Use KM_PUSHPAGE instead of KM_SLEEP

It used to be the case that all KM_SLEEP allocations were GFS_NOFS.
Unfortunately this often resulted in the kernel being unable to
reclaim the ARC, inode, and dentry caches in a timely manor.
The fix was to make KM_SLEEP a GFP_KERNEL allocation in the SPL.

However, this increases the posibility of deadlocking the system
on a zfs write thread.  If a zfs write thread attempts to perform
an allocation it may trigger synchronous reclaim.  This reclaim
may attempt to flush dirty data/inode to disk to free memory.
Unforunately, this write cannot finish because the write thread
which would handle it is holding the previous transaction open.
Deadlock.

To avoid this all allocations in the zfs write thread path must
use KM_PUSHPAGE which prohibits synchronous reclaim for that
thread.  In this way forward progress in ensured.  The risk
with this change is I missed updating an allocation for the
write threads leaving an increased posibility of deadlock.  If
any deadlocks remain they will be unlikely but we'll have to
make sure they all get fixed.

diff --git a/module/zfs/arc.c b/module/zfs/arc.c
index 808c8e8..3a45730 100644 (file)
--- a/module/zfs/arc.c
+++ b/module/zfs/arc.c
@@ -2720,7 +2720,7 @@ top:
                                arc_callback_t  *acb = NULL;
 
                                acb = kmem_zalloc(sizeof (arc_callback_t),
-                                   KM_SLEEP);
+                                   KM_PUSHPAGE);
                                acb->acb_done = done;
                                acb->acb_private = private;
                                if (pio != NULL)
@@ -2836,7 +2836,7 @@ top:
 
                ASSERT(!GHOST_STATE(hdr->b_state));
 
-               acb = kmem_zalloc(sizeof (arc_callback_t), KM_SLEEP);
+               acb = kmem_zalloc(sizeof (arc_callback_t), KM_PUSHPAGE);
                acb->acb_done = done;
                acb->acb_private = private;
 
@@ -2885,7 +2885,7 @@ top:
                                ARCSTAT_BUMP(arcstat_l2_hits);
 
                                cb = kmem_zalloc(sizeof (l2arc_read_callback_t),
-                                   KM_SLEEP);
+                                   KM_PUSHPAGE);
                                cb->l2rcb_buf = buf;
                                cb->l2rcb_spa = spa;
                                cb->l2rcb_bp = *bp;

diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index 87e3a36..1017414 100644 (file)
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -174,7 +174,7 @@ zfs_create_share_dir(zfs_sb_t *zsb, dmu_tx_t *tx)
        vattr.va_uid = crgetuid(kcred);
        vattr.va_gid = crgetgid(kcred);
 
-       sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
+       sharezp = kmem_cache_alloc(znode_cache, KM_PUSHPAGE);
        sharezp->z_moved = 0;
        sharezp->z_unlinked = 0;
        sharezp->z_atime_dirty = 0;
@@ -248,7 +248,7 @@ zfs_inode_alloc(struct super_block *sb, struct inode **ip)
 {
        znode_t *zp;
 
-       zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
+       zp = kmem_cache_alloc(znode_cache, KM_PUSHPAGE);
        *ip = ZTOI(zp);
 
        return (0);

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 3c256aa..2c38e10 100644 (file)
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -255,7 +255,7 @@ static void
 zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize,
        zio_transform_func_t *transform)
 {
-       zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP);
+       zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE);
 
        zt->zt_orig_data = zio->io_data;
        zt->zt_orig_size = zio->io_size;
@@ -370,7 +370,7 @@ zio_unique_parent(zio_t *cio)
 void
 zio_add_child(zio_t *pio, zio_t *cio)
 {
-       zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP);
+       zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE);
        int w;
 
        /*
@@ -491,7 +491,7 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
        ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER));
        ASSERT(vd || stage == ZIO_STAGE_OPEN);
 
-       zio = kmem_cache_alloc(zio_cache, KM_SLEEP);
+       zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE);
        bzero(zio, sizeof (zio_t));
 
        mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
@@ -1503,7 +1503,7 @@ zio_gang_node_alloc(zio_gang_node_t **gnpp)
 
        ASSERT(*gnpp == NULL);
 
-       gn = kmem_zalloc(sizeof (*gn), KM_SLEEP);
+       gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE);
        gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE);
        *gnpp = gn;