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