xfs: remove the i_size field in struct xfs_inode
authorChristoph Hellwig <hch@infradead.org>
Sun, 18 Dec 2011 20:00:11 +0000 (20:00 +0000)
committerBen Myers <bpm@sgi.com>
Tue, 17 Jan 2012 21:08:53 +0000 (15:08 -0600)
There is no fundamental need to keep an in-memory inode size copy in the XFS
inode.  We already have the on-disk value in the dinode, and the separate
in-memory copy that we need for regular files only in the XFS inode.

Remove the xfs_inode i_size field and change the XFS_ISIZE macro to use the
VFS inode i_size field for regular files.  Switch code that was directly
accessing the i_size field in the xfs_inode to XFS_ISIZE, or in cases where
we are limited to regular files direct access of the VFS inode i_size field.

This also allows dropping some fairly complicated code in the write path
which dealt with keeping the xfs_inode i_size uptodate with the VFS i_size
that is getting updated inside ->write_end.

Note that we do not bother resetting the VFS i_size when truncating a file
that gets freed to zero as there is no point in doing so because the VFS inode
is no longer in use at this point.  Just relax the assert in xfs_ifree to
only check the on-disk size instead.

Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Myers <bpm@sgi.com>
12 files changed:
fs/xfs/xfs_aops.c
fs/xfs/xfs_bmap.c
fs/xfs/xfs_file.c
fs/xfs/xfs_fs_subr.c
fs/xfs/xfs_iget.c
fs/xfs/xfs_inode.c
fs/xfs/xfs_inode.h
fs/xfs/xfs_iomap.c
fs/xfs/xfs_iops.c
fs/xfs/xfs_qm_syscalls.c
fs/xfs/xfs_trace.h
fs/xfs/xfs_vnodeops.c

index 574d4ee9b6253ea3d589f23288aa7385907642bb..4d27ea117e0ece3d5b34e20a884002fafda8c8e9 100644 (file)
@@ -111,7 +111,7 @@ xfs_ioend_new_eof(
        xfs_fsize_t             bsize;
 
        bsize = ioend->io_offset + ioend->io_size;
-       isize = MAX(ip->i_size, ip->i_new_size);
+       isize = MAX(i_size_read(VFS_I(ip)), ip->i_new_size);
        isize = MIN(isize, bsize);
        return isize > ip->i_d.di_size ? isize : 0;
 }
index 7a888ca2f7f6a96958cc191db83c1944908ef708..188ef2fbd62880614a29ea0432e20707d5cf45a2 100644 (file)
@@ -3997,11 +3997,8 @@ xfs_bmap_one_block(
        xfs_bmbt_irec_t s;              /* internal version of extent */
 
 #ifndef DEBUG
-       if (whichfork == XFS_DATA_FORK) {
-               return S_ISREG(ip->i_d.di_mode) ?
-                       (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
-                       (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
-       }
+       if (whichfork == XFS_DATA_FORK)
+               return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
 #endif /* !DEBUG */
        if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
                return 0;
@@ -4013,7 +4010,7 @@ xfs_bmap_one_block(
        xfs_bmbt_get_all(ep, &s);
        rval = s.br_startoff == 0 && s.br_blockcount == 1;
        if (rval && whichfork == XFS_DATA_FORK)
-               ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
+               ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
        return rval;
 }
 
@@ -5427,7 +5424,7 @@ xfs_getbmapx_fix_eof_hole(
        if (startblock == HOLESTARTBLOCK) {
                mp = ip->i_mount;
                out->bmv_block = -1;
-               fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
+               fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
                fixlen -= out->bmv_offset;
                if (prealloced && out->bmv_offset + out->bmv_length == end) {
                        /* Came to hole at EOF. Trim it. */
@@ -5515,7 +5512,7 @@ xfs_getbmap(
                        fixlen = XFS_MAXIOFFSET(mp);
                } else {
                        prealloced = 0;
-                       fixlen = ip->i_size;
+                       fixlen = XFS_ISIZE(ip);
                }
        }
 
@@ -5544,7 +5541,7 @@ xfs_getbmap(
 
        xfs_ilock(ip, XFS_IOLOCK_SHARED);
        if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
-               if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
+               if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
                        error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
                        if (error)
                                goto out_unlock_iolock;
index f675f3d9d7b3b3693cfabc0437d52ca31a4f598a..86d5dc2604642d45074f3876fb656116897e2ef7 100644 (file)
@@ -327,7 +327,7 @@ xfs_file_aio_read(
                                mp->m_rtdev_targp : mp->m_ddev_targp;
                if ((iocb->ki_pos & target->bt_smask) ||
                    (size & target->bt_smask)) {
-                       if (iocb->ki_pos == ip->i_size)
+                       if (iocb->ki_pos == i_size_read(inode))
                                return 0;
                        return -XFS_ERROR(EINVAL);
                }
@@ -412,30 +412,6 @@ xfs_file_splice_read(
        return ret;
 }
 
-STATIC void
-xfs_aio_write_isize_update(
-       struct inode    *inode,
-       loff_t          *ppos,
-       ssize_t         bytes_written)
-{
-       struct xfs_inode        *ip = XFS_I(inode);
-       xfs_fsize_t             isize = i_size_read(inode);
-
-       if (bytes_written > 0)
-               XFS_STATS_ADD(xs_write_bytes, bytes_written);
-
-       if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
-                                       *ppos > isize))
-               *ppos = isize;
-
-       if (*ppos > ip->i_size) {
-               xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
-               if (*ppos > ip->i_size)
-                       ip->i_size = *ppos;
-               xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
-       }
-}
-
 /*
  * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
  * part of the I/O may have been written to disk before the error occurred.  In
@@ -451,8 +427,8 @@ xfs_aio_write_newsize_update(
                xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
                if (new_size == ip->i_new_size)
                        ip->i_new_size = 0;
-               if (ip->i_d.di_size > ip->i_size)
-                       ip->i_d.di_size = ip->i_size;
+               if (ip->i_d.di_size > i_size_read(VFS_I(ip)))
+                       ip->i_d.di_size = i_size_read(VFS_I(ip));
                xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
        }
 }
@@ -492,15 +468,16 @@ xfs_file_splice_write(
        new_size = *ppos + count;
 
        xfs_ilock(ip, XFS_ILOCK_EXCL);
-       if (new_size > ip->i_size)
+       if (new_size > i_size_read(inode))
                ip->i_new_size = new_size;
        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 
        trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
 
        ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
+       if (ret > 0)
+               XFS_STATS_ADD(xs_write_bytes, ret);
 
-       xfs_aio_write_isize_update(inode, ppos, ret);
        xfs_aio_write_newsize_update(ip, new_size);
        xfs_iunlock(ip, XFS_IOLOCK_EXCL);
        return ret;
@@ -728,14 +705,14 @@ restart:
         * values are still valid.
         */
        if ((ip->i_new_size && *pos > ip->i_new_size) ||
-           (!ip->i_new_size && *pos > ip->i_size)) {
+           (!ip->i_new_size && *pos > i_size_read(inode))) {
                if (*iolock == XFS_IOLOCK_SHARED) {
                        xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
                        *iolock = XFS_IOLOCK_EXCL;
                        xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
                        goto restart;
                }
-               error = -xfs_zero_eof(ip, *pos, ip->i_size);
+               error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
        }
 
        /*
@@ -744,7 +721,7 @@ restart:
         * ip->i_new_size if this IO ends beyond any other in-flight writes.
         */
        new_size = *pos + *count;
-       if (new_size > ip->i_size) {
+       if (new_size > i_size_read(inode)) {
                if (new_size > ip->i_new_size)
                        ip->i_new_size = new_size;
                *new_sizep = new_size;
@@ -957,11 +934,11 @@ xfs_file_aio_write(
                ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
                                                ocount, &new_size, &iolock);
 
-       xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
-
        if (ret <= 0)
                goto out_unlock;
 
+       XFS_STATS_ADD(xs_write_bytes, ret);
+
        /* Handle various SYNC-type writes */
        if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
                loff_t end = pos + ret - 1;
index ed88ed16811c6ba33c1d211b3195bd1dcea18484..652b875a9d4c441bd3c759df185c153dfee0e8f5 100644 (file)
@@ -90,7 +90,7 @@ xfs_wait_on_pages(
 
        if (mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) {
                return -filemap_fdatawait_range(mapping, first,
-                                       last == -1 ? ip->i_size - 1 : last);
+                                       last == -1 ? XFS_ISIZE(ip) - 1 : last);
        }
        return 0;
 }
index a7cf7139f9ad778b60fdd146fd8c7a404db4d60b..3b5b78aa3b8788a029586894dcb474d41ddf4d65 100644 (file)
@@ -94,7 +94,6 @@ xfs_inode_alloc(
        ip->i_update_core = 0;
        ip->i_delayed_blks = 0;
        memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
-       ip->i_size = 0;
        ip->i_new_size = 0;
 
        return ip;
index 62603369b523289f7a95c03835ef4b09a146f38c..b21022499c2e8f302699f80ca2af344301fee941 100644 (file)
@@ -347,7 +347,6 @@ xfs_iformat(
                        return XFS_ERROR(EFSCORRUPTED);
                }
                ip->i_d.di_size = 0;
-               ip->i_size = 0;
                ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
                break;
 
@@ -853,7 +852,6 @@ xfs_iread(
        }
 
        ip->i_delayed_blks = 0;
-       ip->i_size = ip->i_d.di_size;
 
        /*
         * Mark the buffer containing the inode as something to keep
@@ -1043,7 +1041,6 @@ xfs_ialloc(
        }
 
        ip->i_d.di_size = 0;
-       ip->i_size = 0;
        ip->i_d.di_nextents = 0;
        ASSERT(ip->i_d.di_nblocks == 0);
 
@@ -1198,7 +1195,7 @@ xfs_itruncate_extents(
        int                     done = 0;
 
        ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
-       ASSERT(new_size <= ip->i_size);
+       ASSERT(new_size <= XFS_ISIZE(ip));
        ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
        ASSERT(ip->i_itemp != NULL);
        ASSERT(ip->i_itemp->ili_lock_flags == 0);
@@ -1712,8 +1709,7 @@ xfs_ifree(
        ASSERT(ip->i_d.di_nlink == 0);
        ASSERT(ip->i_d.di_nextents == 0);
        ASSERT(ip->i_d.di_anextents == 0);
-       ASSERT((ip->i_d.di_size == 0 && ip->i_size == 0) ||
-              (!S_ISREG(ip->i_d.di_mode)));
+       ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
        ASSERT(ip->i_d.di_nblocks == 0);
 
        /*
index 4acbe740be4683a9f3265538a6ea460d76d811fa..cd99e43fa8f058e2caf8fd9e1cc5e6a77b944f7f 100644 (file)
@@ -246,16 +246,12 @@ typedef struct xfs_inode {
 
        xfs_icdinode_t          i_d;            /* most of ondisk inode */
 
-       xfs_fsize_t             i_size;         /* in-memory size */
        xfs_fsize_t             i_new_size;     /* size when write completes */
 
        /* VFS inode */
        struct inode            i_vnode;        /* embedded VFS inode */
 } xfs_inode_t;
 
-#define XFS_ISIZE(ip)  S_ISREG((ip)->i_d.di_mode) ? \
-                               (ip)->i_size : (ip)->i_d.di_size;
-
 /* Convert from vfs inode to xfs inode */
 static inline struct xfs_inode *XFS_I(struct inode *inode)
 {
@@ -268,6 +264,18 @@ static inline struct inode *VFS_I(struct xfs_inode *ip)
        return &ip->i_vnode;
 }
 
+/*
+ * For regular files we only update the on-disk filesize when actually
+ * writing data back to disk.  Until then only the copy in the VFS inode
+ * is uptodate.
+ */
+static inline xfs_fsize_t XFS_ISIZE(struct xfs_inode *ip)
+{
+       if (S_ISREG(ip->i_d.di_mode))
+               return i_size_read(VFS_I(ip));
+       return ip->i_d.di_size;
+}
+
 /*
  * i_flags helper functions
  */
index a27a44659da617e270230685a23c03979d660def..246c7d57c6f96c876778128e8d21c90fca692ce9 100644 (file)
@@ -74,7 +74,7 @@ xfs_iomap_eof_align_last_fsb(
                else if (mp->m_dalign)
                        align = mp->m_dalign;
 
-               if (align && ip->i_size >= XFS_FSB_TO_B(mp, align))
+               if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
                        new_last_fsb = roundup_64(*last_fsb, align);
        }
 
@@ -154,7 +154,7 @@ xfs_iomap_write_direct(
 
        offset_fsb = XFS_B_TO_FSBT(mp, offset);
        last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
-       if ((offset + count) > ip->i_size) {
+       if ((offset + count) > XFS_ISIZE(ip)) {
                error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
                if (error)
                        goto error_out;
@@ -211,7 +211,7 @@ xfs_iomap_write_direct(
        xfs_trans_ijoin(tp, ip, 0);
 
        bmapi_flag = 0;
-       if (offset < ip->i_size || extsz)
+       if (offset < XFS_ISIZE(ip) || extsz)
                bmapi_flag |= XFS_BMAPI_PREALLOC;
 
        /*
@@ -286,7 +286,7 @@ xfs_iomap_eof_want_preallocate(
        int             found_delalloc = 0;
 
        *prealloc = 0;
-       if ((offset + count) <= ip->i_size)
+       if (offset + count <= XFS_ISIZE(ip))
                return 0;
 
        /*
@@ -340,7 +340,7 @@ xfs_iomap_prealloc_size(
                 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
                 * ensure we always pass in a non-zero value.
                 */
-               alloc_blocks = XFS_B_TO_FSB(mp, ip->i_size) + 1;
+               alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
                alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
                                        rounddown_pow_of_two(alloc_blocks));
 
@@ -564,7 +564,7 @@ xfs_iomap_write_allocate(
                         * back....
                         */
                        nimaps = 1;
-                       end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
+                       end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
                        error = xfs_bmap_last_offset(NULL, ip, &last_block,
                                                        XFS_DATA_FORK);
                        if (error)
index f02eaa298d3c6878b0c34fcfd2170a710c725ced..ab302539e5b9603b8a67bb9f4399c03625fe1fd7 100644 (file)
@@ -778,7 +778,7 @@ xfs_setattr_size(
                lock_flags |= XFS_IOLOCK_EXCL;
        xfs_ilock(ip, lock_flags);
 
-       oldsize = ip->i_size;
+       oldsize = inode->i_size;
        newsize = iattr->ia_size;
 
        /*
@@ -897,7 +897,6 @@ xfs_setattr_size(
         * they get written to.
         */
        ip->i_d.di_size = newsize;
-       ip->i_size = newsize;
        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
        if (newsize <= oldsize) {
index 27378650b5cbd81e40de45a1df0a22a4f8f86060..eafbcff81f3af43c9dae0a73175bc71abba4a7c5 100644 (file)
@@ -265,7 +265,6 @@ xfs_qm_scall_trunc_qfile(
        xfs_trans_ijoin(tp, ip, 0);
 
        ip->i_d.di_size = 0;
-       ip->i_size = 0;
        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
        error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
index 81efa04161739d2111fc7cddda575b6bcb4ba7a1..2aabcc9c507e1ad85a881261b760b281c6f2c1ea 100644 (file)
@@ -1038,7 +1038,7 @@ DECLARE_EVENT_CLASS(xfs_simple_io_class,
        TP_fast_assign(
                __entry->dev = VFS_I(ip)->i_sb->s_dev;
                __entry->ino = ip->i_ino;
-               __entry->isize = ip->i_size;
+               __entry->isize = VFS_I(ip)->i_size;
                __entry->disize = ip->i_d.di_size;
                __entry->new_size = ip->i_new_size;
                __entry->offset = offset;
index 96ff03421753a9e3a9f183f7d257335b3adbcc5d..0cf52da9d2468a547a614f55cf0191a396f82574 100644 (file)
@@ -175,7 +175,7 @@ xfs_free_eofblocks(
         * Figure out if there are any blocks beyond the end
         * of the file.  If not, then there is nothing to do.
         */
-       end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_size));
+       end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
        last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
        if (last_fsb <= end_fsb)
                return 0;
@@ -233,7 +233,7 @@ xfs_free_eofblocks(
                 * may be full of holes (ie NULL files bug).
                 */
                error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
-                                             ip->i_size);
+                                             XFS_ISIZE(ip));
                if (error) {
                        /*
                         * If we get an error at this point we simply don't
@@ -547,8 +547,8 @@ xfs_release(
                return 0;
 
        if ((S_ISREG(ip->i_d.di_mode) &&
-            ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
-              ip->i_delayed_blks > 0)) &&
+            (VFS_I(ip)->i_size > 0 ||
+             (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
             (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
            (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))) {
 
@@ -625,7 +625,7 @@ xfs_inactive(
         * only one with a reference to the inode.
         */
        truncate = ((ip->i_d.di_nlink == 0) &&
-           ((ip->i_d.di_size != 0) || (ip->i_size != 0) ||
+           ((ip->i_d.di_size != 0) || XFS_ISIZE(ip) != 0 ||
             (ip->i_d.di_nextents > 0) || (ip->i_delayed_blks > 0)) &&
            S_ISREG(ip->i_d.di_mode));
 
@@ -639,12 +639,12 @@ xfs_inactive(
 
        if (ip->i_d.di_nlink != 0) {
                if ((S_ISREG(ip->i_d.di_mode) &&
-                     ((ip->i_size > 0) || (VN_CACHED(VFS_I(ip)) > 0 ||
-                       ip->i_delayed_blks > 0)) &&
-                     (ip->i_df.if_flags & XFS_IFEXTENTS) &&
-                    (!(ip->i_d.di_flags &
+                   (VFS_I(ip)->i_size > 0 ||
+                    (VN_CACHED(VFS_I(ip)) > 0 || ip->i_delayed_blks > 0)) &&
+                   (ip->i_df.if_flags & XFS_IFEXTENTS) &&
+                   (!(ip->i_d.di_flags &
                                (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) ||
-                     (ip->i_delayed_blks != 0)))) {
+                    ip->i_delayed_blks != 0))) {
                        error = xfs_free_eofblocks(mp, ip, 0);
                        if (error)
                                return VN_INACTIVE_CACHE;
@@ -678,7 +678,6 @@ xfs_inactive(
                xfs_trans_ijoin(tp, ip, 0);
 
                ip->i_d.di_size = 0;
-               ip->i_size = 0;
                xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 
                error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
@@ -1974,11 +1973,11 @@ xfs_zero_remaining_bytes(
         * since nothing can read beyond eof.  The space will
         * be zeroed when the file is extended anyway.
         */
-       if (startoff >= ip->i_size)
+       if (startoff >= XFS_ISIZE(ip))
                return 0;
 
-       if (endoff > ip->i_size)
-               endoff = ip->i_size;
+       if (endoff > XFS_ISIZE(ip))
+               endoff = XFS_ISIZE(ip);
 
        bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
                                        mp->m_rtdev_targp : mp->m_ddev_targp,
@@ -2273,7 +2272,7 @@ xfs_change_file_space(
                bf->l_start += offset;
                break;
        case 2: /*SEEK_END*/
-               bf->l_start += ip->i_size;
+               bf->l_start += XFS_ISIZE(ip);
                break;
        default:
                return XFS_ERROR(EINVAL);
@@ -2290,7 +2289,7 @@ xfs_change_file_space(
        bf->l_whence = 0;
 
        startoffset = bf->l_start;
-       fsize = ip->i_size;
+       fsize = XFS_ISIZE(ip);
 
        /*
         * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
This page took 0.175765 seconds and 5 git commands to generate.