src/lxc/storage/storage_utils.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * Copyright © 2017 Canonical Ltd.
   5  *
   6  * Authors:
   7  * Christian Brauner <christian.brauner@ubuntu.com>
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #ifndef _GNU_SOURCE
  25 #define _GNU_SOURCE 1
  26 #endif
  27 #include <ctype.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <inttypes.h>
  33 #include <libgen.h>
  34 #include <sched.h>
  35 #include <stdint.h>
  36 #include <stdio.h>
  37 #include <string.h>
  38 #include <sys/mount.h>
  39 #include <sys/prctl.h>
  40 #include <sys/stat.h>
  41 #include <sys/types.h>
  42 #include <sys/wait.h>
  43 #include <unistd.h>
  44
  45 #include "config.h"
  46 #include "log.h"
  47 #include "nbd.h"
  48 #include "parse.h"
  49 #include "storage.h"
  50 #include "storage_utils.h"
  51 #include "syscall_wrappers.h"
  52 #include "utils.h"
  53
  54 #ifndef HAVE_STRLCPY
  55 #include "include/strlcpy.h"
  56 #endif
  57
  58 #ifndef BLKGETSIZE64
  59 #define BLKGETSIZE64 _IOR(0x12, 114, size_t)
  60 #endif
  61
  62 lxc_log_define(storage_utils, lxc);
  63
  64 /* the bulk of this needs to become a common helper */
  65 char *dir_new_path(char *src, const char *oldname, const char *name,
  66                    const char *oldpath, const char *lxcpath)
  67 {
  68         char *ret, *p, *p2;
  69         int l1, l2, nlen;
  70
  71         nlen = strlen(src) + 1;
  72         l1 = strlen(oldpath);
  73         p = src;
  74         /* if src starts with oldpath, look for oldname only after
  75          * that path */
  76         if (strncmp(src, oldpath, l1) == 0) {
  77                 p += l1;
  78                 nlen += (strlen(lxcpath) - l1);
  79         }
  80         l2 = strlen(oldname);
  81         while ((p = strstr(p, oldname)) != NULL) {
  82                 p += l2;
  83                 nlen += strlen(name) - l2;
  84         }
  85
  86         ret = malloc(nlen);
  87         if (!ret)
  88                 return NULL;
  89
  90         p = ret;
  91         if (strncmp(src, oldpath, l1) == 0) {
  92                 p += sprintf(p, "%s", lxcpath);
  93                 src += l1;
  94         }
  95
  96         while ((p2 = strstr(src, oldname)) != NULL) {
  97                 size_t retlen;
  98
  99                 /* copy text up to oldname */
 100                 retlen = strlcpy(p, src, p2 - src);
 101                 if (retlen >= p2 - src) {
 102                         free(ret);
 103                         return NULL;
 104                 }
 105
 106                 /* move target pointer (p) */
 107                 p += p2 - src;
 108                 /* print new name in place of oldname */
 109                 p += sprintf(p, "%s", name);
 110                 /* move src to end of oldname */
 111                 src = p2 + l2;
 112         }
 113
 114         /* copy the rest of src */
 115         sprintf(p, "%s", src);
 116         return ret;
 117 }
 118
 119 /*
 120  * attach_block_device returns true if all went well,
 121  * meaning either a block device was attached or was not
 122  * needed.  It returns false if something went wrong and
 123  * container startup should be stopped.
 124  */
 125 bool attach_block_device(struct lxc_conf *conf)
 126 {
 127         char *path;
 128
 129         if (!conf->rootfs.path)
 130                 return true;
 131
 132         path = conf->rootfs.path;
 133         if (!requires_nbd(path))
 134                 return true;
 135
 136         path = strchr(path, ':');
 137         if (!path)
 138                 return false;
 139
 140         path++;
 141         if (!attach_nbd(path, conf))
 142                 return false;
 143
 144         return true;
 145 }
 146
 147 /*
 148  * return block size of dev->src in units of bytes
 149  */
 150 int blk_getsize(struct lxc_storage *bdev, uint64_t *size)
 151 {
 152         int fd, ret;
 153         const char *src;
 154
 155         src = lxc_storage_get_path(bdev->src, bdev->type);
 156         fd = open(src, O_RDONLY);
 157         if (fd < 0)
 158                 return -1;
 159
 160         /* size of device in bytes */
 161         ret = ioctl(fd, BLKGETSIZE64, size);
 162         close(fd);
 163         return ret;
 164 }
 165
 166 void detach_block_device(struct lxc_conf *conf)
 167 {
 168         if (conf->nbd_idx != -1)
 169                 detach_nbd_idx(conf->nbd_idx);
 170 }
 171
 172 /*
 173  * Given a lxc_storage (presumably blockdev-based), detect the fstype
 174  * by trying mounting (in a private mntns) it.
 175  * @lxc_storage: bdev to investigate
 176  * @type: preallocated char* in which to write the fstype
 177  * @len: length of passed in char*
 178  * Returns length of fstype, of -1 on error
 179  */
 180 int detect_fs(struct lxc_storage *bdev, char *type, int len)
 181 {
 182         int ret;
 183         int p[2];
 184         size_t linelen;
 185         pid_t pid;
 186         FILE *f;
 187         char *sp1, *sp2, *sp3;
 188         const char *l, *srcdev;
 189         char devpath[PATH_MAX];
 190         char *line = NULL;
 191
 192         if (!bdev || !bdev->src || !bdev->dest)
 193                 return -1;
 194
 195         srcdev = lxc_storage_get_path(bdev->src, bdev->type);
 196
 197         ret = pipe(p);
 198         if (ret < 0)
 199                 return -1;
 200
 201         if ((pid = fork()) < 0)
 202                 return -1;
 203
 204         if (pid > 0) {
 205                 int status;
 206                 close(p[1]);
 207                 memset(type, 0, len);
 208                 ret = read(p[0], type, len - 1);
 209                 close(p[0]);
 210                 if (ret < 0) {
 211                         SYSERROR("error reading from pipe");
 212                         wait(&status);
 213                         return -1;
 214                 } else if (ret == 0) {
 215                         ERROR("child exited early - fstype not found");
 216                         wait(&status);
 217                         return -1;
 218                 }
 219                 wait(&status);
 220                 type[len - 1] = '\0';
 221                 INFO("detected fstype %s for %s", type, srcdev);
 222                 return ret;
 223         }
 224
 225         if (unshare(CLONE_NEWNS) < 0)
 226                 exit(1);
 227
 228         if (detect_shared_rootfs()) {
 229                 if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) {
 230                         SYSERROR("Failed to make / rslave");
 231                         ERROR("Continuing...");
 232                 }
 233         }
 234
 235         ret = mount_unknown_fs(srcdev, bdev->dest, bdev->mntopts);
 236         if (ret < 0) {
 237                 ERROR("failed mounting %s onto %s to detect fstype", srcdev,
 238                       bdev->dest);
 239                 exit(1);
 240         }
 241
 242         l = linkderef(srcdev, devpath);
 243         if (!l)
 244                 exit(1);
 245         f = fopen("/proc/self/mounts", "r");
 246         if (!f)
 247                 exit(1);
 248
 249         while (getline(&line, &linelen, f) != -1) {
 250                 sp1 = strchr(line, ' ');
 251                 if (!sp1)
 252                         exit(1);
 253                 *sp1 = '\0';
 254                 if (strcmp(line, l))
 255                         continue;
 256                 sp2 = strchr(sp1 + 1, ' ');
 257                 if (!sp2)
 258                         exit(1);
 259                 *sp2 = '\0';
 260                 sp3 = strchr(sp2 + 1, ' ');
 261                 if (!sp3)
 262                         exit(1);
 263                 *sp3 = '\0';
 264                 sp2++;
 265                 if (write(p[1], sp2, strlen(sp2)) != strlen(sp2))
 266                         exit(1);
 267
 268                 exit(0);
 269         }
 270
 271         exit(1);
 272 }
 273
 274 int do_mkfs_exec_wrapper(void *args)
 275 {
 276         int ret;
 277         char *mkfs;
 278         char **data = args;
 279         /* strlen("mkfs.")
 280          * +
 281          * strlen(data[0])
 282          * +
 283          * \0
 284          */
 285         size_t len = 5 + strlen(data[0]) + 1;
 286
 287         mkfs = malloc(len);
 288         if (!mkfs)
 289                 return -1;
 290
 291         ret = snprintf(mkfs, len, "mkfs.%s", data[0]);
 292         if (ret < 0 || (size_t)ret >= len) {
 293                 free(mkfs);
 294                 return -1;
 295         }
 296
 297         TRACE("executing \"%s %s\"", mkfs, data[1]);
 298         execlp(mkfs, mkfs, data[1], (char *)NULL);
 299         SYSERROR("failed to run \"%s %s \"", mkfs, data[1]);
 300         free(mkfs);
 301         return -1;
 302 }
 303
 304 /*
 305  * This will return 1 for physical disks, qemu-nbd, loop, etc right now only lvm
 306  * is a block device.
 307  */
 308 int is_blktype(struct lxc_storage *b)
 309 {
 310         if (strcmp(b->type, "lvm") == 0)
 311                 return 1;
 312
 313         return 0;
 314 }
 315
 316 int mount_unknown_fs(const char *rootfs, const char *target,
 317                      const char *options)
 318 {
 319         size_t i;
 320         int ret;
 321         struct cbarg {
 322                 const char *rootfs;
 323                 const char *target;
 324                 const char *options;
 325         } cbarg = {
 326             .rootfs = rootfs,
 327             .target = target,
 328             .options = options,
 329         };
 330
 331         /*
 332          * find the filesystem type with brute force:
 333          * first we check with /etc/filesystems, in case the modules
 334          * are auto-loaded and fall back to the supported kernel fs
 335          */
 336         char *fsfile[] = {
 337             "/etc/filesystems",
 338             "/proc/filesystems",
 339         };
 340
 341         for (i = 0; i < sizeof(fsfile) / sizeof(fsfile[0]); i++) {
 342                 if (access(fsfile[i], F_OK))
 343                         continue;
 344
 345                 ret = lxc_file_for_each_line(fsfile[i], find_fstype_cb, &cbarg);
 346                 if (ret < 0) {
 347                         ERROR("failed to parse '%s'", fsfile[i]);
 348                         return -1;
 349                 }
 350
 351                 if (ret)
 352                         return 0;
 353         }
 354
 355         ERROR("failed to determine fs type for '%s'", rootfs);
 356         return -1;
 357 }
 358
 359 /*
 360  * These are copied from conf.c.  However as conf.c will be moved to using
 361  * the callback system, they can be pulled from there eventually, so we
 362  * don't need to pollute utils.c with these low level functions
 363  */
 364 int find_fstype_cb(char *buffer, void *data)
 365 {
 366         struct cbarg {
 367                 const char *rootfs;
 368                 const char *target;
 369                 const char *options;
 370         } *cbarg = data;
 371
 372         unsigned long mntflags;
 373         char *mntdata;
 374         char *fstype;
 375
 376         /* we don't try 'nodev' entries */
 377         if (strstr(buffer, "nodev"))
 378                 return 0;
 379
 380         fstype = buffer;
 381         fstype += lxc_char_left_gc(fstype, strlen(fstype));
 382         fstype[lxc_char_right_gc(fstype, strlen(fstype))] = '\0';
 383
 384         DEBUG("trying to mount '%s'->'%s' with fstype '%s'", cbarg->rootfs,
 385               cbarg->target, fstype);
 386
 387         if (parse_mntopts(cbarg->options, &mntflags, &mntdata) < 0) {
 388                 free(mntdata);
 389                 return 0;
 390         }
 391
 392         if (mount(cbarg->rootfs, cbarg->target, fstype, mntflags, mntdata)) {
 393                 SYSDEBUG("mount failed with error");
 394                 free(mntdata);
 395                 return 0;
 396         }
 397
 398         free(mntdata);
 399
 400         INFO("mounted '%s' on '%s', with fstype '%s'", cbarg->rootfs,
 401              cbarg->target, fstype);
 402
 403         return 1;
 404 }
 405
 406 const char *linkderef(const char *path, char *dest)
 407 {
 408         struct stat sbuf;
 409         ssize_t ret;
 410
 411         ret = stat(path, &sbuf);
 412         if (ret < 0)
 413                 return NULL;
 414
 415         if (!S_ISLNK(sbuf.st_mode))
 416                 return path;
 417
 418         ret = readlink(path, dest, PATH_MAX);
 419         if (ret < 0) {
 420                 SYSERROR("error reading link %s", path);
 421                 return NULL;
 422         } else if (ret >= PATH_MAX) {
 423                 ERROR("link in %s too long", path);
 424                 return NULL;
 425         }
 426         dest[ret] = '\0';
 427
 428         return dest;
 429 }
 430
 431 /*
 432  * is an unprivileged user allowed to make this kind of snapshot
 433  */
 434 bool unpriv_snap_allowed(struct lxc_storage *b, const char *t, bool snap,
 435                          bool maybesnap)
 436 {
 437         if (!t) {
 438                 /* New type will be same as original (unless snap && b->type ==
 439                  * dir, in which case it will be overlayfs -- which is also
 440                  * allowed).
 441                  */
 442                 if (strcmp(b->type, "dir") == 0 ||
 443                     strcmp(b->type, "overlay") == 0 ||
 444                     strcmp(b->type, "overlayfs") == 0 ||
 445                     strcmp(b->type, "btrfs") == 0 ||
 446                     strcmp(b->type, "loop") == 0)
 447                         return true;
 448
 449                 return false;
 450         }
 451
 452         /* Unprivileged users can copy and snapshot dir, overlayfs, and loop.
 453          * In particular, not zfs, btrfs, or lvm.
 454          */
 455         if (strcmp(t, "dir") == 0 ||
 456             strcmp(t, "overlay") == 0 ||
 457             strcmp(t, "overlayfs") == 0 ||
 458             strcmp(t, "btrfs") == 0 ||
 459             strcmp(t, "loop") == 0)
 460                 return true;
 461
 462         return false;
 463 }
 464
 465 uint64_t get_fssize(char *s)
 466 {
 467         uint64_t ret;
 468         char *end;
 469
 470         ret = strtoull(s, &end, 0);
 471         if (end == s) {
 472                 ERROR("Invalid blockdev size '%s', using default size", s);
 473                 return 0;
 474         }
 475
 476         while (isblank(*end))
 477                 end++;
 478
 479         if (*end == '\0') {
 480                 ret *= 1024ULL * 1024ULL; /* MB by default */
 481         } else if (*end == 'b' || *end == 'B') {
 482                 ret *= 1ULL;
 483         } else if (*end == 'k' || *end == 'K') {
 484                 ret *= 1024ULL;
 485         } else if (*end == 'm' || *end == 'M') {
 486                 ret *= 1024ULL * 1024ULL;
 487         } else if (*end == 'g' || *end == 'G') {
 488                 ret *= 1024ULL * 1024ULL * 1024ULL;
 489         } else if (*end == 't' || *end == 'T') {
 490                 ret *= 1024ULL * 1024ULL * 1024ULL * 1024ULL;
 491         } else {
 492                 ERROR("Invalid blockdev unit size '%c' in '%s', using default size", *end, s);
 493                 return 0;
 494         }
 495
 496         return ret;
 497 }
 498
 499 bool is_valid_storage_type(const char *type)
 500 {
 501         if (strcmp(type, "dir") == 0 ||
 502             strcmp(type, "btrfs") == 0 ||
 503             strcmp(type, "loop") == 0 ||
 504             strcmp(type, "lvm") == 0 ||
 505             strcmp(type, "nbd") == 0 ||
 506             strcmp(type, "overlay") == 0 ||
 507             strcmp(type, "overlayfs") == 0 ||
 508             strcmp(type, "rbd") == 0 ||
 509             strcmp(type, "zfs") == 0)
 510                 return true;
 511
 512         return false;
 513 }
 514
 515 int storage_destroy_wrapper(void *data)
 516 {
 517         struct lxc_conf *conf = data;
 518
 519         if (setgid(0) < 0) {
 520                 ERROR("Failed to setgid to 0");
 521                 return -1;
 522         }
 523
 524         if (setgroups(0, NULL) < 0)
 525                 WARN("Failed to clear groups");
 526
 527         if (setuid(0) < 0) {
 528                 ERROR("Failed to setuid to 0");
 529                 return -1;
 530         }
 531
 532         if (!storage_destroy(conf))
 533                 return -1;
 534
 535         return 0;
 536 }