]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
[XFS] Use KM_NOFS for incore inode extent tree allocation V2
authorDavid Chinner <david@fromorbit.com>
Wed, 13 Aug 2008 06:02:51 +0000 (16:02 +1000)
committerLachlan McIlroy <lachlan@redback.melbourne.sgi.com>
Wed, 13 Aug 2008 06:02:51 +0000 (16:02 +1000)
If we allow incore extent tree allocations to recurse into the
filesystem under memory pressure, new delayed allocations through
xfs_iomap_write_delay() can deadlock on themselves if memory
reclaim tries to write back dirty pages from that inode.

It will deadlock in xfs_iomap_write_allocate() trying to take the
ilock we already hold. This can also show up as complex ABBA deadlocks
when multiple threads are triggering memory reclaim when trying to
allocate extents.

The main cause of this is the fact that delayed allocation is not done in
a transaction, so KM_NOFS is not automatically added to the allocations to
prevent this recursion.

Mark all allocations done for the incore inode extent tree as KM_NOFS to
ensure they never recurse back into the filesystem.

Version 2: o KM_NOFS implies KM_SLEEP, so just use KM_NOFS

SGI-PV: 981498

SGI-Modid: xfs-linux-melb:xfs-kern:31726a

Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Niv Sardi <xaiki@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
fs/xfs/xfs_inode.c

index cfbafc937ee58bcbb997fc573d0f1a914a5fb298..8da67d5717c8272d9da7fb28e591a2664dffab95 100644 (file)
@@ -3707,7 +3707,7 @@ xfs_iext_add_indirect_multi(
         * (all extents past */
        if (nex2) {
                byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
-               nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_SLEEP);
+               nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
                memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
                erp->er_extcount -= nex2;
                xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
@@ -4007,8 +4007,7 @@ xfs_iext_realloc_direct(
                        ifp->if_u1.if_extents =
                                kmem_realloc(ifp->if_u1.if_extents,
                                                rnew_size,
-                                               ifp->if_real_bytes,
-                                               KM_SLEEP);
+                                               ifp->if_real_bytes, KM_NOFS);
                }
                if (rnew_size > ifp->if_real_bytes) {
                        memset(&ifp->if_u1.if_extents[ifp->if_bytes /
@@ -4067,7 +4066,7 @@ xfs_iext_inline_to_direct(
        xfs_ifork_t     *ifp,           /* inode fork pointer */
        int             new_size)       /* number of extents in file */
 {
-       ifp->if_u1.if_extents = kmem_alloc(new_size, KM_SLEEP);
+       ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
        memset(ifp->if_u1.if_extents, 0, new_size);
        if (ifp->if_bytes) {
                memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
@@ -4099,7 +4098,7 @@ xfs_iext_realloc_indirect(
        } else {
                ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
                        kmem_realloc(ifp->if_u1.if_ext_irec,
-                               new_size, size, KM_SLEEP);
+                               new_size, size, KM_NOFS);
        }
 }
 
@@ -4341,11 +4340,10 @@ xfs_iext_irec_init(
        nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
        ASSERT(nextents <= XFS_LINEAR_EXTS);
 
-       erp = (xfs_ext_irec_t *)
-               kmem_alloc(sizeof(xfs_ext_irec_t), KM_SLEEP);
+       erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
 
        if (nextents == 0) {
-               ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
+               ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
        } else if (!ifp->if_real_bytes) {
                xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
        } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
@@ -4393,7 +4391,7 @@ xfs_iext_irec_new(
 
        /* Initialize new extent record */
        erp = ifp->if_u1.if_ext_irec;
-       erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_SLEEP);
+       erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
        ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
        memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
        erp[erp_idx].er_extcount = 0;