fs/notify/fanotify/fanotify_user.c

   1 #include <linux/fcntl.h>
   2 #include <linux/file.h>
   3 #include <linux/fs.h>
   4 #include <linux/anon_inodes.h>
   5 #include <linux/fsnotify_backend.h>
   6 #include <linux/init.h>
   7 #include <linux/mount.h>
   8 #include <linux/namei.h>
   9 #include <linux/poll.h>
  10 #include <linux/security.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/types.h>
  13 #include <linux/uaccess.h>
  14
  15 #include <asm/ioctls.h>
  16
  17 #include "fanotify.h"
  18
  19 static struct kmem_cache *fanotify_mark_cache __read_mostly;
  20
  21 /*
  22  * Get an fsnotify notification event if one exists and is small
  23  * enough to fit in "count". Return an error pointer if the count
  24  * is not large enough.
  25  *
  26  * Called with the group->notification_mutex held.
  27  */
  28 static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
  29                                             size_t count)
  30 {
  31         BUG_ON(!mutex_is_locked(&group->notification_mutex));
  32
  33         pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
  34
  35         if (fsnotify_notify_queue_is_empty(group))
  36                 return NULL;
  37
  38         if (FAN_EVENT_METADATA_LEN > count)
  39                 return ERR_PTR(-EINVAL);
  40
  41         /* held the notification_mutex the whole time, so this is the
  42          * same event we peeked above */
  43         return fsnotify_remove_notify_event(group);
  44 }
  45
  46 static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
  47 {
  48         int client_fd;
  49         struct dentry *dentry;
  50         struct vfsmount *mnt;
  51         struct file *new_file;
  52
  53         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
  54
  55         client_fd = get_unused_fd();
  56         if (client_fd < 0)
  57                 return client_fd;
  58
  59         if (event->data_type != FSNOTIFY_EVENT_PATH) {
  60                 WARN_ON(1);
  61                 put_unused_fd(client_fd);
  62                 return -EINVAL;
  63         }
  64
  65         /*
  66          * we need a new file handle for the userspace program so it can read even if it was
  67          * originally opened O_WRONLY.
  68          */
  69         dentry = dget(event->path.dentry);
  70         mnt = mntget(event->path.mnt);
  71         /* it's possible this event was an overflow event.  in that case dentry and mnt
  72          * are NULL;  That's fine, just don't call dentry open */
  73         if (dentry && mnt)
  74                 new_file = dentry_open(dentry, mnt,
  75                                        O_RDONLY | O_LARGEFILE | FMODE_NONOTIFY,
  76                                        current_cred());
  77         else
  78                 new_file = ERR_PTR(-EOVERFLOW);
  79         if (IS_ERR(new_file)) {
  80                 /*
  81                  * we still send an event even if we can't open the file.  this
  82                  * can happen when say tasks are gone and we try to open their
  83                  * /proc files or we try to open a WRONLY file like in sysfs
  84                  * we just send the errno to userspace since there isn't much
  85                  * else we can do.
  86                  */
  87                 put_unused_fd(client_fd);
  88                 client_fd = PTR_ERR(new_file);
  89         } else {
  90                 fd_install(client_fd, new_file);
  91         }
  92
  93         return client_fd;
  94 }
  95
  96 static ssize_t fill_event_metadata(struct fsnotify_group *group,
  97                                    struct fanotify_event_metadata *metadata,
  98                                    struct fsnotify_event *event)
  99 {
 100         pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
 101                  group, metadata, event);
 102
 103         metadata->event_len = FAN_EVENT_METADATA_LEN;
 104         metadata->vers = FANOTIFY_METADATA_VERSION;
 105         metadata->mask = fanotify_outgoing_mask(event->mask);
 106         metadata->pid = pid_vnr(event->tgid);
 107         metadata->fd = create_fd(group, event);
 108
 109         return metadata->fd;
 110 }
 111
 112 static ssize_t copy_event_to_user(struct fsnotify_group *group,
 113                                   struct fsnotify_event *event,
 114                                   char __user *buf)
 115 {
 116         struct fanotify_event_metadata fanotify_event_metadata;
 117         int ret;
 118
 119         pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 120
 121         ret = fill_event_metadata(group, &fanotify_event_metadata, event);
 122         if (ret < 0)
 123                 return ret;
 124
 125         if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
 126                 return -EFAULT;
 127
 128         return FAN_EVENT_METADATA_LEN;
 129 }
 130
 131 /* intofiy userspace file descriptor functions */
 132 static unsigned int fanotify_poll(struct file *file, poll_table *wait)
 133 {
 134         struct fsnotify_group *group = file->private_data;
 135         int ret = 0;
 136
 137         poll_wait(file, &group->notification_waitq, wait);
 138         mutex_lock(&group->notification_mutex);
 139         if (!fsnotify_notify_queue_is_empty(group))
 140                 ret = POLLIN | POLLRDNORM;
 141         mutex_unlock(&group->notification_mutex);
 142
 143         return ret;
 144 }
 145
 146 static ssize_t fanotify_read(struct file *file, char __user *buf,
 147                              size_t count, loff_t *pos)
 148 {
 149         struct fsnotify_group *group;
 150         struct fsnotify_event *kevent;
 151         char __user *start;
 152         int ret;
 153         DEFINE_WAIT(wait);
 154
 155         start = buf;
 156         group = file->private_data;
 157
 158         pr_debug("%s: group=%p\n", __func__, group);
 159
 160         while (1) {
 161                 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
 162
 163                 mutex_lock(&group->notification_mutex);
 164                 kevent = get_one_event(group, count);
 165                 mutex_unlock(&group->notification_mutex);
 166
 167                 if (kevent) {
 168                         ret = PTR_ERR(kevent);
 169                         if (IS_ERR(kevent))
 170                                 break;
 171                         ret = copy_event_to_user(group, kevent, buf);
 172                         fsnotify_put_event(kevent);
 173                         if (ret < 0)
 174                                 break;
 175                         buf += ret;
 176                         count -= ret;
 177                         continue;
 178                 }
 179
 180                 ret = -EAGAIN;
 181                 if (file->f_flags & O_NONBLOCK)
 182                         break;
 183                 ret = -EINTR;
 184                 if (signal_pending(current))
 185                         break;
 186
 187                 if (start != buf)
 188                         break;
 189
 190                 schedule();
 191         }
 192
 193         finish_wait(&group->notification_waitq, &wait);
 194         if (start != buf && ret != -EFAULT)
 195                 ret = buf - start;
 196         return ret;
 197 }
 198
 199 static int fanotify_release(struct inode *ignored, struct file *file)
 200 {
 201         struct fsnotify_group *group = file->private_data;
 202
 203         pr_debug("%s: file=%p group=%p\n", __func__, file, group);
 204
 205         /* matches the fanotify_init->fsnotify_alloc_group */
 206         fsnotify_put_group(group);
 207
 208         return 0;
 209 }
 210
 211 static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 212 {
 213         struct fsnotify_group *group;
 214         struct fsnotify_event_holder *holder;
 215         void __user *p;
 216         int ret = -ENOTTY;
 217         size_t send_len = 0;
 218
 219         group = file->private_data;
 220
 221         p = (void __user *) arg;
 222
 223         switch (cmd) {
 224         case FIONREAD:
 225                 mutex_lock(&group->notification_mutex);
 226                 list_for_each_entry(holder, &group->notification_list, event_list)
 227                         send_len += FAN_EVENT_METADATA_LEN;
 228                 mutex_unlock(&group->notification_mutex);
 229                 ret = put_user(send_len, (int __user *) p);
 230                 break;
 231         }
 232
 233         return ret;
 234 }
 235
 236 static const struct file_operations fanotify_fops = {
 237         .poll           = fanotify_poll,
 238         .read           = fanotify_read,
 239         .fasync         = NULL,
 240         .release        = fanotify_release,
 241         .unlocked_ioctl = fanotify_ioctl,
 242         .compat_ioctl   = fanotify_ioctl,
 243 };
 244
 245 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
 246 {
 247         kmem_cache_free(fanotify_mark_cache, fsn_mark);
 248 }
 249
 250 static int fanotify_find_path(int dfd, const char __user *filename,
 251                               struct path *path, unsigned int flags)
 252 {
 253         int ret;
 254
 255         pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
 256                  dfd, filename, flags);
 257
 258         if (filename == NULL) {
 259                 struct file *file;
 260                 int fput_needed;
 261
 262                 ret = -EBADF;
 263                 file = fget_light(dfd, &fput_needed);
 264                 if (!file)
 265                         goto out;
 266
 267                 ret = -ENOTDIR;
 268                 if ((flags & FAN_MARK_ONLYDIR) &&
 269                     !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
 270                         fput_light(file, fput_needed);
 271                         goto out;
 272                 }
 273
 274                 *path = file->f_path;
 275                 path_get(path);
 276                 fput_light(file, fput_needed);
 277         } else {
 278                 unsigned int lookup_flags = 0;
 279
 280                 if (!(flags & FAN_MARK_DONT_FOLLOW))
 281                         lookup_flags |= LOOKUP_FOLLOW;
 282                 if (flags & FAN_MARK_ONLYDIR)
 283                         lookup_flags |= LOOKUP_DIRECTORY;
 284
 285                 ret = user_path_at(dfd, filename, lookup_flags, path);
 286                 if (ret)
 287                         goto out;
 288         }
 289
 290         /* you can only watch an inode if you have read permissions on it */
 291         ret = inode_permission(path->dentry->d_inode, MAY_READ);
 292         if (ret)
 293                 path_put(path);
 294 out:
 295         return ret;
 296 }
 297
 298 static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, __u32 mask)
 299 {
 300         __u32 oldmask;
 301
 302         spin_lock(&fsn_mark->lock);
 303         oldmask = fsn_mark->mask;
 304         fsn_mark->mask = oldmask & ~mask;
 305         spin_unlock(&fsn_mark->lock);
 306
 307         if (!(oldmask & ~mask))
 308                 fsnotify_destroy_mark(fsn_mark);
 309
 310         return mask & oldmask;
 311 }
 312
 313 static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
 314                                          struct vfsmount *mnt, __u32 mask)
 315 {
 316         struct fsnotify_mark *fsn_mark = NULL;
 317         __u32 removed;
 318
 319         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 320         if (!fsn_mark)
 321                 return -ENOENT;
 322
 323         removed = fanotify_mark_remove_from_mask(fsn_mark, mask);
 324         fsnotify_put_mark(fsn_mark);
 325         if (removed & group->mask)
 326                 fsnotify_recalc_group_mask(group);
 327         if (removed & mnt->mnt_fsnotify_mask)
 328                 fsnotify_recalc_vfsmount_mask(mnt);
 329
 330         return 0;
 331 }
 332
 333 static int fanotify_remove_inode_mark(struct fsnotify_group *group,
 334                                       struct inode *inode, __u32 mask)
 335 {
 336         struct fsnotify_mark *fsn_mark = NULL;
 337         __u32 removed;
 338
 339         fsn_mark = fsnotify_find_inode_mark(group, inode);
 340         if (!fsn_mark)
 341                 return -ENOENT;
 342
 343         removed = fanotify_mark_remove_from_mask(fsn_mark, mask);
 344         /* matches the fsnotify_find_inode_mark() */
 345         fsnotify_put_mark(fsn_mark);
 346
 347         if (removed & group->mask)
 348                 fsnotify_recalc_group_mask(group);
 349         if (removed & inode->i_fsnotify_mask)
 350                 fsnotify_recalc_inode_mask(inode);
 351
 352         return 0;
 353 }
 354
 355 static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, __u32 mask)
 356 {
 357         __u32 oldmask;
 358
 359         spin_lock(&fsn_mark->lock);
 360         oldmask = fsn_mark->mask;
 361         fsn_mark->mask = oldmask | mask;
 362         spin_unlock(&fsn_mark->lock);
 363
 364         return mask & ~oldmask;
 365 }
 366
 367 static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
 368                                       struct vfsmount *mnt, __u32 mask)
 369 {
 370         struct fsnotify_mark *fsn_mark;
 371         __u32 added;
 372
 373         fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
 374         if (!fsn_mark) {
 375                 int ret;
 376
 377                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 378                 if (!fsn_mark)
 379                         return -ENOMEM;
 380
 381                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 382                 ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
 383                 if (ret) {
 384                         fanotify_free_mark(fsn_mark);
 385                         return ret;
 386                 }
 387         }
 388         added = fanotify_mark_add_to_mask(fsn_mark, mask);
 389         fsnotify_put_mark(fsn_mark);
 390         if (added) {
 391                 if (added & ~group->mask)
 392                         fsnotify_recalc_group_mask(group);
 393                 if (added & ~mnt->mnt_fsnotify_mask)
 394                         fsnotify_recalc_vfsmount_mask(mnt);
 395         }
 396         return 0;
 397 }
 398
 399 static int fanotify_add_inode_mark(struct fsnotify_group *group,
 400                                    struct inode *inode, __u32 mask)
 401 {
 402         struct fsnotify_mark *fsn_mark;
 403         __u32 added;
 404
 405         pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
 406
 407         fsn_mark = fsnotify_find_inode_mark(group, inode);
 408         if (!fsn_mark) {
 409                 int ret;
 410
 411                 fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
 412                 if (!fsn_mark)
 413                         return -ENOMEM;
 414
 415                 fsnotify_init_mark(fsn_mark, fanotify_free_mark);
 416                 ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
 417                 if (ret) {
 418                         fanotify_free_mark(fsn_mark);
 419                         return ret;
 420                 }
 421         }
 422         added = fanotify_mark_add_to_mask(fsn_mark, mask);
 423         fsnotify_put_mark(fsn_mark);
 424         if (added) {
 425                 if (added & ~group->mask)
 426                         fsnotify_recalc_group_mask(group);
 427                 if (added & ~inode->i_fsnotify_mask)
 428                         fsnotify_recalc_inode_mask(inode);
 429         }
 430         return 0;
 431 }
 432
 433 /* fanotify syscalls */
 434 SYSCALL_DEFINE3(fanotify_init, unsigned int, flags, unsigned int, event_f_flags,
 435                 unsigned int, priority)
 436 {
 437         struct fsnotify_group *group;
 438         int f_flags, fd;
 439
 440         pr_debug("%s: flags=%d event_f_flags=%d priority=%d\n",
 441                 __func__, flags, event_f_flags, priority);
 442
 443         if (event_f_flags)
 444                 return -EINVAL;
 445         if (priority)
 446                 return -EINVAL;
 447
 448         if (!capable(CAP_SYS_ADMIN))
 449                 return -EACCES;
 450
 451         if (flags & ~FAN_ALL_INIT_FLAGS)
 452                 return -EINVAL;
 453
 454         f_flags = (O_RDONLY | FMODE_NONOTIFY);
 455         if (flags & FAN_CLOEXEC)
 456                 f_flags |= O_CLOEXEC;
 457         if (flags & FAN_NONBLOCK)
 458                 f_flags |= O_NONBLOCK;
 459
 460         /* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
 461         group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
 462         if (IS_ERR(group))
 463                 return PTR_ERR(group);
 464
 465         fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
 466         if (fd < 0)
 467                 goto out_put_group;
 468
 469         return fd;
 470
 471 out_put_group:
 472         fsnotify_put_group(group);
 473         return fd;
 474 }
 475
 476 SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
 477                               __u64 mask, int dfd,
 478                               const char  __user * pathname)
 479 {
 480         struct inode *inode = NULL;
 481         struct vfsmount *mnt = NULL;
 482         struct fsnotify_group *group;
 483         struct file *filp;
 484         struct path path;
 485         int ret, fput_needed;
 486
 487         pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
 488                  __func__, fanotify_fd, flags, dfd, pathname, mask);
 489
 490         /* we only use the lower 32 bits as of right now. */
 491         if (mask & ((__u64)0xffffffff << 32))
 492                 return -EINVAL;
 493
 494         if (flags & ~FAN_ALL_MARK_FLAGS)
 495                 return -EINVAL;
 496         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
 497         case FAN_MARK_ADD:
 498         case FAN_MARK_REMOVE:
 499                 break;
 500         default:
 501                 return -EINVAL;
 502         }
 503         if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
 504                 return -EINVAL;
 505
 506         filp = fget_light(fanotify_fd, &fput_needed);
 507         if (unlikely(!filp))
 508                 return -EBADF;
 509
 510         /* verify that this is indeed an fanotify instance */
 511         ret = -EINVAL;
 512         if (unlikely(filp->f_op != &fanotify_fops))
 513                 goto fput_and_out;
 514
 515         ret = fanotify_find_path(dfd, pathname, &path, flags);
 516         if (ret)
 517                 goto fput_and_out;
 518
 519         /* inode held in place by reference to path; group by fget on fd */
 520         if (!(flags & FAN_MARK_MOUNT))
 521                 inode = path.dentry->d_inode;
 522         else
 523                 mnt = path.mnt;
 524         group = filp->private_data;
 525
 526         /* create/update an inode mark */
 527         switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) {
 528         case FAN_MARK_ADD:
 529                 if (flags & FAN_MARK_MOUNT)
 530                         ret = fanotify_add_vfsmount_mark(group, mnt, mask);
 531                 else
 532                         ret = fanotify_add_inode_mark(group, inode, mask);
 533                 break;
 534         case FAN_MARK_REMOVE:
 535                 if (flags & FAN_MARK_MOUNT)
 536                         ret = fanotify_remove_vfsmount_mark(group, mnt, mask);
 537                 else
 538                         ret = fanotify_remove_inode_mark(group, inode, mask);
 539                 break;
 540         default:
 541                 ret = -EINVAL;
 542         }
 543
 544         path_put(&path);
 545 fput_and_out:
 546         fput_light(filp, fput_needed);
 547         return ret;
 548 }
 549
 550 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
 551 asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
 552                                   long dfd, long pathname)
 553 {
 554         return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
 555                                   mask, (int) dfd,
 556                                   (const char  __user *) pathname);
 557 }
 558 SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
 559 #endif
 560
 561 /*
 562  * fanotify_user_setup - Our initialization function.  Note that we cannnot return
 563  * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
 564  * must result in panic().
 565  */
 566 static int __init fanotify_user_setup(void)
 567 {
 568         fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
 569
 570         return 0;
 571 }
 572 device_initcall(fanotify_user_setup);