fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fanotify.h>
   2 #include <linux/fcntl.h>
   3 #include <linux/file.h>
   4 #include <linux/fs.h>
   5 #include <linux/anon_inodes.h>
   6 #include <linux/fsnotify_backend.h>
   7 #include <linux/init.h>
   8 #include <linux/mount.h>
   9 #include <linux/namei.h>
  10 #include <linux/poll.h>
  11 #include <linux/security.h>
  12 #include <linux/syscalls.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/compat.h>
  17
  18 #include <asm/ioctls.h>
  19
  20 #include "../../mount.h"
  21 #include "../fdinfo.h"
  22 #include "fanotify.h"
  23
  24 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  25 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  26 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  27
  28 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  29
  30 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  31 static struct kmem_cache *fanotify_response_event_cache __read_mostly;
  32 struct kmem_cache *fanotify_event_cachep __read_mostly;
  33
  34 struct fanotify_response_event {
  35         struct list_head list;
  36         __s32 fd;
  37         struct fanotify_event_info *event;
  38 };
  39
  40 /*
  41  * Get an fsnotify notification event if one exists and is small
  42  * enough to fit in "count". Return an error pointer if the count
  43  * is not large enough.
  44  *
  45  * Called with the group->notification_mutex held.
  46  */
  47 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  48                                             size_t count)
  49 {
  50         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  51
  52         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  53
  54         if (fsnotify_notify_queue_is_empty(group))
  55                 return NULL;
  56
  57         if (FAN_EVENT_METADATA_LEN > count)
  58                 return ERR_PTR(-EINVAL);
  59
  60         /* held the notification_mutex the whole time, so this is the
  61          * same event we peeked above */
  62         return fsnotify_remove_notify_event(group);
  63 }
  64
  65 static int create_fd(struct fsnotify_group *group,
  66                      struct fanotify_event_info *event,
  67                      struct file **file)
  68 {
  69         int client_fd;
  70         struct file *new_file;
  71
  72         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  73
  74         client_fd = get_unused_fd();
  75         if (client_fd < 0)
  76                 return client_fd;
  77
  78         /*
  79          * we need a new file handle for the userspace program so it can read even if it was
  80          * originally opened O_WRONLY.
  81          */
  82         /* it's possible this event was an overflow event.  in that case dentry and mnt
  83          * are NULL;  That's fine, just don't call dentry open */
  84         if (event->path.dentry && event->path.mnt)
  85                 new_file = dentry_open(&event->path,
  86                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  87                                        current_cred());
  88         else
  89                 new_file = ERR_PTR(-EOVERFLOW);
  90         if (IS_ERR(new_file)) {
  91                 /*
  92                  * we still send an event even if we can't open the file.  this
  93                  * can happen when say tasks are gone and we try to open their
  94                  * /proc files or we try to open a WRONLY file like in sysfs
  95                  * we just send the errno to userspace since there isn't much
  96                  * else we can do.
  97                  */
  98                 put_unused_fd(client_fd);
  99                 client_fd = PTR_ERR(new_file);
 100         } else {
 101                 *file = new_file;
 102         }
 103
 104         return client_fd;
 105 }
 106
 107 static int fill_event_metadata(struct fsnotify_group *group,
 108                                struct fanotify_event_metadata *metadata,
 109                                struct fsnotify_event *fsn_event,
 110                                struct file **file)
 111 {
 112         int ret = 0;
 113         struct fanotify_event_info *event;
 114
 115         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 116                  group, metadata, fsn_event);
 117
 118         *file = NULL;
 119         event = container_of(fsn_event, struct fanotify_event_info, fse);
 120         metadata->event_len = FAN_EVENT_METADATA_LEN;
 121         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 122         metadata->vers = FANOTIFY_METADATA_VERSION;
 123         metadata->reserved = 0;
 124         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 125         metadata->pid = pid_vnr(event->tgid);
 126         if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
 127                 metadata->fd = FAN_NOFD;
 128         else {
 129                 metadata->fd = create_fd(group, event, file);
 130                 if (metadata->fd < 0)
 131                         ret = metadata->fd;
 132         }
 133
 134         return ret;
 135 }
 136
 137 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 138 static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
 139                                                   __s32 fd)
 140 {
 141         struct fanotify_response_event *re, *return_re = NULL;
 142
 143         mutex_lock(&group->fanotify_data.access_mutex);
 144         list_for_each_entry(re, &group->fanotify_data.access_list, list) {
 145                 if (re->fd != fd)
 146                         continue;
 147
 148                 list_del_init(&re->list);
 149                 return_re = re;
 150                 break;
 151         }
 152         mutex_unlock(&group->fanotify_data.access_mutex);
 153
 154         pr_debug("%s: found return_re=%p\n", __func__, return_re);
 155
 156         return return_re;
 157 }
 158
 159 static int process_access_response(struct fsnotify_group *group,
 160                                    struct fanotify_response *response_struct)
 161 {
 162         struct fanotify_response_event *re;
 163         __s32 fd = response_struct->fd;
 164         __u32 response = response_struct->response;
 165
 166         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 167                  fd, response);
 168         /*
 169          * make sure the response is valid, if invalid we do nothing and either
 170          * userspace can send a valid response or we will clean it up after the
 171          * timeout
 172          */
 173         switch (response) {
 174         case FAN_ALLOW:
 175         case FAN_DENY:
 176                 break;
 177         default:
 178                 return -EINVAL;
 179         }
 180
 181         if (fd < 0)
 182                 return -EINVAL;
 183
 184         re = dequeue_re(group, fd);
 185         if (!re)
 186                 return -ENOENT;
 187
 188         re->event->response = response;
 189
 190         wake_up(&group->fanotify_data.access_waitq);
 191
 192         kmem_cache_free(fanotify_response_event_cache, re);
 193
 194         return 0;
 195 }
 196
 197 static int prepare_for_access_response(struct fsnotify_group *group,
 198                                        struct fsnotify_event *event,
 199                                        __s32 fd)
 200 {
 201         struct fanotify_response_event *re;
 202
 203         if (!(event->mask & FAN_ALL_PERM_EVENTS))
 204                 return 0;
 205
 206         re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
 207         if (!re)
 208                 return -ENOMEM;
 209
 210         re->event = FANOTIFY_E(event);
 211         re->fd = fd;
 212
 213         mutex_lock(&group->fanotify_data.access_mutex);
 214
 215         if (atomic_read(&group->fanotify_data.bypass_perm)) {
 216                 mutex_unlock(&group->fanotify_data.access_mutex);
 217                 kmem_cache_free(fanotify_response_event_cache, re);
 218                 FANOTIFY_E(event)->response = FAN_ALLOW;
 219                 return 0;
 220         }
 221
 222         list_add_tail(&re->list, &group->fanotify_data.access_list);
 223         mutex_unlock(&group->fanotify_data.access_mutex);
 224
 225         return 0;
 226 }
 227
 228 #else
 229 static int prepare_for_access_response(struct fsnotify_group *group,
 230                                        struct fsnotify_event *event,
 231                                        __s32 fd)
 232 {
 233         return 0;
 234 }
 235
 236 #endif
 237
 238 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 239                                   struct fsnotify_event *event,
 240                                   char __user *buf)
 241 {
 242         struct fanotify_event_metadata fanotify_event_metadata;
 243         struct file *f;
 244         int fd, ret;
 245
 246         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 247
 248         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 249         if (ret < 0)
 250                 goto out;
 251
 252         fd = fanotify_event_metadata.fd;
 253         ret = -EFAULT;
 254         if (copy_to_user(buf, &fanotify_event_metadata,
 255                          fanotify_event_metadata.event_len))
 256                 goto out_close_fd;
 257
 258         ret = prepare_for_access_response(group, event, fd);
 259         if (ret)
 260                 goto out_close_fd;
 261
 262         if (fd != FAN_NOFD)
 263                 fd_install(fd, f);
 264         return fanotify_event_metadata.event_len;
 265
 266 out_close_fd:
 267         if (fd != FAN_NOFD) {
 268                 put_unused_fd(fd);
 269                 fput(f);
 270         }
 271 out:
 272 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 273         if (event->mask & FAN_ALL_PERM_EVENTS) {
 274                 FANOTIFY_E(event)->response = FAN_DENY;
 275                 wake_up(&group->fanotify_data.access_waitq);
 276         }
 277 #endif
 278         return ret;
 279 }
 280
 281 /* intofiy userspace file descriptor functions */
 282 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 283 {
 284         struct fsnotify_group *group = file->private_data;
 285         int ret = 0;
 286
 287         poll_wait(file, &group->notification_waitq, wait);
 288         mutex_lock(&group->notification_mutex);
 289         if (!fsnotify_notify_queue_is_empty(group))
 290                 ret = POLLIN | POLLRDNORM;
 291         mutex_unlock(&group->notification_mutex);
 292
 293         return ret;
 294 }
 295
 296 static ssize_t fanotify_read(struct file *file, char __user *buf,
 297                              size_t count, loff_t *pos)
 298 {
 299         struct fsnotify_group *group;
 300         struct fsnotify_event *kevent;
 301         char __user *start;
 302         int ret;
 303         DEFINE_WAIT(wait);
 304
 305         start = buf;
 306         group = file->private_data;
 307
 308         pr_debug("%s: group=%p\n", __func__, group);
 309
 310         while (1) {
 311                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 312
 313                 mutex_lock(&group->notification_mutex);
 314                 kevent = get_one_event(group, count);
 315                 mutex_unlock(&group->notification_mutex);
 316
 317                 if (kevent) {
 318                         ret = PTR_ERR(kevent);
 319                         if (IS_ERR(kevent))
 320                                 break;
 321                         ret = copy_event_to_user(group, kevent, buf);
 322                         fsnotify_destroy_event(group, kevent);
 323                         if (ret < 0)
 324                                 break;
 325                         buf += ret;
 326                         count -= ret;
 327                         continue;
 328                 }
 329
 330                 ret = -EAGAIN;
 331                 if (file->f_flags & O_NONBLOCK)
 332                         break;
 333                 ret = -ERESTARTSYS;
 334                 if (signal_pending(current))
 335                         break;
 336
 337                 if (start != buf)
 338                         break;
 339
 340                 schedule();
 341         }
 342
 343         finish_wait(&group->notification_waitq, &wait);
 344         if (start != buf && ret != -EFAULT)
 345                 ret = buf - start;
 346         return ret;
 347 }
 348
 349 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 350 {
 351 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 352         struct fanotify_response response = { .fd = -1, .response = -1 };
 353         struct fsnotify_group *group;
 354         int ret;
 355
 356         group = file->private_data;
 357
 358         if (count > sizeof(response))
 359                 count = sizeof(response);
 360
 361         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 362
 363         if (copy_from_user(&response, buf, count))
 364                 return -EFAULT;
 365
 366         ret = process_access_response(group, &response);
 367         if (ret < 0)
 368                 count = ret;
 369
 370         return count;
 371 #else
 372         return -EINVAL;
 373 #endif
 374 }
 375
 376 static int fanotify_release(struct inode *ignored, struct file *file)
 377 {
 378         struct fsnotify_group *group = file->private_data;
 379
 380 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 381         struct fanotify_response_event *re, *lre;
 382
 383         mutex_lock(&group->fanotify_data.access_mutex);
 384
 385         atomic_inc(&group->fanotify_data.bypass_perm);
 386
 387         list_for_each_entry_safe(re, lre, &group->fanotify_data.access_list, list) {
 388                 pr_debug("%s: found group=%p re=%p event=%p\n", __func__, group,
 389                          re, re->event);
 390
 391                 list_del_init(&re->list);
 392                 re->event->response = FAN_ALLOW;
 393
 394                 kmem_cache_free(fanotify_response_event_cache, re);
 395         }
 396         mutex_unlock(&group->fanotify_data.access_mutex);
 397
 398         wake_up(&group->fanotify_data.access_waitq);
 399 #endif
 400
 401         /* matches the fanotify_init->fsnotify_alloc_group */
 402         fsnotify_destroy_group(group);
 403
 404         return 0;
 405 }
 406
 407 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 408 {
 409         struct fsnotify_group *group;
 410         struct fsnotify_event *fsn_event;
 411         void __user *p;
 412         int ret = -ENOTTY;
 413         size_t send_len = 0;
 414
 415         group = file->private_data;
 416
 417         p = (void __user *) arg;
 418
 419         switch (cmd) {
 420         case FIONREAD:
 421                 mutex_lock(&group->notification_mutex);
 422                 list_for_each_entry(fsn_event, &group->notification_list, list)
 423                         send_len += FAN_EVENT_METADATA_LEN;
 424                 mutex_unlock(&group->notification_mutex);
 425                 ret = put_user(send_len, (int __user *) p);
 426                 break;
 427         }
 428
 429         return ret;
 430 }
 431
 432 static const struct file_operations fanotify_fops = {
 433         .show_fdinfo    = fanotify_show_fdinfo,
 434         .poll           = fanotify_poll,
 435         .read           = fanotify_read,
 436         .write          = fanotify_write,
 437         .fasync         = NULL,
 438         .release        = fanotify_release,
 439         .unlocked_ioctl = fanotify_ioctl,
 440         .compat_ioctl   = fanotify_ioctl,
 441         .llseek         = noop_llseek,
 442 };
 443
 444 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 445 {
 446         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 447 }
 448
 449 static int fanotify_find_path(int dfd, const char __user *filename,
 450                               struct path *path, unsigned int flags)
 451 {
 452         int ret;
 453
 454         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 455                  dfd, filename, flags);
 456
 457         if (filename == NULL) {
 458                 struct fd f = fdget(dfd);
 459
 460                 ret = -EBADF;
 461                 if (!f.file)
 462                         goto out;
 463
 464                 ret = -ENOTDIR;
 465                 if ((flags & FAN_MARK_ONLYDIR) &&
 466                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 467                         fdput(f);
 468                         goto out;
 469                 }
 470
 471                 *path = f.file->f_path;
 472                 path_get(path);
 473                 fdput(f);
 474         } else {
 475                 unsigned int lookup_flags = 0;
 476
 477                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 478                         lookup_flags |= LOOKUP_FOLLOW;
 479                 if (flags & FAN_MARK_ONLYDIR)
 480                         lookup_flags |= LOOKUP_DIRECTORY;
 481
 482                 ret = user_path_at(dfd, filename, lookup_flags, path);
 483                 if (ret)
 484                         goto out;
 485         }
 486
 487         /* you can only watch an inode if you have read permissions on it */
 488         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 489         if (ret)
 490                 path_put(path);
 491 out:
 492         return ret;
 493 }
 494
 495 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 496                                             __u32 mask,
 497                                             unsigned int flags,
 498                                             int *destroy)
 499 {
 500         __u32 oldmask;
 501
 502         spin_lock(&fsn_mark->lock);
 503         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 504                 oldmask = fsn_mark->mask;
 505                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
 506         } else {
 507                 oldmask = fsn_mark->ignored_mask;
 508                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
 509         }
 510         spin_unlock(&fsn_mark->lock);
 511
 512         *destroy = !(oldmask & ~mask);
 513
 514         return mask & oldmask;
 515 }
 516
 517 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 518                                          struct vfsmount *mnt, __u32 mask,
 519                                          unsigned int flags)
 520 {
 521         struct fsnotify_mark *fsn_mark = NULL;
 522         __u32 removed;
 523         int destroy_mark;
 524
 525         mutex_lock(&group->mark_mutex);
 526         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 527         if (!fsn_mark) {
 528                 mutex_unlock(&group->mark_mutex);
 529                 return -ENOENT;
 530         }
 531
 532         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 533                                                  &destroy_mark);
 534         if (destroy_mark)
 535                 fsnotify_destroy_mark_locked(fsn_mark, group);
 536         mutex_unlock(&group->mark_mutex);
 537
 538         fsnotify_put_mark(fsn_mark);
 539         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 540                 fsnotify_recalc_vfsmount_mask(mnt);
 541
 542         return 0;
 543 }
 544
 545 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 546                                       struct inode *inode, __u32 mask,
 547                                       unsigned int flags)
 548 {
 549         struct fsnotify_mark *fsn_mark = NULL;
 550         __u32 removed;
 551         int destroy_mark;
 552
 553         mutex_lock(&group->mark_mutex);
 554         fsn_mark = fsnotify_find_inode_mark(group, inode);
 555         if (!fsn_mark) {
 556                 mutex_unlock(&group->mark_mutex);
 557                 return -ENOENT;
 558         }
 559
 560         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 561                                                  &destroy_mark);
 562         if (destroy_mark)
 563                 fsnotify_destroy_mark_locked(fsn_mark, group);
 564         mutex_unlock(&group->mark_mutex);
 565
 566         /* matches the fsnotify_find_inode_mark() */
 567         fsnotify_put_mark(fsn_mark);
 568         if (removed & inode->i_fsnotify_mask)
 569                 fsnotify_recalc_inode_mask(inode);
 570
 571         return 0;
 572 }
 573
 574 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 575                                        __u32 mask,
 576                                        unsigned int flags)
 577 {
 578         __u32 oldmask = -1;
 579
 580         spin_lock(&fsn_mark->lock);
 581         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 582                 oldmask = fsn_mark->mask;
 583                 fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
 584         } else {
 585                 __u32 tmask = fsn_mark->ignored_mask | mask;
 586                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 587                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 588                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 589         }
 590
 591         if (!(flags & FAN_MARK_ONDIR)) {
 592                 __u32 tmask = fsn_mark->ignored_mask | FAN_ONDIR;
 593                 fsnotify_set_mark_ignored_mask_locked(fsn_mark, tmask);
 594         }
 595
 596         spin_unlock(&fsn_mark->lock);
 597
 598         return mask & ~oldmask;
 599 }
 600
 601 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 602                                                    struct inode *inode,
 603                                                    struct vfsmount *mnt)
 604 {
 605         struct fsnotify_mark *mark;
 606         int ret;
 607
 608         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 609                 return ERR_PTR(-ENOSPC);
 610
 611         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 612         if (!mark)
 613                 return ERR_PTR(-ENOMEM);
 614
 615         fsnotify_init_mark(mark, fanotify_free_mark);
 616         ret = fsnotify_add_mark_locked(mark, group, inode, mnt, 0);
 617         if (ret) {
 618                 fsnotify_put_mark(mark);
 619                 return ERR_PTR(ret);
 620         }
 621
 622         return mark;
 623 }
 624
 625
 626 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 627                                       struct vfsmount *mnt, __u32 mask,
 628                                       unsigned int flags)
 629 {
 630         struct fsnotify_mark *fsn_mark;
 631         __u32 added;
 632
 633         mutex_lock(&group->mark_mutex);
 634         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 635         if (!fsn_mark) {
 636                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 637                 if (IS_ERR(fsn_mark)) {
 638                         mutex_unlock(&group->mark_mutex);
 639                         return PTR_ERR(fsn_mark);
 640                 }
 641         }
 642         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 643         mutex_unlock(&group->mark_mutex);
 644
 645         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 646                 fsnotify_recalc_vfsmount_mask(mnt);
 647
 648         fsnotify_put_mark(fsn_mark);
 649         return 0;
 650 }
 651
 652 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 653                                    struct inode *inode, __u32 mask,
 654                                    unsigned int flags)
 655 {
 656         struct fsnotify_mark *fsn_mark;
 657         __u32 added;
 658
 659         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 660
 661         /*
 662          * If some other task has this inode open for write we should not add
 663          * an ignored mark, unless that ignored mark is supposed to survive
 664          * modification changes anyway.
 665          */
 666         if ((flags & FAN_MARK_IGNORED_MASK) &&
 667             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 668             (atomic_read(&inode->i_writecount) > 0))
 669                 return 0;
 670
 671         mutex_lock(&group->mark_mutex);
 672         fsn_mark = fsnotify_find_inode_mark(group, inode);
 673         if (!fsn_mark) {
 674                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 675                 if (IS_ERR(fsn_mark)) {
 676                         mutex_unlock(&group->mark_mutex);
 677                         return PTR_ERR(fsn_mark);
 678                 }
 679         }
 680         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 681         mutex_unlock(&group->mark_mutex);
 682
 683         if (added & ~inode->i_fsnotify_mask)
 684                 fsnotify_recalc_inode_mask(inode);
 685
 686         fsnotify_put_mark(fsn_mark);
 687         return 0;
 688 }
 689
 690 /* fanotify syscalls */
 691 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 692 {
 693         struct fsnotify_group *group;
 694         int f_flags, fd;
 695         struct user_struct *user;
 696
 697         pr_debug("%s: flags=%d event_f_flags=%d\n",
 698                 __func__, flags, event_f_flags);
 699
 700         if (!capable(CAP_SYS_ADMIN))
 701                 return -EPERM;
 702
 703         if (flags & ~FAN_ALL_INIT_FLAGS)
 704                 return -EINVAL;
 705
 706         user = get_current_user();
 707         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 708                 free_uid(user);
 709                 return -EMFILE;
 710         }
 711
 712         f_flags = O_RDWR | FMODE_NONOTIFY;
 713         if (flags & FAN_CLOEXEC)
 714                 f_flags |= O_CLOEXEC;
 715         if (flags & FAN_NONBLOCK)
 716                 f_flags |= O_NONBLOCK;
 717
 718         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 719         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 720         if (IS_ERR(group)) {
 721                 free_uid(user);
 722                 return PTR_ERR(group);
 723         }
 724
 725         group->fanotify_data.user = user;
 726         atomic_inc(&user->fanotify_listeners);
 727
 728         group->fanotify_data.f_flags = event_f_flags;
 729 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 730         mutex_init(&group->fanotify_data.access_mutex);
 731         init_waitqueue_head(&group->fanotify_data.access_waitq);
 732         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 733         atomic_set(&group->fanotify_data.bypass_perm, 0);
 734 #endif
 735         switch (flags & FAN_ALL_CLASS_BITS) {
 736         case FAN_CLASS_NOTIF:
 737                 group->priority = FS_PRIO_0;
 738                 break;
 739         case FAN_CLASS_CONTENT:
 740                 group->priority = FS_PRIO_1;
 741                 break;
 742         case FAN_CLASS_PRE_CONTENT:
 743                 group->priority = FS_PRIO_2;
 744                 break;
 745         default:
 746                 fd = -EINVAL;
 747                 goto out_destroy_group;
 748         }
 749
 750         if (flags & FAN_UNLIMITED_QUEUE) {
 751                 fd = -EPERM;
 752                 if (!capable(CAP_SYS_ADMIN))
 753                         goto out_destroy_group;
 754                 group->max_events = UINT_MAX;
 755         } else {
 756                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 757         }
 758
 759         if (flags & FAN_UNLIMITED_MARKS) {
 760                 fd = -EPERM;
 761                 if (!capable(CAP_SYS_ADMIN))
 762                         goto out_destroy_group;
 763                 group->fanotify_data.max_marks = UINT_MAX;
 764         } else {
 765                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 766         }
 767
 768         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 769         if (fd < 0)
 770                 goto out_destroy_group;
 771
 772         return fd;
 773
 774 out_destroy_group:
 775         fsnotify_destroy_group(group);
 776         return fd;
 777 }
 778
 779 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 780                               __u64, mask, int, dfd,
 781                               const char  __user *, pathname)
 782 {
 783         struct inode *inode = NULL;
 784         struct vfsmount *mnt = NULL;
 785         struct fsnotify_group *group;
 786         struct fd f;
 787         struct path path;
 788         int ret;
 789
 790         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 791                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 792
 793         /* we only use the lower 32 bits as of right now. */
 794         if (mask & ((__u64)0xffffffff << 32))
 795                 return -EINVAL;
 796
 797         if (flags & ~FAN_ALL_MARK_FLAGS)
 798                 return -EINVAL;
 799         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 800         case FAN_MARK_ADD:              /* fallthrough */
 801         case FAN_MARK_REMOVE:
 802                 if (!mask)
 803                         return -EINVAL;
 804         case FAN_MARK_FLUSH:
 805                 break;
 806         default:
 807                 return -EINVAL;
 808         }
 809
 810         if (mask & FAN_ONDIR) {
 811                 flags |= FAN_MARK_ONDIR;
 812                 mask &= ~FAN_ONDIR;
 813         }
 814
 815 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 816         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 817 #else
 818         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 819 #endif
 820                 return -EINVAL;
 821
 822         f = fdget(fanotify_fd);
 823         if (unlikely(!f.file))
 824                 return -EBADF;
 825
 826         /* verify that this is indeed an fanotify instance */
 827         ret = -EINVAL;
 828         if (unlikely(f.file->f_op != &fanotify_fops))
 829                 goto fput_and_out;
 830         group = f.file->private_data;
 831
 832         /*
 833          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 834          * allowed to set permissions events.
 835          */
 836         ret = -EINVAL;
 837         if (mask & FAN_ALL_PERM_EVENTS &&
 838             group->priority == FS_PRIO_0)
 839                 goto fput_and_out;
 840
 841         ret = fanotify_find_path(dfd, pathname, &path, flags);
 842         if (ret)
 843                 goto fput_and_out;
 844
 845         /* inode held in place by reference to path; group by fget on fd */
 846         if (!(flags & FAN_MARK_MOUNT))
 847                 inode = path.dentry->d_inode;
 848         else
 849                 mnt = path.mnt;
 850
 851         /* create/update an inode mark */
 852         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 853         case FAN_MARK_ADD:
 854                 if (flags & FAN_MARK_MOUNT)
 855                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 856                 else
 857                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 858                 break;
 859         case FAN_MARK_REMOVE:
 860                 if (flags & FAN_MARK_MOUNT)
 861                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 862                 else
 863                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 864                 break;
 865         case FAN_MARK_FLUSH:
 866                 if (flags & FAN_MARK_MOUNT)
 867                         fsnotify_clear_vfsmount_marks_by_group(group);
 868                 else
 869                         fsnotify_clear_inode_marks_by_group(group);
 870                 break;
 871         default:
 872                 ret = -EINVAL;
 873         }
 874
 875         path_put(&path);
 876 fput_and_out:
 877         fdput(f);
 878         return ret;
 879 }
 880
 881 #ifdef CONFIG_COMPAT
 882 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 883                                 int, fanotify_fd, unsigned int, flags,
 884                                 __u32, mask0, __u32, mask1, int, dfd,
 885                                 const char  __user *, pathname)
 886 {
 887         return sys_fanotify_mark(fanotify_fd, flags,
 888 #ifdef __BIG_ENDIAN
 889                                 ((__u64)mask0 << 32) | mask1,
 890 #else
 891                                 ((__u64)mask1 << 32) | mask0,
 892 #endif
 893                                  dfd, pathname);
 894 }
 895 #endif
 896
 897 /*
 898  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 899  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 900  * must result in panic().
 901  */
 902 static int __init fanotify_user_setup(void)
 903 {
 904         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 905         fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
 906                                                    SLAB_PANIC);
 907         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 908
 909         return 0;
 910 }
 911 device_initcall(fanotify_user_setup);