nscd/nscd_helper.c

   1 /* Copyright (C) 1998-2002,2003,2004,2005,2006 Free Software Foundation, Inc.
   2    This file is part of the GNU C Library.
   3    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, write to the Free
  17    Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  18    02111-1307 USA.  */
  19
  20 #include <assert.h>
  21 #include <errno.h>
  22 #include <fcntl.h>
  23 #include <stdbool.h>
  24 #include <time.h>
  25 #include <unistd.h>
  26 #include <sys/mman.h>
  27 #include <sys/poll.h>
  28 #include <sys/socket.h>
  29 #include <sys/stat.h>
  30 #include <sys/time.h>
  31 #include <sys/uio.h>
  32 #include <sys/un.h>
  33 #include <not-cancel.h>
  34 #include <nis/rpcsvc/nis.h>
  35
  36 #include "nscd-client.h"
  37
  38
  39 ssize_t
  40 __readall (int fd, void *buf, size_t len)
  41 {
  42   size_t n = len;
  43   ssize_t ret;
  44   do
  45     {
  46       ret = TEMP_FAILURE_RETRY (__read (fd, buf, n));
  47       if (ret <= 0)
  48         break;
  49       buf = (char *) buf + ret;
  50       n -= ret;
  51     }
  52   while (n > 0);
  53   return ret < 0 ? ret : len - n;
  54 }
  55
  56
  57 ssize_t
  58 __readvall (int fd, const struct iovec *iov, int iovcnt)
  59 {
  60   ssize_t ret = TEMP_FAILURE_RETRY (__readv (fd, iov, iovcnt));
  61   if (ret <= 0)
  62     return ret;
  63
  64   size_t total = 0;
  65   for (int i = 0; i < iovcnt; ++i)
  66     total += iov[i].iov_len;
  67
  68   if (ret < total)
  69     {
  70       struct iovec iov_buf[iovcnt];
  71       ssize_t r = ret;
  72
  73       struct iovec *iovp = memcpy (iov_buf, iov, iovcnt * sizeof (*iov));
  74       do
  75         {
  76           while (iovp->iov_len <= r)
  77             {
  78               r -= iovp->iov_len;
  79               --iovcnt;
  80               ++iovp;
  81             }
  82           iovp->iov_base = (char *) iovp->iov_base + r;
  83           iovp->iov_len -= r;
  84           r = TEMP_FAILURE_RETRY (__readv (fd, iovp, iovcnt));
  85           if (r <= 0)
  86             break;
  87           ret += r;
  88         }
  89       while (ret < total);
  90       if (r < 0)
  91         ret = r;
  92     }
  93   return ret;
  94 }
  95
  96
  97 static int
  98 open_socket (void)
  99 {
 100   int sock = __socket (PF_UNIX, SOCK_STREAM, 0);
 101   if (sock < 0)
 102     return -1;
 103
 104   /* Make socket non-blocking.  */
 105   int fl = __fcntl (sock, F_GETFL);
 106   if (fl != -1)
 107     __fcntl (sock, F_SETFL, fl | O_NONBLOCK);
 108
 109   struct sockaddr_un sun;
 110   sun.sun_family = AF_UNIX;
 111   strcpy (sun.sun_path, _PATH_NSCDSOCKET);
 112   if (__connect (sock, (struct sockaddr *) &sun, sizeof (sun)) < 0
 113       && errno != EINPROGRESS)
 114     goto out;
 115
 116   struct pollfd fds[1];
 117   fds[0].fd = sock;
 118   fds[0].events = POLLOUT | POLLERR | POLLHUP;
 119   if (__poll (fds, 1, 5 * 1000) > 0)
 120     /* Success.  We do not check for success of the connect call here.
 121        If it failed, the following operations will fail.  */
 122     return sock;
 123
 124  out:
 125   close_not_cancel_no_status (sock);
 126
 127   return -1;
 128 }
 129
 130
 131 void
 132 __nscd_unmap (struct mapped_database *mapped)
 133 {
 134   assert (mapped->counter == 0);
 135   __munmap ((void *) mapped->head, mapped->mapsize);
 136   free (mapped);
 137 }
 138
 139
 140 static int
 141 wait_on_socket (int sock)
 142 {
 143   struct pollfd fds[1];
 144   fds[0].fd = sock;
 145   fds[0].events = POLLIN | POLLERR | POLLHUP;
 146   int n = __poll (fds, 1, 5 * 1000);
 147   if (n == -1 && __builtin_expect (errno == EINTR, 0))
 148     {
 149       /* Handle the case where the poll() call is interrupted by a
 150          signal.  We cannot just use TEMP_FAILURE_RETRY since it might
 151          lead to infinite loops.  */
 152       struct timeval now;
 153       (void) __gettimeofday (&now, NULL);
 154       long int end = (now.tv_sec + 5) * 1000 + (now.tv_usec + 500) / 1000;
 155       while (1)
 156         {
 157           long int timeout = end - (now.tv_sec * 1000
 158                                     + (now.tv_usec + 500) / 1000);
 159           n = __poll (fds, 1, timeout);
 160           if (n != -1 || errno != EINTR)
 161             break;
 162           (void) __gettimeofday (&now, NULL);
 163         }
 164     }
 165
 166   return n;
 167 }
 168
 169
 170 /* Try to get a file descriptor for the shared meory segment
 171    containing the database.  */
 172 static struct mapped_database *
 173 get_mapping (request_type type, const char *key,
 174              struct mapped_database **mappedp)
 175 {
 176   struct mapped_database *result = NO_MAPPING;
 177 #ifdef SCM_RIGHTS
 178   const size_t keylen = strlen (key) + 1;
 179   int saved_errno = errno;
 180
 181   int mapfd = -1;
 182
 183   /* Send the request.  */
 184   struct
 185   {
 186     request_header req;
 187     char key[keylen];
 188   } reqdata;
 189
 190   int sock = open_socket ();
 191   if (sock < 0)
 192     goto out;
 193
 194   reqdata.req.version = NSCD_VERSION;
 195   reqdata.req.type = type;
 196   reqdata.req.key_len = keylen;
 197   memcpy (reqdata.key, key, keylen);
 198
 199 # ifndef MSG_NOSIGNAL
 200 #  define MSG_NOSIGNAL 0
 201 # endif
 202   if (__builtin_expect (TEMP_FAILURE_RETRY (__send (sock, &reqdata,
 203                                                     sizeof (reqdata),
 204                                                     MSG_NOSIGNAL))
 205                         != sizeof (reqdata), 0))
 206     /* We cannot even write the request.  */
 207     goto out_close2;
 208
 209   /* Room for the data sent along with the file descriptor.  We expect
 210      the key name back.  */
 211 # define resdata reqdata.key
 212   struct iovec iov[1];
 213   iov[0].iov_base = resdata;
 214   iov[0].iov_len = keylen;
 215
 216   union
 217   {
 218     struct cmsghdr hdr;
 219     char bytes[CMSG_SPACE (sizeof (int))];
 220   } buf;
 221   struct msghdr msg = { .msg_iov = iov, .msg_iovlen = 1,
 222                         .msg_control = buf.bytes,
 223                         .msg_controllen = sizeof (buf) };
 224   struct cmsghdr *cmsg = CMSG_FIRSTHDR (&msg);
 225
 226   cmsg->cmsg_level = SOL_SOCKET;
 227   cmsg->cmsg_type = SCM_RIGHTS;
 228   cmsg->cmsg_len = CMSG_LEN (sizeof (int));
 229
 230   /* This access is well-aligned since BUF is correctly aligned for an
 231      int and CMSG_DATA preserves this alignment.  */
 232   *(int *) CMSG_DATA (cmsg) = -1;
 233
 234   msg.msg_controllen = cmsg->cmsg_len;
 235
 236   if (wait_on_socket (sock) <= 0)
 237     goto out_close2;
 238
 239   if (__builtin_expect (TEMP_FAILURE_RETRY (__recvmsg (sock, &msg, 0))
 240                         != keylen, 0))
 241     goto out_close2;
 242
 243   mapfd = *(int *) CMSG_DATA (cmsg);
 244
 245   if (__builtin_expect (CMSG_FIRSTHDR (&msg)->cmsg_len
 246                         != CMSG_LEN (sizeof (int)), 0))
 247     goto out_close;
 248
 249   struct stat64 st;
 250   if (__builtin_expect (strcmp (resdata, key) != 0, 0)
 251       || __builtin_expect (fstat64 (mapfd, &st) != 0, 0)
 252       || __builtin_expect (st.st_size < sizeof (struct database_pers_head), 0))
 253     goto out_close;
 254
 255   struct database_pers_head head;
 256   if (__builtin_expect (TEMP_FAILURE_RETRY (__pread (mapfd, &head,
 257                                                      sizeof (head), 0))
 258                         != sizeof (head), 0))
 259     goto out_close;
 260
 261   if (__builtin_expect (head.version != DB_VERSION, 0)
 262       || __builtin_expect (head.header_size != sizeof (head), 0)
 263       /* This really should not happen but who knows, maybe the update
 264          thread got stuck.  */
 265       || __builtin_expect (! head.nscd_certainly_running
 266                            && head.timestamp + MAPPING_TIMEOUT < time (NULL),
 267                            0))
 268     goto out_close;
 269
 270   size_t size = (sizeof (head) + roundup (head.module * sizeof (ref_t), ALIGN)
 271                  + head.data_size);
 272
 273   if (__builtin_expect (st.st_size < size, 0))
 274     goto out_close;
 275
 276   /* The file is large enough, map it now.  */
 277   void *mapping = __mmap (NULL, size, PROT_READ, MAP_SHARED, mapfd, 0);
 278   if (__builtin_expect (mapping != MAP_FAILED, 1))
 279     {
 280       /* Allocate a record for the mapping.  */
 281       struct mapped_database *newp = malloc (sizeof (*newp));
 282       if (newp == NULL)
 283         {
 284           /* Ugh, after all we went through the memory allocation failed.  */
 285           __munmap (mapping, size);
 286           goto out_close;
 287         }
 288
 289       newp->head = mapping;
 290       newp->data = ((char *) mapping + head.header_size
 291                     + roundup (head.module * sizeof (ref_t), ALIGN));
 292       newp->mapsize = size;
 293       newp->datasize = head.data_size;
 294       /* Set counter to 1 to show it is usable.  */
 295       newp->counter = 1;
 296
 297       result = newp;
 298     }
 299
 300  out_close:
 301   __close (mapfd);
 302  out_close2:
 303   __close (sock);
 304  out:
 305   __set_errno (saved_errno);
 306 #endif  /* SCM_RIGHTS */
 307
 308   struct mapped_database *oldval = *mappedp;
 309   *mappedp = result;
 310
 311   if (oldval != NULL && atomic_decrement_val (&oldval->counter) == 0)
 312     __nscd_unmap (oldval);
 313
 314   return result;
 315 }
 316
 317
 318 struct mapped_database *
 319 __nscd_get_map_ref (request_type type, const char *name,
 320                     volatile struct locked_map_ptr *mapptr, int *gc_cyclep)
 321 {
 322   struct mapped_database *cur = mapptr->mapped;
 323   if (cur == NO_MAPPING)
 324     return cur;
 325
 326   int cnt = 0;
 327   while (__builtin_expect (atomic_compare_and_exchange_val_acq (&mapptr->lock,
 328                                                                 1, 0) != 0, 0))
 329     {
 330       // XXX Best number of rounds?
 331       if (__builtin_expect (++cnt > 5, 0))
 332         return NO_MAPPING;
 333
 334       atomic_delay ();
 335     }
 336
 337   cur = mapptr->mapped;
 338
 339   if (__builtin_expect (cur != NO_MAPPING, 1))
 340     {
 341       /* If not mapped or timestamp not updated, request new map.  */
 342       if (cur == NULL
 343           || (cur->head->nscd_certainly_running == 0
 344               && cur->head->timestamp + MAPPING_TIMEOUT < time (NULL))
 345           || cur->head->data_size > cur->datasize)
 346         cur = get_mapping (type, name,
 347                            (struct mapped_database **) &mapptr->mapped);
 348
 349       if (__builtin_expect (cur != NO_MAPPING, 1))
 350         {
 351           if (__builtin_expect (((*gc_cyclep = cur->head->gc_cycle) & 1) != 0,
 352                                 0))
 353             cur = NO_MAPPING;
 354           else
 355             atomic_increment (&cur->counter);
 356         }
 357     }
 358
 359   mapptr->lock = 0;
 360
 361   return cur;
 362 }
 363
 364
 365 const struct datahead *
 366 __nscd_cache_search (request_type type, const char *key, size_t keylen,
 367                      const struct mapped_database *mapped)
 368 {
 369   unsigned long int hash = __nis_hash (key, keylen) % mapped->head->module;
 370   size_t datasize = mapped->datasize;
 371
 372   ref_t work = mapped->head->array[hash];
 373   while (work != ENDREF && work + sizeof (struct hashentry) <= datasize)
 374     {
 375       struct hashentry *here = (struct hashentry *) (mapped->data + work);
 376
 377       if (type == here->type
 378           && keylen == here->len
 379           && here->key + here->len <= datasize
 380           && memcmp (key, mapped->data + here->key, keylen) == 0
 381           && here->packet + sizeof (struct datahead) <= datasize)
 382         {
 383           /* We found the entry.  Increment the appropriate counter.  */
 384           const struct datahead *dh
 385             = (struct datahead *) (mapped->data + here->packet);
 386
 387           /* See whether we must ignore the entry or whether something
 388              is wrong because garbage collection is in progress.  */
 389           if (dh->usable && here->packet + dh->allocsize <= datasize)
 390             return dh;
 391         }
 392
 393       work = here->next;
 394     }
 395
 396   return NULL;
 397 }
 398
 399
 400 /* Create a socket connected to a name. */
 401 int
 402 __nscd_open_socket (const char *key, size_t keylen, request_type type,
 403                     void *response, size_t responselen)
 404 {
 405   int saved_errno = errno;
 406
 407   int sock = open_socket ();
 408   if (sock >= 0)
 409     {
 410       request_header req;
 411       req.version = NSCD_VERSION;
 412       req.type = type;
 413       req.key_len = keylen;
 414
 415       struct iovec vec[2];
 416       vec[0].iov_base = &req;
 417       vec[0].iov_len = sizeof (request_header);
 418       vec[1].iov_base = (void *) key;
 419       vec[1].iov_len = keylen;
 420
 421       ssize_t nbytes = TEMP_FAILURE_RETRY (__writev (sock, vec, 2));
 422       if (nbytes == (ssize_t) (sizeof (request_header) + keylen)
 423           /* Wait for data.  */
 424           && wait_on_socket (sock) > 0)
 425         {
 426           nbytes = TEMP_FAILURE_RETRY (__read (sock, response, responselen));
 427           if (nbytes == (ssize_t) responselen)
 428             return sock;
 429         }
 430
 431       close_not_cancel_no_status (sock);
 432     }
 433
 434   __set_errno (saved_errno);
 435
 436   return -1;
 437 }