src/test/test-loop-block.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <fcntl.h>
   4 #include <linux/loop.h>
   5 #include <pthread.h>
   6 #include <sys/file.h>
   7 #include <sys/ioctl.h>
   8 #include <sys/mount.h>
   9
  10 #include "alloc-util.h"
  11 #include "capability-util.h"
  12 #include "dissect-image.h"
  13 #include "fd-util.h"
  14 #include "fileio.h"
  15 #include "fs-util.h"
  16 #include "gpt.h"
  17 #include "main-func.h"
  18 #include "missing_loop.h"
  19 #include "mkfs-util.h"
  20 #include "mount-util.h"
  21 #include "namespace-util.h"
  22 #include "parse-util.h"
  23 #include "path-util.h"
  24 #include "string-util.h"
  25 #include "strv.h"
  26 #include "tests.h"
  27 #include "tmpfile-util.h"
  28 #include "user-util.h"
  29 #include "virt.h"
  30
  31 static unsigned arg_n_threads = 5;
  32 static unsigned arg_n_iterations = 3;
  33 static usec_t arg_timeout = 0;
  34
  35 #if HAVE_BLKID
  36 static usec_t end = 0;
  37
  38 static void verify_dissected_image(DissectedImage *dissected) {
  39         assert_se(dissected->partitions[PARTITION_ESP].found);
  40         assert_se(dissected->partitions[PARTITION_ESP].node);
  41         assert_se(dissected->partitions[PARTITION_XBOOTLDR].found);
  42         assert_se(dissected->partitions[PARTITION_XBOOTLDR].node);
  43         assert_se(dissected->partitions[PARTITION_ROOT].found);
  44         assert_se(dissected->partitions[PARTITION_ROOT].node);
  45         assert_se(dissected->partitions[PARTITION_HOME].found);
  46         assert_se(dissected->partitions[PARTITION_HOME].node);
  47 }
  48
  49 static void verify_dissected_image_harder(DissectedImage *dissected) {
  50         verify_dissected_image(dissected);
  51
  52         assert_se(streq(dissected->partitions[PARTITION_ESP].fstype, "vfat"));
  53         assert_se(streq(dissected->partitions[PARTITION_XBOOTLDR].fstype, "vfat"));
  54         assert_se(streq(dissected->partitions[PARTITION_ROOT].fstype, "ext4"));
  55         assert_se(streq(dissected->partitions[PARTITION_HOME].fstype, "ext4"));
  56 }
  57
  58 static void* thread_func(void *ptr) {
  59         int fd = PTR_TO_FD(ptr);
  60         int r;
  61
  62         for (unsigned i = 0; i < arg_n_iterations; i++) {
  63                 _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
  64                 _cleanup_(umount_and_rmdir_and_freep) char *mounted = NULL;
  65                 _cleanup_(dissected_image_unrefp) DissectedImage *dissected = NULL;
  66
  67                 if (now(CLOCK_MONOTONIC) >= end) {
  68                         log_notice("Time's up, exiting thread's loop");
  69                         break;
  70                 }
  71
  72                 log_notice("> Thread iteration #%u.", i);
  73
  74                 assert_se(mkdtemp_malloc(NULL, &mounted) >= 0);
  75
  76                 r = loop_device_make(fd, O_RDONLY, 0, UINT64_MAX, 0, LO_FLAGS_PARTSCAN, LOCK_SH, &loop);
  77                 if (r < 0)
  78                         log_error_errno(r, "Failed to allocate loopback device: %m");
  79                 assert_se(r >= 0);
  80                 assert_se(loop->dev);
  81                 assert_se(loop->backing_file);
  82
  83                 log_notice("Acquired loop device %s, will mount on %s", loop->node, mounted);
  84
  85                 r = dissect_loop_device(loop, NULL, NULL, NULL, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_ADD_PARTITION_DEVICES|DISSECT_IMAGE_PIN_PARTITION_DEVICES, &dissected);
  86                 if (r < 0)
  87                         log_error_errno(r, "Failed dissect loopback device %s: %m", loop->node);
  88                 assert_se(r >= 0);
  89
  90                 log_info("Dissected loop device %s", loop->node);
  91
  92                 for (PartitionDesignator d = 0; d < _PARTITION_DESIGNATOR_MAX; d++) {
  93                         if (!dissected->partitions[d].found)
  94                                 continue;
  95
  96                         log_notice("Found node %s fstype %s designator %s",
  97                                    dissected->partitions[d].node,
  98                                    dissected->partitions[d].fstype,
  99                                    partition_designator_to_string(d));
 100                 }
 101
 102                 verify_dissected_image(dissected);
 103
 104                 r = dissected_image_mount(dissected, mounted, UID_INVALID, UID_INVALID, DISSECT_IMAGE_READ_ONLY);
 105                 log_notice_errno(r, "Mounted %s → %s: %m", loop->node, mounted);
 106                 assert_se(r >= 0);
 107
 108                 /* Now the block device is mounted, we don't need no manual lock anymore, the devices are now
 109                  * pinned by the mounts. */
 110                 assert_se(loop_device_flock(loop, LOCK_UN) >= 0);
 111
 112                 log_notice("Unmounting %s", mounted);
 113                 mounted = umount_and_rmdir_and_free(mounted);
 114
 115                 log_notice("Unmounted.");
 116
 117                 dissected = dissected_image_unref(dissected);
 118
 119                 log_notice("Detaching loop device %s", loop->node);
 120                 loop = loop_device_unref(loop);
 121                 log_notice("Detached loop device.");
 122         }
 123
 124         log_notice("Leaving thread");
 125
 126         return NULL;
 127 }
 128 #endif
 129
 130 static bool have_root_gpt_type(void) {
 131 #ifdef SD_GPT_ROOT_NATIVE
 132         return true;
 133 #else
 134         return false;
 135 #endif
 136 }
 137
 138 static int run(int argc, char *argv[]) {
 139 #if HAVE_BLKID
 140         _cleanup_(dissected_image_unrefp) DissectedImage *dissected = NULL;
 141         _cleanup_(umount_and_rmdir_and_freep) char *mounted = NULL;
 142         pthread_t threads[arg_n_threads];
 143         sd_id128_t id;
 144 #endif
 145         _cleanup_free_ char *p = NULL, *cmd = NULL;
 146         _cleanup_(pclosep) FILE *sfdisk = NULL;
 147         _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
 148         _cleanup_close_ int fd = -EBADF;
 149         int r;
 150
 151         test_setup_logging(LOG_DEBUG);
 152         log_show_tid(true);
 153         log_show_time(true);
 154         log_show_color(true);
 155
 156         if (argc >= 2) {
 157                 r = safe_atou(argv[1], &arg_n_threads);
 158                 if (r < 0)
 159                         return log_error_errno(r, "Failed to parse first argument (number of threads): %s", argv[1]);
 160                 if (arg_n_threads <= 0)
 161                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Number of threads must be at least 1, refusing.");
 162         }
 163
 164         if (argc >= 3) {
 165                 r = safe_atou(argv[2], &arg_n_iterations);
 166                 if (r < 0)
 167                         return log_error_errno(r, "Failed to parse second argument (number of iterations): %s", argv[2]);
 168                 if (arg_n_iterations <= 0)
 169                         return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Number of iterations must be at least 1, refusing.");
 170         }
 171
 172         if (argc >= 4) {
 173                 r = parse_sec(argv[3], &arg_timeout);
 174                 if (r < 0)
 175                         return log_error_errno(r, "Failed to parse third argument (timeout): %s", argv[3]);
 176         }
 177
 178         if (argc >= 5)
 179                 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Too many arguments (expected 3 at max).");
 180
 181         if (!have_root_gpt_type())
 182                 return log_tests_skipped("No root partition GPT defined for this architecture");
 183
 184         r = find_executable("sfdisk", NULL);
 185         if (r < 0)
 186                 return log_tests_skipped_errno(r, "Could not find sfdisk command");
 187
 188         assert_se(tempfn_random_child("/var/tmp", "sfdisk", &p) >= 0);
 189         fd = open(p, O_CREAT|O_EXCL|O_RDWR|O_CLOEXEC|O_NOFOLLOW, 0666);
 190         assert_se(fd >= 0);
 191         assert_se(ftruncate(fd, 256*1024*1024) >= 0);
 192
 193         assert_se(cmd = strjoin("sfdisk ", p));
 194         assert_se(sfdisk = popen(cmd, "we"));
 195
 196         /* A reasonably complex partition table that fits on a 64K disk */
 197         fputs("label: gpt\n"
 198               "size=32M, type=C12A7328-F81F-11D2-BA4B-00A0C93EC93B\n"
 199               "size=32M, type=BC13C2FF-59E6-4262-A352-B275FD6F7172\n"
 200               "size=32M, type=0657FD6D-A4AB-43C4-84E5-0933C84B4F4F\n"
 201               "size=32M, type=", sfdisk);
 202
 203 #ifdef SD_GPT_ROOT_NATIVE
 204         fprintf(sfdisk, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(SD_GPT_ROOT_NATIVE));
 205 #else
 206         fprintf(sfdisk, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(SD_GPT_ROOT_X86_64));
 207 #endif
 208
 209         fputs("\n"
 210               "size=32M, type=933AC7E1-2EB4-4F13-B844-0E14E2AEF915\n", sfdisk);
 211
 212         assert_se(pclose(sfdisk) == 0);
 213         sfdisk = NULL;
 214
 215 #if HAVE_BLKID
 216         assert_se(dissect_image_file(p, NULL, NULL, NULL, 0, &dissected) >= 0);
 217         verify_dissected_image(dissected);
 218         dissected = dissected_image_unref(dissected);
 219 #endif
 220
 221         if (geteuid() != 0 || have_effective_cap(CAP_SYS_ADMIN) <= 0) {
 222                 log_tests_skipped("not running privileged");
 223                 return 0;
 224         }
 225
 226         if (detect_container() > 0) {
 227                 log_tests_skipped("Test not supported in a container, requires udev/uevent notifications");
 228                 return 0;
 229         }
 230
 231         assert_se(loop_device_make(fd, O_RDWR, 0, UINT64_MAX, 0, LO_FLAGS_PARTSCAN, LOCK_EX, &loop) >= 0);
 232
 233 #if HAVE_BLKID
 234         assert_se(dissect_loop_device(loop, NULL, NULL, NULL, DISSECT_IMAGE_ADD_PARTITION_DEVICES|DISSECT_IMAGE_PIN_PARTITION_DEVICES, &dissected) >= 0);
 235         verify_dissected_image(dissected);
 236
 237         FOREACH_STRING(fs, "vfat", "ext4") {
 238                 r = mkfs_exists(fs);
 239                 assert_se(r >= 0);
 240                 if (!r) {
 241                         log_tests_skipped("mkfs.{vfat|ext4} not installed");
 242                         return 0;
 243                 }
 244         }
 245         assert_se(r >= 0);
 246
 247         assert_se(sd_id128_randomize(&id) >= 0);
 248         assert_se(make_filesystem(dissected->partitions[PARTITION_ESP].node, "vfat", "EFI", NULL, id, true, 0, NULL) >= 0);
 249
 250         assert_se(sd_id128_randomize(&id) >= 0);
 251         assert_se(make_filesystem(dissected->partitions[PARTITION_XBOOTLDR].node, "vfat", "xbootldr", NULL, id, true, 0, NULL) >= 0);
 252
 253         assert_se(sd_id128_randomize(&id) >= 0);
 254         assert_se(make_filesystem(dissected->partitions[PARTITION_ROOT].node, "ext4", "root", NULL, id, true, 0, NULL) >= 0);
 255
 256         assert_se(sd_id128_randomize(&id) >= 0);
 257         assert_se(make_filesystem(dissected->partitions[PARTITION_HOME].node, "ext4", "home", NULL, id, true, 0, NULL) >= 0);
 258
 259         dissected = dissected_image_unref(dissected);
 260
 261         /* We created the file systems now via the per-partition block devices. But the dissection code might
 262          * probe them via the whole block device. These block devices have separate buffer caches though,
 263          * hence what was written via the partition device might not appear on the whole block device
 264          * yet. Let's hence explicitly flush the whole block device, so that the read-back definitely
 265          * works. */
 266         assert_se(ioctl(loop->fd, BLKFLSBUF, 0) >= 0);
 267
 268         /* Try to read once, without pinning or adding partitions, i.e. by only accessing the whole block
 269          * device. */
 270         assert_se(dissect_loop_device(loop, NULL, NULL, NULL, 0, &dissected) >= 0);
 271         verify_dissected_image_harder(dissected);
 272         dissected = dissected_image_unref(dissected);
 273
 274         /* Now go via the loopback device after all, but this time add/pin, because now we want to mount it. */
 275         assert_se(dissect_loop_device(loop, NULL, NULL, NULL, DISSECT_IMAGE_ADD_PARTITION_DEVICES|DISSECT_IMAGE_PIN_PARTITION_DEVICES, &dissected) >= 0);
 276         verify_dissected_image_harder(dissected);
 277
 278         assert_se(mkdtemp_malloc(NULL, &mounted) >= 0);
 279
 280         /* We are particularly correct here, and now downgrade LOCK → LOCK_SH. That's because we are done
 281          * with formatting the file systems, so we don't need the exclusive lock anymore. From now on a
 282          * shared one is fine. This way udev can now probe the device if it wants, but still won't call
 283          * BLKRRPART on it, and that's good, because that would destroy our partition table while we are at
 284          * it. */
 285         assert_se(loop_device_flock(loop, LOCK_SH) >= 0);
 286
 287         /* This is a test for the loopback block device setup code and it's use by the image dissection
 288          * logic: since the kernel APIs are hard use and prone to races, let's test this in a heavy duty
 289          * test: we open a bunch of threads and repeatedly allocate and deallocate loopback block devices in
 290          * them in parallel, with an image file with a number of partitions. */
 291         assert_se(detach_mount_namespace() >= 0);
 292
 293         /* This first (writable) mount will initialize the mount point dirs, so that the subsequent read-only ones can work */
 294         assert_se(dissected_image_mount(dissected, mounted, UID_INVALID, UID_INVALID, 0) >= 0);
 295
 296         /* Now we mounted everything, the partitions are pinned. Now it's fine to release the lock
 297          * fully. This means udev could now issue BLKRRPART again, but that's OK given this will fail because
 298          * we now mounted the device. */
 299         assert_se(loop_device_flock(loop, LOCK_UN) >= 0);
 300
 301         assert_se(umount_recursive(mounted, 0) >= 0);
 302         loop = loop_device_unref(loop);
 303
 304         log_notice("Threads are being started now");
 305
 306         /* zero timeout means pick default: let's make sure we run for 10s on slow systems at max */
 307         if (arg_timeout == 0)
 308                 arg_timeout = slow_tests_enabled() ? 5 * USEC_PER_SEC : 1 * USEC_PER_SEC;
 309
 310         end = usec_add(now(CLOCK_MONOTONIC), arg_timeout);
 311
 312         if (arg_n_threads > 1)
 313                 for (unsigned i = 0; i < arg_n_threads; i++)
 314                         assert_se(pthread_create(threads + i, NULL, thread_func, FD_TO_PTR(fd)) == 0);
 315
 316         log_notice("All threads started now.");
 317
 318         if (arg_n_threads == 1)
 319                 assert_se(thread_func(FD_TO_PTR(fd)) == NULL);
 320         else
 321                 for (unsigned i = 0; i < arg_n_threads; i++) {
 322                         log_notice("Joining thread #%u.", i);
 323
 324                         void *k;
 325                         assert_se(pthread_join(threads[i], &k) == 0);
 326                         assert_se(!k);
 327
 328                         log_notice("Joined thread #%u.", i);
 329                 }
 330
 331         log_notice("Threads are all terminated now.");
 332 #else
 333         log_notice("Cutting test short, since we do not have libblkid.");
 334 #endif
 335         return 0;
 336 }
 337
 338 DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);