libringbuffer/shm.c

   1 /*
   2  * libringbuffer/shm.c
   3  *
   4  * Copyright (C) 2005-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; only
   9  * version 2.1 of the License.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "shm.h"
  22 #include <unistd.h>
  23 #include <fcntl.h>
  24 #include <sys/mman.h>
  25 #include <sys/stat.h>   /* For mode constants */
  26 #include <fcntl.h>      /* For O_* constants */
  27 #include <assert.h>
  28 #include <stdio.h>
  29 #include <signal.h>
  30 #include <dirent.h>
  31 #include <lttng/align.h>
  32 #include <helper.h>
  33 #include <limits.h>
  34 #include <helper.h>
  35
  36 /*
  37  * Ensure we have the required amount of space available by writing 0
  38  * into the entire buffer. Not doing so can trigger SIGBUS when going
  39  * beyond the available shm space.
  40  */
  41 static
  42 int zero_file(int fd, size_t len)
  43 {
  44         ssize_t retlen;
  45         size_t written = 0;
  46         char *zeropage;
  47         long pagelen;
  48         int ret;
  49
  50         pagelen = sysconf(_SC_PAGESIZE);
  51         if (pagelen < 0)
  52                 return (int) pagelen;
  53         zeropage = calloc(pagelen, 1);
  54         if (!zeropage)
  55                 return -ENOMEM;
  56
  57         while (len > written) {
  58                 do {
  59                         retlen = write(fd, zeropage,
  60                                 min_t(size_t, pagelen, len - written));
  61                 } while (retlen == -1UL && errno == EINTR);
  62                 if (retlen < 0) {
  63                         ret = (int) retlen;
  64                         goto error;
  65                 }
  66                 written += retlen;
  67         }
  68         ret = 0;
  69 error:
  70         free(zeropage);
  71         return ret;
  72 }
  73
  74 struct shm_object_table *shm_object_table_create(size_t max_nb_obj)
  75 {
  76         struct shm_object_table *table;
  77
  78         table = zmalloc(sizeof(struct shm_object_table) +
  79                         max_nb_obj * sizeof(table->objects[0]));
  80         table->size = max_nb_obj;
  81         return table;
  82 }
  83
  84 struct shm_object *shm_object_table_append(struct shm_object_table *table,
  85                                            size_t memory_map_size)
  86 {
  87         int shmfd, waitfd[2], ret, i, sigblocked = 0;
  88         struct shm_object *obj;
  89         char *memory_map;
  90         char tmp_name[NAME_MAX] = "ust-shm-tmp-XXXXXX";
  91         sigset_t all_sigs, orig_sigs;
  92
  93         if (table->allocated_len >= table->size)
  94                 return NULL;
  95         obj = &table->objects[table->allocated_len];
  96
  97         /* wait_fd: create pipe */
  98         ret = pipe(waitfd);
  99         if (ret < 0) {
 100                 PERROR("pipe");
 101                 goto error_pipe;
 102         }
 103         for (i = 0; i < 2; i++) {
 104                 ret = fcntl(waitfd[i], F_SETFD, FD_CLOEXEC);
 105                 if (ret < 0) {
 106                         PERROR("fcntl");
 107                         goto error_fcntl;
 108                 }
 109         }
 110         /* The write end of the pipe needs to be non-blocking */
 111         ret = fcntl(waitfd[1], F_SETFL, O_NONBLOCK);
 112         if (ret < 0) {
 113                 PERROR("fcntl");
 114                 goto error_fcntl;
 115         }
 116         memcpy(obj->wait_fd, waitfd, sizeof(waitfd));
 117
 118         /* shm_fd: create shm */
 119
 120         /*
 121          * Theoretically, we could leak a shm if the application crashes
 122          * between open and unlink. Disable signals on this thread for
 123          * increased safety against this scenario.
 124          */
 125         sigfillset(&all_sigs);
 126         ret = pthread_sigmask(SIG_BLOCK, &all_sigs, &orig_sigs);
 127         if (ret == -1) {
 128                 PERROR("pthread_sigmask");
 129                 goto error_pthread_sigmask;
 130         }
 131         sigblocked = 1;
 132
 133         /*
 134          * Allocate shm, and immediately unlink its shm oject, keeping
 135          * only the file descriptor as a reference to the object. If it
 136          * already exists (caused by short race window during which the
 137          * global object exists in a concurrent shm_open), simply retry.
 138          * We specifically do _not_ use the / at the beginning of the
 139          * pathname so that some OS implementations can keep it local to
 140          * the process (POSIX leaves this implementation-defined).
 141          */
 142         do {
 143                 /*
 144                  * Using mktemp filename with O_CREAT | O_EXCL open
 145                  * flags.
 146                  */
 147                 mktemp(tmp_name);
 148                 if (tmp_name[0] == '\0') {
 149                         PERROR("mktemp");
 150                         goto error_shm_open;
 151                 }
 152                 shmfd = shm_open(tmp_name,
 153                                  O_CREAT | O_EXCL | O_RDWR, 0700);
 154         } while (shmfd < 0 && (errno == EEXIST || errno == EACCES));
 155         if (shmfd < 0) {
 156                 PERROR("shm_open");
 157                 goto error_shm_open;
 158         }
 159         ret = shm_unlink(tmp_name);
 160         if (ret < 0 && errno != ENOENT) {
 161                 PERROR("shm_unlink");
 162                 goto error_shm_release;
 163         }
 164         sigblocked = 0;
 165         ret = pthread_sigmask(SIG_SETMASK, &orig_sigs, NULL);
 166         if (ret == -1) {
 167                 PERROR("pthread_sigmask");
 168                 goto error_sigmask_release;
 169         }
 170         ret = zero_file(shmfd, memory_map_size);
 171         if (ret) {
 172                 PERROR("zero_file");
 173                 goto error_zero_file;
 174         }
 175         ret = ftruncate(shmfd, memory_map_size);
 176         if (ret) {
 177                 PERROR("ftruncate");
 178                 goto error_ftruncate;
 179         }
 180         obj->shm_fd = shmfd;
 181
 182         /* memory_map: mmap */
 183         memory_map = mmap(NULL, memory_map_size, PROT_READ | PROT_WRITE,
 184                           MAP_SHARED, shmfd, 0);
 185         if (memory_map == MAP_FAILED) {
 186                 PERROR("mmap");
 187                 goto error_mmap;
 188         }
 189         obj->memory_map = memory_map;
 190         obj->memory_map_size = memory_map_size;
 191         obj->allocated_len = 0;
 192         obj->index = table->allocated_len++;
 193
 194         return obj;
 195
 196 error_mmap:
 197 error_ftruncate:
 198 error_shm_release:
 199 error_zero_file:
 200 error_sigmask_release:
 201         ret = close(shmfd);
 202         if (ret) {
 203                 PERROR("close");
 204                 assert(0);
 205         }
 206 error_shm_open:
 207         if (sigblocked) {
 208                 ret = pthread_sigmask(SIG_SETMASK, &orig_sigs, NULL);
 209                 if (ret == -1) {
 210                         PERROR("pthread_sigmask");
 211                 }
 212         }
 213 error_pthread_sigmask:
 214 error_fcntl:
 215         for (i = 0; i < 2; i++) {
 216                 ret = close(waitfd[i]);
 217                 if (ret) {
 218                         PERROR("close");
 219                         assert(0);
 220                 }
 221         }
 222 error_pipe:
 223         return NULL;
 224
 225 }
 226
 227 struct shm_object *shm_object_table_append_shadow(struct shm_object_table *table,
 228                         int shm_fd, int wait_fd, size_t memory_map_size)
 229 {
 230         struct shm_object *obj;
 231         char *memory_map;
 232
 233         if (table->allocated_len >= table->size)
 234                 return NULL;
 235         obj = &table->objects[table->allocated_len];
 236
 237         /* wait_fd: set read end of the pipe. */
 238         obj->wait_fd[0] = wait_fd;
 239         obj->wait_fd[1] = -1;   /* write end is unset. */
 240         obj->shm_fd = shm_fd;
 241
 242         /* memory_map: mmap */
 243         memory_map = mmap(NULL, memory_map_size, PROT_READ | PROT_WRITE,
 244                           MAP_SHARED, shm_fd, 0);
 245         if (memory_map == MAP_FAILED) {
 246                 PERROR("mmap");
 247                 goto error_mmap;
 248         }
 249         obj->memory_map = memory_map;
 250         obj->memory_map_size = memory_map_size;
 251         obj->allocated_len = memory_map_size;
 252         obj->index = table->allocated_len++;
 253
 254         return obj;
 255
 256 error_mmap:
 257         return NULL;
 258 }
 259
 260 static
 261 void shmp_object_destroy(struct shm_object *obj)
 262 {
 263         int ret, i;
 264
 265         if (!obj->is_shadow) {
 266                 ret = munmap(obj->memory_map, obj->memory_map_size);
 267                 if (ret) {
 268                         PERROR("umnmap");
 269                         assert(0);
 270                 }
 271         }
 272         if (obj->shm_fd >= 0) {
 273                 ret = close(obj->shm_fd);
 274                 if (ret) {
 275                         PERROR("close");
 276                         assert(0);
 277                 }
 278         }
 279         for (i = 0; i < 2; i++) {
 280                 if (obj->wait_fd[i] < 0)
 281                         continue;
 282                 ret = close(obj->wait_fd[i]);
 283                 if (ret) {
 284                         PERROR("close");
 285                         assert(0);
 286                 }
 287         }
 288 }
 289
 290 void shm_object_table_destroy(struct shm_object_table *table)
 291 {
 292         int i;
 293
 294         for (i = 0; i < table->allocated_len; i++)
 295                 shmp_object_destroy(&table->objects[i]);
 296         free(table);
 297 }
 298
 299 /*
 300  * zalloc_shm - allocate memory within a shm object.
 301  *
 302  * Shared memory is already zeroed by shmget.
 303  * *NOT* multithread-safe (should be protected by mutex).
 304  * Returns a -1, -1 tuple on error.
 305  */
 306 struct shm_ref zalloc_shm(struct shm_object *obj, size_t len)
 307 {
 308         struct shm_ref ref;
 309         struct shm_ref shm_ref_error = { -1, -1 };
 310
 311         if (obj->memory_map_size - obj->allocated_len < len)
 312                 return shm_ref_error;
 313         ref.index = obj->index;
 314         ref.offset =  obj->allocated_len;
 315         obj->allocated_len += len;
 316         return ref;
 317 }
 318
 319 void align_shm(struct shm_object *obj, size_t align)
 320 {
 321         size_t offset_len = offset_align(obj->allocated_len, align);
 322         obj->allocated_len += offset_len;
 323 }