src/common/shm.c

   1 /*
   2  * Copyright (C) 2011 EfficiOS Inc.
   3  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * SPDX-License-Identifier: GPL-2.0-only
   6  *
   7  */
   8
   9 #define _LGPL_SOURCE
  10 #include <fcntl.h>
  11 #include <limits.h>
  12 #include <sys/mman.h>
  13 #include <sys/stat.h>
  14 #include <sys/types.h>
  15 #include <sys/wait.h>
  16 #include <unistd.h>
  17 #include <urcu.h>
  18
  19 #include <common/error.h>
  20
  21 #include "shm.h"
  22
  23 /*
  24  * We deal with the shm_open vs ftruncate race (happening when the sessiond owns
  25  * the shm and does not let everybody modify it, to ensure safety against
  26  * shm_unlink) by simply letting the mmap fail and retrying after a few
  27  * seconds. For global shm, everybody has rw access to it until the sessiond
  28  * starts.
  29  */
  30 static int get_wait_shm(char *shm_path, size_t mmap_size, int global)
  31 {
  32         int wait_shm_fd, ret;
  33         mode_t mode, old_mode;
  34
  35         assert(shm_path);
  36
  37         /* Default permissions */
  38         mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
  39
  40         /*
  41          * Change owner of the shm path.
  42          */
  43         if (global) {
  44                 /*
  45                  * If global session daemon, any application can
  46                  * register. Make it initially writeable so applications
  47                  * registering concurrently can do ftruncate() by
  48                  * themselves.
  49                  */
  50                 mode |= S_IROTH | S_IWOTH;
  51         }
  52
  53         old_mode = umask(~mode);
  54
  55         /*
  56          * Try creating shm (or get rw access). We don't do an exclusive open,
  57          * because we allow other processes to create+ftruncate it concurrently.
  58          *
  59          * A sysctl, fs.protected_regular may prevent the session daemon from
  60          * opening a previously created shm when the O_CREAT flag is provided.
  61          * Systemd enables this ABI-breaking change by default since v241.
  62          *
  63          * First, attempt to use the create-or-open semantic that is
  64          * desired here. If this fails with EACCES, work around this broken
  65          * behaviour and attempt to open the shm without the O_CREAT flag.
  66          *
  67          * The two attempts are made in this order since applications are
  68          * expected to race with the session daemon to create this shm.
  69          * Attempting an shm_open() without the O_CREAT flag first could fail
  70          * because the file doesn't exist. It could then be created by an
  71          * application, which would cause a second try with the O_CREAT flag to
  72          * fail with EACCES.
  73          *
  74          * Note that this introduces a new failure mode where a user could
  75          * launch an application (creating the shm) and unlink the shm while
  76          * the session daemon is launching, causing the second attempt
  77          * to fail. This is not recovered-from as unlinking the shm will
  78          * prevent userspace tracing from succeeding anyhow: the sessiond would
  79          * use a now-unlinked shm, while the next application would create
  80          * a new named shm.
  81          */
  82         wait_shm_fd = shm_open(shm_path, O_RDWR | O_CREAT, mode);
  83         if (wait_shm_fd < 0) {
  84                 if (errno == EACCES) {
  85                         /* Work around sysctl fs.protected_regular. */
  86                         DBG("shm_open of %s returned EACCES, this may be caused "
  87                                         "by the fs.protected_regular sysctl. "
  88                                         "Attempting to open the shm without "
  89                                         "creating it.", shm_path);
  90                         wait_shm_fd = shm_open(shm_path, O_RDWR, mode);
  91                 }
  92                 if (wait_shm_fd < 0) {
  93                         PERROR("Failed to open \"wait\" shared memory object: path = '%s'", shm_path);
  94                         goto error;
  95                 }
  96         }
  97
  98         ret = ftruncate(wait_shm_fd, mmap_size);
  99         if (ret < 0) {
 100                 PERROR("Failed to truncate \"wait\" shared memory object: fd = %d, size = %zu",
 101                                 wait_shm_fd, mmap_size);
 102                 goto error;
 103         }
 104
 105         if (global) {
 106                 ret = fchown(wait_shm_fd, 0, 0);
 107                 if (ret < 0) {
 108                         PERROR("Failed to set ownership of \"wait\" shared memory object: fd = %d, owner = 0, group = 0",
 109                                         wait_shm_fd);
 110                         goto error;
 111                 }
 112                 /*
 113                  * If global session daemon, any application can
 114                  * register so the shm needs to be set in read-only mode
 115                  * for others.
 116                  */
 117                 mode &= ~S_IWOTH;
 118                 ret = fchmod(wait_shm_fd, mode);
 119                 if (ret < 0) {
 120                         PERROR("Failed to set the mode of the \"wait\" shared memory object: fd = %d, mode = %d",
 121                                         wait_shm_fd, mode);
 122                         goto error;
 123                 }
 124         } else {
 125                 ret = fchown(wait_shm_fd, getuid(), getgid());
 126                 if (ret < 0) {
 127                         PERROR("Failed to set ownership of \"wait\" shared memory object: fd = %d, owner = %d, group = %d",
 128                                         wait_shm_fd, getuid(), getgid());
 129                         goto error;
 130                 }
 131         }
 132
 133         DBG("Wait shared memory file descriptor created successfully: path = '%s', mmap_size = %zu, global = %s, fd = %d",
 134                         shm_path, mmap_size, global ? "true" : "false",
 135                         wait_shm_fd);
 136
 137 end:
 138         (void) umask(old_mode);
 139         return wait_shm_fd;
 140
 141 error:
 142         DBG("Failing to get the wait shm fd");
 143         if (wait_shm_fd >= 0) {
 144                 if (close(wait_shm_fd)) {
 145                         PERROR("Failed to close wait shm file descriptor during error handling");
 146                 }
 147         }
 148
 149         wait_shm_fd = -1;
 150         goto end;
 151 }
 152
 153 /*
 154  * Return the wait shm mmap for UST application notification. The global
 155  * variable is used to indicate if the the session daemon is global
 156  * (root:tracing) or running with an unprivileged user.
 157  *
 158  * This returned value is used by futex_wait_update() in futex.c to WAKE all
 159  * waiters which are UST application waiting for a session daemon.
 160  */
 161 LTTNG_HIDDEN
 162 char *shm_ust_get_mmap(char *shm_path, int global)
 163 {
 164         size_t mmap_size;
 165         int wait_shm_fd, ret;
 166         char *wait_shm_mmap;
 167         long sys_page_size;
 168
 169         assert(shm_path);
 170
 171         sys_page_size = sysconf(_SC_PAGE_SIZE);
 172         if (sys_page_size < 0) {
 173                 PERROR("Failed to get PAGE_SIZE of system");
 174                 goto error;
 175         }
 176         mmap_size = sys_page_size;
 177
 178         wait_shm_fd = get_wait_shm(shm_path, mmap_size, global);
 179         if (wait_shm_fd < 0) {
 180                 goto error;
 181         }
 182
 183         wait_shm_mmap = mmap(NULL, mmap_size, PROT_WRITE | PROT_READ,
 184                         MAP_SHARED, wait_shm_fd, 0);
 185
 186         /* close shm fd immediately after taking the mmap reference */
 187         ret = close(wait_shm_fd);
 188         if (ret) {
 189                 PERROR("Failed to close \"wait\" shared memory object file descriptor: fd = %d",
 190                                 wait_shm_fd);
 191         }
 192
 193         if (wait_shm_mmap == MAP_FAILED) {
 194                 DBG("Failed to mmap the \"wait\" shareed memory object (can be caused by race with ust): path = '%s', global = %s",
 195                                 shm_path, global ? "true" : "false");
 196                 goto error;
 197         }
 198
 199         return wait_shm_mmap;
 200
 201 error:
 202         return NULL;
 203 }
 204
 205 /*
 206  * shm_create_anonymous is never called concurrently within a process.
 207  */
 208 LTTNG_HIDDEN
 209 int shm_create_anonymous(const char *owner_name)
 210 {
 211         char tmp_name[NAME_MAX];
 212         int shmfd, ret;
 213
 214         ret = snprintf(tmp_name, NAME_MAX, "/shm-%s-%d", owner_name, getpid());
 215         if (ret < 0) {
 216                 PERROR("Failed to format shm path: owner_name = '%s', pid = %d",
 217                                 owner_name, getpid());
 218                 return -1;
 219         }
 220
 221         /*
 222          * Allocate shm, and immediately unlink its shm oject, keeping only the
 223          * file descriptor as a reference to the object.
 224          */
 225         shmfd = shm_open(tmp_name, O_CREAT | O_EXCL | O_RDWR, 0700);
 226         if (shmfd < 0) {
 227                 PERROR("Failed to open shared memory object: path = '%s'", tmp_name);
 228                 goto error_shm_open;
 229         }
 230
 231         ret = shm_unlink(tmp_name);
 232         if (ret < 0 && errno != ENOENT) {
 233                 PERROR("Failed to unlink shared memory object: path = '%s'",
 234                                 tmp_name);
 235                 goto error_shm_release;
 236         }
 237
 238         return shmfd;
 239
 240 error_shm_release:
 241         ret = close(shmfd);
 242         if (ret) {
 243                 PERROR("Failed to close shared memory object file descriptor: fd = %d, path = '%s'",
 244                                 shmfd, tmp_name);
 245         }
 246 error_shm_open:
 247         return -1;
 248 }