fs/notify/fanotify/fanotify_user.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/fanotify.h>
   3 #include <linux/fcntl.h>
   4 #include <linux/file.h>
   5 #include <linux/fs.h>
   6 #include <linux/anon_inodes.h>
   7 #include <linux/fsnotify_backend.h>
   8 #include <linux/init.h>
   9 #include <linux/mount.h>
  10 #include <linux/namei.h>
  11 #include <linux/poll.h>
  12 #include <linux/security.h>
  13 #include <linux/syscalls.h>
  14 #include <linux/slab.h>
  15 #include <linux/types.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/compat.h>
  18 #include <linux/sched/signal.h>
  19
  20 #include <asm/ioctls.h>
  21
  22 #include "../../mount.h"
  23 #include "../fdinfo.h"
  24 #include "fanotify.h"
  25
  26 #define FANOTIFY_DEFAULT_MAX_EVENTS     16384
  27 #define FANOTIFY_DEFAULT_MAX_MARKS      8192
  28 #define FANOTIFY_DEFAULT_MAX_LISTENERS  128
  29
  30 /*
  31  * All flags that may be specified in parameter event_f_flags of fanotify_init.
  32  *
  33  * Internal and external open flags are stored together in field f_flags of
  34  * struct file. Only external open flags shall be allowed in event_f_flags.
  35  * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be
  36  * excluded.
  37  */
  38 #define FANOTIFY_INIT_ALL_EVENT_F_BITS                          ( \
  39                 O_ACCMODE       | O_APPEND      | O_NONBLOCK    | \
  40                 __O_SYNC        | O_DSYNC       | O_CLOEXEC     | \
  41                 O_LARGEFILE     | O_NOATIME     )
  42
  43 extern const struct fsnotify_ops fanotify_fsnotify_ops;
  44
  45 struct kmem_cache *fanotify_mark_cache __read_mostly;
  46 struct kmem_cache *fanotify_event_cachep __read_mostly;
  47 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
  48
  49 /*
  50  * Get an fsnotify notification event if one exists and is small
  51  * enough to fit in "count". Return an error pointer if the count
  52  * is not large enough.
  53  *
  54  * Called with the group->notification_lock held.
  55  */
  56 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  57                                             size_t count)
  58 {
  59         assert_spin_locked(&group->notification_lock);
  60
  61         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  62
  63         if (fsnotify_notify_queue_is_empty(group))
  64                 return NULL;
  65
  66         if (FAN_EVENT_METADATA_LEN > count)
  67                 return ERR_PTR(-EINVAL);
  68
  69         /* held the notification_lock the whole time, so this is the
  70          * same event we peeked above */
  71         return fsnotify_remove_first_event(group);
  72 }
  73
  74 static int create_fd(struct fsnotify_group *group,
  75                      struct fanotify_event_info *event,
  76                      struct file **file)
  77 {
  78         int client_fd;
  79         struct file *new_file;
  80
  81         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  82
  83         client_fd = get_unused_fd_flags(group->fanotify_data.f_flags);
  84         if (client_fd < 0)
  85                 return client_fd;
  86
  87         /*
  88          * we need a new file handle for the userspace program so it can read even if it was
  89          * originally opened O_WRONLY.
  90          */
  91         /* it's possible this event was an overflow event.  in that case dentry and mnt
  92          * are NULL;  That's fine, just don't call dentry open */
  93         if (event->path.dentry && event->path.mnt)
  94                 new_file = dentry_open(&event->path,
  95                                        group->fanotify_data.f_flags | FMODE_NONOTIFY,
  96                                        current_cred());
  97         else
  98                 new_file = ERR_PTR(-EOVERFLOW);
  99         if (IS_ERR(new_file)) {
 100                 /*
 101                  * we still send an event even if we can't open the file.  this
 102                  * can happen when say tasks are gone and we try to open their
 103                  * /proc files or we try to open a WRONLY file like in sysfs
 104                  * we just send the errno to userspace since there isn't much
 105                  * else we can do.
 106                  */
 107                 put_unused_fd(client_fd);
 108                 client_fd = PTR_ERR(new_file);
 109         } else {
 110                 *file = new_file;
 111         }
 112
 113         return client_fd;
 114 }
 115
 116 static int fill_event_metadata(struct fsnotify_group *group,
 117                                struct fanotify_event_metadata *metadata,
 118                                struct fsnotify_event *fsn_event,
 119                                struct file **file)
 120 {
 121         int ret = 0;
 122         struct fanotify_event_info *event;
 123
 124         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 125                  group, metadata, fsn_event);
 126
 127         *file = NULL;
 128         event = container_of(fsn_event, struct fanotify_event_info, fse);
 129         metadata->event_len = FAN_EVENT_METADATA_LEN;
 130         metadata->metadata_len = FAN_EVENT_METADATA_LEN;
 131         metadata->vers = FANOTIFY_METADATA_VERSION;
 132         metadata->reserved = 0;
 133         metadata->mask = fsn_event->mask & FAN_ALL_OUTGOING_EVENTS;
 134         metadata->pid = pid_vnr(event->tgid);
 135         if (unlikely(fsn_event->mask & FAN_Q_OVERFLOW))
 136                 metadata->fd = FAN_NOFD;
 137         else {
 138                 metadata->fd = create_fd(group, event, file);
 139                 if (metadata->fd < 0)
 140                         ret = metadata->fd;
 141         }
 142
 143         return ret;
 144 }
 145
 146 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 147 static struct fanotify_perm_event_info *dequeue_event(
 148                                 struct fsnotify_group *group, int fd)
 149 {
 150         struct fanotify_perm_event_info *event, *return_e = NULL;
 151
 152         spin_lock(&group->notification_lock);
 153         list_for_each_entry(event, &group->fanotify_data.access_list,
 154                             fae.fse.list) {
 155                 if (event->fd != fd)
 156                         continue;
 157
 158                 list_del_init(&event->fae.fse.list);
 159                 return_e = event;
 160                 break;
 161         }
 162         spin_unlock(&group->notification_lock);
 163
 164         pr_debug("%s: found return_re=%p\n", __func__, return_e);
 165
 166         return return_e;
 167 }
 168
 169 static int process_access_response(struct fsnotify_group *group,
 170                                    struct fanotify_response *response_struct)
 171 {
 172         struct fanotify_perm_event_info *event;
 173         int fd = response_struct->fd;
 174         int response = response_struct->response;
 175
 176         pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
 177                  fd, response);
 178         /*
 179          * make sure the response is valid, if invalid we do nothing and either
 180          * userspace can send a valid response or we will clean it up after the
 181          * timeout
 182          */
 183         switch (response) {
 184         case FAN_ALLOW:
 185         case FAN_DENY:
 186                 break;
 187         default:
 188                 return -EINVAL;
 189         }
 190
 191         if (fd < 0)
 192                 return -EINVAL;
 193
 194         event = dequeue_event(group, fd);
 195         if (!event)
 196                 return -ENOENT;
 197
 198         event->response = response;
 199         wake_up(&group->fanotify_data.access_waitq);
 200
 201         return 0;
 202 }
 203 #endif
 204
 205 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 206                                   struct fsnotify_event *event,
 207                                   char __user *buf)
 208 {
 209         struct fanotify_event_metadata fanotify_event_metadata;
 210         struct file *f;
 211         int fd, ret;
 212
 213         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 214
 215         ret = fill_event_metadata(group, &fanotify_event_metadata, event, &f);
 216         if (ret < 0)
 217                 return ret;
 218
 219         fd = fanotify_event_metadata.fd;
 220         ret = -EFAULT;
 221         if (copy_to_user(buf, &fanotify_event_metadata,
 222                          fanotify_event_metadata.event_len))
 223                 goto out_close_fd;
 224
 225 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 226         if (event->mask & FAN_ALL_PERM_EVENTS)
 227                 FANOTIFY_PE(event)->fd = fd;
 228 #endif
 229
 230         if (fd != FAN_NOFD)
 231                 fd_install(fd, f);
 232         return fanotify_event_metadata.event_len;
 233
 234 out_close_fd:
 235         if (fd != FAN_NOFD) {
 236                 put_unused_fd(fd);
 237                 fput(f);
 238         }
 239         return ret;
 240 }
 241
 242 /* intofiy userspace file descriptor functions */
 243 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 244 {
 245         struct fsnotify_group *group = file->private_data;
 246         int ret = 0;
 247
 248         poll_wait(file, &group->notification_waitq, wait);
 249         spin_lock(&group->notification_lock);
 250         if (!fsnotify_notify_queue_is_empty(group))
 251                 ret = POLLIN | POLLRDNORM;
 252         spin_unlock(&group->notification_lock);
 253
 254         return ret;
 255 }
 256
 257 static ssize_t fanotify_read(struct file *file, char __user *buf,
 258                              size_t count, loff_t *pos)
 259 {
 260         struct fsnotify_group *group;
 261         struct fsnotify_event *kevent;
 262         char __user *start;
 263         int ret;
 264         DEFINE_WAIT_FUNC(wait, woken_wake_function);
 265
 266         start = buf;
 267         group = file->private_data;
 268
 269         pr_debug("%s: group=%p\n", __func__, group);
 270
 271         add_wait_queue(&group->notification_waitq, &wait);
 272         while (1) {
 273                 spin_lock(&group->notification_lock);
 274                 kevent = get_one_event(group, count);
 275                 spin_unlock(&group->notification_lock);
 276
 277                 if (IS_ERR(kevent)) {
 278                         ret = PTR_ERR(kevent);
 279                         break;
 280                 }
 281
 282                 if (!kevent) {
 283                         ret = -EAGAIN;
 284                         if (file->f_flags & O_NONBLOCK)
 285                                 break;
 286
 287                         ret = -ERESTARTSYS;
 288                         if (signal_pending(current))
 289                                 break;
 290
 291                         if (start != buf)
 292                                 break;
 293
 294                         wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
 295                         continue;
 296                 }
 297
 298                 ret = copy_event_to_user(group, kevent, buf);
 299                 if (unlikely(ret == -EOPENSTALE)) {
 300                         /*
 301                          * We cannot report events with stale fd so drop it.
 302                          * Setting ret to 0 will continue the event loop and
 303                          * do the right thing if there are no more events to
 304                          * read (i.e. return bytes read, -EAGAIN or wait).
 305                          */
 306                         ret = 0;
 307                 }
 308
 309                 /*
 310                  * Permission events get queued to wait for response.  Other
 311                  * events can be destroyed now.
 312                  */
 313                 if (!(kevent->mask & FAN_ALL_PERM_EVENTS)) {
 314                         fsnotify_destroy_event(group, kevent);
 315                 } else {
 316 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 317                         if (ret <= 0) {
 318                                 FANOTIFY_PE(kevent)->response = FAN_DENY;
 319                                 wake_up(&group->fanotify_data.access_waitq);
 320                         } else {
 321                                 spin_lock(&group->notification_lock);
 322                                 list_add_tail(&kevent->list,
 323                                         &group->fanotify_data.access_list);
 324                                 spin_unlock(&group->notification_lock);
 325                         }
 326 #endif
 327                 }
 328                 if (ret < 0)
 329                         break;
 330                 buf += ret;
 331                 count -= ret;
 332         }
 333         remove_wait_queue(&group->notification_waitq, &wait);
 334
 335         if (start != buf && ret != -EFAULT)
 336                 ret = buf - start;
 337         return ret;
 338 }
 339
 340 static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
 341 {
 342 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 343         struct fanotify_response response = { .fd = -1, .response = -1 };
 344         struct fsnotify_group *group;
 345         int ret;
 346
 347         group = file->private_data;
 348
 349         if (count > sizeof(response))
 350                 count = sizeof(response);
 351
 352         pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
 353
 354         if (copy_from_user(&response, buf, count))
 355                 return -EFAULT;
 356
 357         ret = process_access_response(group, &response);
 358         if (ret < 0)
 359                 count = ret;
 360
 361         return count;
 362 #else
 363         return -EINVAL;
 364 #endif
 365 }
 366
 367 static int fanotify_release(struct inode *ignored, struct file *file)
 368 {
 369         struct fsnotify_group *group = file->private_data;
 370
 371 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 372         struct fanotify_perm_event_info *event, *next;
 373         struct fsnotify_event *fsn_event;
 374
 375         /*
 376          * Stop new events from arriving in the notification queue. since
 377          * userspace cannot use fanotify fd anymore, no event can enter or
 378          * leave access_list by now either.
 379          */
 380         fsnotify_group_stop_queueing(group);
 381
 382         /*
 383          * Process all permission events on access_list and notification queue
 384          * and simulate reply from userspace.
 385          */
 386         spin_lock(&group->notification_lock);
 387         list_for_each_entry_safe(event, next, &group->fanotify_data.access_list,
 388                                  fae.fse.list) {
 389                 pr_debug("%s: found group=%p event=%p\n", __func__, group,
 390                          event);
 391
 392                 list_del_init(&event->fae.fse.list);
 393                 event->response = FAN_ALLOW;
 394         }
 395
 396         /*
 397          * Destroy all non-permission events. For permission events just
 398          * dequeue them and set the response. They will be freed once the
 399          * response is consumed and fanotify_get_response() returns.
 400          */
 401         while (!fsnotify_notify_queue_is_empty(group)) {
 402                 fsn_event = fsnotify_remove_first_event(group);
 403                 if (!(fsn_event->mask & FAN_ALL_PERM_EVENTS)) {
 404                         spin_unlock(&group->notification_lock);
 405                         fsnotify_destroy_event(group, fsn_event);
 406                         spin_lock(&group->notification_lock);
 407                 } else
 408                         FANOTIFY_PE(fsn_event)->response = FAN_ALLOW;
 409         }
 410         spin_unlock(&group->notification_lock);
 411
 412         /* Response for all permission events it set, wakeup waiters */
 413         wake_up(&group->fanotify_data.access_waitq);
 414 #endif
 415
 416         /* matches the fanotify_init->fsnotify_alloc_group */
 417         fsnotify_destroy_group(group);
 418
 419         return 0;
 420 }
 421
 422 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 423 {
 424         struct fsnotify_group *group;
 425         struct fsnotify_event *fsn_event;
 426         void __user *p;
 427         int ret = -ENOTTY;
 428         size_t send_len = 0;
 429
 430         group = file->private_data;
 431
 432         p = (void __user *) arg;
 433
 434         switch (cmd) {
 435         case FIONREAD:
 436                 spin_lock(&group->notification_lock);
 437                 list_for_each_entry(fsn_event, &group->notification_list, list)
 438                         send_len += FAN_EVENT_METADATA_LEN;
 439                 spin_unlock(&group->notification_lock);
 440                 ret = put_user(send_len, (int __user *) p);
 441                 break;
 442         }
 443
 444         return ret;
 445 }
 446
 447 static const struct file_operations fanotify_fops = {
 448         .show_fdinfo    = fanotify_show_fdinfo,
 449         .poll           = fanotify_poll,
 450         .read           = fanotify_read,
 451         .write          = fanotify_write,
 452         .fasync         = NULL,
 453         .release        = fanotify_release,
 454         .unlocked_ioctl = fanotify_ioctl,
 455         .compat_ioctl   = fanotify_ioctl,
 456         .llseek         = noop_llseek,
 457 };
 458
 459 static int fanotify_find_path(int dfd, const char __user *filename,
 460                               struct path *path, unsigned int flags)
 461 {
 462         int ret;
 463
 464         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 465                  dfd, filename, flags);
 466
 467         if (filename == NULL) {
 468                 struct fd f = fdget(dfd);
 469
 470                 ret = -EBADF;
 471                 if (!f.file)
 472                         goto out;
 473
 474                 ret = -ENOTDIR;
 475                 if ((flags & FAN_MARK_ONLYDIR) &&
 476                     !(S_ISDIR(file_inode(f.file)->i_mode))) {
 477                         fdput(f);
 478                         goto out;
 479                 }
 480
 481                 *path = f.file->f_path;
 482                 path_get(path);
 483                 fdput(f);
 484         } else {
 485                 unsigned int lookup_flags = 0;
 486
 487                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 488                         lookup_flags |= LOOKUP_FOLLOW;
 489                 if (flags & FAN_MARK_ONLYDIR)
 490                         lookup_flags |= LOOKUP_DIRECTORY;
 491
 492                 ret = user_path_at(dfd, filename, lookup_flags, path);
 493                 if (ret)
 494                         goto out;
 495         }
 496
 497         /* you can only watch an inode if you have read permissions on it */
 498         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 499         if (ret)
 500                 path_put(path);
 501 out:
 502         return ret;
 503 }
 504
 505 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
 506                                             __u32 mask,
 507                                             unsigned int flags,
 508                                             int *destroy)
 509 {
 510         __u32 oldmask = 0;
 511
 512         spin_lock(&fsn_mark->lock);
 513         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 514                 __u32 tmask = fsn_mark->mask & ~mask;
 515
 516                 if (flags & FAN_MARK_ONDIR)
 517                         tmask &= ~FAN_ONDIR;
 518
 519                 oldmask = fsn_mark->mask;
 520                 fsn_mark->mask = tmask;
 521         } else {
 522                 __u32 tmask = fsn_mark->ignored_mask & ~mask;
 523                 if (flags & FAN_MARK_ONDIR)
 524                         tmask &= ~FAN_ONDIR;
 525                 fsn_mark->ignored_mask = tmask;
 526         }
 527         *destroy = !(fsn_mark->mask | fsn_mark->ignored_mask);
 528         spin_unlock(&fsn_mark->lock);
 529
 530         return mask & oldmask;
 531 }
 532
 533 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 534                                          struct vfsmount *mnt, __u32 mask,
 535                                          unsigned int flags)
 536 {
 537         struct fsnotify_mark *fsn_mark = NULL;
 538         __u32 removed;
 539         int destroy_mark;
 540
 541         mutex_lock(&group->mark_mutex);
 542         fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
 543                                       group);
 544         if (!fsn_mark) {
 545                 mutex_unlock(&group->mark_mutex);
 546                 return -ENOENT;
 547         }
 548
 549         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 550                                                  &destroy_mark);
 551         if (removed & real_mount(mnt)->mnt_fsnotify_mask)
 552                 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
 553         if (destroy_mark)
 554                 fsnotify_detach_mark(fsn_mark);
 555         mutex_unlock(&group->mark_mutex);
 556         if (destroy_mark)
 557                 fsnotify_free_mark(fsn_mark);
 558
 559         fsnotify_put_mark(fsn_mark);
 560         return 0;
 561 }
 562
 563 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 564                                       struct inode *inode, __u32 mask,
 565                                       unsigned int flags)
 566 {
 567         struct fsnotify_mark *fsn_mark = NULL;
 568         __u32 removed;
 569         int destroy_mark;
 570
 571         mutex_lock(&group->mark_mutex);
 572         fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 573         if (!fsn_mark) {
 574                 mutex_unlock(&group->mark_mutex);
 575                 return -ENOENT;
 576         }
 577
 578         removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags,
 579                                                  &destroy_mark);
 580         if (removed & inode->i_fsnotify_mask)
 581                 fsnotify_recalc_mask(inode->i_fsnotify_marks);
 582         if (destroy_mark)
 583                 fsnotify_detach_mark(fsn_mark);
 584         mutex_unlock(&group->mark_mutex);
 585         if (destroy_mark)
 586                 fsnotify_free_mark(fsn_mark);
 587
 588         /* matches the fsnotify_find_mark() */
 589         fsnotify_put_mark(fsn_mark);
 590
 591         return 0;
 592 }
 593
 594 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
 595                                        __u32 mask,
 596                                        unsigned int flags)
 597 {
 598         __u32 oldmask = -1;
 599
 600         spin_lock(&fsn_mark->lock);
 601         if (!(flags & FAN_MARK_IGNORED_MASK)) {
 602                 __u32 tmask = fsn_mark->mask | mask;
 603
 604                 if (flags & FAN_MARK_ONDIR)
 605                         tmask |= FAN_ONDIR;
 606
 607                 oldmask = fsn_mark->mask;
 608                 fsn_mark->mask = tmask;
 609         } else {
 610                 __u32 tmask = fsn_mark->ignored_mask | mask;
 611                 if (flags & FAN_MARK_ONDIR)
 612                         tmask |= FAN_ONDIR;
 613
 614                 fsn_mark->ignored_mask = tmask;
 615                 if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
 616                         fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
 617         }
 618         spin_unlock(&fsn_mark->lock);
 619
 620         return mask & ~oldmask;
 621 }
 622
 623 static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group,
 624                                                    struct inode *inode,
 625                                                    struct vfsmount *mnt)
 626 {
 627         struct fsnotify_mark *mark;
 628         int ret;
 629
 630         if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks)
 631                 return ERR_PTR(-ENOSPC);
 632
 633         mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 634         if (!mark)
 635                 return ERR_PTR(-ENOMEM);
 636
 637         fsnotify_init_mark(mark, group);
 638         ret = fsnotify_add_mark_locked(mark, inode, mnt, 0);
 639         if (ret) {
 640                 fsnotify_put_mark(mark);
 641                 return ERR_PTR(ret);
 642         }
 643
 644         return mark;
 645 }
 646
 647
 648 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 649                                       struct vfsmount *mnt, __u32 mask,
 650                                       unsigned int flags)
 651 {
 652         struct fsnotify_mark *fsn_mark;
 653         __u32 added;
 654
 655         mutex_lock(&group->mark_mutex);
 656         fsn_mark = fsnotify_find_mark(&real_mount(mnt)->mnt_fsnotify_marks,
 657                                       group);
 658         if (!fsn_mark) {
 659                 fsn_mark = fanotify_add_new_mark(group, NULL, mnt);
 660                 if (IS_ERR(fsn_mark)) {
 661                         mutex_unlock(&group->mark_mutex);
 662                         return PTR_ERR(fsn_mark);
 663                 }
 664         }
 665         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 666         if (added & ~real_mount(mnt)->mnt_fsnotify_mask)
 667                 fsnotify_recalc_mask(real_mount(mnt)->mnt_fsnotify_marks);
 668         mutex_unlock(&group->mark_mutex);
 669
 670         fsnotify_put_mark(fsn_mark);
 671         return 0;
 672 }
 673
 674 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 675                                    struct inode *inode, __u32 mask,
 676                                    unsigned int flags)
 677 {
 678         struct fsnotify_mark *fsn_mark;
 679         __u32 added;
 680
 681         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 682
 683         /*
 684          * If some other task has this inode open for write we should not add
 685          * an ignored mark, unless that ignored mark is supposed to survive
 686          * modification changes anyway.
 687          */
 688         if ((flags & FAN_MARK_IGNORED_MASK) &&
 689             !(flags & FAN_MARK_IGNORED_SURV_MODIFY) &&
 690             (atomic_read(&inode->i_writecount) > 0))
 691                 return 0;
 692
 693         mutex_lock(&group->mark_mutex);
 694         fsn_mark = fsnotify_find_mark(&inode->i_fsnotify_marks, group);
 695         if (!fsn_mark) {
 696                 fsn_mark = fanotify_add_new_mark(group, inode, NULL);
 697                 if (IS_ERR(fsn_mark)) {
 698                         mutex_unlock(&group->mark_mutex);
 699                         return PTR_ERR(fsn_mark);
 700                 }
 701         }
 702         added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
 703         if (added & ~inode->i_fsnotify_mask)
 704                 fsnotify_recalc_mask(inode->i_fsnotify_marks);
 705         mutex_unlock(&group->mark_mutex);
 706
 707         fsnotify_put_mark(fsn_mark);
 708         return 0;
 709 }
 710
 711 /* fanotify syscalls */
 712 SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
 713 {
 714         struct fsnotify_group *group;
 715         int f_flags, fd;
 716         struct user_struct *user;
 717         struct fanotify_event_info *oevent;
 718
 719         pr_debug("%s: flags=%d event_f_flags=%d\n",
 720                 __func__, flags, event_f_flags);
 721
 722         if (!capable(CAP_SYS_ADMIN))
 723                 return -EPERM;
 724
 725         if (flags & ~FAN_ALL_INIT_FLAGS)
 726                 return -EINVAL;
 727
 728         if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS)
 729                 return -EINVAL;
 730
 731         switch (event_f_flags & O_ACCMODE) {
 732         case O_RDONLY:
 733         case O_RDWR:
 734         case O_WRONLY:
 735                 break;
 736         default:
 737                 return -EINVAL;
 738         }
 739
 740         user = get_current_user();
 741         if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) {
 742                 free_uid(user);
 743                 return -EMFILE;
 744         }
 745
 746         f_flags = O_RDWR | FMODE_NONOTIFY;
 747         if (flags & FAN_CLOEXEC)
 748                 f_flags |= O_CLOEXEC;
 749         if (flags & FAN_NONBLOCK)
 750                 f_flags |= O_NONBLOCK;
 751
 752         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 753         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 754         if (IS_ERR(group)) {
 755                 free_uid(user);
 756                 return PTR_ERR(group);
 757         }
 758
 759         group->fanotify_data.user = user;
 760         atomic_inc(&user->fanotify_listeners);
 761
 762         oevent = fanotify_alloc_event(NULL, FS_Q_OVERFLOW, NULL);
 763         if (unlikely(!oevent)) {
 764                 fd = -ENOMEM;
 765                 goto out_destroy_group;
 766         }
 767         group->overflow_event = &oevent->fse;
 768
 769         if (force_o_largefile())
 770                 event_f_flags |= O_LARGEFILE;
 771         group->fanotify_data.f_flags = event_f_flags;
 772 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 773         init_waitqueue_head(&group->fanotify_data.access_waitq);
 774         INIT_LIST_HEAD(&group->fanotify_data.access_list);
 775 #endif
 776         switch (flags & FAN_ALL_CLASS_BITS) {
 777         case FAN_CLASS_NOTIF:
 778                 group->priority = FS_PRIO_0;
 779                 break;
 780         case FAN_CLASS_CONTENT:
 781                 group->priority = FS_PRIO_1;
 782                 break;
 783         case FAN_CLASS_PRE_CONTENT:
 784                 group->priority = FS_PRIO_2;
 785                 break;
 786         default:
 787                 fd = -EINVAL;
 788                 goto out_destroy_group;
 789         }
 790
 791         if (flags & FAN_UNLIMITED_QUEUE) {
 792                 fd = -EPERM;
 793                 if (!capable(CAP_SYS_ADMIN))
 794                         goto out_destroy_group;
 795                 group->max_events = UINT_MAX;
 796         } else {
 797                 group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS;
 798         }
 799
 800         if (flags & FAN_UNLIMITED_MARKS) {
 801                 fd = -EPERM;
 802                 if (!capable(CAP_SYS_ADMIN))
 803                         goto out_destroy_group;
 804                 group->fanotify_data.max_marks = UINT_MAX;
 805         } else {
 806                 group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS;
 807         }
 808
 809         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 810         if (fd < 0)
 811                 goto out_destroy_group;
 812
 813         return fd;
 814
 815 out_destroy_group:
 816         fsnotify_destroy_group(group);
 817         return fd;
 818 }
 819
 820 SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags,
 821                               __u64, mask, int, dfd,
 822                               const char  __user *, pathname)
 823 {
 824         struct inode *inode = NULL;
 825         struct vfsmount *mnt = NULL;
 826         struct fsnotify_group *group;
 827         struct fd f;
 828         struct path path;
 829         int ret;
 830
 831         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 832                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 833
 834         /* we only use the lower 32 bits as of right now. */
 835         if (mask & ((__u64)0xffffffff << 32))
 836                 return -EINVAL;
 837
 838         if (flags & ~FAN_ALL_MARK_FLAGS)
 839                 return -EINVAL;
 840         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
 841         case FAN_MARK_ADD:              /* fallthrough */
 842         case FAN_MARK_REMOVE:
 843                 if (!mask)
 844                         return -EINVAL;
 845                 break;
 846         case FAN_MARK_FLUSH:
 847                 if (flags & ~(FAN_MARK_MOUNT | FAN_MARK_FLUSH))
 848                         return -EINVAL;
 849                 break;
 850         default:
 851                 return -EINVAL;
 852         }
 853
 854         if (mask & FAN_ONDIR) {
 855                 flags |= FAN_MARK_ONDIR;
 856                 mask &= ~FAN_ONDIR;
 857         }
 858
 859 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 860         if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
 861 #else
 862         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 863 #endif
 864                 return -EINVAL;
 865
 866         f = fdget(fanotify_fd);
 867         if (unlikely(!f.file))
 868                 return -EBADF;
 869
 870         /* verify that this is indeed an fanotify instance */
 871         ret = -EINVAL;
 872         if (unlikely(f.file->f_op != &fanotify_fops))
 873                 goto fput_and_out;
 874         group = f.file->private_data;
 875
 876         /*
 877          * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF.  These are not
 878          * allowed to set permissions events.
 879          */
 880         ret = -EINVAL;
 881         if (mask & FAN_ALL_PERM_EVENTS &&
 882             group->priority == FS_PRIO_0)
 883                 goto fput_and_out;
 884
 885         if (flags & FAN_MARK_FLUSH) {
 886                 ret = 0;
 887                 if (flags & FAN_MARK_MOUNT)
 888                         fsnotify_clear_vfsmount_marks_by_group(group);
 889                 else
 890                         fsnotify_clear_inode_marks_by_group(group);
 891                 goto fput_and_out;
 892         }
 893
 894         ret = fanotify_find_path(dfd, pathname, &path, flags);
 895         if (ret)
 896                 goto fput_and_out;
 897
 898         /* inode held in place by reference to path; group by fget on fd */
 899         if (!(flags & FAN_MARK_MOUNT))
 900                 inode = path.dentry->d_inode;
 901         else
 902                 mnt = path.mnt;
 903
 904         /* create/update an inode mark */
 905         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
 906         case FAN_MARK_ADD:
 907                 if (flags & FAN_MARK_MOUNT)
 908                         ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
 909                 else
 910                         ret = fanotify_add_inode_mark(group, inode, mask, flags);
 911                 break;
 912         case FAN_MARK_REMOVE:
 913                 if (flags & FAN_MARK_MOUNT)
 914                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
 915                 else
 916                         ret = fanotify_remove_inode_mark(group, inode, mask, flags);
 917                 break;
 918         default:
 919                 ret = -EINVAL;
 920         }
 921
 922         path_put(&path);
 923 fput_and_out:
 924         fdput(f);
 925         return ret;
 926 }
 927
 928 #ifdef CONFIG_COMPAT
 929 COMPAT_SYSCALL_DEFINE6(fanotify_mark,
 930                                 int, fanotify_fd, unsigned int, flags,
 931                                 __u32, mask0, __u32, mask1, int, dfd,
 932                                 const char  __user *, pathname)
 933 {
 934         return sys_fanotify_mark(fanotify_fd, flags,
 935 #ifdef __BIG_ENDIAN
 936                                 ((__u64)mask0 << 32) | mask1,
 937 #else
 938                                 ((__u64)mask1 << 32) | mask0,
 939 #endif
 940                                  dfd, pathname);
 941 }
 942 #endif
 943
 944 /*
 945  * fanotify_user_setup - Our initialization function.  Note that we cannot return
 946  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 947  * must result in panic().
 948  */
 949 static int __init fanotify_user_setup(void)
 950 {
 951         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 952         fanotify_event_cachep = KMEM_CACHE(fanotify_event_info, SLAB_PANIC);
 953 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
 954         fanotify_perm_event_cachep = KMEM_CACHE(fanotify_perm_event_info,
 955                                                 SLAB_PANIC);
 956 #endif
 957
 958         return 0;
 959 }
 960 device_initcall(fanotify_user_setup);