2 * linux/net/sunrpc/svc.c
4 * High-level RPC service routines
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
8 * Multiple threads pools and NUMAisation
9 * Copyright (c) 2006 Silicon Graphics, Inc.
10 * by Greg Banks <gnb@melbourne.sgi.com>
13 #include <linux/linkage.h>
14 #include <linux/sched.h>
15 #include <linux/errno.h>
16 #include <linux/net.h>
19 #include <linux/interrupt.h>
20 #include <linux/module.h>
21 #include <linux/kthread.h>
22 #include <linux/slab.h>
24 #include <linux/sunrpc/types.h>
25 #include <linux/sunrpc/xdr.h>
26 #include <linux/sunrpc/stats.h>
27 #include <linux/sunrpc/svcsock.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/sunrpc/bc_xprt.h>
31 #include <trace/events/sunrpc.h>
33 #define RPCDBG_FACILITY RPCDBG_SVCDSP
35 static void svc_unregister(const struct svc_serv
*serv
, struct net
*net
);
37 #define svc_serv_is_pooled(serv) ((serv)->sv_ops->svo_function)
39 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
42 * Structure for mapping cpus to pools and vice versa.
43 * Setup once during sunrpc initialisation.
45 struct svc_pool_map svc_pool_map
= {
46 .mode
= SVC_POOL_DEFAULT
48 EXPORT_SYMBOL_GPL(svc_pool_map
);
50 static DEFINE_MUTEX(svc_pool_map_mutex
);/* protects svc_pool_map.count only */
53 param_set_pool_mode(const char *val
, struct kernel_param
*kp
)
55 int *ip
= (int *)kp
->arg
;
56 struct svc_pool_map
*m
= &svc_pool_map
;
59 mutex_lock(&svc_pool_map_mutex
);
66 if (!strncmp(val
, "auto", 4))
68 else if (!strncmp(val
, "global", 6))
69 *ip
= SVC_POOL_GLOBAL
;
70 else if (!strncmp(val
, "percpu", 6))
71 *ip
= SVC_POOL_PERCPU
;
72 else if (!strncmp(val
, "pernode", 7))
73 *ip
= SVC_POOL_PERNODE
;
78 mutex_unlock(&svc_pool_map_mutex
);
83 param_get_pool_mode(char *buf
, struct kernel_param
*kp
)
85 int *ip
= (int *)kp
->arg
;
90 return strlcpy(buf
, "auto", 20);
92 return strlcpy(buf
, "global", 20);
94 return strlcpy(buf
, "percpu", 20);
95 case SVC_POOL_PERNODE
:
96 return strlcpy(buf
, "pernode", 20);
98 return sprintf(buf
, "%d", *ip
);
102 module_param_call(pool_mode
, param_set_pool_mode
, param_get_pool_mode
,
103 &svc_pool_map
.mode
, 0644);
106 * Detect best pool mapping mode heuristically,
107 * according to the machine's topology.
110 svc_pool_map_choose_mode(void)
114 if (nr_online_nodes
> 1) {
116 * Actually have multiple NUMA nodes,
117 * so split pools on NUMA node boundaries
119 return SVC_POOL_PERNODE
;
122 node
= first_online_node
;
123 if (nr_cpus_node(node
) > 2) {
125 * Non-trivial SMP, or CONFIG_NUMA on
126 * non-NUMA hardware, e.g. with a generic
127 * x86_64 kernel on Xeons. In this case we
128 * want to divide the pools on cpu boundaries.
130 return SVC_POOL_PERCPU
;
133 /* default: one global pool */
134 return SVC_POOL_GLOBAL
;
138 * Allocate the to_pool[] and pool_to[] arrays.
139 * Returns 0 on success or an errno.
142 svc_pool_map_alloc_arrays(struct svc_pool_map
*m
, unsigned int maxpools
)
144 m
->to_pool
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
147 m
->pool_to
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
161 * Initialise the pool map for SVC_POOL_PERCPU mode.
162 * Returns number of pools or <0 on error.
165 svc_pool_map_init_percpu(struct svc_pool_map
*m
)
167 unsigned int maxpools
= nr_cpu_ids
;
168 unsigned int pidx
= 0;
172 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
176 for_each_online_cpu(cpu
) {
177 BUG_ON(pidx
>= maxpools
);
178 m
->to_pool
[cpu
] = pidx
;
179 m
->pool_to
[pidx
] = cpu
;
182 /* cpus brought online later all get mapped to pool0, sorry */
189 * Initialise the pool map for SVC_POOL_PERNODE mode.
190 * Returns number of pools or <0 on error.
193 svc_pool_map_init_pernode(struct svc_pool_map
*m
)
195 unsigned int maxpools
= nr_node_ids
;
196 unsigned int pidx
= 0;
200 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
204 for_each_node_with_cpus(node
) {
205 /* some architectures (e.g. SN2) have cpuless nodes */
206 BUG_ON(pidx
> maxpools
);
207 m
->to_pool
[node
] = pidx
;
208 m
->pool_to
[pidx
] = node
;
211 /* nodes brought online later all get mapped to pool0, sorry */
218 * Add a reference to the global map of cpus to pools (and
219 * vice versa). Initialise the map if we're the first user.
220 * Returns the number of pools.
223 svc_pool_map_get(void)
225 struct svc_pool_map
*m
= &svc_pool_map
;
228 mutex_lock(&svc_pool_map_mutex
);
231 mutex_unlock(&svc_pool_map_mutex
);
235 if (m
->mode
== SVC_POOL_AUTO
)
236 m
->mode
= svc_pool_map_choose_mode();
239 case SVC_POOL_PERCPU
:
240 npools
= svc_pool_map_init_percpu(m
);
242 case SVC_POOL_PERNODE
:
243 npools
= svc_pool_map_init_pernode(m
);
248 /* default, or memory allocation failure */
250 m
->mode
= SVC_POOL_GLOBAL
;
254 mutex_unlock(&svc_pool_map_mutex
);
257 EXPORT_SYMBOL_GPL(svc_pool_map_get
);
260 * Drop a reference to the global map of cpus to pools.
261 * When the last reference is dropped, the map data is
262 * freed; this allows the sysadmin to change the pool
263 * mode using the pool_mode module option without
264 * rebooting or re-loading sunrpc.ko.
267 svc_pool_map_put(void)
269 struct svc_pool_map
*m
= &svc_pool_map
;
271 mutex_lock(&svc_pool_map_mutex
);
281 mutex_unlock(&svc_pool_map_mutex
);
283 EXPORT_SYMBOL_GPL(svc_pool_map_put
);
285 static int svc_pool_map_get_node(unsigned int pidx
)
287 const struct svc_pool_map
*m
= &svc_pool_map
;
290 if (m
->mode
== SVC_POOL_PERCPU
)
291 return cpu_to_node(m
->pool_to
[pidx
]);
292 if (m
->mode
== SVC_POOL_PERNODE
)
293 return m
->pool_to
[pidx
];
298 * Set the given thread's cpus_allowed mask so that it
299 * will only run on cpus in the given pool.
302 svc_pool_map_set_cpumask(struct task_struct
*task
, unsigned int pidx
)
304 struct svc_pool_map
*m
= &svc_pool_map
;
305 unsigned int node
= m
->pool_to
[pidx
];
308 * The caller checks for sv_nrpools > 1, which
309 * implies that we've been initialized.
311 WARN_ON_ONCE(m
->count
== 0);
316 case SVC_POOL_PERCPU
:
318 set_cpus_allowed_ptr(task
, cpumask_of(node
));
321 case SVC_POOL_PERNODE
:
323 set_cpus_allowed_ptr(task
, cpumask_of_node(node
));
330 * Use the mapping mode to choose a pool for a given CPU.
331 * Used when enqueueing an incoming RPC. Always returns
332 * a non-NULL pool pointer.
335 svc_pool_for_cpu(struct svc_serv
*serv
, int cpu
)
337 struct svc_pool_map
*m
= &svc_pool_map
;
338 unsigned int pidx
= 0;
341 * An uninitialised map happens in a pure client when
342 * lockd is brought up, so silently treat it the
343 * same as SVC_POOL_GLOBAL.
345 if (svc_serv_is_pooled(serv
)) {
347 case SVC_POOL_PERCPU
:
348 pidx
= m
->to_pool
[cpu
];
350 case SVC_POOL_PERNODE
:
351 pidx
= m
->to_pool
[cpu_to_node(cpu
)];
355 return &serv
->sv_pools
[pidx
% serv
->sv_nrpools
];
358 int svc_rpcb_setup(struct svc_serv
*serv
, struct net
*net
)
362 err
= rpcb_create_local(net
);
366 /* Remove any stale portmap registrations */
367 svc_unregister(serv
, net
);
370 EXPORT_SYMBOL_GPL(svc_rpcb_setup
);
372 void svc_rpcb_cleanup(struct svc_serv
*serv
, struct net
*net
)
374 svc_unregister(serv
, net
);
377 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup
);
379 static int svc_uses_rpcbind(struct svc_serv
*serv
)
381 struct svc_program
*progp
;
384 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
385 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
386 if (progp
->pg_vers
[i
] == NULL
)
388 if (!progp
->pg_vers
[i
]->vs_hidden
)
396 int svc_bind(struct svc_serv
*serv
, struct net
*net
)
398 if (!svc_uses_rpcbind(serv
))
400 return svc_rpcb_setup(serv
, net
);
402 EXPORT_SYMBOL_GPL(svc_bind
);
404 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
406 __svc_init_bc(struct svc_serv
*serv
)
408 INIT_LIST_HEAD(&serv
->sv_cb_list
);
409 spin_lock_init(&serv
->sv_cb_lock
);
410 init_waitqueue_head(&serv
->sv_cb_waitq
);
414 __svc_init_bc(struct svc_serv
*serv
)
420 * Create an RPC service
422 static struct svc_serv
*
423 __svc_create(struct svc_program
*prog
, unsigned int bufsize
, int npools
,
424 struct svc_serv_ops
*ops
)
426 struct svc_serv
*serv
;
428 unsigned int xdrsize
;
431 if (!(serv
= kzalloc(sizeof(*serv
), GFP_KERNEL
)))
433 serv
->sv_name
= prog
->pg_name
;
434 serv
->sv_program
= prog
;
435 serv
->sv_nrthreads
= 1;
436 serv
->sv_stats
= prog
->pg_stats
;
437 if (bufsize
> RPCSVC_MAXPAYLOAD
)
438 bufsize
= RPCSVC_MAXPAYLOAD
;
439 serv
->sv_max_payload
= bufsize
? bufsize
: 4096;
440 serv
->sv_max_mesg
= roundup(serv
->sv_max_payload
+ PAGE_SIZE
, PAGE_SIZE
);
444 prog
->pg_lovers
= prog
->pg_nvers
-1;
445 for (vers
=0; vers
<prog
->pg_nvers
; vers
++)
446 if (prog
->pg_vers
[vers
]) {
447 prog
->pg_hivers
= vers
;
448 if (prog
->pg_lovers
> vers
)
449 prog
->pg_lovers
= vers
;
450 if (prog
->pg_vers
[vers
]->vs_xdrsize
> xdrsize
)
451 xdrsize
= prog
->pg_vers
[vers
]->vs_xdrsize
;
453 prog
= prog
->pg_next
;
455 serv
->sv_xdrsize
= xdrsize
;
456 INIT_LIST_HEAD(&serv
->sv_tempsocks
);
457 INIT_LIST_HEAD(&serv
->sv_permsocks
);
458 init_timer(&serv
->sv_temptimer
);
459 spin_lock_init(&serv
->sv_lock
);
463 serv
->sv_nrpools
= npools
;
465 kcalloc(serv
->sv_nrpools
, sizeof(struct svc_pool
),
467 if (!serv
->sv_pools
) {
472 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
473 struct svc_pool
*pool
= &serv
->sv_pools
[i
];
475 dprintk("svc: initialising pool %u for %s\n",
479 INIT_LIST_HEAD(&pool
->sp_sockets
);
480 INIT_LIST_HEAD(&pool
->sp_all_threads
);
481 spin_lock_init(&pool
->sp_lock
);
488 svc_create(struct svc_program
*prog
, unsigned int bufsize
,
489 struct svc_serv_ops
*ops
)
491 return __svc_create(prog
, bufsize
, /*npools*/1, ops
);
493 EXPORT_SYMBOL_GPL(svc_create
);
496 svc_create_pooled(struct svc_program
*prog
, unsigned int bufsize
,
497 struct svc_serv_ops
*ops
)
499 struct svc_serv
*serv
;
500 unsigned int npools
= svc_pool_map_get();
502 serv
= __svc_create(prog
, bufsize
, npools
, ops
);
510 EXPORT_SYMBOL_GPL(svc_create_pooled
);
512 void svc_shutdown_net(struct svc_serv
*serv
, struct net
*net
)
514 svc_close_net(serv
, net
);
516 if (serv
->sv_ops
->svo_shutdown
)
517 serv
->sv_ops
->svo_shutdown(serv
, net
);
519 EXPORT_SYMBOL_GPL(svc_shutdown_net
);
522 * Destroy an RPC service. Should be called with appropriate locking to
523 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
526 svc_destroy(struct svc_serv
*serv
)
528 dprintk("svc: svc_destroy(%s, %d)\n",
529 serv
->sv_program
->pg_name
,
532 if (serv
->sv_nrthreads
) {
533 if (--(serv
->sv_nrthreads
) != 0) {
534 svc_sock_update_bufs(serv
);
538 printk("svc_destroy: no threads for serv=%p!\n", serv
);
540 del_timer_sync(&serv
->sv_temptimer
);
543 * The last user is gone and thus all sockets have to be destroyed to
544 * the point. Check this.
546 BUG_ON(!list_empty(&serv
->sv_permsocks
));
547 BUG_ON(!list_empty(&serv
->sv_tempsocks
));
549 cache_clean_deferred(serv
);
551 if (svc_serv_is_pooled(serv
))
554 kfree(serv
->sv_pools
);
557 EXPORT_SYMBOL_GPL(svc_destroy
);
560 * Allocate an RPC server's buffer space.
561 * We allocate pages and place them in rq_argpages.
564 svc_init_buffer(struct svc_rqst
*rqstp
, unsigned int size
, int node
)
566 unsigned int pages
, arghi
;
568 /* bc_xprt uses fore channel allocated buffers */
569 if (svc_is_backchannel(rqstp
))
572 pages
= size
/ PAGE_SIZE
+ 1; /* extra page as we hold both request and reply.
573 * We assume one is at most one page
576 WARN_ON_ONCE(pages
> RPCSVC_MAXPAGES
);
577 if (pages
> RPCSVC_MAXPAGES
)
578 pages
= RPCSVC_MAXPAGES
;
580 struct page
*p
= alloc_pages_node(node
, GFP_KERNEL
, 0);
583 rqstp
->rq_pages
[arghi
++] = p
;
590 * Release an RPC server buffer
593 svc_release_buffer(struct svc_rqst
*rqstp
)
597 for (i
= 0; i
< ARRAY_SIZE(rqstp
->rq_pages
); i
++)
598 if (rqstp
->rq_pages
[i
])
599 put_page(rqstp
->rq_pages
[i
]);
603 svc_rqst_alloc(struct svc_serv
*serv
, struct svc_pool
*pool
, int node
)
605 struct svc_rqst
*rqstp
;
607 rqstp
= kzalloc_node(sizeof(*rqstp
), GFP_KERNEL
, node
);
611 __set_bit(RQ_BUSY
, &rqstp
->rq_flags
);
612 spin_lock_init(&rqstp
->rq_lock
);
613 rqstp
->rq_server
= serv
;
614 rqstp
->rq_pool
= pool
;
616 rqstp
->rq_argp
= kmalloc_node(serv
->sv_xdrsize
, GFP_KERNEL
, node
);
620 rqstp
->rq_resp
= kmalloc_node(serv
->sv_xdrsize
, GFP_KERNEL
, node
);
624 if (!svc_init_buffer(rqstp
, serv
->sv_max_mesg
, node
))
629 svc_rqst_free(rqstp
);
632 EXPORT_SYMBOL_GPL(svc_rqst_alloc
);
635 svc_prepare_thread(struct svc_serv
*serv
, struct svc_pool
*pool
, int node
)
637 struct svc_rqst
*rqstp
;
639 rqstp
= svc_rqst_alloc(serv
, pool
, node
);
641 return ERR_PTR(-ENOMEM
);
643 serv
->sv_nrthreads
++;
644 spin_lock_bh(&pool
->sp_lock
);
645 pool
->sp_nrthreads
++;
646 list_add_rcu(&rqstp
->rq_all
, &pool
->sp_all_threads
);
647 spin_unlock_bh(&pool
->sp_lock
);
650 EXPORT_SYMBOL_GPL(svc_prepare_thread
);
653 * Choose a pool in which to create a new thread, for svc_set_num_threads
655 static inline struct svc_pool
*
656 choose_pool(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
661 return &serv
->sv_pools
[(*state
)++ % serv
->sv_nrpools
];
665 * Choose a thread to kill, for svc_set_num_threads
667 static inline struct task_struct
*
668 choose_victim(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
671 struct task_struct
*task
= NULL
;
674 spin_lock_bh(&pool
->sp_lock
);
676 /* choose a pool in round-robin fashion */
677 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
678 pool
= &serv
->sv_pools
[--(*state
) % serv
->sv_nrpools
];
679 spin_lock_bh(&pool
->sp_lock
);
680 if (!list_empty(&pool
->sp_all_threads
))
682 spin_unlock_bh(&pool
->sp_lock
);
688 if (!list_empty(&pool
->sp_all_threads
)) {
689 struct svc_rqst
*rqstp
;
692 * Remove from the pool->sp_all_threads list
693 * so we don't try to kill it again.
695 rqstp
= list_entry(pool
->sp_all_threads
.next
, struct svc_rqst
, rq_all
);
696 set_bit(RQ_VICTIM
, &rqstp
->rq_flags
);
697 list_del_rcu(&rqstp
->rq_all
);
698 task
= rqstp
->rq_task
;
700 spin_unlock_bh(&pool
->sp_lock
);
706 * Create or destroy enough new threads to make the number
707 * of threads the given number. If `pool' is non-NULL, applies
708 * only to threads in that pool, otherwise round-robins between
709 * all pools. Caller must ensure that mutual exclusion between this and
710 * server startup or shutdown.
712 * Destroying threads relies on the service threads filling in
713 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
714 * has been created using svc_create_pooled().
716 * Based on code that used to be in nfsd_svc() but tweaked
720 svc_set_num_threads(struct svc_serv
*serv
, struct svc_pool
*pool
, int nrservs
)
722 struct svc_rqst
*rqstp
;
723 struct task_struct
*task
;
724 struct svc_pool
*chosen_pool
;
726 unsigned int state
= serv
->sv_nrthreads
-1;
730 /* The -1 assumes caller has done a svc_get() */
731 nrservs
-= (serv
->sv_nrthreads
-1);
733 spin_lock_bh(&pool
->sp_lock
);
734 nrservs
-= pool
->sp_nrthreads
;
735 spin_unlock_bh(&pool
->sp_lock
);
738 /* create new threads */
739 while (nrservs
> 0) {
741 chosen_pool
= choose_pool(serv
, pool
, &state
);
743 node
= svc_pool_map_get_node(chosen_pool
->sp_id
);
744 rqstp
= svc_prepare_thread(serv
, chosen_pool
, node
);
746 error
= PTR_ERR(rqstp
);
750 __module_get(serv
->sv_ops
->svo_module
);
751 task
= kthread_create_on_node(serv
->sv_ops
->svo_function
, rqstp
,
752 node
, "%s", serv
->sv_name
);
754 error
= PTR_ERR(task
);
755 module_put(serv
->sv_ops
->svo_module
);
756 svc_exit_thread(rqstp
);
760 rqstp
->rq_task
= task
;
761 if (serv
->sv_nrpools
> 1)
762 svc_pool_map_set_cpumask(task
, chosen_pool
->sp_id
);
764 svc_sock_update_bufs(serv
);
765 wake_up_process(task
);
767 /* destroy old threads */
768 while (nrservs
< 0 &&
769 (task
= choose_victim(serv
, pool
, &state
)) != NULL
) {
770 send_sig(SIGINT
, task
, 1);
776 EXPORT_SYMBOL_GPL(svc_set_num_threads
);
779 * Called from a server thread as it's exiting. Caller must hold the "service
780 * mutex" for the service.
783 svc_rqst_free(struct svc_rqst
*rqstp
)
785 svc_release_buffer(rqstp
);
786 kfree(rqstp
->rq_resp
);
787 kfree(rqstp
->rq_argp
);
788 kfree(rqstp
->rq_auth_data
);
789 kfree_rcu(rqstp
, rq_rcu_head
);
791 EXPORT_SYMBOL_GPL(svc_rqst_free
);
794 svc_exit_thread(struct svc_rqst
*rqstp
)
796 struct svc_serv
*serv
= rqstp
->rq_server
;
797 struct svc_pool
*pool
= rqstp
->rq_pool
;
799 spin_lock_bh(&pool
->sp_lock
);
800 pool
->sp_nrthreads
--;
801 if (!test_and_set_bit(RQ_VICTIM
, &rqstp
->rq_flags
))
802 list_del_rcu(&rqstp
->rq_all
);
803 spin_unlock_bh(&pool
->sp_lock
);
805 svc_rqst_free(rqstp
);
807 /* Release the server */
811 EXPORT_SYMBOL_GPL(svc_exit_thread
);
814 * Register an "inet" protocol family netid with the local
815 * rpcbind daemon via an rpcbind v4 SET request.
817 * No netconfig infrastructure is available in the kernel, so
818 * we map IP_ protocol numbers to netids by hand.
820 * Returns zero on success; a negative errno value is returned
821 * if any error occurs.
823 static int __svc_rpcb_register4(struct net
*net
, const u32 program
,
825 const unsigned short protocol
,
826 const unsigned short port
)
828 const struct sockaddr_in sin
= {
829 .sin_family
= AF_INET
,
830 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
831 .sin_port
= htons(port
),
838 netid
= RPCBIND_NETID_UDP
;
841 netid
= RPCBIND_NETID_TCP
;
847 error
= rpcb_v4_register(net
, program
, version
,
848 (const struct sockaddr
*)&sin
, netid
);
851 * User space didn't support rpcbind v4, so retry this
852 * registration request with the legacy rpcbind v2 protocol.
854 if (error
== -EPROTONOSUPPORT
)
855 error
= rpcb_register(net
, program
, version
, protocol
, port
);
860 #if IS_ENABLED(CONFIG_IPV6)
862 * Register an "inet6" protocol family netid with the local
863 * rpcbind daemon via an rpcbind v4 SET request.
865 * No netconfig infrastructure is available in the kernel, so
866 * we map IP_ protocol numbers to netids by hand.
868 * Returns zero on success; a negative errno value is returned
869 * if any error occurs.
871 static int __svc_rpcb_register6(struct net
*net
, const u32 program
,
873 const unsigned short protocol
,
874 const unsigned short port
)
876 const struct sockaddr_in6 sin6
= {
877 .sin6_family
= AF_INET6
,
878 .sin6_addr
= IN6ADDR_ANY_INIT
,
879 .sin6_port
= htons(port
),
886 netid
= RPCBIND_NETID_UDP6
;
889 netid
= RPCBIND_NETID_TCP6
;
895 error
= rpcb_v4_register(net
, program
, version
,
896 (const struct sockaddr
*)&sin6
, netid
);
899 * User space didn't support rpcbind version 4, so we won't
900 * use a PF_INET6 listener.
902 if (error
== -EPROTONOSUPPORT
)
903 error
= -EAFNOSUPPORT
;
907 #endif /* IS_ENABLED(CONFIG_IPV6) */
910 * Register a kernel RPC service via rpcbind version 4.
912 * Returns zero on success; a negative errno value is returned
913 * if any error occurs.
915 static int __svc_register(struct net
*net
, const char *progname
,
916 const u32 program
, const u32 version
,
918 const unsigned short protocol
,
919 const unsigned short port
)
921 int error
= -EAFNOSUPPORT
;
925 error
= __svc_rpcb_register4(net
, program
, version
,
928 #if IS_ENABLED(CONFIG_IPV6)
930 error
= __svc_rpcb_register6(net
, program
, version
,
939 * svc_register - register an RPC service with the local portmapper
940 * @serv: svc_serv struct for the service to register
941 * @net: net namespace for the service to register
942 * @family: protocol family of service's listener socket
943 * @proto: transport protocol number to advertise
944 * @port: port to advertise
946 * Service is registered for any address in the passed-in protocol family
948 int svc_register(const struct svc_serv
*serv
, struct net
*net
,
949 const int family
, const unsigned short proto
,
950 const unsigned short port
)
952 struct svc_program
*progp
;
953 struct svc_version
*vers
;
957 WARN_ON_ONCE(proto
== 0 && port
== 0);
958 if (proto
== 0 && port
== 0)
961 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
962 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
963 vers
= progp
->pg_vers
[i
];
967 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
970 proto
== IPPROTO_UDP
? "udp" : "tcp",
974 " (but not telling portmap)" : "");
980 * Don't register a UDP port if we need congestion
983 if (vers
->vs_need_cong_ctrl
&& proto
== IPPROTO_UDP
)
986 error
= __svc_register(net
, progp
->pg_name
, progp
->pg_prog
,
987 i
, family
, proto
, port
);
989 if (vers
->vs_rpcb_optnl
) {
995 printk(KERN_WARNING
"svc: failed to register "
996 "%sv%u RPC service (errno %d).\n",
997 progp
->pg_name
, i
, -error
);
1007 * If user space is running rpcbind, it should take the v4 UNSET
1008 * and clear everything for this [program, version]. If user space
1009 * is running portmap, it will reject the v4 UNSET, but won't have
1010 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
1011 * in this case to clear all existing entries for [program, version].
1013 static void __svc_unregister(struct net
*net
, const u32 program
, const u32 version
,
1014 const char *progname
)
1018 error
= rpcb_v4_register(net
, program
, version
, NULL
, "");
1021 * User space didn't support rpcbind v4, so retry this
1022 * request with the legacy rpcbind v2 protocol.
1024 if (error
== -EPROTONOSUPPORT
)
1025 error
= rpcb_register(net
, program
, version
, 0, 0);
1027 dprintk("svc: %s(%sv%u), error %d\n",
1028 __func__
, progname
, version
, error
);
1032 * All netids, bind addresses and ports registered for [program, version]
1033 * are removed from the local rpcbind database (if the service is not
1034 * hidden) to make way for a new instance of the service.
1036 * The result of unregistration is reported via dprintk for those who want
1037 * verification of the result, but is otherwise not important.
1039 static void svc_unregister(const struct svc_serv
*serv
, struct net
*net
)
1041 struct svc_program
*progp
;
1042 unsigned long flags
;
1045 clear_thread_flag(TIF_SIGPENDING
);
1047 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
1048 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
1049 if (progp
->pg_vers
[i
] == NULL
)
1051 if (progp
->pg_vers
[i
]->vs_hidden
)
1054 dprintk("svc: attempting to unregister %sv%u\n",
1056 __svc_unregister(net
, progp
->pg_prog
, i
, progp
->pg_name
);
1060 spin_lock_irqsave(¤t
->sighand
->siglock
, flags
);
1061 recalc_sigpending();
1062 spin_unlock_irqrestore(¤t
->sighand
->siglock
, flags
);
1066 * dprintk the given error with the address of the client that caused it.
1068 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
1069 static __printf(2, 3)
1070 void svc_printk(struct svc_rqst
*rqstp
, const char *fmt
, ...)
1072 struct va_format vaf
;
1074 char buf
[RPC_MAX_ADDRBUFLEN
];
1076 va_start(args
, fmt
);
1081 dprintk("svc: %s: %pV", svc_print_addr(rqstp
, buf
, sizeof(buf
)), &vaf
);
1086 static __printf(2,3) void svc_printk(struct svc_rqst
*rqstp
, const char *fmt
, ...) {}
1090 * Common routine for processing the RPC request.
1093 svc_process_common(struct svc_rqst
*rqstp
, struct kvec
*argv
, struct kvec
*resv
)
1095 struct svc_program
*progp
;
1096 struct svc_version
*versp
= NULL
; /* compiler food */
1097 struct svc_procedure
*procp
= NULL
;
1098 struct svc_serv
*serv
= rqstp
->rq_server
;
1101 u32 prog
, vers
, proc
;
1102 __be32 auth_stat
, rpc_stat
;
1104 __be32
*reply_statp
;
1106 rpc_stat
= rpc_success
;
1108 if (argv
->iov_len
< 6*4)
1111 /* Will be turned off only in gss privacy case: */
1112 set_bit(RQ_SPLICE_OK
, &rqstp
->rq_flags
);
1113 /* Will be turned off only when NFSv4 Sessions are used */
1114 set_bit(RQ_USEDEFERRAL
, &rqstp
->rq_flags
);
1115 clear_bit(RQ_DROPME
, &rqstp
->rq_flags
);
1117 /* Setup reply header */
1118 rqstp
->rq_xprt
->xpt_ops
->xpo_prep_reply_hdr(rqstp
);
1120 svc_putu32(resv
, rqstp
->rq_xid
);
1122 vers
= svc_getnl(argv
);
1124 /* First words of reply: */
1125 svc_putnl(resv
, 1); /* REPLY */
1127 if (vers
!= 2) /* RPC version number */
1130 /* Save position in case we later decide to reject: */
1131 reply_statp
= resv
->iov_base
+ resv
->iov_len
;
1133 svc_putnl(resv
, 0); /* ACCEPT */
1135 rqstp
->rq_prog
= prog
= svc_getnl(argv
); /* program number */
1136 rqstp
->rq_vers
= vers
= svc_getnl(argv
); /* version number */
1137 rqstp
->rq_proc
= proc
= svc_getnl(argv
); /* procedure number */
1139 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
)
1140 if (prog
== progp
->pg_prog
)
1144 * Decode auth data, and add verifier to reply buffer.
1145 * We do this before anything else in order to get a decent
1148 auth_res
= svc_authenticate(rqstp
, &auth_stat
);
1149 /* Also give the program a chance to reject this call: */
1150 if (auth_res
== SVC_OK
&& progp
) {
1151 auth_stat
= rpc_autherr_badcred
;
1152 auth_res
= progp
->pg_authenticate(rqstp
);
1160 rpc_stat
= rpc_system_err
;
1175 if (vers
>= progp
->pg_nvers
||
1176 !(versp
= progp
->pg_vers
[vers
]))
1180 * Some protocol versions (namely NFSv4) require some form of
1181 * congestion control. (See RFC 7530 section 3.1 paragraph 2)
1182 * In other words, UDP is not allowed. We mark those when setting
1183 * up the svc_xprt, and verify that here.
1185 * The spec is not very clear about what error should be returned
1186 * when someone tries to access a server that is listening on UDP
1187 * for lower versions. RPC_PROG_MISMATCH seems to be the closest
1190 if (versp
->vs_need_cong_ctrl
&&
1191 !test_bit(XPT_CONG_CTRL
, &rqstp
->rq_xprt
->xpt_flags
))
1194 procp
= versp
->vs_proc
+ proc
;
1195 if (proc
>= versp
->vs_nproc
|| !procp
->pc_func
)
1197 rqstp
->rq_procinfo
= procp
;
1199 /* Syntactic check complete */
1200 serv
->sv_stats
->rpccnt
++;
1202 /* Build the reply header. */
1203 statp
= resv
->iov_base
+resv
->iov_len
;
1204 svc_putnl(resv
, RPC_SUCCESS
);
1206 /* Bump per-procedure stats counter */
1209 /* Initialize storage for argp and resp */
1210 memset(rqstp
->rq_argp
, 0, procp
->pc_argsize
);
1211 memset(rqstp
->rq_resp
, 0, procp
->pc_ressize
);
1213 /* un-reserve some of the out-queue now that we have a
1214 * better idea of reply size
1216 if (procp
->pc_xdrressize
)
1217 svc_reserve_auth(rqstp
, procp
->pc_xdrressize
<<2);
1219 /* Call the function that processes the request. */
1220 if (!versp
->vs_dispatch
) {
1221 /* Decode arguments */
1222 xdr
= procp
->pc_decode
;
1223 if (xdr
&& !xdr(rqstp
, argv
->iov_base
, rqstp
->rq_argp
))
1226 *statp
= procp
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
1229 if (*statp
== rpc_drop_reply
||
1230 test_bit(RQ_DROPME
, &rqstp
->rq_flags
)) {
1231 if (procp
->pc_release
)
1232 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1235 if (*statp
== rpc_autherr_badcred
) {
1236 if (procp
->pc_release
)
1237 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1240 if (*statp
== rpc_success
&&
1241 (xdr
= procp
->pc_encode
) &&
1242 !xdr(rqstp
, resv
->iov_base
+resv
->iov_len
, rqstp
->rq_resp
)) {
1243 dprintk("svc: failed to encode reply\n");
1244 /* serv->sv_stats->rpcsystemerr++; */
1245 *statp
= rpc_system_err
;
1248 dprintk("svc: calling dispatcher\n");
1249 if (!versp
->vs_dispatch(rqstp
, statp
)) {
1250 /* Release reply info */
1251 if (procp
->pc_release
)
1252 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1257 /* Check RPC status result */
1258 if (*statp
!= rpc_success
)
1259 resv
->iov_len
= ((void*)statp
) - resv
->iov_base
+ 4;
1261 /* Release reply info */
1262 if (procp
->pc_release
)
1263 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1265 if (procp
->pc_encode
== NULL
)
1269 if (svc_authorise(rqstp
))
1271 return 1; /* Caller can now send it */
1274 svc_authorise(rqstp
); /* doesn't hurt to call this twice */
1275 dprintk("svc: svc_process dropit\n");
1279 if (test_bit(XPT_TEMP
, &rqstp
->rq_xprt
->xpt_flags
))
1280 svc_close_xprt(rqstp
->rq_xprt
);
1281 dprintk("svc: svc_process close\n");
1285 svc_printk(rqstp
, "short len %zd, dropping request\n",
1290 serv
->sv_stats
->rpcbadfmt
++;
1291 svc_putnl(resv
, 1); /* REJECT */
1292 svc_putnl(resv
, 0); /* RPC_MISMATCH */
1293 svc_putnl(resv
, 2); /* Only RPCv2 supported */
1298 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat
));
1299 serv
->sv_stats
->rpcbadauth
++;
1300 /* Restore write pointer to location of accept status: */
1301 xdr_ressize_check(rqstp
, reply_statp
);
1302 svc_putnl(resv
, 1); /* REJECT */
1303 svc_putnl(resv
, 1); /* AUTH_ERROR */
1304 svc_putnl(resv
, ntohl(auth_stat
)); /* status */
1308 dprintk("svc: unknown program %d\n", prog
);
1309 serv
->sv_stats
->rpcbadfmt
++;
1310 svc_putnl(resv
, RPC_PROG_UNAVAIL
);
1314 svc_printk(rqstp
, "unknown version (%d for prog %d, %s)\n",
1315 vers
, prog
, progp
->pg_name
);
1317 serv
->sv_stats
->rpcbadfmt
++;
1318 svc_putnl(resv
, RPC_PROG_MISMATCH
);
1319 svc_putnl(resv
, progp
->pg_lovers
);
1320 svc_putnl(resv
, progp
->pg_hivers
);
1324 svc_printk(rqstp
, "unknown procedure (%d)\n", proc
);
1326 serv
->sv_stats
->rpcbadfmt
++;
1327 svc_putnl(resv
, RPC_PROC_UNAVAIL
);
1331 svc_printk(rqstp
, "failed to decode args\n");
1333 rpc_stat
= rpc_garbage_args
;
1335 serv
->sv_stats
->rpcbadfmt
++;
1336 svc_putnl(resv
, ntohl(rpc_stat
));
1341 * Process the RPC request.
1344 svc_process(struct svc_rqst
*rqstp
)
1346 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1347 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1348 struct svc_serv
*serv
= rqstp
->rq_server
;
1352 * Setup response xdr_buf.
1353 * Initially it has just one page
1355 rqstp
->rq_next_page
= &rqstp
->rq_respages
[1];
1356 resv
->iov_base
= page_address(rqstp
->rq_respages
[0]);
1358 rqstp
->rq_res
.pages
= rqstp
->rq_respages
+ 1;
1359 rqstp
->rq_res
.len
= 0;
1360 rqstp
->rq_res
.page_base
= 0;
1361 rqstp
->rq_res
.page_len
= 0;
1362 rqstp
->rq_res
.buflen
= PAGE_SIZE
;
1363 rqstp
->rq_res
.tail
[0].iov_base
= NULL
;
1364 rqstp
->rq_res
.tail
[0].iov_len
= 0;
1366 dir
= svc_getnl(argv
);
1368 /* direction != CALL */
1369 svc_printk(rqstp
, "bad direction %d, dropping request\n", dir
);
1370 serv
->sv_stats
->rpcbadfmt
++;
1374 /* Returns 1 for send, 0 for drop */
1375 if (likely(svc_process_common(rqstp
, argv
, resv
))) {
1376 int ret
= svc_send(rqstp
);
1378 trace_svc_process(rqstp
, ret
);
1382 trace_svc_process(rqstp
, 0);
1386 EXPORT_SYMBOL_GPL(svc_process
);
1388 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
1390 * Process a backchannel RPC request that arrived over an existing
1391 * outbound connection
1394 bc_svc_process(struct svc_serv
*serv
, struct rpc_rqst
*req
,
1395 struct svc_rqst
*rqstp
)
1397 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1398 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1399 struct rpc_task
*task
;
1403 dprintk("svc: %s(%p)\n", __func__
, req
);
1405 /* Build the svc_rqst used by the common processing routine */
1406 rqstp
->rq_xprt
= serv
->sv_bc_xprt
;
1407 rqstp
->rq_xid
= req
->rq_xid
;
1408 rqstp
->rq_prot
= req
->rq_xprt
->prot
;
1409 rqstp
->rq_server
= serv
;
1411 rqstp
->rq_addrlen
= sizeof(req
->rq_xprt
->addr
);
1412 memcpy(&rqstp
->rq_addr
, &req
->rq_xprt
->addr
, rqstp
->rq_addrlen
);
1413 memcpy(&rqstp
->rq_arg
, &req
->rq_rcv_buf
, sizeof(rqstp
->rq_arg
));
1414 memcpy(&rqstp
->rq_res
, &req
->rq_snd_buf
, sizeof(rqstp
->rq_res
));
1416 /* Adjust the argument buffer length */
1417 rqstp
->rq_arg
.len
= req
->rq_private_buf
.len
;
1418 if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
) {
1419 rqstp
->rq_arg
.head
[0].iov_len
= rqstp
->rq_arg
.len
;
1420 rqstp
->rq_arg
.page_len
= 0;
1421 } else if (rqstp
->rq_arg
.len
<= rqstp
->rq_arg
.head
[0].iov_len
+
1422 rqstp
->rq_arg
.page_len
)
1423 rqstp
->rq_arg
.page_len
= rqstp
->rq_arg
.len
-
1424 rqstp
->rq_arg
.head
[0].iov_len
;
1426 rqstp
->rq_arg
.len
= rqstp
->rq_arg
.head
[0].iov_len
+
1427 rqstp
->rq_arg
.page_len
;
1429 /* reset result send buffer "put" position */
1433 * Skip the next two words because they've already been
1434 * processed in the transport
1436 svc_getu32(argv
); /* XID */
1437 svc_getnl(argv
); /* CALLDIR */
1439 /* Parse and execute the bc call */
1440 proc_error
= svc_process_common(rqstp
, argv
, resv
);
1442 atomic_inc(&req
->rq_xprt
->bc_free_slots
);
1444 /* Processing error: drop the request */
1445 xprt_free_bc_request(req
);
1449 /* Finally, send the reply synchronously */
1450 memcpy(&req
->rq_snd_buf
, &rqstp
->rq_res
, sizeof(req
->rq_snd_buf
));
1451 task
= rpc_run_bc_task(req
);
1453 error
= PTR_ERR(task
);
1457 WARN_ON_ONCE(atomic_read(&task
->tk_count
) != 1);
1458 error
= task
->tk_status
;
1462 dprintk("svc: %s(), error=%d\n", __func__
, error
);
1465 EXPORT_SYMBOL_GPL(bc_svc_process
);
1466 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
1469 * Return (transport-specific) limit on the rpc payload.
1471 u32
svc_max_payload(const struct svc_rqst
*rqstp
)
1473 u32 max
= rqstp
->rq_xprt
->xpt_class
->xcl_max_payload
;
1475 if (rqstp
->rq_server
->sv_max_payload
< max
)
1476 max
= rqstp
->rq_server
->sv_max_payload
;
1479 EXPORT_SYMBOL_GPL(svc_max_payload
);