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 #define _LGPL_SOURCE
  22 #include "shm.h"
  23 #include <unistd.h>
  24 #include <fcntl.h>
  25 #include <sys/mman.h>
  26 #include <sys/types.h>
  27 #include <sys/stat.h>   /* For mode constants */
  28 #include <fcntl.h>      /* For O_* constants */
  29 #include <assert.h>
  30 #include <stdio.h>
  31 #include <signal.h>
  32 #include <dirent.h>
  33 #include <lttng/align.h>
  34 #include <limits.h>
  35 #include <helper.h>
  36 #include <ust-fd.h>
  37
  38 /*
  39  * Ensure we have the required amount of space available by writing 0
  40  * into the entire buffer. Not doing so can trigger SIGBUS when going
  41  * beyond the available shm space.
  42  *
  43  * Also ensure the file metadata is synced with the storage by using
  44  * fsync(2).
  45  */
  46 static
  47 int zero_file(int fd, size_t len)
  48 {
  49         ssize_t retlen;
  50         size_t written = 0;
  51         char *zeropage;
  52         long pagelen;
  53         int ret;
  54
  55         pagelen = sysconf(_SC_PAGESIZE);
  56         if (pagelen < 0)
  57                 return (int) pagelen;
  58         zeropage = calloc(pagelen, 1);
  59         if (!zeropage)
  60                 return -ENOMEM;
  61
  62         while (len > written) {
  63                 do {
  64                         retlen = write(fd, zeropage,
  65                                 min_t(size_t, pagelen, len - written));
  66                 } while (retlen == -1UL && errno == EINTR);
  67                 if (retlen < 0) {
  68                         ret = (int) retlen;
  69                         goto error;
  70                 }
  71                 written += retlen;
  72         }
  73         ret = fsync(fd);
  74         if (ret) {
  75                 ret = (int) -errno;
  76                 goto error;
  77         }
  78         ret = 0;
  79 error:
  80         free(zeropage);
  81         return ret;
  82 }
  83
  84 struct shm_object_table *shm_object_table_create(size_t max_nb_obj)
  85 {
  86         struct shm_object_table *table;
  87
  88         table = zmalloc(sizeof(struct shm_object_table) +
  89                         max_nb_obj * sizeof(table->objects[0]));
  90         if (!table)
  91                 return NULL;
  92         table->size = max_nb_obj;
  93         return table;
  94 }
  95
  96 static
  97 struct shm_object *_shm_object_table_alloc_shm(struct shm_object_table *table,
  98                                            size_t memory_map_size,
  99                                            int stream_fd)
 100 {
 101         int shmfd, waitfd[2], ret, i;
 102         struct shm_object *obj;
 103         char *memory_map;
 104
 105         if (stream_fd < 0)
 106                 return NULL;
 107         if (table->allocated_len >= table->size)
 108                 return NULL;
 109         obj = &table->objects[table->allocated_len];
 110
 111         /* wait_fd: create pipe */
 112         ret = pipe(waitfd);
 113         if (ret < 0) {
 114                 PERROR("pipe");
 115                 goto error_pipe;
 116         }
 117         for (i = 0; i < 2; i++) {
 118                 ret = fcntl(waitfd[i], F_SETFD, FD_CLOEXEC);
 119                 if (ret < 0) {
 120                         PERROR("fcntl");
 121                         goto error_fcntl;
 122                 }
 123         }
 124         /* The write end of the pipe needs to be non-blocking */
 125         ret = fcntl(waitfd[1], F_SETFL, O_NONBLOCK);
 126         if (ret < 0) {
 127                 PERROR("fcntl");
 128                 goto error_fcntl;
 129         }
 130         memcpy(obj->wait_fd, waitfd, sizeof(waitfd));
 131
 132         /* create shm */
 133
 134         shmfd = stream_fd;
 135         ret = zero_file(shmfd, memory_map_size);
 136         if (ret) {
 137                 PERROR("zero_file");
 138                 goto error_zero_file;
 139         }
 140         ret = ftruncate(shmfd, memory_map_size);
 141         if (ret) {
 142                 PERROR("ftruncate");
 143                 goto error_ftruncate;
 144         }
 145         obj->shm_fd_ownership = 0;
 146         obj->shm_fd = shmfd;
 147
 148         /* memory_map: mmap */
 149         memory_map = mmap(NULL, memory_map_size, PROT_READ | PROT_WRITE,
 150                           MAP_SHARED, shmfd, 0);
 151         if (memory_map == MAP_FAILED) {
 152                 PERROR("mmap");
 153                 goto error_mmap;
 154         }
 155         obj->type = SHM_OBJECT_SHM;
 156         obj->memory_map = memory_map;
 157         obj->memory_map_size = memory_map_size;
 158         obj->allocated_len = 0;
 159         obj->index = table->allocated_len++;
 160
 161         return obj;
 162
 163 error_mmap:
 164 error_ftruncate:
 165 error_zero_file:
 166 error_fcntl:
 167         for (i = 0; i < 2; i++) {
 168                 ret = close(waitfd[i]);
 169                 if (ret) {
 170                         PERROR("close");
 171                         assert(0);
 172                 }
 173         }
 174 error_pipe:
 175         return NULL;
 176 }
 177
 178 static
 179 struct shm_object *_shm_object_table_alloc_mem(struct shm_object_table *table,
 180                                            size_t memory_map_size)
 181 {
 182         struct shm_object *obj;
 183         void *memory_map;
 184         int waitfd[2], i, ret;
 185
 186         if (table->allocated_len >= table->size)
 187                 return NULL;
 188         obj = &table->objects[table->allocated_len];
 189
 190         memory_map = zmalloc(memory_map_size);
 191         if (!memory_map)
 192                 goto alloc_error;
 193
 194         /* wait_fd: create pipe */
 195         ret = pipe(waitfd);
 196         if (ret < 0) {
 197                 PERROR("pipe");
 198                 goto error_pipe;
 199         }
 200         for (i = 0; i < 2; i++) {
 201                 ret = fcntl(waitfd[i], F_SETFD, FD_CLOEXEC);
 202                 if (ret < 0) {
 203                         PERROR("fcntl");
 204                         goto error_fcntl;
 205                 }
 206         }
 207         /* The write end of the pipe needs to be non-blocking */
 208         ret = fcntl(waitfd[1], F_SETFL, O_NONBLOCK);
 209         if (ret < 0) {
 210                 PERROR("fcntl");
 211                 goto error_fcntl;
 212         }
 213         memcpy(obj->wait_fd, waitfd, sizeof(waitfd));
 214
 215         /* no shm_fd */
 216         obj->shm_fd = -1;
 217         obj->shm_fd_ownership = 0;
 218
 219         obj->type = SHM_OBJECT_MEM;
 220         obj->memory_map = memory_map;
 221         obj->memory_map_size = memory_map_size;
 222         obj->allocated_len = 0;
 223         obj->index = table->allocated_len++;
 224
 225         return obj;
 226
 227 error_fcntl:
 228         for (i = 0; i < 2; i++) {
 229                 ret = close(waitfd[i]);
 230                 if (ret) {
 231                         PERROR("close");
 232                         assert(0);
 233                 }
 234         }
 235 error_pipe:
 236         free(memory_map);
 237 alloc_error:
 238         return NULL;
 239 }
 240
 241 struct shm_object *shm_object_table_alloc(struct shm_object_table *table,
 242                         size_t memory_map_size,
 243                         enum shm_object_type type,
 244                         int stream_fd)
 245 {
 246         switch (type) {
 247         case SHM_OBJECT_SHM:
 248                 return _shm_object_table_alloc_shm(table, memory_map_size,
 249                                 stream_fd);
 250         case SHM_OBJECT_MEM:
 251                 return _shm_object_table_alloc_mem(table, memory_map_size);
 252         default:
 253                 assert(0);
 254         }
 255         return NULL;
 256 }
 257
 258 struct shm_object *shm_object_table_append_shm(struct shm_object_table *table,
 259                         int shm_fd, int wakeup_fd, uint32_t stream_nr,
 260                         size_t memory_map_size)
 261 {
 262         struct shm_object *obj;
 263         char *memory_map;
 264         int ret;
 265
 266         if (table->allocated_len >= table->size)
 267                 return NULL;
 268         /* streams _must_ be received in sequential order, else fail. */
 269         if (stream_nr + 1 != table->allocated_len)
 270                 return NULL;
 271
 272         obj = &table->objects[table->allocated_len];
 273
 274         /* wait_fd: set write end of the pipe. */
 275         obj->wait_fd[0] = -1;   /* read end is unset */
 276         obj->wait_fd[1] = wakeup_fd;
 277         obj->shm_fd = shm_fd;
 278         obj->shm_fd_ownership = 1;
 279
 280         ret = fcntl(obj->wait_fd[1], F_SETFD, FD_CLOEXEC);
 281         if (ret < 0) {
 282                 PERROR("fcntl");
 283                 goto error_fcntl;
 284         }
 285         /* The write end of the pipe needs to be non-blocking */
 286         ret = fcntl(obj->wait_fd[1], F_SETFL, O_NONBLOCK);
 287         if (ret < 0) {
 288                 PERROR("fcntl");
 289                 goto error_fcntl;
 290         }
 291
 292         /* memory_map: mmap */
 293         memory_map = mmap(NULL, memory_map_size, PROT_READ | PROT_WRITE,
 294                           MAP_SHARED, shm_fd, 0);
 295         if (memory_map == MAP_FAILED) {
 296                 PERROR("mmap");
 297                 goto error_mmap;
 298         }
 299         obj->type = SHM_OBJECT_SHM;
 300         obj->memory_map = memory_map;
 301         obj->memory_map_size = memory_map_size;
 302         obj->allocated_len = memory_map_size;
 303         obj->index = table->allocated_len++;
 304
 305         return obj;
 306
 307 error_fcntl:
 308 error_mmap:
 309         return NULL;
 310 }
 311
 312 /*
 313  * Passing ownership of mem to object.
 314  */
 315 struct shm_object *shm_object_table_append_mem(struct shm_object_table *table,
 316                         void *mem, size_t memory_map_size, int wakeup_fd)
 317 {
 318         struct shm_object *obj;
 319         int ret;
 320
 321         if (table->allocated_len >= table->size)
 322                 return NULL;
 323         obj = &table->objects[table->allocated_len];
 324
 325         obj->wait_fd[0] = -1;   /* read end is unset */
 326         obj->wait_fd[1] = wakeup_fd;
 327         obj->shm_fd = -1;
 328         obj->shm_fd_ownership = 0;
 329
 330         ret = fcntl(obj->wait_fd[1], F_SETFD, FD_CLOEXEC);
 331         if (ret < 0) {
 332                 PERROR("fcntl");
 333                 goto error_fcntl;
 334         }
 335         /* The write end of the pipe needs to be non-blocking */
 336         ret = fcntl(obj->wait_fd[1], F_SETFL, O_NONBLOCK);
 337         if (ret < 0) {
 338                 PERROR("fcntl");
 339                 goto error_fcntl;
 340         }
 341
 342         obj->type = SHM_OBJECT_MEM;
 343         obj->memory_map = mem;
 344         obj->memory_map_size = memory_map_size;
 345         obj->allocated_len = memory_map_size;
 346         obj->index = table->allocated_len++;
 347
 348         return obj;
 349
 350 error_fcntl:
 351         return NULL;
 352 }
 353
 354 static
 355 void shmp_object_destroy(struct shm_object *obj, int consumer)
 356 {
 357         switch (obj->type) {
 358         case SHM_OBJECT_SHM:
 359         {
 360                 int ret, i;
 361
 362                 ret = munmap(obj->memory_map, obj->memory_map_size);
 363                 if (ret) {
 364                         PERROR("umnmap");
 365                         assert(0);
 366                 }
 367
 368                 if (obj->shm_fd_ownership) {
 369                         /* Delete FDs only if called from app (not consumer). */
 370                         if (!consumer) {
 371                                 lttng_ust_lock_fd_tracker();
 372                                 ret = close(obj->shm_fd);
 373                                 if (!ret) {
 374                                         lttng_ust_delete_fd_from_tracker(obj->shm_fd);
 375                                 } else {
 376                                         PERROR("close");
 377                                         assert(0);
 378                                 }
 379                                 lttng_ust_unlock_fd_tracker();
 380                         } else {
 381                                 ret = close(obj->shm_fd);
 382                                 if (ret) {
 383                                         PERROR("close");
 384                                         assert(0);
 385                                 }
 386                         }
 387                 }
 388                 for (i = 0; i < 2; i++) {
 389                         if (obj->wait_fd[i] < 0)
 390                                 continue;
 391                         if (!consumer) {
 392                                 lttng_ust_lock_fd_tracker();
 393                                 ret = close(obj->wait_fd[i]);
 394                                 if (!ret) {
 395                                         lttng_ust_delete_fd_from_tracker(obj->wait_fd[i]);
 396                                 } else {
 397                                         PERROR("close");
 398                                         assert(0);
 399                                 }
 400                                 lttng_ust_unlock_fd_tracker();
 401                         } else {
 402                                 ret = close(obj->wait_fd[i]);
 403                                 if (ret) {
 404                                         PERROR("close");
 405                                         assert(0);
 406                                 }
 407                         }
 408                 }
 409                 break;
 410         }
 411         case SHM_OBJECT_MEM:
 412         {
 413                 int ret, i;
 414
 415                 for (i = 0; i < 2; i++) {
 416                         if (obj->wait_fd[i] < 0)
 417                                 continue;
 418                         if (!consumer) {
 419                                 lttng_ust_lock_fd_tracker();
 420                                 ret = close(obj->wait_fd[i]);
 421                                 if (!ret) {
 422                                         lttng_ust_delete_fd_from_tracker(obj->wait_fd[i]);
 423                                 } else {
 424                                         PERROR("close");
 425                                         assert(0);
 426                                 }
 427                                 lttng_ust_unlock_fd_tracker();
 428                         } else {
 429                                 ret = close(obj->wait_fd[i]);
 430                                 if (ret) {
 431                                         PERROR("close");
 432                                         assert(0);
 433                                 }
 434                         }
 435                 }
 436                 free(obj->memory_map);
 437                 break;
 438         }
 439         default:
 440                 assert(0);
 441         }
 442 }
 443
 444 void shm_object_table_destroy(struct shm_object_table *table, int consumer)
 445 {
 446         int i;
 447
 448         for (i = 0; i < table->allocated_len; i++)
 449                 shmp_object_destroy(&table->objects[i], consumer);
 450         free(table);
 451 }
 452
 453 /*
 454  * zalloc_shm - allocate memory within a shm object.
 455  *
 456  * Shared memory is already zeroed by shmget.
 457  * *NOT* multithread-safe (should be protected by mutex).
 458  * Returns a -1, -1 tuple on error.
 459  */
 460 struct shm_ref zalloc_shm(struct shm_object *obj, size_t len)
 461 {
 462         struct shm_ref ref;
 463         struct shm_ref shm_ref_error = { -1, -1 };
 464
 465         if (obj->memory_map_size - obj->allocated_len < len)
 466                 return shm_ref_error;
 467         ref.index = obj->index;
 468         ref.offset =  obj->allocated_len;
 469         obj->allocated_len += len;
 470         return ref;
 471 }
 472
 473 void align_shm(struct shm_object *obj, size_t align)
 474 {
 475         size_t offset_len = offset_align(obj->allocated_len, align);
 476         obj->allocated_len += offset_len;
 477 }