From 996d5b4db4b191f2676cf8775565cab8a5e2753b Mon Sep 17 00:00:00 2001 From: Matthew Wilcox Date: Wed, 11 Jul 2018 14:02:24 -0700 Subject: 9p: Use a slab for allocating requests Replace the custom batch allocation with a slab. Use an IDR to store pointers to the active requests instead of an array. We don't try to handle P9_NOTAG specially; the IDR will happily shrink all the way back once the TVERSION call has completed. Link: http://lkml.kernel.org/r/20180711210225.19730-6-willy@infradead.org Signed-off-by: Matthew Wilcox Cc: Eric Van Hensbergen Cc: Ron Minnich Cc: Latchesar Ionkov Signed-off-by: Dominique Martinet --- include/net/9p/client.h | 51 +++++++++---------------------------------------- 1 file changed, 9 insertions(+), 42 deletions(-) (limited to 'include') diff --git a/include/net/9p/client.h b/include/net/9p/client.h index 0fa0fbab33b0..a4dc42c53d18 100644 --- a/include/net/9p/client.h +++ b/include/net/9p/client.h @@ -64,22 +64,15 @@ enum p9_trans_status { /** * enum p9_req_status_t - status of a request - * @REQ_STATUS_IDLE: request slot unused * @REQ_STATUS_ALLOC: request has been allocated but not sent * @REQ_STATUS_UNSENT: request waiting to be sent * @REQ_STATUS_SENT: request sent to server * @REQ_STATUS_RCVD: response received from server * @REQ_STATUS_FLSHD: request has been flushed * @REQ_STATUS_ERROR: request encountered an error on the client side - * - * The @REQ_STATUS_IDLE state is used to mark a request slot as unused - * but use is actually tracked by the idpool structure which handles tag - * id allocation. - * */ enum p9_req_status_t { - REQ_STATUS_IDLE, REQ_STATUS_ALLOC, REQ_STATUS_UNSENT, REQ_STATUS_SENT, @@ -92,24 +85,12 @@ enum p9_req_status_t { * struct p9_req_t - request slots * @status: status of this request slot * @t_err: transport error - * @flush_tag: tag of request being flushed (for flush requests) * @wq: wait_queue for the client to block on for this request * @tc: the request fcall structure * @rc: the response fcall structure * @aux: transport specific data (provided for trans_fd migration) * @req_list: link for higher level objects to chain requests - * - * Transport use an array to track outstanding requests - * instead of a list. While this may incurr overhead during initial - * allocation or expansion, it makes request lookup much easier as the - * tag id is a index into an array. (We use tag+1 so that we can accommodate - * the -1 tag for the T_VERSION request). - * This also has the nice effect of only having to allocate wait_queues - * once, instead of constantly allocating and freeing them. Its possible - * other resources could benefit from this scheme as well. - * */ - struct p9_req_t { int status; int t_err; @@ -117,40 +98,26 @@ struct p9_req_t { struct p9_fcall *tc; struct p9_fcall *rc; void *aux; - struct list_head req_list; }; /** * struct p9_client - per client instance state - * @lock: protect @fidlist + * @lock: protect @fids and @reqs * @msize: maximum data size negotiated by protocol - * @dotu: extension flags negotiated by protocol * @proto_version: 9P protocol version to use * @trans_mod: module API instantiated with this client + * @status: connection state * @trans: tranport instance state and API * @fids: All active FID handles - * @tagpool - transaction id accounting for session - * @reqs - 2D array of requests - * @max_tag - current maximum tag id allocated - * @name - node name used as client id + * @reqs: All active requests. + * @name: node name used as client id * * The client structure is used to keep track of various per-client * state that has been instantiated. - * In order to minimize per-transaction overhead we use a - * simple array to lookup requests instead of a hash table - * or linked list. In order to support larger number of - * transactions, we make this a 2D array, allocating new rows - * when we need to grow the total number of the transactions. - * - * Each row is 256 requests and we'll support up to 256 rows for - * a total of 64k concurrent requests per session. - * - * Bugs: duplicated data and potentially unnecessary elements. */ - struct p9_client { - spinlock_t lock; /* protect client structure */ + spinlock_t lock; unsigned int msize; unsigned char proto_version; struct p9_trans_module *trans_mod; @@ -170,10 +137,7 @@ struct p9_client { } trans_opts; struct idr fids; - - struct p9_idpool *tagpool; - struct p9_req_t *reqs[P9_ROW_MAXTAG]; - int max_tag; + struct idr reqs; char name[__NEW_UTS_LEN + 1]; }; @@ -279,4 +243,7 @@ struct p9_fid *p9_client_xattrwalk(struct p9_fid *, const char *, u64 *); int p9_client_xattrcreate(struct p9_fid *, const char *, u64, int); int p9_client_readlink(struct p9_fid *fid, char **target); +int p9_client_init(void); +void p9_client_exit(void); + #endif /* NET_9P_CLIENT_H */ -- cgit v1.2.3 From 6348b903d79119a8157aace08ab99521f5dba139 Mon Sep 17 00:00:00 2001 From: Matthew Wilcox Date: Wed, 11 Jul 2018 14:02:25 -0700 Subject: 9p: Remove p9_idpool There are no more users left of the p9_idpool; delete it. Link: http://lkml.kernel.org/r/20180711210225.19730-7-willy@infradead.org Signed-off-by: Matthew Wilcox Cc: Eric Van Hensbergen Cc: Ron Minnich Cc: Latchesar Ionkov Signed-off-by: Dominique Martinet --- include/net/9p/9p.h | 8 -------- 1 file changed, 8 deletions(-) (limited to 'include') diff --git a/include/net/9p/9p.h b/include/net/9p/9p.h index b8eb51a661e5..e23896116d9a 100644 --- a/include/net/9p/9p.h +++ b/include/net/9p/9p.h @@ -561,16 +561,8 @@ struct p9_fcall { u8 *sdata; }; -struct p9_idpool; - int p9_errstr2errno(char *errstr, int len); -struct p9_idpool *p9_idpool_create(void); -void p9_idpool_destroy(struct p9_idpool *); -int p9_idpool_get(struct p9_idpool *p); -void p9_idpool_put(int id, struct p9_idpool *p); -int p9_idpool_check(int id, struct p9_idpool *p); - int p9_error_init(void); int p9_trans_fd_init(void); void p9_trans_fd_exit(void); -- cgit v1.2.3 From 523adb6cc10b48655c0abe556505240741425b49 Mon Sep 17 00:00:00 2001 From: Dominique Martinet Date: Mon, 30 Jul 2018 05:55:19 +0000 Subject: 9p: embed fcall in req to round down buffer allocs 'msize' is often a power of two, or at least page-aligned, so avoiding an overhead of two dozen bytes for each allocation will help the allocator do its work and reduce memory fragmentation. Link: http://lkml.kernel.org/r/1533825236-22896-1-git-send-email-asmadeus@codewreck.org Suggested-by: Matthew Wilcox Signed-off-by: Dominique Martinet Reviewed-by: Greg Kurz Acked-by: Jun Piao Cc: Matthew Wilcox --- include/net/9p/client.h | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/include/net/9p/client.h b/include/net/9p/client.h index a4dc42c53d18..c2671d40bb6b 100644 --- a/include/net/9p/client.h +++ b/include/net/9p/client.h @@ -95,8 +95,8 @@ struct p9_req_t { int status; int t_err; wait_queue_head_t wq; - struct p9_fcall *tc; - struct p9_fcall *rc; + struct p9_fcall tc; + struct p9_fcall rc; void *aux; struct list_head req_list; }; @@ -230,6 +230,7 @@ int p9_client_mkdir_dotl(struct p9_fid *fid, const char *name, int mode, kgid_t gid, struct p9_qid *); int p9_client_lock_dotl(struct p9_fid *fid, struct p9_flock *flock, u8 *status); int p9_client_getlock_dotl(struct p9_fid *fid, struct p9_getlock *fl); +void p9_fcall_fini(struct p9_fcall *fc); struct p9_req_t *p9_tag_lookup(struct p9_client *, u16); void p9_client_cb(struct p9_client *c, struct p9_req_t *req, int status); -- cgit v1.2.3 From 91a76be37ff89795526c452a6799576b03bec501 Mon Sep 17 00:00:00 2001 From: Dominique Martinet Date: Mon, 30 Jul 2018 15:14:37 +0900 Subject: 9p: add a per-client fcall kmem_cache Having a specific cache for the fcall allocations helps speed up end-to-end latency. The caches will automatically be merged if there are multiple caches of items with the same size so we do not need to try to share a cache between different clients of the same size. Since the msize is negotiated with the server, only allocate the cache after that negotiation has happened - previous allocations or allocations of different sizes (e.g. zero-copy fcall) are made with kmalloc directly. Some figures on two beefy VMs with Connect-IB (sriov) / trans=rdma, with ior running 32 processes in parallel doing small 32 bytes IOs: - no alloc (4.18-rc7 request cache): 65.4k req/s - non-power of two alloc, no patch: 61.6k req/s - power of two alloc, no patch: 62.2k req/s - non-power of two alloc, with patch: 64.7k req/s - power of two alloc, with patch: 65.1k req/s Link: http://lkml.kernel.org/r/1532943263-24378-2-git-send-email-asmadeus@codewreck.org Signed-off-by: Dominique Martinet Acked-by: Jun Piao Cc: Matthew Wilcox Cc: Greg Kurz --- include/net/9p/9p.h | 4 ++++ include/net/9p/client.h | 1 + 2 files changed, 5 insertions(+) (limited to 'include') diff --git a/include/net/9p/9p.h b/include/net/9p/9p.h index e23896116d9a..beede1e1a919 100644 --- a/include/net/9p/9p.h +++ b/include/net/9p/9p.h @@ -336,6 +336,9 @@ enum p9_qid_t { #define P9_NOFID (u32)(~0) #define P9_MAXWELEM 16 +/* Minimal header size: size[4] type[1] tag[2] */ +#define P9_HDRSZ 7 + /* ample room for Twrite/Rread header */ #define P9_IOHDRSZ 24 @@ -558,6 +561,7 @@ struct p9_fcall { size_t offset; size_t capacity; + struct kmem_cache *cache; u8 *sdata; }; diff --git a/include/net/9p/client.h b/include/net/9p/client.h index c2671d40bb6b..735f3979d559 100644 --- a/include/net/9p/client.h +++ b/include/net/9p/client.h @@ -123,6 +123,7 @@ struct p9_client { struct p9_trans_module *trans_mod; enum p9_trans_status status; void *trans; + struct kmem_cache *fcall_cache; union { struct { -- cgit v1.2.3 From 728356dedeff8ef999cb436c71333ef4ac51a81c Mon Sep 17 00:00:00 2001 From: Tomas Bortoli Date: Tue, 14 Aug 2018 19:43:42 +0200 Subject: 9p: Add refcount to p9_req_t To avoid use-after-free(s), use a refcount to keep track of the usable references to any instantiated struct p9_req_t. This commit adds p9_req_put(), p9_req_get() and p9_req_try_get() as wrappers to kref_put(), kref_get() and kref_get_unless_zero(). These are used by the client and the transports to keep track of valid requests' references. p9_free_req() is added back and used as callback by kref_put(). Add SLAB_TYPESAFE_BY_RCU as it ensures that the memory freed by kmem_cache_free() will not be reused for another type until the rcu synchronisation period is over, so an address gotten under rcu read lock is safe to inc_ref() without corrupting random memory while the lock is held. Link: http://lkml.kernel.org/r/1535626341-20693-1-git-send-email-asmadeus@codewreck.org Co-developed-by: Dominique Martinet Signed-off-by: Tomas Bortoli Reported-by: syzbot+467050c1ce275af2a5b8@syzkaller.appspotmail.com Signed-off-by: Dominique Martinet --- include/net/9p/client.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'include') diff --git a/include/net/9p/client.h b/include/net/9p/client.h index 735f3979d559..947a570307a6 100644 --- a/include/net/9p/client.h +++ b/include/net/9p/client.h @@ -94,6 +94,7 @@ enum p9_req_status_t { struct p9_req_t { int status; int t_err; + struct kref refcount; wait_queue_head_t wq; struct p9_fcall tc; struct p9_fcall rc; @@ -233,6 +234,19 @@ int p9_client_lock_dotl(struct p9_fid *fid, struct p9_flock *flock, u8 *status); int p9_client_getlock_dotl(struct p9_fid *fid, struct p9_getlock *fl); void p9_fcall_fini(struct p9_fcall *fc); struct p9_req_t *p9_tag_lookup(struct p9_client *, u16); + +static inline void p9_req_get(struct p9_req_t *r) +{ + kref_get(&r->refcount); +} + +static inline int p9_req_try_get(struct p9_req_t *r) +{ + return kref_get_unless_zero(&r->refcount); +} + +int p9_req_put(struct p9_req_t *r); + void p9_client_cb(struct p9_client *c, struct p9_req_t *req, int status); int p9_parse_header(struct p9_fcall *, int32_t *, int8_t *, int16_t *, int); -- cgit v1.2.3