src/common/compat/compat-epoll.c

   1 /*
   2  * Copyright (C)  2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _LGPL_SOURCE
  19 #include <assert.h>
  20 #include <fcntl.h>
  21 #include <limits.h>
  22 #include <stdlib.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <unistd.h>
  26
  27 #include <common/error.h>
  28 #include <common/defaults.h>
  29 #include <common/macros.h>
  30 #include <common/utils.h>
  31
  32 #include "poll.h"
  33
  34 unsigned int poll_max_size;
  35
  36 /*
  37  * Resize the epoll events structure of the new size.
  38  *
  39  * Return 0 on success or else -1 with the current events pointer untouched.
  40  */
  41 static int resize_poll_event(struct lttng_poll_event *events,
  42                 uint32_t new_size)
  43 {
  44         struct epoll_event *ptr;
  45
  46         assert(events);
  47
  48         ptr = realloc(events->events, new_size * sizeof(*ptr));
  49         if (ptr == NULL) {
  50                 PERROR("realloc epoll add");
  51                 goto error;
  52         }
  53         if (new_size > events->alloc_size) {
  54                 /* Zero newly allocated memory */
  55                 memset(ptr + events->alloc_size, 0,
  56                         (new_size - events->alloc_size) * sizeof(*ptr));
  57         }
  58         events->events = ptr;
  59         events->alloc_size = new_size;
  60
  61         return 0;
  62
  63 error:
  64         return -1;
  65 }
  66
  67 /*
  68  * Create epoll set and allocate returned events structure.
  69  */
  70 int compat_epoll_create(struct lttng_poll_event *events, int size, int flags)
  71 {
  72         int ret;
  73
  74         if (events == NULL || size <= 0) {
  75                 goto error;
  76         }
  77
  78         if (!poll_max_size) {
  79                 if (lttng_poll_set_max_size()) {
  80                         goto error;
  81                 }
  82         }
  83
  84         /* Don't bust the limit here */
  85         if (size > poll_max_size) {
  86                 size = poll_max_size;
  87         }
  88
  89         ret = compat_glibc_epoll_create(size, flags);
  90         if (ret < 0) {
  91                 /* At this point, every error is fatal */
  92                 PERROR("epoll_create1");
  93                 goto error;
  94         }
  95
  96         events->epfd = ret;
  97
  98         /* This *must* be freed by using lttng_poll_free() */
  99         events->events = zmalloc(size * sizeof(struct epoll_event));
 100         if (events->events == NULL) {
 101                 PERROR("zmalloc epoll set");
 102                 goto error_close;
 103         }
 104
 105         events->alloc_size = events->init_size = size;
 106         events->nb_fd = 0;
 107
 108         return 0;
 109
 110 error_close:
 111         ret = close(events->epfd);
 112         if (ret) {
 113                 PERROR("close");
 114         }
 115 error:
 116         return -1;
 117 }
 118
 119 /*
 120  * Add a fd to the epoll set with requesting events.
 121  */
 122 int compat_epoll_add(struct lttng_poll_event *events, int fd, uint32_t req_events)
 123 {
 124         int ret;
 125         struct epoll_event ev;
 126
 127         if (events == NULL || events->events == NULL || fd < 0) {
 128                 ERR("Bad compat epoll add arguments");
 129                 goto error;
 130         }
 131
 132         /*
 133          * Zero struct epoll_event to ensure all representations of its
 134          * union are zeroed.
 135          */
 136         memset(&ev, 0, sizeof(ev));
 137         ev.events = req_events;
 138         ev.data.fd = fd;
 139
 140         ret = epoll_ctl(events->epfd, EPOLL_CTL_ADD, fd, &ev);
 141         if (ret < 0) {
 142                 switch (errno) {
 143                 case EEXIST:
 144                         /* If exist, it's OK. */
 145                         goto end;
 146                 case ENOSPC:
 147                 case EPERM:
 148                         /* Print PERROR and goto end not failing. Show must go on. */
 149                         PERROR("epoll_ctl ADD");
 150                         goto end;
 151                 default:
 152                         PERROR("epoll_ctl ADD fatal");
 153                         goto error;
 154                 }
 155         }
 156
 157         events->nb_fd++;
 158
 159 end:
 160         return 0;
 161
 162 error:
 163         return -1;
 164 }
 165
 166 /*
 167  * Remove a fd from the epoll set.
 168  */
 169 int compat_epoll_del(struct lttng_poll_event *events, int fd)
 170 {
 171         int ret;
 172
 173         if (events == NULL || fd < 0 || events->nb_fd == 0) {
 174                 goto error;
 175         }
 176
 177         ret = epoll_ctl(events->epfd, EPOLL_CTL_DEL, fd, NULL);
 178         if (ret < 0) {
 179                 switch (errno) {
 180                 case ENOENT:
 181                 case EPERM:
 182                         /* Print PERROR and goto end not failing. Show must go on. */
 183                         PERROR("epoll_ctl DEL");
 184                         goto end;
 185                 default:
 186                         PERROR("epoll_ctl DEL fatal");
 187                         goto error;
 188                 }
 189         }
 190
 191         events->nb_fd--;
 192
 193 end:
 194         return 0;
 195
 196 error:
 197         return -1;
 198 }
 199
 200 /*
 201  * Set an fd's events.
 202  */
 203 int compat_epoll_mod(struct lttng_poll_event *events, int fd, uint32_t req_events)
 204 {
 205         int ret;
 206         struct epoll_event ev;
 207
 208         if (events == NULL || fd < 0 || events->nb_fd == 0) {
 209                 goto error;
 210         }
 211
 212         /*
 213          * Zero struct epoll_event to ensure all representations of its
 214          * union are zeroed.
 215          */
 216         memset(&ev, 0, sizeof(ev));
 217         ev.events = req_events;
 218         ev.data.fd = fd;
 219
 220         ret = epoll_ctl(events->epfd, EPOLL_CTL_MOD, fd, &ev);
 221         if (ret < 0) {
 222                 switch (errno) {
 223                 case ENOENT:
 224                 case EPERM:
 225                         /* Print PERROR and goto end not failing. Show must go on. */
 226                         PERROR("epoll_ctl MOD");
 227                         goto end;
 228                 default:
 229                         PERROR("epoll_ctl MOD fatal");
 230                         goto error;
 231                 }
 232         }
 233
 234 end:
 235         return 0;
 236
 237 error:
 238         return -1;
 239 }
 240
 241 /*
 242  * Wait on epoll set. This is a blocking call of timeout value.
 243  */
 244 int compat_epoll_wait(struct lttng_poll_event *events, int timeout)
 245 {
 246         int ret;
 247         uint32_t new_size;
 248
 249         if (events == NULL || events->events == NULL) {
 250                 ERR("Wrong arguments in compat_epoll_wait");
 251                 goto error;
 252         }
 253
 254         if (events->nb_fd == 0) {
 255                 errno = EINVAL;
 256                 return -1;
 257         }
 258
 259         /*
 260          * Resize if needed before waiting. We could either expand the array or
 261          * shrink it down. It's important to note that after this step, we are
 262          * ensured that the events argument of the epoll_wait call will be large
 263          * enough to hold every possible returned events.
 264          */
 265         new_size = 1U << utils_get_count_order_u32(events->nb_fd);
 266         if (new_size != events->alloc_size && new_size >= events->init_size) {
 267                 ret = resize_poll_event(events, new_size);
 268                 if (ret < 0) {
 269                         /* ENOMEM problem at this point. */
 270                         goto error;
 271                 }
 272         }
 273
 274         do {
 275                 ret = epoll_wait(events->epfd, events->events, events->nb_fd, timeout);
 276         } while (ret == -1 && errno == EINTR);
 277         if (ret < 0) {
 278                 /* At this point, every error is fatal */
 279                 PERROR("epoll_wait");
 280                 goto error;
 281         }
 282
 283         /*
 284          * Since the returned events are set sequentially in the "events" structure
 285          * we only need to return the epoll_wait value and iterate over it.
 286          */
 287         return ret;
 288
 289 error:
 290         return -1;
 291 }
 292
 293 /*
 294  * Setup poll set maximum size.
 295  */
 296 int compat_epoll_set_max_size(void)
 297 {
 298         int ret, fd, retval = 0;
 299         ssize_t size_ret;
 300         char buf[64];
 301
 302         fd = open(COMPAT_EPOLL_PROC_PATH, O_RDONLY);
 303         if (fd < 0) {
 304                 /*
 305                  * Failing on opening [1] is not an error per see. [1] was
 306                  * introduced in Linux 2.6.28 but epoll is available since
 307                  * 2.5.44. Hence, goto end and set a default value without
 308                  * setting an error return value.
 309                  *
 310                  * [1] /proc/sys/fs/epoll/max_user_watches
 311                  */
 312                 retval = 0;
 313                 goto end;
 314         }
 315
 316         size_ret = lttng_read(fd, buf, sizeof(buf));
 317         /*
 318          * Allow reading a file smaller than buf, but keep space for
 319          * final \0.
 320          */
 321         if (size_ret < 0 || size_ret >= sizeof(buf)) {
 322                 PERROR("read set max size");
 323                 retval = -1;
 324                 goto end_read;
 325         }
 326         buf[size_ret] = '\0';
 327         poll_max_size = atoi(buf);
 328 end_read:
 329         ret = close(fd);
 330         if (ret) {
 331                 PERROR("close");
 332         }
 333 end:
 334         if (!poll_max_size) {
 335                 poll_max_size = DEFAULT_POLL_SIZE;
 336         }
 337         DBG("epoll set max size is %d", poll_max_size);
 338         return retval;
 339 }