From 2a4c8994eeef50796015f8a2005e4a75c1929166 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 14 Jun 2012 13:08:38 -0400 Subject: NFSv4.1: Fix umount when filelayout DS is also the MDS Currently there is a 'chicken and egg' issue when the DS is also the mounted MDS. The nfs_match_client() reference from nfs4_set_ds_client bumps the cl_count, the nfs_client is not freed at umount, and nfs4_deviceid_purge_client is not called to dereference the MDS usage of a deviceid which holds a reference to the DS nfs_client. The result is the umount program returns, but the nfs_client is not freed, and the cl_session hearbeat continues. The MDS (and all other nfs mounts) lose their last nfs_client reference in nfs_free_server when the last nfs_server (fsid) is umounted. The file layout DS lose their last nfs_client reference in destroy_ds when the last deviceid referencing the data server is put and destroy_ds is called. This is triggered by a call to nfs4_deviceid_purge_client which removes references to a pNFS deviceid used by an MDS mount. The fix is to track how many pnfs enabled filesystems are mounted from this server, and then to purge the device id cache once that count reaches zero. Reported-by: Jorge Mora Reported-by: Andy Adamson Signed-off-by: Trond Myklebust diff --git a/fs/nfs/client.c b/fs/nfs/client.c index 17ba6b9..f005b5b 100644 --- a/fs/nfs/client.c +++ b/fs/nfs/client.c @@ -207,7 +207,6 @@ error_0: static void nfs4_shutdown_session(struct nfs_client *clp) { if (nfs4_has_session(clp)) { - nfs4_deviceid_purge_client(clp); nfs4_destroy_session(clp->cl_session); nfs4_destroy_clientid(clp); } diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index b8323aa..bdf7e52 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -80,6 +80,9 @@ unset_pnfs_layoutdriver(struct nfs_server *nfss) if (nfss->pnfs_curr_ld) { if (nfss->pnfs_curr_ld->clear_layoutdriver) nfss->pnfs_curr_ld->clear_layoutdriver(nfss); + /* Decrement the MDS count. Purge the deviceid cache if zero */ + if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count)) + nfs4_deviceid_purge_client(nfss->nfs_client); module_put(nfss->pnfs_curr_ld->owner); } nfss->pnfs_curr_ld = NULL; @@ -127,6 +130,8 @@ set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, module_put(ld_type->owner); goto out_no_driver; } + /* Bump the MDS count */ + atomic_inc(&server->nfs_client->cl_mds_count); dprintk("%s: pNFS module for %u set\n", __func__, id); return; diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h index fbb78fb..f58325a 100644 --- a/include/linux/nfs_fs_sb.h +++ b/include/linux/nfs_fs_sb.h @@ -25,6 +25,7 @@ struct nfs41_impl_id; */ struct nfs_client { atomic_t cl_count; + atomic_t cl_mds_count; int cl_cons_state; /* current construction state (-ve: init error) */ #define NFS_CS_READY 0 /* ready to be used */ #define NFS_CS_INITING 1 /* busy initialising */ -- cgit v0.10.2 From 0a9c63fae7df086ff5e107273c3cce8642430974 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Fri, 15 Jun 2012 13:02:58 -0400 Subject: NFSv4.1: Fix a race in set_pnfs_layoutdriver The call to try_module_get() dereferences ld_type outside the spin locks, which means that it may be pointing to garbage if a module unload was in progress. Signed-off-by: Trond Myklebust diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index bdf7e52..bbc49ca 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -70,6 +70,10 @@ find_pnfs_driver(u32 id) spin_lock(&pnfs_spinlock); local = find_pnfs_driver_locked(id); + if (local != NULL && !try_module_get(local->owner)) { + dprintk("%s: Could not grab reference on module\n", __func__); + local = NULL; + } spin_unlock(&pnfs_spinlock); return local; } @@ -118,10 +122,6 @@ set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh, goto out_no_driver; } } - if (!try_module_get(ld_type->owner)) { - dprintk("%s: Could not grab reference on module\n", __func__); - goto out_no_driver; - } server->pnfs_curr_ld = ld_type; if (ld_type->set_layoutdriver && ld_type->set_layoutdriver(server, mntfh)) { -- cgit v0.10.2 From 5a695da26367f53a368c90435e0b883c12f02791 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 19 Jun 2012 13:39:14 -0400 Subject: NFS: Fix a refcounting issue in O_DIRECT In nfs_direct_write_reschedule(), the requests from nfs_scan_commit_list have a refcount of 2, whereas the operations in nfs_direct_write_completion_ops expect them to have a refcount of 1. This patch adds a call to release the extra references. Signed-off-by: Trond Myklebust Cc: Fred Isaman diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 3168f6e..9a4cbfc 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -490,6 +490,7 @@ static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq) dreq->error = -EIO; spin_unlock(cinfo.lock); } + nfs_release_request(req); } nfs_pageio_complete(&desc); -- cgit v0.10.2 From 1a0de48ae56b5cdb9a46b3d3a0b578dd7f787f22 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 19 Jun 2012 18:38:56 -0400 Subject: NFS: Initialise commit_info.rpc_out when !defined(CONFIG_NFS_V4) Signed-off-by: Trond Myklebust Cc: Fred Isaman diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c index e605d69..f729698 100644 --- a/fs/nfs/inode.c +++ b/fs/nfs/inode.c @@ -1530,7 +1530,6 @@ static inline void nfs4_init_once(struct nfs_inode *nfsi) nfsi->delegation_state = 0; init_rwsem(&nfsi->rwsem); nfsi->layout = NULL; - atomic_set(&nfsi->commit_info.rpcs_out, 0); #endif } @@ -1545,6 +1544,7 @@ static void init_once(void *foo) INIT_LIST_HEAD(&nfsi->commit_info.list); nfsi->npages = 0; nfsi->commit_info.ncommit = 0; + atomic_set(&nfsi->commit_info.rpcs_out, 0); atomic_set(&nfsi->silly_count, 1); INIT_HLIST_HEAD(&nfsi->silly_list); init_waitqueue_head(&nfsi->waitqueue); -- cgit v0.10.2 From b1027439dff844675f6c0df97a1b1d190791a699 Mon Sep 17 00:00:00 2001 From: Bryan Schumaker Date: Wed, 20 Jun 2012 14:35:28 -0400 Subject: NFS: Force the legacy idmapper to be single threaded It was initially coded under the assumption that there would only be one request at a time, so use a lock to enforce this requirement.. Signed-off-by: Bryan Schumaker CC: stable@vger.kernel.org [3.4+] Signed-off-by: Trond Myklebust diff --git a/fs/nfs/idmap.c b/fs/nfs/idmap.c index b5b86a0..864c51e 100644 --- a/fs/nfs/idmap.c +++ b/fs/nfs/idmap.c @@ -57,6 +57,11 @@ unsigned int nfs_idmap_cache_timeout = 600; static const struct cred *id_resolver_cache; static struct key_type key_type_id_resolver_legacy; +struct idmap { + struct rpc_pipe *idmap_pipe; + struct key_construction *idmap_key_cons; + struct mutex idmap_mutex; +}; /** * nfs_fattr_init_names - initialise the nfs_fattr owner_name/group_name fields @@ -310,9 +315,11 @@ static ssize_t nfs_idmap_get_key(const char *name, size_t namelen, name, namelen, type, data, data_size, NULL); if (ret < 0) { + mutex_lock(&idmap->idmap_mutex); ret = nfs_idmap_request_key(&key_type_id_resolver_legacy, name, namelen, type, data, data_size, idmap); + mutex_unlock(&idmap->idmap_mutex); } return ret; } @@ -354,11 +361,6 @@ static int nfs_idmap_lookup_id(const char *name, size_t namelen, const char *typ /* idmap classic begins here */ module_param(nfs_idmap_cache_timeout, int, 0644); -struct idmap { - struct rpc_pipe *idmap_pipe; - struct key_construction *idmap_key_cons; -}; - enum { Opt_find_uid, Opt_find_gid, Opt_find_user, Opt_find_group, Opt_find_err }; @@ -469,6 +471,7 @@ nfs_idmap_new(struct nfs_client *clp) return error; } idmap->idmap_pipe = pipe; + mutex_init(&idmap->idmap_mutex); clp->cl_idmap = idmap; return 0; -- cgit v0.10.2