src/bin/lttng-sessiond/shm.c

   1 /*
   2  * Copyright (C)  2011 - David Goulet <david.goulet@polymtl.ca>
   3  *                       Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _LGPL_SOURCE
  20 #include <fcntl.h>
  21 #include <limits.h>
  22 #include <sys/mman.h>
  23 #include <sys/stat.h>
  24 #include <sys/types.h>
  25 #include <sys/wait.h>
  26 #include <unistd.h>
  27 #include <urcu.h>
  28
  29 #include <common/error.h>
  30
  31 #include "shm.h"
  32
  33 /*
  34  * Using fork to set umask in the child process (not multi-thread safe). We
  35  * deal with the shm_open vs ftruncate race (happening when the sessiond owns
  36  * the shm and does not let everybody modify it, to ensure safety against
  37  * shm_unlink) by simply letting the mmap fail and retrying after a few
  38  * seconds. For global shm, everybody has rw access to it until the sessiond
  39  * starts.
  40  */
  41 static int get_wait_shm(char *shm_path, size_t mmap_size, int global)
  42 {
  43         int wait_shm_fd, ret;
  44         mode_t mode;
  45
  46         assert(shm_path);
  47
  48         /* Default permissions */
  49         mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
  50
  51         /*
  52          * Change owner of the shm path.
  53          */
  54         if (global) {
  55                 /*
  56                  * If global session daemon, any application can
  57                  * register. Make it initially writeable so applications
  58                  * registering concurrently can do ftruncate() by
  59                  * themselves.
  60                  */
  61                 mode |= S_IROTH | S_IWOTH;
  62         }
  63
  64         /*
  65          * We're alone in a child process, so we can modify the process-wide
  66          * umask.
  67          */
  68         umask(~mode);
  69
  70         /*
  71          * Try creating shm (or get rw access). We don't do an exclusive open,
  72          * because we allow other processes to create+ftruncate it concurrently.
  73          *
  74          * A sysctl, fs.protected_regular may prevent the session daemon from
  75          * opening a previously created shm when the O_CREAT flag is provided.
  76          * Systemd enables this ABI-breaking change by default since v241.
  77          *
  78          * First, attempt to use the create-or-open semantic that is
  79          * desired here. If this fails with EACCES, work around this broken
  80          * behaviour and attempt to open the shm without the O_CREAT flag.
  81          *
  82          * The two attempts are made in this order since applications are
  83          * expected to race with the session daemon to create this shm.
  84          * Attempting an shm_open() without the O_CREAT flag first could fail
  85          * because the file doesn't exist. It could then be created by an
  86          * application, which would cause a second try with the O_CREAT flag to
  87          * fail with EACCES.
  88          *
  89          * Note that this introduces a new failure mode where a user could
  90          * launch an application (creating the shm) and unlink the shm while
  91          * the session daemon is launching, causing the second attempt
  92          * to fail. This is not recovered-from as unlinking the shm will
  93          * prevent userspace tracing from succeeding anyhow: the sessiond would
  94          * use a now-unlinked shm, while the next application would create
  95          * a new named shm.
  96          */
  97         wait_shm_fd = shm_open(shm_path, O_RDWR | O_CREAT, mode);
  98         if (wait_shm_fd < 0) {
  99                 if (errno == EACCES) {
 100                         /* Work around sysctl fs.protected_regular. */
 101                         DBG("shm_open of %s returned EACCES, this may be caused "
 102                                         "by the fs.protected_regular sysctl. "
 103                                         "Attempting to open the shm without "
 104                                         "creating it.", shm_path);
 105                         wait_shm_fd = shm_open(shm_path, O_RDWR, mode);
 106                 }
 107                 if (wait_shm_fd < 0) {
 108                         PERROR("Failed to open wait shm at %s", shm_path);
 109                         goto error;
 110                 }
 111         }
 112
 113         ret = ftruncate(wait_shm_fd, mmap_size);
 114         if (ret < 0) {
 115                 PERROR("ftruncate wait shm");
 116                 exit(EXIT_FAILURE);
 117         }
 118
 119 #ifndef __FreeBSD__
 120         if (global) {
 121                 ret = fchown(wait_shm_fd, 0, 0);
 122                 if (ret < 0) {
 123                         PERROR("fchown");
 124                         exit(EXIT_FAILURE);
 125                 }
 126                 /*
 127                  * If global session daemon, any application can
 128                  * register so the shm needs to be set in read-only mode
 129                  * for others.
 130                  */
 131                 mode &= ~S_IWOTH;
 132                 ret = fchmod(wait_shm_fd, mode);
 133                 if (ret < 0) {
 134                         PERROR("fchmod");
 135                         exit(EXIT_FAILURE);
 136                 }
 137         } else {
 138                 ret = fchown(wait_shm_fd, getuid(), getgid());
 139                 if (ret < 0) {
 140                         PERROR("fchown");
 141                         exit(EXIT_FAILURE);
 142                 }
 143         }
 144 #else
 145 #warning "FreeBSD does not support setting file mode on shm FD."
 146 #endif
 147
 148         DBG("Got the wait shm fd %d", wait_shm_fd);
 149
 150         return wait_shm_fd;
 151
 152 error:
 153         DBG("Failing to get the wait shm fd");
 154
 155         return -1;
 156 }
 157
 158 /*
 159  * Return the wait shm mmap for UST application notification. The global
 160  * variable is used to indicate if the the session daemon is global
 161  * (root:tracing) or running with an unprivileged user.
 162  *
 163  * This returned value is used by futex_wait_update() in futex.c to WAKE all
 164  * waiters which are UST application waiting for a session daemon.
 165  */
 166 char *shm_ust_get_mmap(char *shm_path, int global)
 167 {
 168         size_t mmap_size;
 169         int wait_shm_fd, ret;
 170         char *wait_shm_mmap;
 171         long sys_page_size;
 172
 173         assert(shm_path);
 174
 175         sys_page_size = sysconf(_SC_PAGE_SIZE);
 176         if (sys_page_size < 0) {
 177                 PERROR("sysconf PAGE_SIZE");
 178                 goto error;
 179         }
 180         mmap_size = sys_page_size;
 181
 182         wait_shm_fd = get_wait_shm(shm_path, mmap_size, global);
 183         if (wait_shm_fd < 0) {
 184                 goto error;
 185         }
 186
 187         wait_shm_mmap = mmap(NULL, mmap_size, PROT_WRITE | PROT_READ,
 188                         MAP_SHARED, wait_shm_fd, 0);
 189
 190         /* close shm fd immediately after taking the mmap reference */
 191         ret = close(wait_shm_fd);
 192         if (ret) {
 193                 PERROR("Error closing fd");
 194         }
 195
 196         if (wait_shm_mmap == MAP_FAILED) {
 197                 DBG("mmap error (can be caused by race with ust).");
 198                 goto error;
 199         }
 200
 201         return wait_shm_mmap;
 202
 203 error:
 204         return NULL;
 205 }