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 #define RPCDBG_FACILITY RPCDBG_SVCDSP
33 static void svc_unregister(const struct svc_serv
*serv
);
35 #define svc_serv_is_pooled(serv) ((serv)->sv_function)
38 * Mode for mapping cpus to pools.
41 SVC_POOL_AUTO
= -1, /* choose one of the others */
42 SVC_POOL_GLOBAL
, /* no mapping, just a single global pool
43 * (legacy & UP mode) */
44 SVC_POOL_PERCPU
, /* one pool per cpu */
45 SVC_POOL_PERNODE
/* one pool per numa node */
47 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL
50 * Structure for mapping cpus to pools and vice versa.
51 * Setup once during sunrpc initialisation.
53 static struct svc_pool_map
{
54 int count
; /* How many svc_servs use us */
55 int mode
; /* Note: int not enum to avoid
56 * warnings about "enumeration value
57 * not handled in switch" */
59 unsigned int *pool_to
; /* maps pool id to cpu or node */
60 unsigned int *to_pool
; /* maps cpu or node to pool id */
63 .mode
= SVC_POOL_DEFAULT
65 static DEFINE_MUTEX(svc_pool_map_mutex
);/* protects svc_pool_map.count only */
68 param_set_pool_mode(const char *val
, struct kernel_param
*kp
)
70 int *ip
= (int *)kp
->arg
;
71 struct svc_pool_map
*m
= &svc_pool_map
;
74 mutex_lock(&svc_pool_map_mutex
);
81 if (!strncmp(val
, "auto", 4))
83 else if (!strncmp(val
, "global", 6))
84 *ip
= SVC_POOL_GLOBAL
;
85 else if (!strncmp(val
, "percpu", 6))
86 *ip
= SVC_POOL_PERCPU
;
87 else if (!strncmp(val
, "pernode", 7))
88 *ip
= SVC_POOL_PERNODE
;
93 mutex_unlock(&svc_pool_map_mutex
);
98 param_get_pool_mode(char *buf
, struct kernel_param
*kp
)
100 int *ip
= (int *)kp
->arg
;
105 return strlcpy(buf
, "auto", 20);
106 case SVC_POOL_GLOBAL
:
107 return strlcpy(buf
, "global", 20);
108 case SVC_POOL_PERCPU
:
109 return strlcpy(buf
, "percpu", 20);
110 case SVC_POOL_PERNODE
:
111 return strlcpy(buf
, "pernode", 20);
113 return sprintf(buf
, "%d", *ip
);
117 module_param_call(pool_mode
, param_set_pool_mode
, param_get_pool_mode
,
118 &svc_pool_map
.mode
, 0644);
121 * Detect best pool mapping mode heuristically,
122 * according to the machine's topology.
125 svc_pool_map_choose_mode(void)
129 if (nr_online_nodes
> 1) {
131 * Actually have multiple NUMA nodes,
132 * so split pools on NUMA node boundaries
134 return SVC_POOL_PERNODE
;
137 node
= first_online_node
;
138 if (nr_cpus_node(node
) > 2) {
140 * Non-trivial SMP, or CONFIG_NUMA on
141 * non-NUMA hardware, e.g. with a generic
142 * x86_64 kernel on Xeons. In this case we
143 * want to divide the pools on cpu boundaries.
145 return SVC_POOL_PERCPU
;
148 /* default: one global pool */
149 return SVC_POOL_GLOBAL
;
153 * Allocate the to_pool[] and pool_to[] arrays.
154 * Returns 0 on success or an errno.
157 svc_pool_map_alloc_arrays(struct svc_pool_map
*m
, unsigned int maxpools
)
159 m
->to_pool
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
162 m
->pool_to
= kcalloc(maxpools
, sizeof(unsigned int), GFP_KERNEL
);
175 * Initialise the pool map for SVC_POOL_PERCPU mode.
176 * Returns number of pools or <0 on error.
179 svc_pool_map_init_percpu(struct svc_pool_map
*m
)
181 unsigned int maxpools
= nr_cpu_ids
;
182 unsigned int pidx
= 0;
186 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
190 for_each_online_cpu(cpu
) {
191 BUG_ON(pidx
> maxpools
);
192 m
->to_pool
[cpu
] = pidx
;
193 m
->pool_to
[pidx
] = cpu
;
196 /* cpus brought online later all get mapped to pool0, sorry */
203 * Initialise the pool map for SVC_POOL_PERNODE mode.
204 * Returns number of pools or <0 on error.
207 svc_pool_map_init_pernode(struct svc_pool_map
*m
)
209 unsigned int maxpools
= nr_node_ids
;
210 unsigned int pidx
= 0;
214 err
= svc_pool_map_alloc_arrays(m
, maxpools
);
218 for_each_node_with_cpus(node
) {
219 /* some architectures (e.g. SN2) have cpuless nodes */
220 BUG_ON(pidx
> maxpools
);
221 m
->to_pool
[node
] = pidx
;
222 m
->pool_to
[pidx
] = node
;
225 /* nodes brought online later all get mapped to pool0, sorry */
232 * Add a reference to the global map of cpus to pools (and
233 * vice versa). Initialise the map if we're the first user.
234 * Returns the number of pools.
237 svc_pool_map_get(void)
239 struct svc_pool_map
*m
= &svc_pool_map
;
242 mutex_lock(&svc_pool_map_mutex
);
245 mutex_unlock(&svc_pool_map_mutex
);
249 if (m
->mode
== SVC_POOL_AUTO
)
250 m
->mode
= svc_pool_map_choose_mode();
253 case SVC_POOL_PERCPU
:
254 npools
= svc_pool_map_init_percpu(m
);
256 case SVC_POOL_PERNODE
:
257 npools
= svc_pool_map_init_pernode(m
);
262 /* default, or memory allocation failure */
264 m
->mode
= SVC_POOL_GLOBAL
;
268 mutex_unlock(&svc_pool_map_mutex
);
274 * Drop a reference to the global map of cpus to pools.
275 * When the last reference is dropped, the map data is
276 * freed; this allows the sysadmin to change the pool
277 * mode using the pool_mode module option without
278 * rebooting or re-loading sunrpc.ko.
281 svc_pool_map_put(void)
283 struct svc_pool_map
*m
= &svc_pool_map
;
285 mutex_lock(&svc_pool_map_mutex
);
288 m
->mode
= SVC_POOL_DEFAULT
;
294 mutex_unlock(&svc_pool_map_mutex
);
299 * Set the given thread's cpus_allowed mask so that it
300 * will only run on cpus in the given pool.
303 svc_pool_map_set_cpumask(struct task_struct
*task
, unsigned int pidx
)
305 struct svc_pool_map
*m
= &svc_pool_map
;
306 unsigned int node
= m
->pool_to
[pidx
];
309 * The caller checks for sv_nrpools > 1, which
310 * implies that we've been initialized.
312 BUG_ON(m
->count
== 0);
315 case SVC_POOL_PERCPU
:
317 set_cpus_allowed_ptr(task
, cpumask_of(node
));
320 case SVC_POOL_PERNODE
:
322 set_cpus_allowed_ptr(task
, cpumask_of_node(node
));
329 * Use the mapping mode to choose a pool for a given CPU.
330 * Used when enqueueing an incoming RPC. Always returns
331 * a non-NULL pool pointer.
334 svc_pool_for_cpu(struct svc_serv
*serv
, int cpu
)
336 struct svc_pool_map
*m
= &svc_pool_map
;
337 unsigned int pidx
= 0;
340 * An uninitialised map happens in a pure client when
341 * lockd is brought up, so silently treat it the
342 * same as SVC_POOL_GLOBAL.
344 if (svc_serv_is_pooled(serv
)) {
346 case SVC_POOL_PERCPU
:
347 pidx
= m
->to_pool
[cpu
];
349 case SVC_POOL_PERNODE
:
350 pidx
= m
->to_pool
[cpu_to_node(cpu
)];
354 return &serv
->sv_pools
[pidx
% serv
->sv_nrpools
];
359 * Create an RPC service
361 static struct svc_serv
*
362 __svc_create(struct svc_program
*prog
, unsigned int bufsize
, int npools
,
363 void (*shutdown
)(struct svc_serv
*serv
))
365 struct svc_serv
*serv
;
367 unsigned int xdrsize
;
370 if (!(serv
= kzalloc(sizeof(*serv
), GFP_KERNEL
)))
372 serv
->sv_name
= prog
->pg_name
;
373 serv
->sv_program
= prog
;
374 serv
->sv_nrthreads
= 1;
375 serv
->sv_stats
= prog
->pg_stats
;
376 if (bufsize
> RPCSVC_MAXPAYLOAD
)
377 bufsize
= RPCSVC_MAXPAYLOAD
;
378 serv
->sv_max_payload
= bufsize
? bufsize
: 4096;
379 serv
->sv_max_mesg
= roundup(serv
->sv_max_payload
+ PAGE_SIZE
, PAGE_SIZE
);
380 serv
->sv_shutdown
= shutdown
;
383 prog
->pg_lovers
= prog
->pg_nvers
-1;
384 for (vers
=0; vers
<prog
->pg_nvers
; vers
++)
385 if (prog
->pg_vers
[vers
]) {
386 prog
->pg_hivers
= vers
;
387 if (prog
->pg_lovers
> vers
)
388 prog
->pg_lovers
= vers
;
389 if (prog
->pg_vers
[vers
]->vs_xdrsize
> xdrsize
)
390 xdrsize
= prog
->pg_vers
[vers
]->vs_xdrsize
;
392 prog
= prog
->pg_next
;
394 serv
->sv_xdrsize
= xdrsize
;
395 INIT_LIST_HEAD(&serv
->sv_tempsocks
);
396 INIT_LIST_HEAD(&serv
->sv_permsocks
);
397 init_timer(&serv
->sv_temptimer
);
398 spin_lock_init(&serv
->sv_lock
);
400 serv
->sv_nrpools
= npools
;
402 kcalloc(serv
->sv_nrpools
, sizeof(struct svc_pool
),
404 if (!serv
->sv_pools
) {
409 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
410 struct svc_pool
*pool
= &serv
->sv_pools
[i
];
412 dprintk("svc: initialising pool %u for %s\n",
416 INIT_LIST_HEAD(&pool
->sp_threads
);
417 INIT_LIST_HEAD(&pool
->sp_sockets
);
418 INIT_LIST_HEAD(&pool
->sp_all_threads
);
419 spin_lock_init(&pool
->sp_lock
);
422 /* Remove any stale portmap registrations */
423 svc_unregister(serv
);
429 svc_create(struct svc_program
*prog
, unsigned int bufsize
,
430 void (*shutdown
)(struct svc_serv
*serv
))
432 return __svc_create(prog
, bufsize
, /*npools*/1, shutdown
);
434 EXPORT_SYMBOL_GPL(svc_create
);
437 svc_create_pooled(struct svc_program
*prog
, unsigned int bufsize
,
438 void (*shutdown
)(struct svc_serv
*serv
),
439 svc_thread_fn func
, struct module
*mod
)
441 struct svc_serv
*serv
;
442 unsigned int npools
= svc_pool_map_get();
444 serv
= __svc_create(prog
, bufsize
, npools
, shutdown
);
447 serv
->sv_function
= func
;
448 serv
->sv_module
= mod
;
453 EXPORT_SYMBOL_GPL(svc_create_pooled
);
456 * Destroy an RPC service. Should be called with appropriate locking to
457 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks.
460 svc_destroy(struct svc_serv
*serv
)
462 dprintk("svc: svc_destroy(%s, %d)\n",
463 serv
->sv_program
->pg_name
,
466 if (serv
->sv_nrthreads
) {
467 if (--(serv
->sv_nrthreads
) != 0) {
468 svc_sock_update_bufs(serv
);
472 printk("svc_destroy: no threads for serv=%p!\n", serv
);
474 del_timer_sync(&serv
->sv_temptimer
);
476 svc_close_all(&serv
->sv_tempsocks
);
478 if (serv
->sv_shutdown
)
479 serv
->sv_shutdown(serv
);
481 svc_close_all(&serv
->sv_permsocks
);
483 BUG_ON(!list_empty(&serv
->sv_permsocks
));
484 BUG_ON(!list_empty(&serv
->sv_tempsocks
));
486 cache_clean_deferred(serv
);
488 if (svc_serv_is_pooled(serv
))
491 #if defined(CONFIG_NFS_V4_1)
492 svc_sock_destroy(serv
->bc_xprt
);
493 #endif /* CONFIG_NFS_V4_1 */
495 svc_unregister(serv
);
496 kfree(serv
->sv_pools
);
499 EXPORT_SYMBOL_GPL(svc_destroy
);
502 * Allocate an RPC server's buffer space.
503 * We allocate pages and place them in rq_argpages.
506 svc_init_buffer(struct svc_rqst
*rqstp
, unsigned int size
)
508 unsigned int pages
, arghi
;
510 /* bc_xprt uses fore channel allocated buffers */
511 if (svc_is_backchannel(rqstp
))
514 pages
= size
/ PAGE_SIZE
+ 1; /* extra page as we hold both request and reply.
515 * We assume one is at most one page
518 BUG_ON(pages
> RPCSVC_MAXPAGES
);
520 struct page
*p
= alloc_page(GFP_KERNEL
);
523 rqstp
->rq_pages
[arghi
++] = p
;
530 * Release an RPC server buffer
533 svc_release_buffer(struct svc_rqst
*rqstp
)
537 for (i
= 0; i
< ARRAY_SIZE(rqstp
->rq_pages
); i
++)
538 if (rqstp
->rq_pages
[i
])
539 put_page(rqstp
->rq_pages
[i
]);
543 svc_prepare_thread(struct svc_serv
*serv
, struct svc_pool
*pool
)
545 struct svc_rqst
*rqstp
;
547 rqstp
= kzalloc(sizeof(*rqstp
), GFP_KERNEL
);
551 init_waitqueue_head(&rqstp
->rq_wait
);
553 serv
->sv_nrthreads
++;
554 spin_lock_bh(&pool
->sp_lock
);
555 pool
->sp_nrthreads
++;
556 list_add(&rqstp
->rq_all
, &pool
->sp_all_threads
);
557 spin_unlock_bh(&pool
->sp_lock
);
558 rqstp
->rq_server
= serv
;
559 rqstp
->rq_pool
= pool
;
561 rqstp
->rq_argp
= kmalloc(serv
->sv_xdrsize
, GFP_KERNEL
);
565 rqstp
->rq_resp
= kmalloc(serv
->sv_xdrsize
, GFP_KERNEL
);
569 if (!svc_init_buffer(rqstp
, serv
->sv_max_mesg
))
574 svc_exit_thread(rqstp
);
576 return ERR_PTR(-ENOMEM
);
578 EXPORT_SYMBOL_GPL(svc_prepare_thread
);
581 * Choose a pool in which to create a new thread, for svc_set_num_threads
583 static inline struct svc_pool
*
584 choose_pool(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
589 return &serv
->sv_pools
[(*state
)++ % serv
->sv_nrpools
];
593 * Choose a thread to kill, for svc_set_num_threads
595 static inline struct task_struct
*
596 choose_victim(struct svc_serv
*serv
, struct svc_pool
*pool
, unsigned int *state
)
599 struct task_struct
*task
= NULL
;
602 spin_lock_bh(&pool
->sp_lock
);
604 /* choose a pool in round-robin fashion */
605 for (i
= 0; i
< serv
->sv_nrpools
; i
++) {
606 pool
= &serv
->sv_pools
[--(*state
) % serv
->sv_nrpools
];
607 spin_lock_bh(&pool
->sp_lock
);
608 if (!list_empty(&pool
->sp_all_threads
))
610 spin_unlock_bh(&pool
->sp_lock
);
616 if (!list_empty(&pool
->sp_all_threads
)) {
617 struct svc_rqst
*rqstp
;
620 * Remove from the pool->sp_all_threads list
621 * so we don't try to kill it again.
623 rqstp
= list_entry(pool
->sp_all_threads
.next
, struct svc_rqst
, rq_all
);
624 list_del_init(&rqstp
->rq_all
);
625 task
= rqstp
->rq_task
;
627 spin_unlock_bh(&pool
->sp_lock
);
633 * Create or destroy enough new threads to make the number
634 * of threads the given number. If `pool' is non-NULL, applies
635 * only to threads in that pool, otherwise round-robins between
636 * all pools. Must be called with a svc_get() reference and
637 * the BKL or another lock to protect access to svc_serv fields.
639 * Destroying threads relies on the service threads filling in
640 * rqstp->rq_task, which only the nfs ones do. Assumes the serv
641 * has been created using svc_create_pooled().
643 * Based on code that used to be in nfsd_svc() but tweaked
647 svc_set_num_threads(struct svc_serv
*serv
, struct svc_pool
*pool
, int nrservs
)
649 struct svc_rqst
*rqstp
;
650 struct task_struct
*task
;
651 struct svc_pool
*chosen_pool
;
653 unsigned int state
= serv
->sv_nrthreads
-1;
656 /* The -1 assumes caller has done a svc_get() */
657 nrservs
-= (serv
->sv_nrthreads
-1);
659 spin_lock_bh(&pool
->sp_lock
);
660 nrservs
-= pool
->sp_nrthreads
;
661 spin_unlock_bh(&pool
->sp_lock
);
664 /* create new threads */
665 while (nrservs
> 0) {
667 chosen_pool
= choose_pool(serv
, pool
, &state
);
669 rqstp
= svc_prepare_thread(serv
, chosen_pool
);
671 error
= PTR_ERR(rqstp
);
675 __module_get(serv
->sv_module
);
676 task
= kthread_create(serv
->sv_function
, rqstp
, serv
->sv_name
);
678 error
= PTR_ERR(task
);
679 module_put(serv
->sv_module
);
680 svc_exit_thread(rqstp
);
684 rqstp
->rq_task
= task
;
685 if (serv
->sv_nrpools
> 1)
686 svc_pool_map_set_cpumask(task
, chosen_pool
->sp_id
);
688 svc_sock_update_bufs(serv
);
689 wake_up_process(task
);
691 /* destroy old threads */
692 while (nrservs
< 0 &&
693 (task
= choose_victim(serv
, pool
, &state
)) != NULL
) {
694 send_sig(SIGINT
, task
, 1);
700 EXPORT_SYMBOL_GPL(svc_set_num_threads
);
703 * Called from a server thread as it's exiting. Caller must hold the BKL or
704 * the "service mutex", whichever is appropriate for the service.
707 svc_exit_thread(struct svc_rqst
*rqstp
)
709 struct svc_serv
*serv
= rqstp
->rq_server
;
710 struct svc_pool
*pool
= rqstp
->rq_pool
;
712 svc_release_buffer(rqstp
);
713 kfree(rqstp
->rq_resp
);
714 kfree(rqstp
->rq_argp
);
715 kfree(rqstp
->rq_auth_data
);
717 spin_lock_bh(&pool
->sp_lock
);
718 pool
->sp_nrthreads
--;
719 list_del(&rqstp
->rq_all
);
720 spin_unlock_bh(&pool
->sp_lock
);
724 /* Release the server */
728 EXPORT_SYMBOL_GPL(svc_exit_thread
);
731 * Register an "inet" protocol family netid with the local
732 * rpcbind daemon via an rpcbind v4 SET request.
734 * No netconfig infrastructure is available in the kernel, so
735 * we map IP_ protocol numbers to netids by hand.
737 * Returns zero on success; a negative errno value is returned
738 * if any error occurs.
740 static int __svc_rpcb_register4(const u32 program
, const u32 version
,
741 const unsigned short protocol
,
742 const unsigned short port
)
744 const struct sockaddr_in sin
= {
745 .sin_family
= AF_INET
,
746 .sin_addr
.s_addr
= htonl(INADDR_ANY
),
747 .sin_port
= htons(port
),
754 netid
= RPCBIND_NETID_UDP
;
757 netid
= RPCBIND_NETID_TCP
;
763 error
= rpcb_v4_register(program
, version
,
764 (const struct sockaddr
*)&sin
, netid
);
767 * User space didn't support rpcbind v4, so retry this
768 * registration request with the legacy rpcbind v2 protocol.
770 if (error
== -EPROTONOSUPPORT
)
771 error
= rpcb_register(program
, version
, protocol
, port
);
776 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
778 * Register an "inet6" protocol family netid with the local
779 * rpcbind daemon via an rpcbind v4 SET request.
781 * No netconfig infrastructure is available in the kernel, so
782 * we map IP_ protocol numbers to netids by hand.
784 * Returns zero on success; a negative errno value is returned
785 * if any error occurs.
787 static int __svc_rpcb_register6(const u32 program
, const u32 version
,
788 const unsigned short protocol
,
789 const unsigned short port
)
791 const struct sockaddr_in6 sin6
= {
792 .sin6_family
= AF_INET6
,
793 .sin6_addr
= IN6ADDR_ANY_INIT
,
794 .sin6_port
= htons(port
),
801 netid
= RPCBIND_NETID_UDP6
;
804 netid
= RPCBIND_NETID_TCP6
;
810 error
= rpcb_v4_register(program
, version
,
811 (const struct sockaddr
*)&sin6
, netid
);
814 * User space didn't support rpcbind version 4, so we won't
815 * use a PF_INET6 listener.
817 if (error
== -EPROTONOSUPPORT
)
818 error
= -EAFNOSUPPORT
;
822 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
825 * Register a kernel RPC service via rpcbind version 4.
827 * Returns zero on success; a negative errno value is returned
828 * if any error occurs.
830 static int __svc_register(const char *progname
,
831 const u32 program
, const u32 version
,
833 const unsigned short protocol
,
834 const unsigned short port
)
836 int error
= -EAFNOSUPPORT
;
840 error
= __svc_rpcb_register4(program
, version
,
843 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
845 error
= __svc_rpcb_register6(program
, version
,
847 #endif /* defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) */
851 printk(KERN_WARNING
"svc: failed to register %sv%u RPC "
852 "service (errno %d).\n", progname
, version
, -error
);
857 * svc_register - register an RPC service with the local portmapper
858 * @serv: svc_serv struct for the service to register
859 * @family: protocol family of service's listener socket
860 * @proto: transport protocol number to advertise
861 * @port: port to advertise
863 * Service is registered for any address in the passed-in protocol family
865 int svc_register(const struct svc_serv
*serv
, const int family
,
866 const unsigned short proto
, const unsigned short port
)
868 struct svc_program
*progp
;
872 BUG_ON(proto
== 0 && port
== 0);
874 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
875 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
876 if (progp
->pg_vers
[i
] == NULL
)
879 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n",
882 proto
== IPPROTO_UDP
? "udp" : "tcp",
885 progp
->pg_vers
[i
]->vs_hidden
?
886 " (but not telling portmap)" : "");
888 if (progp
->pg_vers
[i
]->vs_hidden
)
891 error
= __svc_register(progp
->pg_name
, progp
->pg_prog
,
892 i
, family
, proto
, port
);
902 * If user space is running rpcbind, it should take the v4 UNSET
903 * and clear everything for this [program, version]. If user space
904 * is running portmap, it will reject the v4 UNSET, but won't have
905 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient
906 * in this case to clear all existing entries for [program, version].
908 static void __svc_unregister(const u32 program
, const u32 version
,
909 const char *progname
)
913 error
= rpcb_v4_register(program
, version
, NULL
, "");
916 * User space didn't support rpcbind v4, so retry this
917 * request with the legacy rpcbind v2 protocol.
919 if (error
== -EPROTONOSUPPORT
)
920 error
= rpcb_register(program
, version
, 0, 0);
922 dprintk("svc: %s(%sv%u), error %d\n",
923 __func__
, progname
, version
, error
);
927 * All netids, bind addresses and ports registered for [program, version]
928 * are removed from the local rpcbind database (if the service is not
929 * hidden) to make way for a new instance of the service.
931 * The result of unregistration is reported via dprintk for those who want
932 * verification of the result, but is otherwise not important.
934 static void svc_unregister(const struct svc_serv
*serv
)
936 struct svc_program
*progp
;
940 clear_thread_flag(TIF_SIGPENDING
);
942 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
) {
943 for (i
= 0; i
< progp
->pg_nvers
; i
++) {
944 if (progp
->pg_vers
[i
] == NULL
)
946 if (progp
->pg_vers
[i
]->vs_hidden
)
949 __svc_unregister(progp
->pg_prog
, i
, progp
->pg_name
);
953 spin_lock_irqsave(¤t
->sighand
->siglock
, flags
);
955 spin_unlock_irqrestore(¤t
->sighand
->siglock
, flags
);
959 * Printk the given error with the address of the client that caused it.
962 __attribute__ ((format (printf
, 2, 3)))
963 svc_printk(struct svc_rqst
*rqstp
, const char *fmt
, ...)
967 char buf
[RPC_MAX_ADDRBUFLEN
];
969 if (!net_ratelimit())
972 printk(KERN_WARNING
"svc: %s: ",
973 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
976 r
= vprintk(fmt
, args
);
983 * Common routine for processing the RPC request.
986 svc_process_common(struct svc_rqst
*rqstp
, struct kvec
*argv
, struct kvec
*resv
)
988 struct svc_program
*progp
;
989 struct svc_version
*versp
= NULL
; /* compiler food */
990 struct svc_procedure
*procp
= NULL
;
991 struct svc_serv
*serv
= rqstp
->rq_server
;
994 u32 prog
, vers
, proc
;
995 __be32 auth_stat
, rpc_stat
;
999 rpc_stat
= rpc_success
;
1001 if (argv
->iov_len
< 6*4)
1004 /* Will be turned off only in gss privacy case: */
1005 rqstp
->rq_splice_ok
= 1;
1006 /* Will be turned off only when NFSv4 Sessions are used */
1007 rqstp
->rq_usedeferral
= 1;
1009 /* Setup reply header */
1010 rqstp
->rq_xprt
->xpt_ops
->xpo_prep_reply_hdr(rqstp
);
1012 svc_putu32(resv
, rqstp
->rq_xid
);
1014 vers
= svc_getnl(argv
);
1016 /* First words of reply: */
1017 svc_putnl(resv
, 1); /* REPLY */
1019 if (vers
!= 2) /* RPC version number */
1022 /* Save position in case we later decide to reject: */
1023 reply_statp
= resv
->iov_base
+ resv
->iov_len
;
1025 svc_putnl(resv
, 0); /* ACCEPT */
1027 rqstp
->rq_prog
= prog
= svc_getnl(argv
); /* program number */
1028 rqstp
->rq_vers
= vers
= svc_getnl(argv
); /* version number */
1029 rqstp
->rq_proc
= proc
= svc_getnl(argv
); /* procedure number */
1031 progp
= serv
->sv_program
;
1033 for (progp
= serv
->sv_program
; progp
; progp
= progp
->pg_next
)
1034 if (prog
== progp
->pg_prog
)
1038 * Decode auth data, and add verifier to reply buffer.
1039 * We do this before anything else in order to get a decent
1042 auth_res
= svc_authenticate(rqstp
, &auth_stat
);
1043 /* Also give the program a chance to reject this call: */
1044 if (auth_res
== SVC_OK
&& progp
) {
1045 auth_stat
= rpc_autherr_badcred
;
1046 auth_res
= progp
->pg_authenticate(rqstp
);
1054 rpc_stat
= rpc_system_err
;
1059 if (test_bit(XPT_TEMP
, &rqstp
->rq_xprt
->xpt_flags
))
1060 svc_close_xprt(rqstp
->rq_xprt
);
1070 if (vers
>= progp
->pg_nvers
||
1071 !(versp
= progp
->pg_vers
[vers
]))
1074 procp
= versp
->vs_proc
+ proc
;
1075 if (proc
>= versp
->vs_nproc
|| !procp
->pc_func
)
1077 rqstp
->rq_procinfo
= procp
;
1079 /* Syntactic check complete */
1080 serv
->sv_stats
->rpccnt
++;
1082 /* Build the reply header. */
1083 statp
= resv
->iov_base
+resv
->iov_len
;
1084 svc_putnl(resv
, RPC_SUCCESS
);
1086 /* Bump per-procedure stats counter */
1089 /* Initialize storage for argp and resp */
1090 memset(rqstp
->rq_argp
, 0, procp
->pc_argsize
);
1091 memset(rqstp
->rq_resp
, 0, procp
->pc_ressize
);
1093 /* un-reserve some of the out-queue now that we have a
1094 * better idea of reply size
1096 if (procp
->pc_xdrressize
)
1097 svc_reserve_auth(rqstp
, procp
->pc_xdrressize
<<2);
1099 /* Call the function that processes the request. */
1100 if (!versp
->vs_dispatch
) {
1101 /* Decode arguments */
1102 xdr
= procp
->pc_decode
;
1103 if (xdr
&& !xdr(rqstp
, argv
->iov_base
, rqstp
->rq_argp
))
1106 *statp
= procp
->pc_func(rqstp
, rqstp
->rq_argp
, rqstp
->rq_resp
);
1109 if (*statp
== rpc_drop_reply
) {
1110 if (procp
->pc_release
)
1111 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1114 if (*statp
== rpc_success
&&
1115 (xdr
= procp
->pc_encode
) &&
1116 !xdr(rqstp
, resv
->iov_base
+resv
->iov_len
, rqstp
->rq_resp
)) {
1117 dprintk("svc: failed to encode reply\n");
1118 /* serv->sv_stats->rpcsystemerr++; */
1119 *statp
= rpc_system_err
;
1122 dprintk("svc: calling dispatcher\n");
1123 if (!versp
->vs_dispatch(rqstp
, statp
)) {
1124 /* Release reply info */
1125 if (procp
->pc_release
)
1126 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1131 /* Check RPC status result */
1132 if (*statp
!= rpc_success
)
1133 resv
->iov_len
= ((void*)statp
) - resv
->iov_base
+ 4;
1135 /* Release reply info */
1136 if (procp
->pc_release
)
1137 procp
->pc_release(rqstp
, NULL
, rqstp
->rq_resp
);
1139 if (procp
->pc_encode
== NULL
)
1143 if (svc_authorise(rqstp
))
1145 return 1; /* Caller can now send it */
1148 svc_authorise(rqstp
); /* doesn't hurt to call this twice */
1149 dprintk("svc: svc_process dropit\n");
1154 svc_printk(rqstp
, "short len %Zd, dropping request\n",
1157 goto dropit
; /* drop request */
1160 serv
->sv_stats
->rpcbadfmt
++;
1161 svc_putnl(resv
, 1); /* REJECT */
1162 svc_putnl(resv
, 0); /* RPC_MISMATCH */
1163 svc_putnl(resv
, 2); /* Only RPCv2 supported */
1168 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat
));
1169 serv
->sv_stats
->rpcbadauth
++;
1170 /* Restore write pointer to location of accept status: */
1171 xdr_ressize_check(rqstp
, reply_statp
);
1172 svc_putnl(resv
, 1); /* REJECT */
1173 svc_putnl(resv
, 1); /* AUTH_ERROR */
1174 svc_putnl(resv
, ntohl(auth_stat
)); /* status */
1178 dprintk("svc: unknown program %d\n", prog
);
1179 serv
->sv_stats
->rpcbadfmt
++;
1180 svc_putnl(resv
, RPC_PROG_UNAVAIL
);
1184 svc_printk(rqstp
, "unknown version (%d for prog %d, %s)\n",
1185 vers
, prog
, progp
->pg_name
);
1187 serv
->sv_stats
->rpcbadfmt
++;
1188 svc_putnl(resv
, RPC_PROG_MISMATCH
);
1189 svc_putnl(resv
, progp
->pg_lovers
);
1190 svc_putnl(resv
, progp
->pg_hivers
);
1194 svc_printk(rqstp
, "unknown procedure (%d)\n", proc
);
1196 serv
->sv_stats
->rpcbadfmt
++;
1197 svc_putnl(resv
, RPC_PROC_UNAVAIL
);
1201 svc_printk(rqstp
, "failed to decode args\n");
1203 rpc_stat
= rpc_garbage_args
;
1205 serv
->sv_stats
->rpcbadfmt
++;
1206 svc_putnl(resv
, ntohl(rpc_stat
));
1209 EXPORT_SYMBOL_GPL(svc_process
);
1212 * Process the RPC request.
1215 svc_process(struct svc_rqst
*rqstp
)
1217 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1218 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1219 struct svc_serv
*serv
= rqstp
->rq_server
;
1224 * Setup response xdr_buf.
1225 * Initially it has just one page
1227 rqstp
->rq_resused
= 1;
1228 resv
->iov_base
= page_address(rqstp
->rq_respages
[0]);
1230 rqstp
->rq_res
.pages
= rqstp
->rq_respages
+ 1;
1231 rqstp
->rq_res
.len
= 0;
1232 rqstp
->rq_res
.page_base
= 0;
1233 rqstp
->rq_res
.page_len
= 0;
1234 rqstp
->rq_res
.buflen
= PAGE_SIZE
;
1235 rqstp
->rq_res
.tail
[0].iov_base
= NULL
;
1236 rqstp
->rq_res
.tail
[0].iov_len
= 0;
1238 rqstp
->rq_xid
= svc_getu32(argv
);
1240 dir
= svc_getnl(argv
);
1242 /* direction != CALL */
1243 svc_printk(rqstp
, "bad direction %d, dropping request\n", dir
);
1244 serv
->sv_stats
->rpcbadfmt
++;
1249 error
= svc_process_common(rqstp
, argv
, resv
);
1253 return svc_send(rqstp
);
1256 #if defined(CONFIG_NFS_V4_1)
1258 * Process a backchannel RPC request that arrived over an existing
1259 * outbound connection
1262 bc_svc_process(struct svc_serv
*serv
, struct rpc_rqst
*req
,
1263 struct svc_rqst
*rqstp
)
1265 struct kvec
*argv
= &rqstp
->rq_arg
.head
[0];
1266 struct kvec
*resv
= &rqstp
->rq_res
.head
[0];
1269 /* Build the svc_rqst used by the common processing routine */
1270 rqstp
->rq_xprt
= serv
->bc_xprt
;
1271 rqstp
->rq_xid
= req
->rq_xid
;
1272 rqstp
->rq_prot
= req
->rq_xprt
->prot
;
1273 rqstp
->rq_server
= serv
;
1275 rqstp
->rq_addrlen
= sizeof(req
->rq_xprt
->addr
);
1276 memcpy(&rqstp
->rq_addr
, &req
->rq_xprt
->addr
, rqstp
->rq_addrlen
);
1277 memcpy(&rqstp
->rq_arg
, &req
->rq_rcv_buf
, sizeof(rqstp
->rq_arg
));
1278 memcpy(&rqstp
->rq_res
, &req
->rq_snd_buf
, sizeof(rqstp
->rq_res
));
1280 /* reset result send buffer "put" position */
1283 if (rqstp
->rq_prot
!= IPPROTO_TCP
) {
1284 printk(KERN_ERR
"No support for Non-TCP transports!\n");
1289 * Skip the next two words because they've already been
1290 * processed in the trasport
1292 svc_getu32(argv
); /* XID */
1293 svc_getnl(argv
); /* CALLDIR */
1295 error
= svc_process_common(rqstp
, argv
, resv
);
1299 memcpy(&req
->rq_snd_buf
, &rqstp
->rq_res
, sizeof(req
->rq_snd_buf
));
1300 return bc_send(req
);
1302 EXPORT_SYMBOL(bc_svc_process
);
1303 #endif /* CONFIG_NFS_V4_1 */
1306 * Return (transport-specific) limit on the rpc payload.
1308 u32
svc_max_payload(const struct svc_rqst
*rqstp
)
1310 u32 max
= rqstp
->rq_xprt
->xpt_class
->xcl_max_payload
;
1312 if (rqstp
->rq_server
->sv_max_payload
< max
)
1313 max
= rqstp
->rq_server
->sv_max_payload
;
1316 EXPORT_SYMBOL_GPL(svc_max_payload
);