lib/lockfile.c

   1  /* Copyright (c) 2008, 2009, 2010 Nicira Networks
   2  *
   3  * Licensed under the Apache License, Version 2.0 (the "License");
   4  * you may not use this file except in compliance with the License.
   5  * You may obtain a copy of the License at:
   6  *
   7  *     http://www.apache.org/licenses/LICENSE-2.0
   8  *
   9  * Unless required by applicable law or agreed to in writing, software
  10  * distributed under the License is distributed on an "AS IS" BASIS,
  11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12  * See the License for the specific language governing permissions and
  13  * limitations under the License.
  14  */
  15
  16 #include <config.h>
  17
  18 #include "lockfile.h"
  19
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <stdlib.h>
  23 #include <string.h>
  24 #include <sys/stat.h>
  25 #include <unistd.h>
  26
  27 #include "coverage.h"
  28 #include "hash.h"
  29 #include "hmap.h"
  30 #include "timeval.h"
  31 #include "util.h"
  32
  33 #define THIS_MODULE VLM_lockfile
  34 #include "vlog.h"
  35
  36 struct lockfile {
  37     struct hmap_node hmap_node;
  38     char *name;
  39     dev_t device;
  40     ino_t inode;
  41     int fd;
  42 };
  43
  44 /* Lock table.
  45  *
  46  * We have to do this stupid dance because POSIX says that closing *any* file
  47  * descriptor for a file on which a process holds a lock drops *all* locks on
  48  * that file.  That means that we can't afford to open a lockfile more than
  49  * once. */
  50 static struct hmap lock_table = HMAP_INITIALIZER(&lock_table);
  51
  52 static void lockfile_unhash(struct lockfile *);
  53 static int lockfile_try_lock(const char *name, bool block,
  54                              struct lockfile **lockfilep);
  55
  56 /* Returns the name of the lockfile that would be created for locking a file
  57  * named 'file_name'.  The caller is responsible for freeing the returned
  58  * name, with free(), when it is no longer needed. */
  59 char *
  60 lockfile_name(const char *file_name)
  61 {
  62     const char *slash = strrchr(file_name, '/');
  63     return (slash
  64             ? xasprintf("%.*s/.%s.~lock~",
  65                         (int) (slash - file_name), file_name, slash + 1)
  66             : xasprintf(".%s.~lock~", file_name));
  67 }
  68
  69 /* Locks the configuration file against modification by other processes and
  70  * re-reads it from disk.
  71  *
  72  * The 'timeout' specifies the maximum number of milliseconds to wait for the
  73  * config file to become free.  Use 0 to avoid waiting or INT_MAX to wait
  74  * forever.
  75  *
  76  * Returns 0 on success, otherwise a positive errno value.  On success,
  77  * '*lockfilep' is set to point to a new "struct lockfile *" that may be
  78  * unlocked with lockfile_unlock().  On failure, '*lockfilep' is set to
  79  * NULL. */
  80 int
  81 lockfile_lock(const char *file, int timeout, struct lockfile **lockfilep)
  82 {
  83     /* Only exclusive ("write") locks are supported.  This is not a problem
  84      * because the Open vSwitch code that currently uses lock files does so in
  85      * stylized ways such that any number of readers may access a file while it
  86      * is being written. */
  87     long long int warn_elapsed = 1000;
  88     long long int start, elapsed;
  89     char *lock_name;
  90     int error;
  91
  92     COVERAGE_INC(lockfile_lock);
  93
  94     lock_name = lockfile_name(file);
  95     time_refresh();
  96     start = time_msec();
  97
  98     do {
  99         error = lockfile_try_lock(lock_name, timeout > 0, lockfilep);
 100         time_refresh();
 101         elapsed = time_msec() - start;
 102         if (elapsed > warn_elapsed) {
 103             warn_elapsed *= 2;
 104             VLOG_WARN("%s: waiting for lock file, %lld ms elapsed",
 105                       lock_name, elapsed);
 106         }
 107     } while (error == EINTR && (timeout == INT_MAX || elapsed < timeout));
 108
 109     if (!error) {
 110         if (elapsed) {
 111             VLOG_WARN("%s: waited %lld ms for lock file",
 112                       lock_name, elapsed);
 113         }
 114     } else if (error == EINTR) {
 115         COVERAGE_INC(lockfile_timeout);
 116         VLOG_WARN("%s: giving up on lock file after %lld ms",
 117                   lock_name, elapsed);
 118         error = ETIMEDOUT;
 119     } else {
 120         COVERAGE_INC(lockfile_error);
 121         if (error == EACCES) {
 122             error = EAGAIN;
 123         }
 124         VLOG_WARN("%s: failed to lock file "
 125                   "(after %lld ms, with %d-ms timeout): %s",
 126                   lock_name, elapsed, timeout, strerror(error));
 127     }
 128
 129     free(lock_name);
 130     return error;
 131 }
 132
 133 /* Unlocks 'lockfile', which must have been created by a call to
 134  * lockfile_lock(), and frees 'lockfile'. */
 135 void
 136 lockfile_unlock(struct lockfile *lockfile)
 137 {
 138     if (lockfile) {
 139         COVERAGE_INC(lockfile_unlock);
 140         lockfile_unhash(lockfile);
 141         free(lockfile->name);
 142         free(lockfile);
 143     }
 144 }
 145
 146 /* Marks all the currently locked lockfiles as no longer locked.  It makes
 147  * sense to call this function after fork(), because a child created by fork()
 148  * does not hold its parents' locks. */
 149 void
 150 lockfile_postfork(void)
 151 {
 152     struct lockfile *lockfile;
 153
 154     HMAP_FOR_EACH (lockfile, struct lockfile, hmap_node, &lock_table) {
 155         if (lockfile->fd >= 0) {
 156             VLOG_WARN("%s: child does not inherit lock", lockfile->name);
 157             lockfile_unhash(lockfile);
 158         }
 159     }
 160 }
 161 \f
 162 static uint32_t
 163 lockfile_hash(dev_t device, ino_t inode)
 164 {
 165     return hash_bytes(&device, sizeof device,
 166                       hash_bytes(&inode, sizeof inode, 0));
 167 }
 168
 169 static struct lockfile *
 170 lockfile_find(dev_t device, ino_t inode)
 171 {
 172     struct lockfile *lockfile;
 173
 174     HMAP_FOR_EACH_WITH_HASH (lockfile, struct lockfile, hmap_node,
 175                              lockfile_hash(device, inode), &lock_table) {
 176         if (lockfile->device == device && lockfile->inode == inode) {
 177             return lockfile;
 178         }
 179     }
 180     return NULL;
 181 }
 182
 183 static void
 184 lockfile_unhash(struct lockfile *lockfile)
 185 {
 186     if (lockfile->fd >= 0) {
 187         close(lockfile->fd);
 188         lockfile->fd = -1;
 189         hmap_remove(&lock_table, &lockfile->hmap_node);
 190     }
 191 }
 192
 193 static struct lockfile *
 194 lockfile_register(const char *name, dev_t device, ino_t inode, int fd)
 195 {
 196     struct lockfile *lockfile;
 197
 198     lockfile = lockfile_find(device, inode);
 199     if (lockfile) {
 200         VLOG_ERR("%s: lock file disappeared and reappeared!", name);
 201         lockfile_unhash(lockfile);
 202     }
 203
 204     lockfile = xmalloc(sizeof *lockfile);
 205     lockfile->name = xstrdup(name);
 206     lockfile->device = device;
 207     lockfile->inode = inode;
 208     lockfile->fd = fd;
 209     hmap_insert(&lock_table, &lockfile->hmap_node,
 210                 lockfile_hash(device, inode));
 211     return lockfile;
 212 }
 213
 214 static int
 215 lockfile_try_lock(const char *name, bool block, struct lockfile **lockfilep)
 216 {
 217     struct flock l;
 218     struct stat s;
 219     int error;
 220     int fd;
 221
 222     *lockfilep = NULL;
 223
 224     /* Open the lock file, first creating it if necessary. */
 225     for (;;) {
 226         /* Check whether we've already got a lock on that file. */
 227         if (!stat(name, &s)) {
 228             if (lockfile_find(s.st_dev, s.st_ino)) {
 229                 return EDEADLK;
 230             }
 231         } else if (errno != ENOENT) {
 232             VLOG_WARN("%s: failed to stat lock file: %s",
 233                       name, strerror(errno));
 234             return errno;
 235         }
 236
 237         /* Try to open an existing lock file. */
 238         fd = open(name, O_RDWR);
 239         if (fd >= 0) {
 240             break;
 241         } else if (errno != ENOENT) {
 242             VLOG_WARN("%s: failed to open lock file: %s",
 243                       name, strerror(errno));
 244             return errno;
 245         }
 246
 247         /* Try to create a new lock file. */
 248         VLOG_INFO("%s: lock file does not exist, creating", name);
 249         fd = open(name, O_RDWR | O_CREAT | O_EXCL, 0600);
 250         if (fd >= 0) {
 251             break;
 252         } else if (errno != EEXIST) {
 253             VLOG_WARN("%s: failed to create lock file: %s",
 254                       name, strerror(errno));
 255             return errno;
 256         }
 257
 258         /* Someone else created the lock file.  Try again. */
 259     }
 260
 261     /* Get the inode and device number for the lock table. */
 262     if (fstat(fd, &s)) {
 263         VLOG_ERR("%s: failed to fstat lock file: %s", name, strerror(errno));
 264         close(fd);
 265         return errno;
 266     }
 267
 268     /* Try to lock the file. */
 269     memset(&l, 0, sizeof l);
 270     l.l_type = F_WRLCK;
 271     l.l_whence = SEEK_SET;
 272     l.l_start = 0;
 273     l.l_len = 0;
 274
 275     time_disable_restart();
 276     error = fcntl(fd, block ? F_SETLKW : F_SETLK, &l) == -1 ? errno : 0;
 277     time_enable_restart();
 278
 279     if (!error) {
 280         *lockfilep = lockfile_register(name, s.st_dev, s.st_ino, fd);
 281     } else {
 282         close(fd);
 283     }
 284     return error;
 285 }
 286