jail/fs.c

   1 /*
   2  * Copyright (C) 2015 John Crispin <blogic@openwrt.org>
   3  * Copyright (C) 2015 Etienne Champetier <champetier.etienne@gmail.com>
   4  * Copyright (C) 2020 Daniel Golle <daniel@makrotopia.org>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU Lesser General Public License version 2.1
   8  * as published by the Free Software Foundation
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  */
  15
  16 #define _GNU_SOURCE
  17
  18 #include <assert.h>
  19 #include <elf.h>
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <linux/limits.h>
  23 #include <stdlib.h>
  24 #include <stdio.h>
  25 #include <string.h>
  26 #include <sys/stat.h>
  27 #include <sys/mman.h>
  28 #include <unistd.h>
  29 #include <libgen.h>
  30
  31 #include <libubox/avl.h>
  32 #include <libubox/avl-cmp.h>
  33 #include <libubox/blobmsg.h>
  34 #include <libubox/list.h>
  35 #include <libubox/utils.h>
  36
  37 #include "elf.h"
  38 #include "fs.h"
  39 #include "jail.h"
  40 #include "log.h"
  41
  42 #define UJAIL_NOAFILE "/tmp/.ujailnoafile"
  43
  44 struct mount {
  45         struct avl_node avl;
  46         const char *source;
  47         const char *target;
  48         const char *filesystemtype;
  49         unsigned long mountflags;
  50         unsigned long propflags;
  51         const char *optstr;
  52         int error;
  53         bool inner;
  54 };
  55
  56 struct avl_tree mounts;
  57
  58 static int do_mount(const char *root, const char *orig_source, const char *target, const char *filesystemtype,
  59                     unsigned long orig_mountflags, unsigned long propflags, const char *optstr, int error, bool inner)
  60 {
  61         struct stat s;
  62         char new[PATH_MAX];
  63         char *source = (char *)orig_source;
  64         int fd;
  65         bool is_bind = (orig_mountflags & MS_BIND);
  66         bool is_mask = (source == (void *)(-1));
  67         unsigned long mountflags = orig_mountflags;
  68
  69         if (source && is_bind && stat(source, &s)) {
  70                 ERROR("stat(%s) failed: %m\n", source);
  71                 return error;
  72         }
  73
  74         if (!is_mask && orig_source && inner) {
  75                 if (asprintf(&source, "%s%s", root, orig_source) < 0)
  76                         return ENOMEM;
  77         }
  78
  79         snprintf(new, sizeof(new), "%s%s", root, target?target:source);
  80
  81         if (is_mask) {
  82                 if (stat(new, &s))
  83                         return 0; /* doesn't exists, nothing to mask */
  84
  85                 if (S_ISDIR(s.st_mode)) {/* use empty 0-sized tmpfs for directories */
  86                         if (mount(NULL, new, "tmpfs", MS_RDONLY | MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_NOATIME, "size=0,mode=000"))
  87                                 return error;
  88                 } else {
  89                         /* mount-bind 0-sized file having mode 000 */
  90                         if (mount(UJAIL_NOAFILE, new, "bind", MS_BIND, NULL))
  91                                 return error;
  92
  93                         if (mount(UJAIL_NOAFILE, new, "bind", MS_REMOUNT | MS_BIND | MS_RDONLY | MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_NOATIME, NULL))
  94                                 return error;
  95                 }
  96
  97                 DEBUG("masked path %s\n", new);
  98                 return 0;
  99         }
 100
 101
 102         if (!is_bind || (source && S_ISDIR(s.st_mode))) {
 103                 mkdir_p(new, 0755);
 104         } else if (is_bind && source) {
 105                 mkdir_p(dirname(new), 0755);
 106                 snprintf(new, sizeof(new), "%s%s", root, target?target:source);
 107                 fd = creat(new, 0644);
 108                 if (fd == -1) {
 109                         ERROR("creat(%s) failed: %m\n", new);
 110                         return error;
 111                 }
 112                 close(fd);
 113         }
 114
 115         if (is_bind) {
 116                 if (mount(source?:new, new, filesystemtype?:"bind", MS_BIND | (mountflags & MS_REC), optstr)) {
 117                         if (error)
 118                                 ERROR("failed to mount -B %s %s: %m\n", source, new);
 119
 120                         if (inner)
 121                                 free(source);
 122
 123                         return error;
 124                 }
 125                 mountflags |= MS_REMOUNT;
 126         }
 127
 128         const char *hack_fstype = ((!filesystemtype || strcmp(filesystemtype, "cgroup"))?filesystemtype:"cgroup2");
 129         if (mount(source?:(is_bind?new:NULL), new, hack_fstype?:"none", mountflags, optstr)) {
 130                 if (error)
 131                         ERROR("failed to mount %s %s: %m\n", source, new);
 132
 133                 if (inner)
 134                         free(source);
 135
 136                 return error;
 137         }
 138
 139         DEBUG("mount %s%s %s (%s)\n", (mountflags & MS_BIND)?"-B ":"", source, new,
 140               (mountflags & MS_RDONLY)?"ro":"rw");
 141
 142         if (propflags && mount("none", new, "none", propflags, NULL)) {
 143                 if (error)
 144                         ERROR("failed to mount --make-... %s \n", new);
 145
 146                 if (inner)
 147                         free(source);
 148
 149                 return error;
 150         }
 151
 152         if (inner)
 153                 free(source);
 154
 155         return 0;
 156 }
 157
 158 static int _add_mount(const char *source, const char *target, const char *filesystemtype,
 159                       unsigned long mountflags, unsigned long propflags, const char *optstr,
 160                       int error, bool inner)
 161 {
 162         assert(target != NULL);
 163
 164         if (avl_find(&mounts, target))
 165                 return 1;
 166
 167         struct mount *m;
 168         m = calloc(1, sizeof(struct mount));
 169         if (!m)
 170                 return ENOMEM;
 171
 172         m->avl.key = m->target = strdup(target);
 173         if (source) {
 174                 if (source != (void*)(-1))
 175                         m->source = strdup(source);
 176                 else
 177                         m->source = (void*)(-1);
 178         }
 179         if (filesystemtype)
 180                 m->filesystemtype = strdup(filesystemtype);
 181
 182         if (optstr)
 183                 m->optstr = strdup(optstr);
 184
 185         m->mountflags = mountflags;
 186         m->propflags = propflags;
 187         m->error = error;
 188         m->inner = inner;
 189
 190         avl_insert(&mounts, &m->avl);
 191         DEBUG("adding mount %s %s bind(%d) ro(%d) err(%d)\n", (m->source == (void*)(-1))?"mask":m->source, m->target,
 192                 !!(m->mountflags & MS_BIND), !!(m->mountflags & MS_RDONLY), m->error != 0);
 193
 194         return 0;
 195 }
 196
 197 int add_mount(const char *source, const char *target, const char *filesystemtype,
 198               unsigned long mountflags, unsigned long propflags, const char *optstr, int error)
 199 {
 200         return _add_mount(source, target, filesystemtype, mountflags, propflags, optstr, error, false);
 201 }
 202
 203 int add_mount_inner(const char *source, const char *target, const char *filesystemtype,
 204               unsigned long mountflags, unsigned long propflags, const char *optstr, int error)
 205 {
 206         return _add_mount(source, target, filesystemtype, mountflags, propflags, optstr, error, true);
 207 }
 208
 209 int add_mount_bind(const char *path, int readonly, int error)
 210 {
 211         unsigned long mountflags = MS_BIND;
 212
 213         if (readonly)
 214                 mountflags |= MS_RDONLY;
 215
 216         return add_mount(path, path, NULL, mountflags, 0, NULL, error);
 217 }
 218
 219 enum {
 220         OCI_MOUNT_SOURCE,
 221         OCI_MOUNT_DESTINATION,
 222         OCI_MOUNT_TYPE,
 223         OCI_MOUNT_OPTIONS,
 224         __OCI_MOUNT_MAX,
 225 };
 226
 227 static const struct blobmsg_policy oci_mount_policy[] = {
 228         [OCI_MOUNT_SOURCE] = { "source", BLOBMSG_TYPE_STRING },
 229         [OCI_MOUNT_DESTINATION] = { "destination", BLOBMSG_TYPE_STRING },
 230         [OCI_MOUNT_TYPE] = { "type", BLOBMSG_TYPE_STRING },
 231         [OCI_MOUNT_OPTIONS] = { "options", BLOBMSG_TYPE_ARRAY },
 232 };
 233
 234 struct mount_opt {
 235         struct list_head list;
 236         char *optstr;
 237 };
 238
 239 #ifndef MS_LAZYTIME
 240 #define MS_LAZYTIME (1 << 25)
 241 #endif
 242
 243 static int parseOCImountopts(struct blob_attr *msg, unsigned long *mount_flags, unsigned long *propagation_flags, char **mount_data, int *error)
 244 {
 245         struct blob_attr *cur;
 246         int rem;
 247         unsigned long mf = 0;
 248         unsigned long pf = 0;
 249         char *tmp;
 250         struct list_head fsopts = LIST_HEAD_INIT(fsopts);
 251         size_t len = 0;
 252         struct mount_opt *opt, *tmpopt;
 253
 254         blobmsg_for_each_attr(cur, msg, rem) {
 255                 tmp = blobmsg_get_string(cur);
 256                 if (!strcmp("ro", tmp))
 257                         mf |= MS_RDONLY;
 258                 else if (!strcmp("rw", tmp))
 259                         mf &= ~MS_RDONLY;
 260                 else if (!strcmp("bind", tmp))
 261                         mf = MS_BIND;
 262                 else if (!strcmp("rbind", tmp))
 263                         mf |= MS_BIND | MS_REC;
 264                 else if (!strcmp("sync", tmp))
 265                         mf |= MS_SYNCHRONOUS;
 266                 else if (!strcmp("async", tmp))
 267                         mf &= ~MS_SYNCHRONOUS;
 268                 else if (!strcmp("atime", tmp))
 269                         mf &= ~MS_NOATIME;
 270                 else if (!strcmp("noatime", tmp))
 271                         mf |= MS_NOATIME;
 272                 else if (!strcmp("defaults", tmp))
 273                         mf = 0; /* rw, suid, dev, exec, auto, nouser, and async */
 274                 else if (!strcmp("dev", tmp))
 275                         mf &= ~MS_NODEV;
 276                 else if (!strcmp("nodev", tmp))
 277                         mf |= MS_NODEV;
 278                 else if (!strcmp("iversion", tmp))
 279                         mf |= MS_I_VERSION;
 280                 else if (!strcmp("noiversion", tmp))
 281                         mf &= ~MS_I_VERSION;
 282                 else if (!strcmp("diratime", tmp))
 283                         mf &= ~MS_NODIRATIME;
 284                 else if (!strcmp("nodiratime", tmp))
 285                         mf |= MS_NODIRATIME;
 286                 else if (!strcmp("dirsync", tmp))
 287                         mf |= MS_DIRSYNC;
 288                 else if (!strcmp("exec", tmp))
 289                         mf &= ~MS_NOEXEC;
 290                 else if (!strcmp("noexec", tmp))
 291                         mf |= MS_NOEXEC;
 292                 else if (!strcmp("mand", tmp))
 293                         mf |= MS_MANDLOCK;
 294                 else if (!strcmp("nomand", tmp))
 295                         mf &= ~MS_MANDLOCK;
 296                 else if (!strcmp("relatime", tmp))
 297                         mf |= MS_RELATIME;
 298                 else if (!strcmp("norelatime", tmp))
 299                         mf &= ~MS_RELATIME;
 300                 else if (!strcmp("strictatime", tmp))
 301                         mf |= MS_STRICTATIME;
 302                 else if (!strcmp("nostrictatime", tmp))
 303                         mf &= ~MS_STRICTATIME;
 304                 else if (!strcmp("lazytime", tmp))
 305                         mf |= MS_LAZYTIME;
 306                 else if (!strcmp("nolazytime", tmp))
 307                         mf &= ~MS_LAZYTIME;
 308                 else if (!strcmp("suid", tmp))
 309                         mf &= ~MS_NOSUID;
 310                 else if (!strcmp("nosuid", tmp))
 311                         mf |= MS_NOSUID;
 312                 else if (!strcmp("remount", tmp))
 313                         mf |= MS_REMOUNT;
 314                 /* propagation flags */
 315                 else if (!strcmp("private", tmp))
 316                         pf |= MS_PRIVATE;
 317                 else if (!strcmp("rprivate", tmp))
 318                         pf |= MS_PRIVATE | MS_REC;
 319                 else if (!strcmp("slave", tmp))
 320                         pf |= MS_SLAVE;
 321                 else if (!strcmp("rslave", tmp))
 322                         pf |= MS_SLAVE | MS_REC;
 323                 else if (!strcmp("shared", tmp))
 324                         pf |= MS_SHARED;
 325                 else if (!strcmp("rshared", tmp))
 326                         pf |= MS_SHARED | MS_REC;
 327                 else if (!strcmp("unbindable", tmp))
 328                         pf |= MS_UNBINDABLE;
 329                 else if (!strcmp("runbindable", tmp))
 330                         pf |= MS_UNBINDABLE | MS_REC;
 331                 /* special case: 'nofail' */
 332                 else if(!strcmp("nofail", tmp))
 333                         *error = 0;
 334                 else if (!strcmp("auto", tmp) ||
 335                          !strcmp("noauto", tmp) ||
 336                          !strcmp("user", tmp) ||
 337                          !strcmp("group", tmp) ||
 338                          !strcmp("_netdev", tmp))
 339                         DEBUG("ignoring built-in mount option %s\n", tmp);
 340                 else {
 341                         /* filesystem-specific free-form option */
 342                         opt = calloc(1, sizeof(*opt));
 343                         opt->optstr = tmp;
 344                         list_add_tail(&opt->list, &fsopts);
 345                 }
 346         };
 347
 348         *mount_flags = mf;
 349         *propagation_flags = pf;
 350
 351         list_for_each_entry(opt, &fsopts, list) {
 352                 if (len)
 353                         ++len;
 354
 355                 len += strlen(opt->optstr);
 356         };
 357
 358         if (len) {
 359                 *mount_data = calloc(len + 1, sizeof(char));
 360                 if (!mount_data)
 361                         return ENOMEM;
 362
 363                 len = 0;
 364                 list_for_each_entry(opt, &fsopts, list) {
 365                         if (len)
 366                                 strcat(*mount_data, ",");
 367
 368                         strcat(*mount_data, opt->optstr);
 369                         ++len;
 370                 }
 371
 372                 list_for_each_entry_safe(opt, tmpopt, &fsopts, list) {
 373                         list_del(&opt->list);
 374                         free(opt);
 375                 }
 376         }
 377
 378         DEBUG("mount flags(%08lx) propagation(%08lx) fsopts(\"%s\")\n", mf, pf, *mount_data?:"");
 379
 380         return 0;
 381 }
 382
 383 int parseOCImount(struct blob_attr *msg)
 384 {
 385         struct blob_attr *tb[__OCI_MOUNT_MAX];
 386         unsigned long mount_flags = 0;
 387         unsigned long propagation_flags = 0;
 388         char *mount_data = NULL;
 389         int ret, err = -1;
 390
 391         blobmsg_parse(oci_mount_policy, __OCI_MOUNT_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
 392
 393         if (!tb[OCI_MOUNT_DESTINATION])
 394                 return EINVAL;
 395
 396         if (tb[OCI_MOUNT_OPTIONS]) {
 397                 ret = parseOCImountopts(tb[OCI_MOUNT_OPTIONS], &mount_flags, &propagation_flags, &mount_data, &err);
 398                 if (ret)
 399                         return ret;
 400         }
 401
 402         ret = add_mount(tb[OCI_MOUNT_SOURCE] ? blobmsg_get_string(tb[OCI_MOUNT_SOURCE]) : NULL,
 403                   blobmsg_get_string(tb[OCI_MOUNT_DESTINATION]),
 404                   tb[OCI_MOUNT_TYPE] ? blobmsg_get_string(tb[OCI_MOUNT_TYPE]) : NULL,
 405                   mount_flags, propagation_flags, mount_data, err);
 406
 407         if (mount_data)
 408                 free(mount_data);
 409
 410         return ret;
 411 }
 412
 413 static void build_noafile(void) {
 414         int fd;
 415
 416         fd = creat(UJAIL_NOAFILE, 0000);
 417         if (fd == -1)
 418                 return;
 419
 420         close(fd);
 421         return;
 422 }
 423
 424 int mount_all(const char *jailroot) {
 425         struct library *l;
 426         struct mount *m;
 427
 428         build_noafile();
 429
 430         avl_for_each_element(&libraries, l, avl)
 431                 add_mount_bind(l->path, 1, -1);
 432
 433         avl_for_each_element(&mounts, m, avl)
 434                 if (do_mount(jailroot, m->source, m->target, m->filesystemtype, m->mountflags,
 435                              m->propflags, m->optstr, m->error, m->inner))
 436                         return -1;
 437
 438         return 0;
 439 }
 440
 441 void mount_free(void) {
 442         struct mount *m, *tmp;
 443
 444         avl_remove_all_elements(&mounts, m, avl, tmp) {
 445                 if (m->source != (void*)(-1))
 446                         free((void*)m->source);
 447                 free((void*)m->target);
 448                 free((void*)m->filesystemtype);
 449                 free((void*)m->optstr);
 450                 free(m);
 451         }
 452 }
 453
 454 void mount_list_init(void) {
 455         avl_init(&mounts, avl_strcmp, false, NULL);
 456 }
 457
 458 static int add_script_interp(const char *path, const char *map, int size)
 459 {
 460         int start = 2;
 461         while (start < size && map[start] != '/') {
 462                 start++;
 463         }
 464         if (start >= size) {
 465                 ERROR("bad script interp (%s)\n", path);
 466                 return -1;
 467         }
 468         int stop = start + 1;
 469         while (stop < size && map[stop] > 0x20 && map[stop] <= 0x7e) {
 470                 stop++;
 471         }
 472         if (stop >= size || (stop-start) > PATH_MAX) {
 473                 ERROR("bad script interp (%s)\n", path);
 474                 return -1;
 475         }
 476         char buf[PATH_MAX];
 477         strncpy(buf, map+start, stop-start);
 478         return add_path_and_deps(buf, 1, -1, 0);
 479 }
 480
 481 int add_path_and_deps(const char *path, int readonly, int error, int lib)
 482 {
 483         assert(path != NULL);
 484
 485         if (lib == 0 && path[0] != '/') {
 486                 ERROR("%s is not an absolute path\n", path);
 487                 return error;
 488         }
 489
 490         char *map = NULL;
 491         int fd, ret = -1;
 492         if (path[0] == '/') {
 493                 if (avl_find(&mounts, path))
 494                         return 0;
 495                 fd = open(path, O_RDONLY|O_CLOEXEC);
 496                 if (fd == -1)
 497                         return error;
 498                 add_mount_bind(path, readonly, error);
 499         } else {
 500                 if (avl_find(&libraries, path))
 501                         return 0;
 502                 char *fullpath;
 503                 fd = lib_open(&fullpath, path);
 504                 if (fd == -1)
 505                         return error;
 506                 if (fullpath) {
 507                         alloc_library(fullpath, path);
 508                         free(fullpath);
 509                 }
 510         }
 511
 512         struct stat s;
 513         if (fstat(fd, &s) == -1) {
 514                 ERROR("fstat(%s) failed: %m\n", path);
 515                 ret = error;
 516                 goto out;
 517         }
 518
 519         if (!S_ISREG(s.st_mode)) {
 520                 ret = 0;
 521                 goto out;
 522         }
 523
 524         /* too small to be an ELF or a script -> "normal" file */
 525         if (s.st_size < 4) {
 526                 ret = 0;
 527                 goto out;
 528         }
 529
 530         map = mmap(NULL, s.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
 531         if (map == MAP_FAILED) {
 532                 ERROR("failed to mmap %s: %m\n", path);
 533                 ret = -1;
 534                 goto out;
 535         }
 536
 537         if (map[0] == '#' && map[1] == '!') {
 538                 ret = add_script_interp(path, map, s.st_size);
 539                 goto out;
 540         }
 541
 542         if (map[0] == ELFMAG0 && map[1] == ELFMAG1 && map[2] == ELFMAG2 && map[3] == ELFMAG3) {
 543                 ret = elf_load_deps(path, map);
 544                 goto out;
 545         }
 546
 547         ret = 0;
 548
 549 out:
 550         if (fd >= 0)
 551                 close(fd);
 552         if (map)
 553                 munmap(map, s.st_size);
 554
 555         return ret;
 556 }