       XFS_CORRUPTION_ERROR("xfs_iformat(8)",
+                                            XFS_ERRLEVEL_LOW,
+                                            ip->i_mount, dip);
+                       return XFS_ERROR(EFSCORRUPTED);
+               }
+
                error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
                break;
        case XFS_DINODE_FMT_EXTENTS:
@@ -635,7 +651,7 @@ xfs_iformat_btree(
        return 0;
 }
 
-void
+STATIC void
 xfs_dinode_from_disk(
        xfs_icdinode_t          *to,
        xfs_dinode_t            *from)
@@ -764,127 +780,37 @@ xfs_dic2xflags(
 }
 
 /*
- * Allocate and initialise an xfs_inode.
- */
-STATIC struct xfs_inode *
-xfs_inode_alloc(
-       struct xfs_mount        *mp,
-       xfs_ino_t               ino)
-{
-       struct xfs_inode        *ip;
-
-       /*
-        * if this didn't occur in transactions, we could use
-        * KM_MAYFAIL and return NULL here on ENOMEM. Set the
-        * code up to do this anyway.
-        */
-       ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
-       if (!ip)
-               return NULL;
-
-       ASSERT(atomic_read(&ip->i_iocount) == 0);
-       ASSERT(atomic_read(&ip->i_pincount) == 0);
-       ASSERT(!spin_is_locked(&ip->i_flags_lock));
-       ASSERT(completion_done(&ip->i_flush));
-
-       /*
-        * initialise the VFS inode here to get failures
-        * out of the way early.
-        */
-       if (!inode_init_always(mp->m_super, VFS_I(ip))) {
-               kmem_zone_free(xfs_inode_zone, ip);
-               return NULL;
-       }
-
-       /* initialise the xfs inode */
-       ip->i_ino = ino;
-       ip->i_mount = mp;
-       ip->i_blkno = 0;
-       ip->i_len = 0;
-       ip->i_boffset =0;
-       ip->i_afp = NULL;
-       memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
-       ip->i_flags = 0;
-       ip->i_update_core = 0;
-       ip->i_update_size = 0;
-       ip->i_delayed_blks = 0;
-       memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
-       ip->i_size = 0;
-       ip->i_new_size = 0;
-
-       /*
-        * Initialize inode's trace buffers.
-        */
-#ifdef XFS_INODE_TRACE
-       ip->i_trace = ktrace_alloc(INODE_TRACE_SIZE, KM_NOFS);
-#endif
-#ifdef XFS_BMAP_TRACE
-       ip->i_xtrace = ktrace_alloc(XFS_BMAP_KTRACE_SIZE, KM_NOFS);
-#endif
-#ifdef XFS_BTREE_TRACE
-       ip->i_btrace = ktrace_alloc(XFS_BMBT_KTRACE_SIZE, KM_NOFS);
-#endif
-#ifdef XFS_RW_TRACE
-       ip->i_rwtrace = ktrace_alloc(XFS_RW_KTRACE_SIZE, KM_NOFS);
-#endif
-#ifdef XFS_ILOCK_TRACE
-       ip->i_lock_trace = ktrace_alloc(XFS_ILOCK_KTRACE_SIZE, KM_NOFS);
-#endif
-#ifdef XFS_DIR2_TRACE
-       ip->i_dir_trace = ktrace_alloc(XFS_DIR2_KTRACE_SIZE, KM_NOFS);
-#endif
-
-       return ip;
-}
-
-/*
- * Given a mount structure and an inode number, return a pointer
- * to a newly allocated in-core inode corresponding to the given
- * inode number.
- *
- * Initialize the inode's attributes and extent pointers if it
- * already has them (it will not if the inode has no links).
+ * Read the disk inode attributes into the in-core inode structure.
  */
 int
 xfs_iread(
        xfs_mount_t     *mp,
        xfs_trans_t     *tp,
-       xfs_ino_t       ino,
-       xfs_inode_t     **ipp,
+       xfs_inode_t     *ip,
        xfs_daddr_t     bno,
-       uint            imap_flags)
+       uint            iget_flags)
 {
        xfs_buf_t       *bp;
        xfs_dinode_t    *dip;
-       xfs_inode_t     *ip;
-       xfs_imap_t      imap;
        int             error;
 
-       ip = xfs_inode_alloc(mp, ino);
-       if (!ip)
-               return ENOMEM;
-
        /*
-        * Get pointers to the on-disk inode and the buffer containing it.
+        * Fill in the location information in the in-core inode.
         */
-       imap.im_blkno = bno;
-       error = xfs_imap(mp, tp, ip->i_ino, &imap, imap_flags);
+       ip->i_imap.im_blkno = bno;
+       error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags);
        if (error)
-               goto out_destroy_inode;
+               return error;
+       ASSERT(bno == 0 || bno == ip->i_imap.im_blkno);
 
        /*
-        * Fill in the fields in the inode that will be used to
-        * map the inode to its buffer from now on.
+        * Get pointers to the on-disk inode and the buffer containing it.
         */
-       ip->i_blkno = imap.im_blkno;
-       ip->i_len = imap.im_len;
-       ip->i_boffset = imap.im_boffset;
-       ASSERT(bno == 0 || bno == imap.im_blkno);
-
-       error = xfs_imap_to_bp(mp, tp, &imap, &bp, XFS_BUF_LOCK, imap_flags);
+       error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &bp,
+                              XFS_BUF_LOCK, iget_flags);
        if (error)
-               goto out_destroy_inode;
-       dip = (xfs_dinode_t *)xfs_buf_offset(bp, imap.im_boffset);
+               return error;
+       dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
 
        /*
         * If we got something that isn't an inode it means someone
@@ -966,7 +892,7 @@ xfs_iread(
         * around for a while.  This helps to keep recently accessed
         * meta-data in-core longer.
         */
-        XFS_BUF_SET_REF(bp, XFS_INO_REF);
+       XFS_BUF_SET_REF(bp, XFS_INO_REF);
 
        /*
         * Use xfs_trans_brelse() to release the buffer containing the
@@ -981,14 +907,8 @@ xfs_iread(
         * to worry about the inode being changed just because we released
         * the buffer.
         */
-       xfs_trans_brelse(tp, bp);
-       *ipp = ip;
-       return 0;
-
  out_brelse:
        xfs_trans_brelse(tp, bp);
- out_destroy_inode:
-       xfs_destroy_inode(ip);
        return error;
 }
 
@@ -1327,7 +1247,7 @@ xfs_isize_check(
  * In that case the pages will still be in memory, but the inode size
  * will never have been updated.
  */
-xfs_fsize_t
+STATIC xfs_fsize_t
 xfs_file_last_byte(
        xfs_inode_t     *ip)
 {
@@ -1347,8 +1267,10 @@ xfs_file_last_byte(
         * necessary.
         */
        if (ip->i_df.if_flags & XFS_IFEXTENTS) {
+               xfs_ilock(ip, XFS_ILOCK_SHARED);
                error = xfs_bmap_last_offset(NULL, ip, &last_block,
                        XFS_DATA_FORK);
+               xfs_iunlock(ip, XFS_ILOCK_SHARED);
                if (error) {
                        last_block = 0;
                }
@@ -1424,8 +1346,8 @@ xfs_itrunc_trace(
  * direct I/O with the truncate operation.  Also, because we hold
  * the IOLOCK in exclusive mode, we prevent new direct I/Os from being
  * started until the truncate completes and drops the lock. Essentially,
- * the vn_iowait() call forms an I/O barrier that provides strict ordering
- * between direct I/Os and the truncate operation.
+ * the xfs_ioend_wait() call forms an I/O barrier that provides strict
+ * ordering between direct I/Os and the truncate operation.
  *
  * The flags parameter can have either the value XFS_ITRUNC_DEFINITE
  * or XFS_ITRUNC_MAYBE.  The XFS_ITRUNC_MAYBE value should be used
@@ -1456,7 +1378,7 @@ xfs_itruncate_start(
 
        /* wait for the completion of any pending DIOs */
        if (new_size == 0 || new_size < ip->i_size)
-               vn_iowait(ip);
+               xfs_ioend_wait(ip);
 
        /*
         * Call toss_pages or flushinval_pages to get rid of pages
@@ -1703,10 +1625,10 @@ xfs_itruncate_finish(
                 * in this file with garbage in them once recovery
                 * runs.
                 */
-               XFS_BMAP_INIT(&free_list, &first_block);
+               xfs_bmap_init(&free_list, &first_block);
                error = xfs_bunmapi(ntp, ip,
                                    first_unmap_block, unmap_len,
-                                   XFS_BMAPI_AFLAG(fork) |
+                                   xfs_bmapi_aflag(fork) |
                                      (sync ? 0 : XFS_BMAPI_ASYNC),
                                    XFS_ITRUNC_MAX_EXTENTS,
                                    &first_block, &free_list,
@@ -1872,7 +1794,7 @@ xfs_iunlink(
                ASSERT(be32_to_cpu(dip->di_next_unlinked) == NULLAGINO);
                /* both on-disk, don't endian flip twice */
                dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
-               offset = ip->i_boffset +
+               offset = ip->i_imap.im_boffset +
                        offsetof(xfs_dinode_t, di_next_unlinked);
                xfs_trans_inode_buf(tp, ibp);
                xfs_trans_log_buf(tp, ibp, offset,
@@ -1958,7 +1880,7 @@ xfs_iunlink_remove(
                ASSERT(next_agino != 0);
                if (next_agino != NULLAGINO) {
                        dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
-                       offset = ip->i_boffset +
+                       offset = ip->i_imap.im_boffset +
                                offsetof(xfs_dinode_t, di_next_unlinked);
                        xfs_trans_inode_buf(tp, ibp);
                        xfs_trans_log_buf(tp, ibp, offset,
@@ -2021,7 +1943,7 @@ xfs_iunlink_remove(
                ASSERT(next_agino != agino);
                if (next_agino != NULLAGINO) {
                        dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
-                       offset = ip->i_boffset +
+                       offset = ip->i_imap.im_boffset +
                                offsetof(xfs_dinode_t, di_next_unlinked);
                        xfs_trans_inode_buf(tp, ibp);
                        xfs_trans_log_buf(tp, ibp, offset,
@@ -2508,64 +2430,6 @@ xfs_idata_realloc(
        ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
 }
 
-
-
-
-/*
- * Map inode to disk block and offset.
- *
- * mp -- the mount point structure for the current file system
- * tp -- the current transaction
- * ino -- the inode number of the inode to be located
- * imap -- this structure is filled in with the information necessary
- *      to retrieve the given inode from disk
- * flags -- flags to pass to xfs_dilocate indicating whether or not
- *      lookups in the inode btree were OK or not
- */
-int
-xfs_imap(
-       xfs_mount_t     *mp,
-       xfs_trans_t     *tp,
-       xfs_ino_t       ino,
-       xfs_imap_t      *imap,
-       uint            flags)
-{
-       xfs_fsblock_t   fsbno;
-       int             len;
-       int             off;
-       int             error;
-
-       fsbno = imap->im_blkno ?
-               XFS_DADDR_TO_FSB(mp, imap->im_blkno) : NULLFSBLOCK;
-       error = xfs_dilocate(mp, tp, ino, &fsbno, &len, &off, flags);
-       if (error)
-               return error;
-
-       imap->im_blkno = XFS_FSB_TO_DADDR(mp, fsbno);
-       imap->im_len = XFS_FSB_TO_BB(mp, len);
-       imap->im_agblkno = XFS_FSB_TO_AGBNO(mp, fsbno);
-       imap->im_ioffset = (ushort)off;
-       imap->im_boffset = (ushort)(off << mp->m_sb.sb_inodelog);
-
-       /*
-        * If the inode number maps to a block outside the bounds
-        * of the file system then return NULL rather than calling
-        * read_buf and panicing when we get an error from the
-        * driver.
-        */
-       if ((imap->im_blkno + imap->im_len) >
-           XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
-               xfs_fs_cmn_err(CE_ALERT, mp, "xfs_imap: "
-                       "(imap->im_blkno (0x%llx) + imap->im_len (0x%llx)) > "
-                       " XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) (0x%llx)",
-                       (unsigned long long) imap->im_blkno,
-                       (unsigned long long) imap->im_len,
-                       XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
-               return EINVAL;
-       }
-       return 0;
-}
-
 void
 xfs_idestroy_fork(
        xfs_inode_t     *ip,
@@ -2610,78 +2474,6 @@ xfs_idestroy_fork(
 }
 
 /*
- * This is called free all the memory associated with an inode.
- * It must free the inode itself and any buffers allocated for
- * if_extents/if_data and if_broot.  It must also free the lock
- * associated with the inode.
- *
- * Note: because we don't initialise everything on reallocation out
- * of the zone, we must ensure we nullify everything correctly before
- * freeing the structure.
- */
-void
-xfs_idestroy(
-       xfs_inode_t     *ip)
-{
-       switch (ip->i_d.di_mode & S_IFMT) {
-       case S_IFREG:
-       case S_IFDIR:
-       case S_IFLNK:
-               xfs_idestroy_fork(ip, XFS_DATA_FORK);
-               break;
-       }
-       if (ip->i_afp)
-               xfs_idestroy_fork(ip, XFS_ATTR_FORK);
-
-#ifdef XFS_INODE_TRACE
-       ktrace_free(ip->i_trace);
-#endif
-#ifdef XFS_BMAP_TRACE
-       ktrace_free(ip->i_xtrace);
-#endif
-#ifdef XFS_BTREE_TRACE
-       ktrace_free(ip->i_btrace);
-#endif
-#ifdef XFS_RW_TRACE
-       ktrace_free(ip->i_rwtrace);
-#endif
-#ifdef XFS_ILOCK_TRACE
-       ktrace_free(ip->i_lock_trace);
-#endif
-#ifdef XFS_DIR2_TRACE
-       ktrace_free(ip->i_dir_trace);
-#endif
-       if (ip->i_itemp) {
-               /*
-                * Only if we are shutting down the fs will we see an
-                * inode still in the AIL. If it is there, we should remove
-                * it to prevent a use-after-free from occurring.
-                */
-               xfs_log_item_t  *lip = &ip->i_itemp->ili_item;
-               struct xfs_ail  *ailp = lip->li_ailp;
-
-               ASSERT(((lip->li_flags & XFS_LI_IN_AIL) == 0) ||
-                                      XFS_FORCED_SHUTDOWN(ip->i_mount));
-               if (lip->li_flags & XFS_LI_IN_AIL) {
-                       spin_lock(&ailp->xa_lock);
-                       if (lip->li_flags & XFS_LI_IN_AIL)
-                               xfs_trans_ail_delete(ailp, lip);
-                       else
-                               spin_unlock(&ailp->xa_lock);
-               }
-               xfs_inode_item_destroy(ip);
-               ip->i_itemp = NULL;
-       }
-       /* asserts to verify all state is correct here */
-       ASSERT(atomic_read(&ip->i_iocount) == 0);
-       ASSERT(atomic_read(&ip->i_pincount) == 0);
-       ASSERT(!spin_is_locked(&ip->i_flags_lock));
-       ASSERT(completion_done(&ip->i_flush));
-       kmem_zone_free(xfs_inode_zone, ip);
-}
-
-
-/*
  * Increment the pin count of the given buffer.
  * This value is protected by ipinlock spinlock in the mount structure.
  */
@@ -2789,7 +2581,7 @@ xfs_iextents_copy(
        for (i = 0; i < nrecs; i++) {
                xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
                start_block = xfs_bmbt_get_startblock(ep);
-               if (ISNULLSTARTBLOCK(start_block)) {
+               if (isnullstartblock(start_block)) {
                        /*
                         * It's a delayed allocation extent, so skip it.
                         */
@@ -3036,7 +2828,6 @@ cluster_corrupt_out:
                        XFS_BUF_CLR_BDSTRAT_FUNC(bp);
                        XFS_BUF_UNDONE(bp);
                        XFS_BUF_STALE(bp);
-                       XFS_BUF_SHUT(bp);
                        XFS_BUF_ERROR(bp,EIO);
                        xfs_biodone(bp);
                } else {
@@ -3259,7 +3050,7 @@ xfs_iflush_int(
        }
 
        /* set *dip = inode's place in the buffer */
-       dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_boffset);
+       dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
 
        /*
         * Clear i_update_core before copying out the data.
@@ -3277,9 +3068,9 @@ xfs_iflush_int(
        SYNCHRONIZE();
 
        /*
-        * Make sure to get the latest atime from the Linux inode.
+        * Make sure to get the latest timestamps from the Linux inode.
         */
-       xfs_synchronize_atime(ip);
+       xfs_synchronize_times(ip);
 
        if (XFS_TEST_ERROR(be16_to_cpu(dip->di_magic) != XFS_DINODE_MAGIC,
                               mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
@@ -3464,8 +3255,6 @@ corrupt_out:
 
 
 #ifdef XFS_ILOCK_TRACE
-ktrace_t       *xfs_ilock_trace_buf;
-
 void
 xfs_ilock_trace(xfs_inode_t *ip, int lock, unsigned int lockflags, inst_t *ra)
 {
@@ -4048,7 +3837,7 @@ xfs_iext_inline_to_direct(
 /*
  * Resize an extent indirection array to new_size bytes.
  */
-void
+STATIC void
 xfs_iext_realloc_indirect(
        xfs_ifork_t     *ifp,           /* inode fork pointer */
        int             new_size)       /* new indirection array size */
@@ -4073,7 +3862,7 @@ xfs_iext_realloc_indirect(
 /*
  * Switch from indirection array to linear (direct) extent allocations.
  */
-void
+STATIC void
 xfs_iext_indirect_to_direct(
         xfs_ifork_t    *ifp)           /* inode fork pointer */
 {