From: Fred Isaman <fred.isaman@xxxxxxxxx>
We encountered a deadlock where the SEQUENCE that accompanied the
LAYOUTGET triggered a session drain, while ff_layout_alloc_lseg
triggered a GETDEVICEINFO. The GETDEVICEINFO hung waiting for the
session drain, while the LAYOUTGET held the slot waiting for
alloc_lseg to finish.
Avoid this by moving the call to nfs4_find_get_deviceid out of
ff_layout_alloc_lseg and into nfs4_ff_layout_prepare_ds.
Signed-off-by: Fred Isaman <fred.isaman@xxxxxxxxx>
[dros@xxxxxxxxxxxxxxx: pNFS/flexfiles: fix races in ff_layout_mirror_valid]
Signed-off-by: Weston Andros Adamson <dros@xxxxxxxxxxxxxxx>
Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
---
fs/nfs/flexfilelayout/flexfilelayout.c | 37 ++---------------------
fs/nfs/flexfilelayout/flexfilelayout.h | 2 +-
fs/nfs/flexfilelayout/flexfilelayoutdev.c | 50 ++++++++++++++++++++++++-------
3 files changed, 43 insertions(+), 46 deletions(-)
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
index ca1012a42e14..ef4c9d17d4a5 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -183,7 +183,7 @@ ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
spin_lock(&inode->i_lock);
list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
- if (mirror->mirror_ds != pos->mirror_ds)
+ if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
continue;
if (!ff_mirror_match_fh(mirror, pos))
continue;
@@ -360,19 +360,6 @@ static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
}
}
-static void ff_layout_mark_devices_valid(struct nfs4_ff_layout_segment *fls)
-{
- struct nfs4_deviceid_node *node;
- int i;
-
- if (!(fls->flags & FF_FLAGS_NO_IO_THRU_MDS))
- return;
- for (i = 0; i < fls->mirror_array_cnt; i++) {
- node = &fls->mirror_array[i]->mirror_ds->id_node;
- clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
- }
-}
-
static struct pnfs_layout_segment *
ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
struct nfs4_layoutget_res *lgr,
@@ -426,8 +413,6 @@ ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
for (i = 0; i < fls->mirror_array_cnt; i++) {
struct nfs4_ff_layout_mirror *mirror;
- struct nfs4_deviceid devid;
- struct nfs4_deviceid_node *idnode;
struct auth_cred acred = { .group_info = ff_zero_group };
struct rpc_cred __rcu *cred;
u32 ds_count, fh_count, id;
@@ -452,24 +437,10 @@ ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
fls->mirror_array[i]->ds_count = ds_count;
/* deviceid */
- rc = decode_deviceid(&stream, &devid);
+ rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
if (rc)
goto out_err_free;
- idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
- &devid, lh->plh_lc_cred,
- gfp_flags);
- /*
- * upon success, mirror_ds is allocated by previous
- * getdeviceinfo, or newly by .alloc_deviceid_node
- * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
- */
- if (idnode)
- fls->mirror_array[i]->mirror_ds =
- FF_LAYOUT_MIRROR_DS(idnode);
- else
- goto out_err_free;
-
/* efficiency */
rc = -EIO;
p = xdr_inline_decode(&stream, 4);
@@ -567,8 +538,6 @@ ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
rc = ff_layout_check_layout(lgr);
if (rc)
goto out_err_free;
- ff_layout_mark_devices_valid(fls);
-
ret = &fls->generic_hdr;
dprintk("<-- %s (success)\n", __func__);
out_free_page:
@@ -2332,7 +2301,7 @@ ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
if (i >= dev_limit)
break;
- if (!mirror->mirror_ds)
+ if (IS_ERR_OR_NULL(mirror->mirror_ds))
continue;
if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
continue;
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.h b/fs/nfs/flexfilelayout/flexfilelayout.h
index 35221fe390c5..7223c4ea8cde 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.h
+++ b/fs/nfs/flexfilelayout/flexfilelayout.h
@@ -74,6 +74,7 @@ struct nfs4_ff_layout_mirror {
struct list_head mirrors;
u32 ds_count;
u32 efficiency;
+ struct nfs4_deviceid devid;
struct nfs4_ff_layout_ds *mirror_ds;
u32 fh_versions_cnt;
struct nfs_fh *fh_versions;
@@ -211,7 +212,6 @@ nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg,
struct inode *inode);
struct rpc_cred *ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg,
u32 ds_idx, struct rpc_cred *mdscred);
-bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg);
bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg);
diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
index 142bfd0b1663..3cc39d1c1206 100644
--- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
+++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
@@ -20,9 +20,11 @@
static unsigned int dataserver_timeo = NFS_DEF_TCP_RETRANS;
static unsigned int dataserver_retrans;
+static bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
+
void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
{
- if (mirror_ds)
+ if (!IS_ERR_OR_NULL(mirror_ds))
nfs4_put_deviceid_node(&mirror_ds->id_node);
}
@@ -182,12 +184,29 @@ static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
}
static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
- struct nfs4_ff_layout_mirror *mirror)
+ struct nfs4_ff_layout_mirror *mirror,
+ bool create)
{
- if (mirror == NULL || mirror->mirror_ds == NULL) {
- pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
- lseg);
- return false;
+ if (mirror == NULL || IS_ERR(mirror->mirror_ds))
+ goto outerr;
+ if (mirror->mirror_ds == NULL) {
+ if (create) {
+ struct nfs4_deviceid_node *node;
+ struct pnfs_layout_hdr *lh = lseg->pls_layout;
+ struct nfs4_ff_layout_ds *mirror_ds = ERR_PTR(-ENODEV);
+
+ node = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
+ &mirror->devid, lh->plh_lc_cred,
+ GFP_KERNEL);
+ if (node)
+ mirror_ds = FF_LAYOUT_MIRROR_DS(node);
+
+ /* check for race with another call to this function */
+ if (cmpxchg(&mirror->mirror_ds, NULL, mirror_ds) &&
+ mirror_ds != ERR_PTR(-ENODEV))
+ nfs4_put_deviceid_node(node);
+ } else
+ goto outerr;
}
if (mirror->mirror_ds->ds == NULL) {
struct nfs4_deviceid_node *devid;
@@ -196,6 +215,9 @@ static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
return false;
}
return true;
+outerr:
+ pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
+ return false;
}
static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
@@ -323,7 +345,7 @@ nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
struct nfs_fh *fh = NULL;
- if (!ff_layout_mirror_valid(lseg, mirror)) {
+ if (!ff_layout_mirror_valid(lseg, mirror, false)) {
pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
__func__, mirror_idx);
goto out;
@@ -363,7 +385,7 @@ nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
struct nfs_server *s = NFS_SERVER(ino);
unsigned int max_payload;
- if (!ff_layout_mirror_valid(lseg, mirror)) {
+ if (!ff_layout_mirror_valid(lseg, mirror, true)) {
pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
__func__, ds_idx);
goto out;
@@ -547,7 +569,11 @@ static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
mirror = FF_LAYOUT_COMP(lseg, idx);
- if (mirror && mirror->mirror_ds) {
+ if (mirror) {
+ if (!mirror->mirror_ds)
+ return true;
+ if (IS_ERR(mirror->mirror_ds))
+ continue;
devid = &mirror->mirror_ds->id_node;
if (!ff_layout_test_devid_unavailable(devid))
return true;
@@ -565,8 +591,10 @@ static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
mirror = FF_LAYOUT_COMP(lseg, idx);
- if (!mirror || !mirror->mirror_ds)
+ if (!mirror || IS_ERR(mirror->mirror_ds))
return false;
+ if (!mirror->mirror_ds)
+ continue;
devid = &mirror->mirror_ds->id_node;
if (ff_layout_test_devid_unavailable(devid))
return false;
@@ -575,7 +603,7 @@ static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
}
-bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
+static bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
{
if (lseg->pls_range.iomode == IOMODE_READ)
return ff_read_layout_has_available_ds(lseg);
--
2.9.3
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
