lxcfs.c

   1 /* lxcfs
   2  *
   3  * Copyright © 2014 Canonical, Inc
   4  * Author: Serge Hallyn <serge.hallyn@ubuntu.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2, as
   8  * 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  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 /*
  21  * NOTES - make sure to run this as -s to avoid threading.
  22  * TODO - can we enforce that here from the code?
  23  */
  24 #define FUSE_USE_VERSION 26
  25
  26 #include <stdio.h>
  27 #include <dirent.h>
  28 #include <fcntl.h>
  29 #include <fuse.h>
  30 #include <unistd.h>
  31 #include <errno.h>
  32 #include <stdbool.h>
  33 #include <time.h>
  34 #include <string.h>
  35 #include <stdlib.h>
  36 #include <libgen.h>
  37
  38 #include <nih/alloc.h>
  39 #include <nih/string.h>
  40
  41 #include "cgmanager.h"
  42
  43 struct lxcfs_state {
  44         /*
  45          * a null-terminated, nih-allocated list of the mounted subsystems.  We
  46          * detect this at startup.
  47          */
  48         char **subsystems;
  49 };
  50 #define LXCFS_DATA ((struct lxcfs_state *) fuse_get_context()->private_data)
  51
  52 /*
  53  * Given a open file * to /proc/pid/{u,g}id_map, and an id
  54  * valid in the caller's namespace, return the id mapped into
  55  * pid's namespace.
  56  * Returns the mapped id, or -1 on error.
  57  */
  58 unsigned int
  59 convert_id_to_ns(FILE *idfile, unsigned int in_id)
  60 {
  61         unsigned int nsuid,   // base id for a range in the idfile's namespace
  62                      hostuid, // base id for a range in the caller's namespace
  63                      count;   // number of ids in this range
  64         char line[400];
  65         int ret;
  66
  67         fseek(idfile, 0L, SEEK_SET);
  68         while (fgets(line, 400, idfile)) {
  69                 ret = sscanf(line, "%u %u %u\n", &nsuid, &hostuid, &count);
  70                 if (ret != 3)
  71                         continue;
  72                 if (hostuid + count < hostuid || nsuid + count < nsuid) {
  73                         /*
  74                          * uids wrapped around - unexpected as this is a procfile,
  75                          * so just bail.
  76                          */
  77                         fprintf(stderr, "pid wrapparound at entry %u %u %u in %s",
  78                                 nsuid, hostuid, count, line);
  79                         return -1;
  80                 }
  81                 if (hostuid <= in_id && hostuid+count > in_id) {
  82                         /*
  83                          * now since hostuid <= in_id < hostuid+count, and
  84                          * hostuid+count and nsuid+count do not wrap around,
  85                          * we know that nsuid+(in_id-hostuid) which must be
  86                          * less that nsuid+(count) must not wrap around
  87                          */
  88                         return (in_id - hostuid) + nsuid;
  89                 }
  90         }
  91
  92         // no answer found
  93         return -1;
  94 }
  95
  96 /*
  97  * for is_privileged_over,
  98  * specify whether we require the calling uid to be root in his
  99  * namespace
 100  */
 101 #define NS_ROOT_REQD true
 102 #define NS_ROOT_OPT false
 103
 104 static bool is_privileged_over(pid_t pid, uid_t uid, uid_t victim, bool req_ns_root)
 105 {
 106         nih_local char *fpath = NULL;
 107         bool answer = false;
 108         uid_t nsuid;
 109
 110         if (victim == -1 || uid == -1)
 111                 return false;
 112
 113         /*
 114          * If the request is one not requiring root in the namespace,
 115          * then having the same uid suffices.  (i.e. uid 1000 has write
 116          * access to files owned by uid 1000
 117          */
 118         if (!req_ns_root && uid == victim)
 119                 return true;
 120
 121         fpath = NIH_MUST( nih_sprintf(NULL, "/proc/%d/uid_map", pid) );
 122         FILE *f = fopen(fpath, "r");
 123         if (!f)
 124                 return false;
 125
 126         /* if caller's not root in his namespace, reject */
 127         nsuid = convert_id_to_ns(f, uid);
 128         if (nsuid)
 129                 goto out;
 130
 131         /*
 132          * If victim is not mapped into caller's ns, reject.
 133          * XXX I'm not sure this check is needed given that fuse
 134          * will be sending requests where the vfs has converted
 135          */
 136         nsuid = convert_id_to_ns(f, victim);
 137         if (nsuid == -1)
 138                 goto out;
 139
 140         answer = true;
 141
 142 out:
 143         fclose(f);
 144         return answer;
 145 }
 146
 147 static bool perms_include(int fmode, mode_t req_mode)
 148 {
 149         mode_t r;
 150
 151         switch (req_mode & O_ACCMODE) {
 152         case O_RDONLY:
 153                 r = S_IROTH;
 154                 break;
 155         case O_WRONLY:
 156                 r = S_IWOTH;
 157                 break;
 158         case O_RDWR:
 159                 r = S_IROTH | S_IWOTH;
 160                 break;
 161         default:
 162                 return false;
 163         }
 164         return ((fmode & r) == r);
 165 }
 166
 167 /*
 168  * check whether a fuse context may access a cgroup dir or file
 169  *
 170  * If file is not null, it is a cgroup file to check under cg.
 171  * If file is null, then we are checking perms on cg itself.
 172  *
 173  * For files we can check the mode of the list_keys result.
 174  * For cgroups, we must make assumptions based on the files under the
 175  * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups
 176  * yet.
 177  */
 178 static bool fc_may_access(struct fuse_context *fc, const char *contrl, const char *cg, const char *file, mode_t mode)
 179 {
 180         nih_local struct cgm_keys **list = NULL;
 181         int i;
 182
 183         if (!file)
 184                 file = "tasks";
 185
 186         if (*file == '/')
 187                 file++;
 188
 189         if (!cgm_list_keys(contrl, cg, &list))
 190                 return false;
 191         for (i = 0; list[i]; i++) {
 192                 if (strcmp(list[i]->name, file) == 0) {
 193                         struct cgm_keys *k = list[i];
 194                         if (is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_OPT)) {
 195                                 if (perms_include(k->mode >> 6, mode))
 196                                         return true;
 197                         }
 198                         if (fc->gid == k->gid) {
 199                                 if (perms_include(k->mode >> 3, mode))
 200                                         return true;
 201                         }
 202                         return perms_include(k->mode, mode);
 203                 }
 204         }
 205
 206         return false;
 207 }
 208
 209 /*
 210  * given /cgroup/freezer/a/b, return "freezer".  this will be nih-allocated
 211  * and needs to be nih_freed.
 212  */
 213 static char *pick_controller_from_path(struct fuse_context *fc, const char *path)
 214 {
 215         const char *p1;
 216         char *ret, *slash;
 217
 218         if (strlen(path) < 9)
 219                 return NULL;
 220         p1 = path+8;
 221         ret = nih_strdup(NULL, p1);
 222         if (!ret)
 223                 return ret;
 224         slash = strstr(ret, "/");
 225         if (slash)
 226                 *slash = '\0';
 227
 228         /* verify that it is a subsystem */
 229         char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL;
 230         int i;
 231         if (!list) {
 232                 nih_free(ret);
 233                 return NULL;
 234         }
 235         for (i = 0;  list[i];  i++) {
 236                 if (strcmp(list[i], ret) == 0)
 237                         return ret;
 238         }
 239         nih_free(ret);
 240         return NULL;
 241 }
 242
 243 /*
 244  * Find the start of cgroup in /cgroup/controller/the/cgroup/path
 245  * Note that the returned value may include files (keynames) etc
 246  */
 247 static const char *find_cgroup_in_path(const char *path)
 248 {
 249         const char *p1;
 250
 251         if (strlen(path) < 9)
 252                 return NULL;
 253         p1 = strstr(path+8, "/");
 254         if (!p1)
 255                 return NULL;
 256         return p1+1;
 257 }
 258
 259 static bool is_child_cgroup(const char *contr, const char *dir, const char *f)
 260 {
 261         nih_local char **list = NULL;
 262         int i;
 263
 264         if (!f)
 265                 return false;
 266         if (*f == '/')
 267                 f++;
 268
 269         if (!cgm_list_children(contr, dir, &list))
 270                 return false;
 271         for (i = 0; list[i]; i++) {
 272                 if (strcmp(list[i], f) == 0)
 273                         return true;
 274         }
 275
 276         return false;
 277 }
 278
 279 static struct cgm_keys *get_cgroup_key(const char *contr, const char *dir, const char *f)
 280 {
 281         nih_local struct cgm_keys **list = NULL;
 282         struct cgm_keys *k;
 283         int i;
 284
 285         if (!f)
 286                 return NULL;
 287         if (*f == '/')
 288                 f++;
 289         if (!cgm_list_keys(contr, dir, &list))
 290                 return NULL;
 291         for (i = 0; list[i]; i++) {
 292                 if (strcmp(list[i]->name, f) == 0) {
 293                         k = NIH_MUST( nih_alloc(NULL, (sizeof(*k))) );
 294                         k->name = NIH_MUST( nih_strdup(k, list[i]->name) );
 295                         k->uid = list[i]->uid;
 296                         k->gid = list[i]->gid;
 297                         k->mode = list[i]->mode;
 298                         return k;
 299                 }
 300         }
 301
 302         return NULL;
 303 }
 304
 305 static void get_cgdir_and_path(const char *cg, char **dir, char **file)
 306 {
 307         char *p;
 308
 309         *dir = NIH_MUST( nih_strdup(NULL, cg) );
 310         *file = strrchr(cg, '/');
 311         if (!*file) {
 312                 *file = NULL;
 313                 return;
 314         }
 315         p = strrchr(*dir, '/');
 316         *p = '\0';
 317 }
 318
 319 static size_t get_file_size(const char *contrl, const char *cg, const char *f)
 320 {
 321         nih_local char *data = NULL;
 322         size_t s;
 323         if (!cgm_get_value(contrl, cg, f, &data))
 324                 return -EINVAL;
 325         s = strlen(data);
 326         return s;
 327 }
 328
 329 /*
 330  * FUSE ops for /cgroup
 331  */
 332
 333 static int cg_getattr(const char *path, struct stat *sb)
 334 {
 335         struct timespec now;
 336         struct fuse_context *fc = fuse_get_context();
 337         nih_local char * cgdir = NULL;
 338         char *fpath = NULL, *path1, *path2;
 339         nih_local struct cgm_keys *k = NULL;
 340         const char *cgroup;
 341         nih_local char *controller = NULL;
 342
 343
 344         if (!fc)
 345                 return -EIO;
 346
 347         memset(sb, 0, sizeof(struct stat));
 348
 349         if (clock_gettime(CLOCK_REALTIME, &now) < 0)
 350                 return -EINVAL;
 351
 352         sb->st_uid = sb->st_gid = 0;
 353         sb->st_atim = sb->st_mtim = sb->st_ctim = now;
 354         sb->st_size = 0;
 355
 356         if (strcmp(path, "/cgroup") == 0) {
 357                 sb->st_mode = S_IFDIR | 00755;
 358                 sb->st_nlink = 2;
 359                 return 0;
 360         }
 361
 362         controller = pick_controller_from_path(fc, path);
 363         if (!controller)
 364                 return -EIO;
 365         cgroup = find_cgroup_in_path(path);
 366         if (!cgroup) {
 367                 /* this is just /cgroup/controller, return it as a dir */
 368                 sb->st_mode = S_IFDIR | 00755;
 369                 sb->st_nlink = 2;
 370                 return 0;
 371         }
 372
 373         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 374
 375         if (!fpath) {
 376                 path1 = "/";
 377                 path2 = cgdir;
 378         } else {
 379                 path1 = cgdir;
 380                 path2 = fpath;
 381         }
 382
 383         /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
 384          * Then check that caller's cgroup is under path if fpath is a child
 385          * cgroup, or cgdir if fpath is a file */
 386
 387         if (is_child_cgroup(controller, path1, path2)) {
 388                 if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY))
 389                         return -EPERM;
 390
 391                 // get uid, gid, from '/tasks' file and make up a mode
 392                 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
 393                 sb->st_mode = S_IFDIR | 00755;
 394                 k = get_cgroup_key(controller, cgroup, "tasks");
 395                 if (!k) {
 396                         fprintf(stderr, "Failed to find a tasks file for %s\n", cgroup);
 397                         sb->st_uid = sb->st_gid = 0;
 398                 } else {
 399                         fprintf(stderr, "found a tasks file for %s\n", cgroup);
 400                         sb->st_uid = k->uid;
 401                         sb->st_gid = k->gid;
 402                 }
 403                 sb->st_nlink = 2;
 404                 return 0;
 405         }
 406
 407         if ((k = get_cgroup_key(controller, path1, path2)) != NULL) {
 408                 if (!fc_may_access(fc, controller, path1, path2, O_RDONLY))
 409                         return -EPERM;
 410
 411                 sb->st_mode = S_IFREG | k->mode;
 412                 sb->st_nlink = 1;
 413                 sb->st_uid = k->uid;
 414                 sb->st_gid = k->gid;
 415                 sb->st_size = get_file_size(controller, path1, path2);
 416                 return 0;
 417         }
 418
 419         return -ENOENT;
 420 }
 421
 422 static int cg_opendir(const char *path, struct fuse_file_info *fi)
 423 {
 424         return 0;
 425 }
 426
 427 static int cg_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset,
 428                 struct fuse_file_info *fi)
 429 {
 430         struct fuse_context *fc = fuse_get_context();
 431
 432         if (!fc)
 433                 return -EIO;
 434
 435         if (strcmp(path, "/cgroup") == 0) {
 436                 // get list of controllers
 437                 char **list = LXCFS_DATA ? LXCFS_DATA->subsystems : NULL;
 438                 int i;
 439
 440                 if (!list)
 441                         return -EIO;
 442                 /* TODO - collect the list of controllers at fuse_init */
 443                 for (i = 0;  list[i]; i++) {
 444                         if (filler(buf, list[i], NULL, 0) != 0) {
 445                                 return -EIO;
 446                         }
 447                 }
 448                 return 0;
 449         }
 450
 451         // return list of keys for the controller, and list of child cgroups
 452         nih_local struct cgm_keys **list = NULL;
 453         const char *cgroup;
 454         nih_local char *controller = NULL;
 455         int i;
 456
 457         controller = pick_controller_from_path(fc, path);
 458         if (!controller)
 459                 return -EIO;
 460
 461         cgroup = find_cgroup_in_path(path);
 462         if (!cgroup) {
 463                 /* this is just /cgroup/controller, return its contents */
 464                 cgroup = "/";
 465         }
 466
 467         if (!fc_may_access(fc, controller, cgroup, NULL, O_RDONLY))
 468                 return -EPERM;
 469
 470         if (!cgm_list_keys(controller, cgroup, &list))
 471                 return -EINVAL;
 472         for (i = 0; list[i]; i++) {
 473                 if (filler(buf, list[i]->name, NULL, 0) != 0) {
 474                         return -EIO;
 475                 }
 476         }
 477
 478         // now get the list of child cgroups
 479         nih_local char **clist;
 480
 481         if (!cgm_list_children(controller, cgroup, &clist))
 482                 return 0;
 483         for (i = 0; clist[i]; i++) {
 484                 if (filler(buf, clist[i], NULL, 0) != 0) {
 485                         return -EIO;
 486                 }
 487         }
 488         return 0;
 489 }
 490
 491 static int cg_releasedir(const char *path, struct fuse_file_info *fi)
 492 {
 493         return 0;
 494 }
 495
 496 static int cg_open(const char *path, struct fuse_file_info *fi)
 497 {
 498         nih_local char *controller = NULL;
 499         const char *cgroup;
 500         char *fpath = NULL, *path1, *path2;
 501         nih_local char * cgdir = NULL;
 502         nih_local struct cgm_keys *k = NULL;
 503         struct fuse_context *fc = fuse_get_context();
 504
 505         if (!fc)
 506                 return -EIO;
 507
 508         controller = pick_controller_from_path(fc, path);
 509         if (!controller)
 510                 return -EIO;
 511         cgroup = find_cgroup_in_path(path);
 512         if (!cgroup)
 513                 return -EINVAL;
 514
 515         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 516         if (!fpath) {
 517                 path1 = "/";
 518                 path2 = cgdir;
 519         } else {
 520                 path1 = cgdir;
 521                 path2 = fpath;
 522         }
 523
 524         if ((k = get_cgroup_key(controller, path1, path2)) != NULL) {
 525                 if (!fc_may_access(fc, controller, path1, path2, fi->flags))
 526                         return -EPERM;
 527
 528                 /* TODO - we want to cache this info for read/write */
 529                 return 0;
 530         }
 531
 532         return -EINVAL;
 533 }
 534
 535 static int cg_read(const char *path, char *buf, size_t size, off_t offset,
 536                 struct fuse_file_info *fi)
 537 {
 538         nih_local char *controller = NULL;
 539         const char *cgroup;
 540         char *fpath = NULL, *path1, *path2;
 541         struct fuse_context *fc = fuse_get_context();
 542         nih_local char * cgdir = NULL;
 543         nih_local struct cgm_keys *k = NULL;
 544
 545         if (offset)
 546                 return -EIO;
 547
 548         if (!fc)
 549                 return -EIO;
 550
 551         controller = pick_controller_from_path(fc, path);
 552         if (!controller)
 553                 return -EIO;
 554         cgroup = find_cgroup_in_path(path);
 555         if (!cgroup)
 556                 return -EINVAL;
 557
 558         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 559         if (!fpath) {
 560                 path1 = "/";
 561                 path2 = cgdir;
 562         } else {
 563                 path1 = cgdir;
 564                 path2 = fpath;
 565         }
 566
 567         if ((k = get_cgroup_key(controller, path1, path2)) != NULL) {
 568                 nih_local char *data = NULL;
 569                 int s;
 570
 571                 if (!fc_may_access(fc, controller, path1, path2, O_RDONLY))
 572                         return -EPERM;
 573
 574                 if (!cgm_get_value(controller, path1, path2, &data))
 575                         return -EINVAL;
 576
 577                 s = strlen(data);
 578                 if (s > size)
 579                         s = size;
 580                 memcpy(buf, data, s);
 581
 582                 return s;
 583         }
 584
 585         return -EINVAL;
 586 }
 587
 588 int cg_write(const char *path, const char *buf, size_t size, off_t offset,
 589              struct fuse_file_info *fi)
 590 {
 591         nih_local char *controller = NULL;
 592         const char *cgroup;
 593         char *fpath = NULL, *path1, *path2;
 594         struct fuse_context *fc = fuse_get_context();
 595         nih_local char * cgdir = NULL;
 596         nih_local struct cgm_keys *k = NULL;
 597
 598 fprintf(stderr, "cg_write: starting\n");
 599
 600         if (offset)
 601                 return -EIO;
 602
 603         if (!fc)
 604                 return -EIO;
 605
 606         controller = pick_controller_from_path(fc, path);
 607         if (!controller)
 608                 return -EIO;
 609         cgroup = find_cgroup_in_path(path);
 610         if (!cgroup)
 611                 return -EINVAL;
 612
 613         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 614         if (!fpath) {
 615                 path1 = "/";
 616                 path2 = cgdir;
 617         } else {
 618                 path1 = cgdir;
 619                 path2 = fpath;
 620         }
 621
 622         if ((k = get_cgroup_key(controller, path1, path2)) != NULL) {
 623                 if (!fc_may_access(fc, controller, path1, path2, O_WRONLY))
 624                         return -EPERM;
 625
 626                 if (!cgm_set_value(controller, path1, path2, buf))
 627                         return -EINVAL;
 628
 629                 return size;
 630         }
 631
 632         return -EINVAL;
 633 }
 634
 635 int cg_chown(const char *path, uid_t uid, gid_t gid)
 636 {
 637         struct fuse_context *fc = fuse_get_context();
 638         nih_local char * cgdir = NULL;
 639         char *fpath = NULL, *path1, *path2;
 640         nih_local struct cgm_keys *k = NULL;
 641         const char *cgroup;
 642         nih_local char *controller = NULL;
 643
 644
 645         if (!fc)
 646                 return -EIO;
 647
 648         if (strcmp(path, "/cgroup") == 0)
 649                 return -EINVAL;
 650
 651         controller = pick_controller_from_path(fc, path);
 652         if (!controller)
 653                 return -EIO;
 654         cgroup = find_cgroup_in_path(path);
 655         if (!cgroup)
 656                 /* this is just /cgroup/controller */
 657                 return -EINVAL;
 658
 659         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 660
 661         if (!fpath) {
 662                 path1 = "/";
 663                 path2 = cgdir;
 664         } else {
 665                 path1 = cgdir;
 666                 path2 = fpath;
 667         }
 668
 669         if (is_child_cgroup(controller, path1, path2)) {
 670                 // get uid, gid, from '/tasks' file and make up a mode
 671                 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
 672                 k = get_cgroup_key(controller, cgroup, "tasks");
 673
 674         } else
 675                 k = get_cgroup_key(controller, path1, path2);
 676
 677         if (!k)
 678                 return -EINVAL;
 679
 680         /*
 681          * This being a fuse request, the uid and gid must be valid
 682          * in the caller's namespace.  So we can just check to make
 683          * sure that the caller is root in his uid, and privileged
 684          * over the file's current owner.
 685          */
 686         if (!is_privileged_over(fc->pid, fc->uid, k->uid, NS_ROOT_REQD))
 687                 return -EPERM;
 688
 689         if (!cgm_chown_file(controller, cgroup, uid, gid))
 690                 return -EINVAL;
 691         return 0;
 692 }
 693
 694 int cg_mkdir(const char *path, mode_t mode)
 695 {
 696         struct fuse_context *fc = fuse_get_context();
 697         nih_local struct cgm_keys **list = NULL;
 698         char *fpath = NULL, *path1;
 699         nih_local char * cgdir = NULL;
 700         const char *cgroup;
 701         nih_local char *controller = NULL;
 702
 703         if (!fc)
 704                 return -EIO;
 705
 706
 707         controller = pick_controller_from_path(fc, path);
 708         if (!controller)
 709                 return -EIO;
 710
 711         cgroup = find_cgroup_in_path(path);
 712         if (!cgroup)
 713                 return -EIO;
 714
 715         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 716         if (!fpath)
 717                 path1 = "/";
 718         else
 719                 path1 = cgdir;
 720
 721         if (!fc_may_access(fc, controller, path1, NULL, O_RDWR))
 722                 return -EPERM;
 723
 724
 725         if (!cgm_create(controller, cgroup, fc->uid, fc->gid))
 726                 return -EINVAL;
 727
 728         return 0;
 729 }
 730
 731 static int cg_rmdir(const char *path)
 732 {
 733         struct fuse_context *fc = fuse_get_context();
 734         nih_local struct cgm_keys **list = NULL;
 735         char *fpath = NULL;
 736         nih_local char * cgdir = NULL;
 737         const char *cgroup;
 738         nih_local char *controller = NULL;
 739
 740         if (!fc)
 741                 return -EIO;
 742
 743
 744         controller = pick_controller_from_path(fc, path);
 745         if (!controller)
 746                 return -EIO;
 747
 748         cgroup = find_cgroup_in_path(path);
 749         if (!cgroup)
 750                 return -EIO;
 751
 752         get_cgdir_and_path(cgroup, &cgdir, &fpath);
 753         if (!fpath)
 754                 return -EINVAL;
 755
 756         if (!fc_may_access(fc, controller, cgdir, NULL, O_WRONLY))
 757                 return -EPERM;
 758
 759         if (!cgm_remove(controller, cgroup))
 760                 return -EINVAL;
 761
 762         return 0;
 763 }
 764
 765 /*
 766  * FUSE ops for /proc
 767  */
 768
 769 static int proc_getattr(const char *path, struct stat *sb)
 770 {
 771         if (strcmp(path, "/proc") != 0)
 772                 return -EINVAL;
 773         sb->st_mode = S_IFDIR | 00755;
 774         sb->st_nlink = 2;
 775         return 0;
 776 }
 777
 778 /*
 779  * FUSE ops for /
 780  * these just delegate to the /proc and /cgroup ops as
 781  * needed
 782  */
 783
 784 static int lxcfs_getattr(const char *path, struct stat *sb)
 785 {
 786         if (strcmp(path, "/") == 0) {
 787                 sb->st_mode = S_IFDIR | 00755;
 788                 sb->st_nlink = 2;
 789                 return 0;
 790         }
 791         if (strncmp(path, "/cgroup", 7) == 0) {
 792                 return cg_getattr(path, sb);
 793         }
 794         if (strncmp(path, "/proc", 7) == 0) {
 795                 return proc_getattr(path, sb);
 796         }
 797         return -EINVAL;
 798 }
 799
 800 static int lxcfs_opendir(const char *path, struct fuse_file_info *fi)
 801 {
 802         if (strcmp(path, "/") == 0)
 803                 return 0;
 804
 805         if (strncmp(path, "/cgroup", 7) == 0) {
 806                 return cg_opendir(path, fi);
 807         }
 808         return -EINVAL;
 809 }
 810
 811 static int lxcfs_readdir(const char *path, void *buf, fuse_fill_dir_t filler, off_t offset,
 812                 struct fuse_file_info *fi)
 813 {
 814         if (strcmp(path, "/") == 0) {
 815                 if (filler(buf, "proc", NULL, 0) != 0 ||
 816                                 filler(buf, "cgroup", NULL, 0) != 0)
 817                         return -EINVAL;
 818                 return 0;
 819         }
 820         if (strncmp(path, "/cgroup", 7) == 0) {
 821                 return cg_readdir(path, buf, filler, offset, fi);
 822         }
 823         return -EINVAL;
 824 }
 825
 826 static int lxcfs_releasedir(const char *path, struct fuse_file_info *fi)
 827 {
 828         if (strcmp(path, "/") == 0)
 829                 return 0;
 830         if (strncmp(path, "/cgroup", 7) == 0) {
 831                 return cg_releasedir(path, fi);
 832         }
 833         return -EINVAL;
 834 }
 835
 836 static int lxcfs_open(const char *path, struct fuse_file_info *fi)
 837 {
 838         if (strncmp(path, "/cgroup", 7) == 0) {
 839                 return cg_open(path, fi);
 840         }
 841
 842         return -EINVAL;
 843 }
 844
 845 static int lxcfs_read(const char *path, char *buf, size_t size, off_t offset,
 846                 struct fuse_file_info *fi)
 847 {
 848         if (strncmp(path, "/cgroup", 7) == 0) {
 849                 return cg_read(path, buf, size, offset, fi);
 850         }
 851
 852         return -EINVAL;
 853 }
 854
 855 int lxcfs_write(const char *path, const char *buf, size_t size, off_t offset,
 856              struct fuse_file_info *fi)
 857 {
 858         if (strncmp(path, "/cgroup", 7) == 0) {
 859                 return cg_write(path, buf, size, offset, fi);
 860         }
 861
 862         return -EINVAL;
 863 }
 864
 865 static int lxcfs_flush(const char *path, struct fuse_file_info *fi)
 866 {
 867         return 0;
 868 }
 869
 870 static int lxcfs_release(const char *path, struct fuse_file_info *fi)
 871 {
 872         return 0;
 873 }
 874
 875 static int lxcfs_fsync(const char *path, int datasync, struct fuse_file_info *fi)
 876 {
 877         return 0;
 878 }
 879
 880 int lxcfs_mkdir(const char *path, mode_t mode)
 881 {
 882         if (strncmp(path, "/cgroup", 7) == 0)
 883                 return cg_mkdir(path, mode);
 884
 885         return -EINVAL;
 886 }
 887
 888 int lxcfs_chown(const char *path, uid_t uid, gid_t gid)
 889 {
 890         if (strncmp(path, "/cgroup", 7) == 0)
 891                 return cg_chown(path, uid, gid);
 892
 893         return -EINVAL;
 894 }
 895
 896 /*
 897  * cat first does a truncate before doing ops->write.  This doesn't
 898  * really make sense for cgroups.  So just return 0 always but do
 899  * nothing.
 900  */
 901 int lxcfs_truncate(const char *path, off_t newsize)
 902 {
 903         if (strncmp(path, "/cgroup", 7) == 0)
 904                 return 0;
 905         return -EINVAL;
 906 }
 907
 908 int lxcfs_rmdir(const char *path)
 909 {
 910         if (strncmp(path, "/cgroup", 7) == 0)
 911                 return cg_rmdir(path);
 912         return -EINVAL;
 913 }
 914
 915 const struct fuse_operations lxcfs_ops = {
 916         .getattr = lxcfs_getattr,
 917         .readlink = NULL,
 918         .getdir = NULL,
 919         .mknod = NULL,
 920         .mkdir = lxcfs_mkdir,
 921         .unlink = NULL,
 922         .rmdir = lxcfs_rmdir,
 923         .symlink = NULL,
 924         .rename = NULL,
 925         .link = NULL,
 926         .chmod = NULL,
 927         .chown = lxcfs_chown,
 928         .truncate = lxcfs_truncate,
 929         .utime = NULL,
 930
 931         .open = lxcfs_open,
 932         .read = lxcfs_read,
 933         .release = lxcfs_release,
 934         .write = lxcfs_write,
 935
 936         .statfs = NULL,
 937         .flush = lxcfs_flush,
 938         .fsync = lxcfs_fsync,
 939
 940         .setxattr = NULL,
 941         .getxattr = NULL,
 942         .listxattr = NULL,
 943         .removexattr = NULL,
 944
 945         .opendir = lxcfs_opendir,
 946         .readdir = lxcfs_readdir,
 947         .releasedir = lxcfs_releasedir,
 948
 949         .fsyncdir = NULL,
 950         .init = NULL,
 951         .destroy = NULL,
 952         .access = NULL,
 953         .create = NULL,
 954         .ftruncate = NULL,
 955         .fgetattr = NULL,
 956 };
 957
 958 static void usage(const char *me)
 959 {
 960         fprintf(stderr, "Usage:\n");
 961         fprintf(stderr, "\n");
 962         fprintf(stderr, "%s [FUSE and mount options] mountpoint\n", me);
 963         exit(1);
 964 }
 965
 966 static bool is_help(char *w)
 967 {
 968         if (strcmp(w, "-h") == 0 ||
 969                         strcmp(w, "--help") == 0 ||
 970                         strcmp(w, "-help") == 0 ||
 971                         strcmp(w, "help") == 0)
 972                 return true;
 973         return false;
 974 }
 975
 976 int main(int argc, char *argv[])
 977 {
 978         int ret;
 979         struct lxcfs_state *d;
 980
 981         if (argc < 2 || is_help(argv[1]))
 982                 usage(argv[0]);
 983
 984         d = malloc(sizeof(*d));
 985         if (!d)
 986                 return -1;
 987
 988         if (!cgm_escape_cgroup())
 989                 fprintf(stderr, "WARNING: failed to escape to root cgroup\n");
 990
 991         if (!cgm_get_controllers(&d->subsystems))
 992                 return -1;
 993
 994         ret = fuse_main(argc, argv, &lxcfs_ops, d);
 995
 996         return ret;
 997 }