nscd/nscd_helper.c

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