src/shared/machine-pool.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2015 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   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 License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <linux/loop.h>
  23 #include <signal.h>
  24 #include <stdbool.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <sys/ioctl.h>
  28 #include <sys/mount.h>
  29 #include <sys/prctl.h>
  30 #include <sys/stat.h>
  31 #include <sys/statfs.h>
  32 #include <sys/statvfs.h>
  33 #include <unistd.h>
  34
  35 #include "sd-bus-protocol.h"
  36 #include "sd-bus.h"
  37
  38 #include "alloc-util.h"
  39 #include "btrfs-util.h"
  40 #include "fd-util.h"
  41 #include "fileio.h"
  42 #include "fs-util.h"
  43 #include "lockfile-util.h"
  44 #include "log.h"
  45 #include "machine-pool.h"
  46 #include "macro.h"
  47 #include "missing.h"
  48 #include "mkdir.h"
  49 #include "mount-util.h"
  50 #include "parse-util.h"
  51 #include "path-util.h"
  52 #include "process-util.h"
  53 #include "signal-util.h"
  54 #include "stat-util.h"
  55 #include "string-util.h"
  56
  57 #define VAR_LIB_MACHINES_SIZE_START (1024UL*1024UL*500UL)
  58 #define VAR_LIB_MACHINES_FREE_MIN (1024UL*1024UL*750UL)
  59
  60 static int check_btrfs(void) {
  61         struct statfs sfs;
  62
  63         if (statfs("/var/lib/machines", &sfs) < 0) {
  64                 if (errno != ENOENT)
  65                         return -errno;
  66
  67                 if (statfs("/var/lib", &sfs) < 0)
  68                         return -errno;
  69         }
  70
  71         return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC);
  72 }
  73
  74 static int setup_machine_raw(uint64_t size, sd_bus_error *error) {
  75         _cleanup_free_ char *tmp = NULL;
  76         _cleanup_close_ int fd = -1;
  77         struct statvfs ss;
  78         pid_t pid = 0;
  79         siginfo_t si;
  80         int r;
  81
  82         /* We want to be able to make use of btrfs-specific file
  83          * system features, in particular subvolumes, reflinks and
  84          * quota. Hence, if we detect that /var/lib/machines.raw is
  85          * not located on btrfs, let's create a loopback file, place a
  86          * btrfs file system into it, and mount it to
  87          * /var/lib/machines. */
  88
  89         fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
  90         if (fd >= 0) {
  91                 r = fd;
  92                 fd = -1;
  93                 return r;
  94         }
  95
  96         if (errno != ENOENT)
  97                 return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m");
  98
  99         r = tempfn_xxxxxx("/var/lib/machines.raw", NULL, &tmp);
 100         if (r < 0)
 101                 return r;
 102
 103         (void) mkdir_p_label("/var/lib", 0755);
 104         fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600);
 105         if (fd < 0)
 106                 return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m");
 107
 108         if (fstatvfs(fd, &ss) < 0) {
 109                 r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m");
 110                 goto fail;
 111         }
 112
 113         if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) {
 114                 r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines.");
 115                 goto fail;
 116         }
 117
 118         if (ftruncate(fd, size) < 0) {
 119                 r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m");
 120                 goto fail;
 121         }
 122
 123         pid = fork();
 124         if (pid < 0) {
 125                 r = sd_bus_error_set_errnof(error, errno, "Failed to fork mkfs.btrfs: %m");
 126                 goto fail;
 127         }
 128
 129         if (pid == 0) {
 130
 131                 /* Child */
 132
 133                 (void) reset_all_signal_handlers();
 134                 (void) reset_signal_mask();
 135                 assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0);
 136
 137                 fd = safe_close(fd);
 138
 139                 execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL);
 140                 if (errno == ENOENT)
 141                         return 99;
 142
 143                 _exit(EXIT_FAILURE);
 144         }
 145
 146         r = wait_for_terminate(pid, &si);
 147         if (r < 0) {
 148                 sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m");
 149                 goto fail;
 150         }
 151
 152         pid = 0;
 153
 154         if (si.si_code != CLD_EXITED) {
 155                 r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs died abnormally.");
 156                 goto fail;
 157         }
 158         if (si.si_status == 99) {
 159                 r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing");
 160                 goto fail;
 161         }
 162         if (si.si_status != 0) {
 163                 r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", si.si_status);
 164                 goto fail;
 165         }
 166
 167         r = rename_noreplace(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw");
 168         if (r < 0) {
 169                 sd_bus_error_set_errnof(error, r, "Failed to move /var/lib/machines.raw into place: %m");
 170                 goto fail;
 171         }
 172
 173         r = fd;
 174         fd = -1;
 175
 176         return r;
 177
 178 fail:
 179         unlink_noerrno(tmp);
 180
 181         if (pid > 1)
 182                 kill_and_sigcont(pid, SIGKILL);
 183
 184         return r;
 185 }
 186
 187 int setup_machine_directory(uint64_t size, sd_bus_error *error) {
 188         _cleanup_release_lock_file_ LockFile lock_file = LOCK_FILE_INIT;
 189         struct loop_info64 info = {
 190                 .lo_flags = LO_FLAGS_AUTOCLEAR,
 191         };
 192         _cleanup_close_ int fd = -1, control = -1, loop = -1;
 193         _cleanup_free_ char* loopdev = NULL;
 194         char tmpdir[] = "/tmp/machine-pool.XXXXXX", *mntdir = NULL;
 195         bool tmpdir_made = false, mntdir_made = false, mntdir_mounted = false;
 196         char buf[FORMAT_BYTES_MAX];
 197         int r, nr = -1;
 198
 199         /* btrfs cannot handle file systems < 16M, hence use this as minimum */
 200         if (size == (uint64_t) -1)
 201                 size = VAR_LIB_MACHINES_SIZE_START;
 202         else if (size < 16*1024*1024)
 203                 size = 16*1024*1024;
 204
 205         /* Make sure we only set the directory up once at a time */
 206         r = make_lock_file("/run/systemd/machines.lock", LOCK_EX, &lock_file);
 207         if (r < 0)
 208                 return r;
 209
 210         r = check_btrfs();
 211         if (r < 0)
 212                 return sd_bus_error_set_errnof(error, r, "Failed to determine whether /var/lib/machines is located on btrfs: %m");
 213         if (r > 0) {
 214                 (void) btrfs_subvol_make_label("/var/lib/machines");
 215
 216                 r = btrfs_quota_enable("/var/lib/machines", true);
 217                 if (r < 0)
 218                         log_warning_errno(r, "Failed to enable quota for /var/lib/machines, ignoring: %m");
 219
 220                 r = btrfs_subvol_auto_qgroup("/var/lib/machines", 0, true);
 221                 if (r < 0)
 222                         log_warning_errno(r, "Failed to set up default quota hierarchy for /var/lib/machines, ignoring: %m");
 223
 224                 return 1;
 225         }
 226
 227         if (path_is_mount_point("/var/lib/machines", AT_SYMLINK_FOLLOW) > 0) {
 228                 log_debug("/var/lib/machines is already a mount point, not creating loopback file for it.");
 229                 return 0;
 230         }
 231
 232         r = dir_is_populated("/var/lib/machines");
 233         if (r < 0 && r != -ENOENT)
 234                 return r;
 235         if (r > 0) {
 236                 log_debug("/var/log/machines is already populated, not creating loopback file for it.");
 237                 return 0;
 238         }
 239
 240         r = mkfs_exists("btrfs");
 241         if (r == -ENOENT) {
 242                 log_debug("mkfs.btrfs is missing, cannot create loopback file for /var/lib/machines.");
 243                 return 0;
 244         }
 245         if (r < 0)
 246                 return r;
 247
 248         fd = setup_machine_raw(size, error);
 249         if (fd < 0)
 250                 return fd;
 251
 252         control = open("/dev/loop-control", O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK);
 253         if (control < 0)
 254                 return sd_bus_error_set_errnof(error, errno, "Failed to open /dev/loop-control: %m");
 255
 256         nr = ioctl(control, LOOP_CTL_GET_FREE);
 257         if (nr < 0)
 258                 return sd_bus_error_set_errnof(error, errno, "Failed to allocate loop device: %m");
 259
 260         if (asprintf(&loopdev, "/dev/loop%i", nr) < 0) {
 261                 r = -ENOMEM;
 262                 goto fail;
 263         }
 264
 265         loop = open(loopdev, O_CLOEXEC|O_RDWR|O_NOCTTY|O_NONBLOCK);
 266         if (loop < 0) {
 267                 r = sd_bus_error_set_errnof(error, errno, "Failed to open loopback device: %m");
 268                 goto fail;
 269         }
 270
 271         if (ioctl(loop, LOOP_SET_FD, fd) < 0) {
 272                 r = sd_bus_error_set_errnof(error, errno, "Failed to bind loopback device: %m");
 273                 goto fail;
 274         }
 275
 276         if (ioctl(loop, LOOP_SET_STATUS64, &info) < 0) {
 277                 r = sd_bus_error_set_errnof(error, errno, "Failed to enable auto-clear for loopback device: %m");
 278                 goto fail;
 279         }
 280
 281         /* We need to make sure the new /var/lib/machines directory
 282          * has an access mode of 0700 at the time it is first made
 283          * available. mkfs will create it with 0755 however. Hence,
 284          * let's mount the directory into an inaccessible directory
 285          * below /tmp first, fix the access mode, and move it to the
 286          * public place then. */
 287
 288         if (!mkdtemp(tmpdir)) {
 289                 r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount parent directory: %m");
 290                 goto fail;
 291         }
 292         tmpdir_made = true;
 293
 294         mntdir = strjoina(tmpdir, "/mnt");
 295         if (mkdir(mntdir, 0700) < 0) {
 296                 r = sd_bus_error_set_errnof(error, errno, "Failed to create temporary mount directory: %m");
 297                 goto fail;
 298         }
 299         mntdir_made = true;
 300
 301         if (mount(loopdev, mntdir, "btrfs", 0, NULL) < 0) {
 302                 r = sd_bus_error_set_errnof(error, errno, "Failed to mount loopback device: %m");
 303                 goto fail;
 304         }
 305         mntdir_mounted = true;
 306
 307         r = btrfs_quota_enable(mntdir, true);
 308         if (r < 0)
 309                 log_warning_errno(r, "Failed to enable quota, ignoring: %m");
 310
 311         r = btrfs_subvol_auto_qgroup(mntdir, 0, true);
 312         if (r < 0)
 313                 log_warning_errno(r, "Failed to set up default quota hierarchy, ignoring: %m");
 314
 315         if (chmod(mntdir, 0700) < 0) {
 316                 r = sd_bus_error_set_errnof(error, errno, "Failed to fix owner: %m");
 317                 goto fail;
 318         }
 319
 320         (void) mkdir_p_label("/var/lib/machines", 0700);
 321
 322         if (mount(mntdir, "/var/lib/machines", NULL, MS_BIND, NULL) < 0) {
 323                 r = sd_bus_error_set_errnof(error, errno, "Failed to mount directory into right place: %m");
 324                 goto fail;
 325         }
 326
 327         (void) syncfs(fd);
 328
 329         log_info("Set up /var/lib/machines as btrfs loopback file system of size %s mounted on /var/lib/machines.raw.", format_bytes(buf, sizeof(buf), size));
 330
 331         (void) umount2(mntdir, MNT_DETACH);
 332         (void) rmdir(mntdir);
 333         (void) rmdir(tmpdir);
 334
 335         return 1;
 336
 337 fail:
 338         if (mntdir_mounted)
 339                 (void) umount2(mntdir, MNT_DETACH);
 340
 341         if (mntdir_made)
 342                 (void) rmdir(mntdir);
 343         if (tmpdir_made)
 344                 (void) rmdir(tmpdir);
 345
 346         if (loop >= 0) {
 347                 (void) ioctl(loop, LOOP_CLR_FD);
 348                 loop = safe_close(loop);
 349         }
 350
 351         if (control >= 0 && nr >= 0)
 352                 (void) ioctl(control, LOOP_CTL_REMOVE, nr);
 353
 354         return r;
 355 }
 356
 357 static int sync_path(const char *p) {
 358         _cleanup_close_ int fd = -1;
 359
 360         fd = open(p, O_RDONLY|O_CLOEXEC|O_NOCTTY);
 361         if (fd < 0)
 362                 return -errno;
 363
 364         if (syncfs(fd) < 0)
 365                 return -errno;
 366
 367         return 0;
 368 }
 369
 370 int grow_machine_directory(void) {
 371         char buf[FORMAT_BYTES_MAX];
 372         struct statvfs a, b;
 373         uint64_t old_size, new_size, max_add;
 374         int r;
 375
 376         /* Ensure the disk space data is accurate */
 377         sync_path("/var/lib/machines");
 378         sync_path("/var/lib/machines.raw");
 379
 380         if (statvfs("/var/lib/machines.raw", &a) < 0)
 381                 return -errno;
 382
 383         if (statvfs("/var/lib/machines", &b) < 0)
 384                 return -errno;
 385
 386         /* Don't grow if not enough disk space is available on the host */
 387         if (((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) <= VAR_LIB_MACHINES_FREE_MIN)
 388                 return 0;
 389
 390         /* Don't grow if at least 1/3th of the fs is still free */
 391         if (b.f_bavail > b.f_blocks / 3)
 392                 return 0;
 393
 394         /* Calculate how much we are willing to add at most */
 395         max_add = ((uint64_t) a.f_bavail * (uint64_t) a.f_bsize) - VAR_LIB_MACHINES_FREE_MIN;
 396
 397         /* Calculate the old size */
 398         old_size = (uint64_t) b.f_blocks * (uint64_t) b.f_bsize;
 399
 400         /* Calculate the new size as three times the size of what is used right now */
 401         new_size = ((uint64_t) b.f_blocks - (uint64_t) b.f_bavail) * (uint64_t) b.f_bsize * 3;
 402
 403         /* Always, grow at least to the start size */
 404         if (new_size < VAR_LIB_MACHINES_SIZE_START)
 405                 new_size = VAR_LIB_MACHINES_SIZE_START;
 406
 407         /* If the new size is smaller than the old size, don't grow */
 408         if (new_size < old_size)
 409                 return 0;
 410
 411         /* Ensure we never add more than the maximum */
 412         if (new_size > old_size + max_add)
 413                 new_size = old_size + max_add;
 414
 415         r = btrfs_resize_loopback("/var/lib/machines", new_size, true);
 416         if (r <= 0)
 417                 return r;
 418
 419         /* Also bump the quota, of both the subvolume leaf qgroup, as
 420          * well as of any subtree quota group by the same id but a
 421          * higher level, if it exists. */
 422         (void) btrfs_qgroup_set_limit("/var/lib/machines", 0, new_size);
 423         (void) btrfs_subvol_set_subtree_quota_limit("/var/lib/machines", 0, new_size);
 424
 425         log_info("Grew /var/lib/machines btrfs loopback file system to %s.", format_bytes(buf, sizeof(buf), new_size));
 426         return 1;
 427 }