include/linux/sunrpc/svc.h

   1 /*
   2  * linux/include/linux/sunrpc/svc.h
   3  *
   4  * RPC server declarations.
   5  *
   6  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
   7  */
   8
   9
  10 #ifndef SUNRPC_SVC_H
  11 #define SUNRPC_SVC_H
  12
  13 #include <linux/in.h>
  14 #include <linux/in6.h>
  15 #include <linux/sunrpc/types.h>
  16 #include <linux/sunrpc/xdr.h>
  17 #include <linux/sunrpc/auth.h>
  18 #include <linux/sunrpc/svcauth.h>
  19 #include <linux/wait.h>
  20 #include <linux/mm.h>
  21
  22 /*
  23  * This is the RPC server thread function prototype
  24  */
  25 typedef int             (*svc_thread_fn)(void *);
  26
  27 /* statistics for svc_pool structures */
  28 struct svc_pool_stats {
  29         unsigned long   packets;
  30         unsigned long   sockets_queued;
  31         unsigned long   threads_woken;
  32         unsigned long   threads_timedout;
  33 };
  34
  35 /*
  36  *
  37  * RPC service thread pool.
  38  *
  39  * Pool of threads and temporary sockets.  Generally there is only
  40  * a single one of these per RPC service, but on NUMA machines those
  41  * services that can benefit from it (i.e. nfs but not lockd) will
  42  * have one pool per NUMA node.  This optimisation reduces cross-
  43  * node traffic on multi-node NUMA NFS servers.
  44  */
  45 struct svc_pool {
  46         unsigned int            sp_id;          /* pool id; also node id on NUMA */
  47         spinlock_t              sp_lock;        /* protects all fields */
  48         struct list_head        sp_threads;     /* idle server threads */
  49         struct list_head        sp_sockets;     /* pending sockets */
  50         unsigned int            sp_nrthreads;   /* # of threads in pool */
  51         struct list_head        sp_all_threads; /* all server threads */
  52         struct svc_pool_stats   sp_stats;       /* statistics on pool operation */
  53 } ____cacheline_aligned_in_smp;
  54
  55 /*
  56  * RPC service.
  57  *
  58  * An RPC service is a ``daemon,'' possibly multithreaded, which
  59  * receives and processes incoming RPC messages.
  60  * It has one or more transport sockets associated with it, and maintains
  61  * a list of idle threads waiting for input.
  62  *
  63  * We currently do not support more than one RPC program per daemon.
  64  */
  65 struct svc_serv {
  66         struct svc_program *    sv_program;     /* RPC program */
  67         struct svc_stat *       sv_stats;       /* RPC statistics */
  68         spinlock_t              sv_lock;
  69         unsigned int            sv_nrthreads;   /* # of server threads */
  70         unsigned int            sv_maxconn;     /* max connections allowed or
  71                                                  * '0' causing max to be based
  72                                                  * on number of threads. */
  73
  74         unsigned int            sv_max_payload; /* datagram payload size */
  75         unsigned int            sv_max_mesg;    /* max_payload + 1 page for overheads */
  76         unsigned int            sv_xdrsize;     /* XDR buffer size */
  77         struct list_head        sv_permsocks;   /* all permanent sockets */
  78         struct list_head        sv_tempsocks;   /* all temporary sockets */
  79         int                     sv_tmpcnt;      /* count of temporary sockets */
  80         struct timer_list       sv_temptimer;   /* timer for aging temporary sockets */
  81
  82         char *                  sv_name;        /* service name */
  83
  84         unsigned int            sv_nrpools;     /* number of thread pools */
  85         struct svc_pool *       sv_pools;       /* array of thread pools */
  86
  87         void                    (*sv_shutdown)(struct svc_serv *serv);
  88                                                 /* Callback to use when last thread
  89                                                  * exits.
  90                                                  */
  91
  92         struct module *         sv_module;      /* optional module to count when
  93                                                  * adding threads */
  94         svc_thread_fn           sv_function;    /* main function for threads */
  95 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
  96         struct list_head        sv_cb_list;     /* queue for callback requests
  97                                                  * that arrive over the same
  98                                                  * connection */
  99         spinlock_t              sv_cb_lock;     /* protects the svc_cb_list */
 100         wait_queue_head_t       sv_cb_waitq;    /* sleep here if there are no
 101                                                  * entries in the svc_cb_list */
 102         struct svc_xprt         *sv_bc_xprt;    /* callback on fore channel */
 103 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 104 };
 105
 106 /*
 107  * We use sv_nrthreads as a reference count.  svc_destroy() drops
 108  * this refcount, so we need to bump it up around operations that
 109  * change the number of threads.  Horrible, but there it is.
 110  * Should be called with the BKL held.
 111  */
 112 static inline void svc_get(struct svc_serv *serv)
 113 {
 114         serv->sv_nrthreads++;
 115 }
 116
 117 /*
 118  * Maximum payload size supported by a kernel RPC server.
 119  * This is use to determine the max number of pages nfsd is
 120  * willing to return in a single READ operation.
 121  *
 122  * These happen to all be powers of 2, which is not strictly
 123  * necessary but helps enforce the real limitation, which is
 124  * that they should be multiples of PAGE_CACHE_SIZE.
 125  *
 126  * For UDP transports, a block plus NFS,RPC, and UDP headers
 127  * has to fit into the IP datagram limit of 64K.  The largest
 128  * feasible number for all known page sizes is probably 48K,
 129  * but we choose 32K here.  This is the same as the historical
 130  * Linux limit; someone who cares more about NFS/UDP performance
 131  * can test a larger number.
 132  *
 133  * For TCP transports we have more freedom.  A size of 1MB is
 134  * chosen to match the client limit.  Other OSes are known to
 135  * have larger limits, but those numbers are probably beyond
 136  * the point of diminishing returns.
 137  */
 138 #define RPCSVC_MAXPAYLOAD       (1*1024*1024u)
 139 #define RPCSVC_MAXPAYLOAD_TCP   RPCSVC_MAXPAYLOAD
 140 #define RPCSVC_MAXPAYLOAD_UDP   (32*1024u)
 141
 142 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
 143
 144 /*
 145  * RPC Requsts and replies are stored in one or more pages.
 146  * We maintain an array of pages for each server thread.
 147  * Requests are copied into these pages as they arrive.  Remaining
 148  * pages are available to write the reply into.
 149  *
 150  * Pages are sent using ->sendpage so each server thread needs to
 151  * allocate more to replace those used in sending.  To help keep track
 152  * of these pages we have a receive list where all pages initialy live,
 153  * and a send list where pages are moved to when there are to be part
 154  * of a reply.
 155  *
 156  * We use xdr_buf for holding responses as it fits well with NFS
 157  * read responses (that have a header, and some data pages, and possibly
 158  * a tail) and means we can share some client side routines.
 159  *
 160  * The xdr_buf.head kvec always points to the first page in the rq_*pages
 161  * list.  The xdr_buf.pages pointer points to the second page on that
 162  * list.  xdr_buf.tail points to the end of the first page.
 163  * This assumes that the non-page part of an rpc reply will fit
 164  * in a page - NFSd ensures this.  lockd also has no trouble.
 165  *
 166  * Each request/reply pair can have at most one "payload", plus two pages,
 167  * one for the request, and one for the reply.
 168  * We using ->sendfile to return read data, we might need one extra page
 169  * if the request is not page-aligned.  So add another '1'.
 170  */
 171 #define RPCSVC_MAXPAGES         ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
 172                                 + 2 + 1)
 173
 174 static inline u32 svc_getnl(struct kvec *iov)
 175 {
 176         __be32 val, *vp;
 177         vp = iov->iov_base;
 178         val = *vp++;
 179         iov->iov_base = (void*)vp;
 180         iov->iov_len -= sizeof(__be32);
 181         return ntohl(val);
 182 }
 183
 184 static inline void svc_putnl(struct kvec *iov, u32 val)
 185 {
 186         __be32 *vp = iov->iov_base + iov->iov_len;
 187         *vp = htonl(val);
 188         iov->iov_len += sizeof(__be32);
 189 }
 190
 191 static inline __be32 svc_getu32(struct kvec *iov)
 192 {
 193         __be32 val, *vp;
 194         vp = iov->iov_base;
 195         val = *vp++;
 196         iov->iov_base = (void*)vp;
 197         iov->iov_len -= sizeof(__be32);
 198         return val;
 199 }
 200
 201 static inline void svc_ungetu32(struct kvec *iov)
 202 {
 203         __be32 *vp = (__be32 *)iov->iov_base;
 204         iov->iov_base = (void *)(vp - 1);
 205         iov->iov_len += sizeof(*vp);
 206 }
 207
 208 static inline void svc_putu32(struct kvec *iov, __be32 val)
 209 {
 210         __be32 *vp = iov->iov_base + iov->iov_len;
 211         *vp = val;
 212         iov->iov_len += sizeof(__be32);
 213 }
 214
 215 union svc_addr_u {
 216     struct in_addr      addr;
 217     struct in6_addr     addr6;
 218 };
 219
 220 /*
 221  * The context of a single thread, including the request currently being
 222  * processed.
 223  */
 224 struct svc_rqst {
 225         struct list_head        rq_list;        /* idle list */
 226         struct list_head        rq_all;         /* all threads list */
 227         struct svc_xprt *       rq_xprt;        /* transport ptr */
 228         struct sockaddr_storage rq_addr;        /* peer address */
 229         size_t                  rq_addrlen;
 230
 231         struct svc_serv *       rq_server;      /* RPC service definition */
 232         struct svc_pool *       rq_pool;        /* thread pool */
 233         struct svc_procedure *  rq_procinfo;    /* procedure info */
 234         struct auth_ops *       rq_authop;      /* authentication flavour */
 235         u32                     rq_flavor;      /* pseudoflavor */
 236         struct svc_cred         rq_cred;        /* auth info */
 237         void *                  rq_xprt_ctxt;   /* transport specific context ptr */
 238         struct svc_deferred_req*rq_deferred;    /* deferred request we are replaying */
 239         int                     rq_usedeferral; /* use deferral */
 240
 241         size_t                  rq_xprt_hlen;   /* xprt header len */
 242         struct xdr_buf          rq_arg;
 243         struct xdr_buf          rq_res;
 244         struct page *           rq_pages[RPCSVC_MAXPAGES];
 245         struct page *           *rq_respages;   /* points into rq_pages */
 246         int                     rq_resused;     /* number of pages used for result */
 247
 248         struct kvec             rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
 249
 250         __be32                  rq_xid;         /* transmission id */
 251         u32                     rq_prog;        /* program number */
 252         u32                     rq_vers;        /* program version */
 253         u32                     rq_proc;        /* procedure number */
 254         u32                     rq_prot;        /* IP protocol */
 255         unsigned short
 256                                 rq_secure  : 1; /* secure port */
 257
 258         union svc_addr_u        rq_daddr;       /* dest addr of request
 259                                                  *  - reply from here */
 260
 261         void *                  rq_argp;        /* decoded arguments */
 262         void *                  rq_resp;        /* xdr'd results */
 263         void *                  rq_auth_data;   /* flavor-specific data */
 264
 265         int                     rq_reserved;    /* space on socket outq
 266                                                  * reserved for this request
 267                                                  */
 268
 269         struct cache_req        rq_chandle;     /* handle passed to caches for
 270                                                  * request delaying
 271                                                  */
 272         bool                    rq_dropme;
 273         /* Catering to nfsd */
 274         struct auth_domain *    rq_client;      /* RPC peer info */
 275         struct auth_domain *    rq_gssclient;   /* "gss/"-style peer info */
 276         int                     rq_cachetype;
 277         struct svc_cacherep *   rq_cacherep;    /* cache info */
 278         int                     rq_splice_ok;   /* turned off in gss privacy
 279                                                  * to prevent encrypting page
 280                                                  * cache pages */
 281         wait_queue_head_t       rq_wait;        /* synchronization */
 282         struct task_struct      *rq_task;       /* service thread */
 283 };
 284
 285 /*
 286  * Rigorous type checking on sockaddr type conversions
 287  */
 288 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
 289 {
 290         return (struct sockaddr_in *) &rqst->rq_addr;
 291 }
 292
 293 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
 294 {
 295         return (struct sockaddr_in6 *) &rqst->rq_addr;
 296 }
 297
 298 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
 299 {
 300         return (struct sockaddr *) &rqst->rq_addr;
 301 }
 302
 303 /*
 304  * Check buffer bounds after decoding arguments
 305  */
 306 static inline int
 307 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
 308 {
 309         char *cp = (char *)p;
 310         struct kvec *vec = &rqstp->rq_arg.head[0];
 311         return cp >= (char*)vec->iov_base
 312                 && cp <= (char*)vec->iov_base + vec->iov_len;
 313 }
 314
 315 static inline int
 316 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
 317 {
 318         struct kvec *vec = &rqstp->rq_res.head[0];
 319         char *cp = (char*)p;
 320
 321         vec->iov_len = cp - (char*)vec->iov_base;
 322
 323         return vec->iov_len <= PAGE_SIZE;
 324 }
 325
 326 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
 327 {
 328         while (rqstp->rq_resused) {
 329                 struct page **pp = (rqstp->rq_respages +
 330                                     --rqstp->rq_resused);
 331                 if (*pp) {
 332                         put_page(*pp);
 333                         *pp = NULL;
 334                 }
 335         }
 336 }
 337
 338 struct svc_deferred_req {
 339         u32                     prot;   /* protocol (UDP or TCP) */
 340         struct svc_xprt         *xprt;
 341         struct sockaddr_storage addr;   /* where reply must go */
 342         size_t                  addrlen;
 343         union svc_addr_u        daddr;  /* where reply must come from */
 344         struct cache_deferred_req handle;
 345         size_t                  xprt_hlen;
 346         int                     argslen;
 347         __be32                  args[0];
 348 };
 349
 350 /*
 351  * List of RPC programs on the same transport endpoint
 352  */
 353 struct svc_program {
 354         struct svc_program *    pg_next;        /* other programs (same xprt) */
 355         u32                     pg_prog;        /* program number */
 356         unsigned int            pg_lovers;      /* lowest version */
 357         unsigned int            pg_hivers;      /* lowest version */
 358         unsigned int            pg_nvers;       /* number of versions */
 359         struct svc_version **   pg_vers;        /* version array */
 360         char *                  pg_name;        /* service name */
 361         char *                  pg_class;       /* class name: services sharing authentication */
 362         struct svc_stat *       pg_stats;       /* rpc statistics */
 363         int                     (*pg_authenticate)(struct svc_rqst *);
 364 };
 365
 366 /*
 367  * RPC program version
 368  */
 369 struct svc_version {
 370         u32                     vs_vers;        /* version number */
 371         u32                     vs_nproc;       /* number of procedures */
 372         struct svc_procedure *  vs_proc;        /* per-procedure info */
 373         u32                     vs_xdrsize;     /* xdrsize needed for this version */
 374
 375         unsigned int            vs_hidden : 1;  /* Don't register with portmapper.
 376                                                  * Only used for nfsacl so far. */
 377
 378         /* Override dispatch function (e.g. when caching replies).
 379          * A return value of 0 means drop the request.
 380          * vs_dispatch == NULL means use default dispatcher.
 381          */
 382         int                     (*vs_dispatch)(struct svc_rqst *, __be32 *);
 383 };
 384
 385 /*
 386  * RPC procedure info
 387  */
 388 typedef __be32  (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
 389 struct svc_procedure {
 390         svc_procfunc            pc_func;        /* process the request */
 391         kxdrproc_t              pc_decode;      /* XDR decode args */
 392         kxdrproc_t              pc_encode;      /* XDR encode result */
 393         kxdrproc_t              pc_release;     /* XDR free result */
 394         unsigned int            pc_argsize;     /* argument struct size */
 395         unsigned int            pc_ressize;     /* result struct size */
 396         unsigned int            pc_count;       /* call count */
 397         unsigned int            pc_cachetype;   /* cache info (NFS) */
 398         unsigned int            pc_xdrressize;  /* maximum size of XDR reply */
 399 };
 400
 401 /*
 402  * Function prototypes.
 403  */
 404 struct svc_serv *svc_create(struct svc_program *, unsigned int,
 405                             void (*shutdown)(struct svc_serv *));
 406 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
 407                                         struct svc_pool *pool, int node);
 408 void               svc_exit_thread(struct svc_rqst *);
 409 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
 410                         void (*shutdown)(struct svc_serv *),
 411                         svc_thread_fn, struct module *);
 412 int                svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 413 int                svc_pool_stats_open(struct svc_serv *serv, struct file *file);
 414 void               svc_destroy(struct svc_serv *);
 415 int                svc_process(struct svc_rqst *);
 416 int                bc_svc_process(struct svc_serv *, struct rpc_rqst *,
 417                         struct svc_rqst *);
 418 int                svc_register(const struct svc_serv *, const int,
 419                                 const unsigned short, const unsigned short);
 420
 421 void               svc_wake_up(struct svc_serv *);
 422 void               svc_reserve(struct svc_rqst *rqstp, int space);
 423 struct svc_pool *  svc_pool_for_cpu(struct svc_serv *serv, int cpu);
 424 char *             svc_print_addr(struct svc_rqst *, char *, size_t);
 425
 426 #define RPC_MAX_ADDRBUFLEN      (63U)
 427
 428 /*
 429  * When we want to reduce the size of the reserved space in the response
 430  * buffer, we need to take into account the size of any checksum data that
 431  * may be at the end of the packet. This is difficult to determine exactly
 432  * for all cases without actually generating the checksum, so we just use a
 433  * static value.
 434  */
 435 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
 436 {
 437         int added_space = 0;
 438
 439         if (rqstp->rq_authop->flavour)
 440                 added_space = RPC_MAX_AUTH_SIZE;
 441         svc_reserve(rqstp, space + added_space);
 442 }
 443
 444 #endif /* SUNRPC_SVC_H */