1/* 2 * fs/inotify_user.c - inotify support for userspace 3 * 4 * Authors: 5 * John McCutchan <ttb@tentacle.dhs.org> 6 * Robert Love <rml@novell.com> 7 * 8 * Copyright (C) 2005 John McCutchan 9 * Copyright 2006 Hewlett-Packard Development Company, L.P. 10 * 11 * Copyright (C) 2009 Eric Paris <Red Hat Inc> 12 * inotify was largely rewriten to make use of the fsnotify infrastructure 13 * 14 * This program is free software; you can redistribute it and/or modify it 15 * under the terms of the GNU General Public License as published by the 16 * Free Software Foundation; either version 2, or (at your option) any 17 * later version. 18 * 19 * This program is distributed in the hope that it will be useful, but 20 * WITHOUT ANY WARRANTY; without even the implied warranty of 21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 22 * General Public License for more details. 23 */ 24 25#include <linux/file.h> 26#include <linux/fs.h> /* struct inode */ 27#include <linux/fsnotify_backend.h> 28#include <linux/idr.h> 29#include <linux/init.h> /* module_init */ 30#include <linux/inotify.h> 31#include <linux/kernel.h> /* roundup() */ 32#include <linux/namei.h> /* LOOKUP_FOLLOW */ 33#include <linux/sched.h> /* struct user */ 34#include <linux/slab.h> /* struct kmem_cache */ 35#include <linux/syscalls.h> 36#include <linux/types.h> 37#include <linux/anon_inodes.h> 38#include <linux/uaccess.h> 39#include <linux/poll.h> 40#include <linux/wait.h> 41 42#include "inotify.h" 43 44#include <asm/ioctls.h> 45 46/* these are configurable via /proc/sys/fs/inotify/ */ 47static int inotify_max_user_instances __read_mostly; 48static int inotify_max_queued_events __read_mostly; 49static int inotify_max_user_watches __read_mostly; 50 51static struct kmem_cache *inotify_inode_mark_cachep __read_mostly; 52struct kmem_cache *event_priv_cachep __read_mostly; 53 54#ifdef CONFIG_SYSCTL 55 56#include <linux/sysctl.h> 57 58static int zero; 59 60ctl_table inotify_table[] = { 61 { 62 .procname = "max_user_instances", 63 .data = &inotify_max_user_instances, 64 .maxlen = sizeof(int), 65 .mode = 0644, 66 .proc_handler = proc_dointvec_minmax, 67 .extra1 = &zero, 68 }, 69 { 70 .procname = "max_user_watches", 71 .data = &inotify_max_user_watches, 72 .maxlen = sizeof(int), 73 .mode = 0644, 74 .proc_handler = proc_dointvec_minmax, 75 .extra1 = &zero, 76 }, 77 { 78 .procname = "max_queued_events", 79 .data = &inotify_max_queued_events, 80 .maxlen = sizeof(int), 81 .mode = 0644, 82 .proc_handler = proc_dointvec_minmax, 83 .extra1 = &zero 84 }, 85 { } 86}; 87#endif /* CONFIG_SYSCTL */ 88 89static inline __u32 inotify_arg_to_mask(u32 arg) 90{ 91 __u32 mask; 92 93 /* 94 * everything should accept their own ignored, cares about children, 95 * and should receive events when the inode is unmounted 96 */ 97 mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT); 98 99 /* mask off the flags used to open the fd */ 100 mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK)); 101 102 return mask; 103} 104 105static inline u32 inotify_mask_to_arg(__u32 mask) 106{ 107 return mask & (IN_ALL_EVENTS | IN_ISDIR | IN_UNMOUNT | IN_IGNORED | 108 IN_Q_OVERFLOW); 109} 110 111/* intofiy userspace file descriptor functions */ 112static unsigned int inotify_poll(struct file *file, poll_table *wait) 113{ 114 struct fsnotify_group *group = file->private_data; 115 int ret = 0; 116 117 poll_wait(file, &group->notification_waitq, wait); 118 mutex_lock(&group->notification_mutex); 119 if (!fsnotify_notify_queue_is_empty(group)) 120 ret = POLLIN | POLLRDNORM; 121 mutex_unlock(&group->notification_mutex); 122 123 return ret; 124} 125 126/* 127 * Get an inotify_kernel_event if one exists and is small 128 * enough to fit in "count". Return an error pointer if 129 * not large enough. 130 * 131 * Called with the group->notification_mutex held. 132 */ 133static struct fsnotify_event *get_one_event(struct fsnotify_group *group, 134 size_t count) 135{ 136 size_t event_size = sizeof(struct inotify_event); 137 struct fsnotify_event *event; 138 139 if (fsnotify_notify_queue_is_empty(group)) 140 return NULL; 141 142 event = fsnotify_peek_notify_event(group); 143 144 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 145 146 if (event->name_len) 147 event_size += roundup(event->name_len + 1, event_size); 148 149 if (event_size > count) 150 return ERR_PTR(-EINVAL); 151 152 /* held the notification_mutex the whole time, so this is the 153 * same event we peeked above */ 154 fsnotify_remove_notify_event(group); 155 156 return event; 157} 158 159/* 160 * Copy an event to user space, returning how much we copied. 161 * 162 * We already checked that the event size is smaller than the 163 * buffer we had in "get_one_event()" above. 164 */ 165static ssize_t copy_event_to_user(struct fsnotify_group *group, 166 struct fsnotify_event *event, 167 char __user *buf) 168{ 169 struct inotify_event inotify_event; 170 struct fsnotify_event_private_data *fsn_priv; 171 struct inotify_event_private_data *priv; 172 size_t event_size = sizeof(struct inotify_event); 173 size_t name_len = 0; 174 175 pr_debug("%s: group=%p event=%p\n", __func__, group, event); 176 177 /* we get the inotify watch descriptor from the event private data */ 178 spin_lock(&event->lock); 179 fsn_priv = fsnotify_remove_priv_from_event(group, event); 180 spin_unlock(&event->lock); 181 182 if (!fsn_priv) 183 inotify_event.wd = -1; 184 else { 185 priv = container_of(fsn_priv, struct inotify_event_private_data, 186 fsnotify_event_priv_data); 187 inotify_event.wd = priv->wd; 188 inotify_free_event_priv(fsn_priv); 189 } 190 191 /* 192 * round up event->name_len so it is a multiple of event_size 193 * plus an extra byte for the terminating '\0'. 194 */ 195 if (event->name_len) 196 name_len = roundup(event->name_len + 1, event_size); 197 inotify_event.len = name_len; 198 199 inotify_event.mask = inotify_mask_to_arg(event->mask); 200 inotify_event.cookie = event->sync_cookie; 201 202 /* send the main event */ 203 if (copy_to_user(buf, &inotify_event, event_size)) 204 return -EFAULT; 205 206 buf += event_size; 207 208 /* 209 * fsnotify only stores the pathname, so here we have to send the pathname 210 * and then pad that pathname out to a multiple of sizeof(inotify_event) 211 * with zeros. I get my zeros from the nul_inotify_event. 212 */ 213 if (name_len) { 214 unsigned int len_to_zero = name_len - event->name_len; 215 /* copy the path name */ 216 if (copy_to_user(buf, event->file_name, event->name_len)) 217 return -EFAULT; 218 buf += event->name_len; 219 220 /* fill userspace with 0's */ 221 if (clear_user(buf, len_to_zero)) 222 return -EFAULT; 223 buf += len_to_zero; 224 event_size += name_len; 225 } 226 227 return event_size; 228} 229 230static ssize_t inotify_read(struct file *file, char __user *buf, 231 size_t count, loff_t *pos) 232{ 233 struct fsnotify_group *group; 234 struct fsnotify_event *kevent; 235 char __user *start; 236 int ret; 237 DEFINE_WAIT(wait); 238 239 start = buf; 240 group = file->private_data; 241 242 while (1) { 243 prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE); 244 245 mutex_lock(&group->notification_mutex); 246 kevent = get_one_event(group, count); 247 mutex_unlock(&group->notification_mutex); 248 249 pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent); 250 251 if (kevent) { 252 ret = PTR_ERR(kevent); 253 if (IS_ERR(kevent)) 254 break; 255 ret = copy_event_to_user(group, kevent, buf); 256 fsnotify_put_event(kevent); 257 if (ret < 0) 258 break; 259 buf += ret; 260 count -= ret; 261 continue; 262 } 263 264 ret = -EAGAIN; 265 if (file->f_flags & O_NONBLOCK) 266 break; 267 ret = -EINTR; 268 if (signal_pending(current)) 269 break; 270 271 if (start != buf) 272 break; 273 274 schedule(); 275 } 276 277 finish_wait(&group->notification_waitq, &wait); 278 if (start != buf && ret != -EFAULT) 279 ret = buf - start; 280 return ret; 281} 282 283static int inotify_fasync(int fd, struct file *file, int on) 284{ 285 struct fsnotify_group *group = file->private_data; 286 287 return fasync_helper(fd, file, on, &group->inotify_data.fa) >= 0 ? 0 : -EIO; 288} 289 290static int inotify_release(struct inode *ignored, struct file *file) 291{ 292 struct fsnotify_group *group = file->private_data; 293 struct user_struct *user = group->inotify_data.user; 294 295 pr_debug("%s: group=%p\n", __func__, group); 296 297 fsnotify_clear_marks_by_group(group); 298 299 /* free this group, matching get was inotify_init->fsnotify_obtain_group */ 300 fsnotify_put_group(group); 301 302 atomic_dec(&user->inotify_devs); 303 304 return 0; 305} 306 307static long inotify_ioctl(struct file *file, unsigned int cmd, 308 unsigned long arg) 309{ 310 struct fsnotify_group *group; 311 struct fsnotify_event_holder *holder; 312 struct fsnotify_event *event; 313 void __user *p; 314 int ret = -ENOTTY; 315 size_t send_len = 0; 316 317 group = file->private_data; 318 p = (void __user *) arg; 319 320 pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd); 321 322 switch (cmd) { 323 case FIONREAD: 324 mutex_lock(&group->notification_mutex); 325 list_for_each_entry(holder, &group->notification_list, event_list) { 326 event = holder->event; 327 send_len += sizeof(struct inotify_event); 328 if (event->name_len) 329 send_len += roundup(event->name_len + 1, 330 sizeof(struct inotify_event)); 331 } 332 mutex_unlock(&group->notification_mutex); 333 ret = put_user(send_len, (int __user *) p); 334 break; 335 } 336 337 return ret; 338} 339 340static const struct file_operations inotify_fops = { 341 .poll = inotify_poll, 342 .read = inotify_read, 343 .fasync = inotify_fasync, 344 .release = inotify_release, 345 .unlocked_ioctl = inotify_ioctl, 346 .compat_ioctl = inotify_ioctl, 347}; 348 349 350/* 351 * find_inode - resolve a user-given path to a specific inode 352 */ 353static int inotify_find_inode(const char __user *dirname, struct path *path, unsigned flags) 354{ 355 int error; 356 357 error = user_path_at(AT_FDCWD, dirname, flags, path); 358 if (error) 359 return error; 360 /* you can only watch an inode if you have read permissions on it */ 361 error = inode_permission(path->dentry->d_inode, MAY_READ); 362 if (error) 363 path_put(path); 364 return error; 365} 366 367static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock, 368 int *last_wd, 369 struct inotify_inode_mark *i_mark) 370{ 371 int ret; 372 373 do { 374 if (unlikely(!idr_pre_get(idr, GFP_KERNEL))) 375 return -ENOMEM; 376 377 spin_lock(idr_lock); 378 ret = idr_get_new_above(idr, i_mark, *last_wd + 1, 379 &i_mark->wd); 380 /* we added the mark to the idr, take a reference */ 381 if (!ret) { 382 *last_wd = i_mark->wd; 383 fsnotify_get_mark(&i_mark->fsn_mark); 384 } 385 spin_unlock(idr_lock); 386 } while (ret == -EAGAIN); 387 388 return ret; 389} 390 391static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group, 392 int wd) 393{ 394 struct idr *idr = &group->inotify_data.idr; 395 spinlock_t *idr_lock = &group->inotify_data.idr_lock; 396 struct inotify_inode_mark *i_mark; 397 398 assert_spin_locked(idr_lock); 399 400 i_mark = idr_find(idr, wd); 401 if (i_mark) { 402 struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark; 403 404 fsnotify_get_mark(fsn_mark); 405 /* One ref for being in the idr, one ref we just took */ 406 BUG_ON(atomic_read(&fsn_mark->refcnt) < 2); 407 } 408 409 return i_mark; 410} 411 412static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group, 413 int wd) 414{ 415 struct inotify_inode_mark *i_mark; 416 spinlock_t *idr_lock = &group->inotify_data.idr_lock; 417 418 spin_lock(idr_lock); 419 i_mark = inotify_idr_find_locked(group, wd); 420 spin_unlock(idr_lock); 421 422 return i_mark; 423} 424 425static void do_inotify_remove_from_idr(struct fsnotify_group *group, 426 struct inotify_inode_mark *i_mark) 427{ 428 struct idr *idr = &group->inotify_data.idr; 429 spinlock_t *idr_lock = &group->inotify_data.idr_lock; 430 int wd = i_mark->wd; 431 432 assert_spin_locked(idr_lock); 433 434 idr_remove(idr, wd); 435 436 /* removed from the idr, drop that ref */ 437 fsnotify_put_mark(&i_mark->fsn_mark); 438} 439 440/* 441 * Remove the mark from the idr (if present) and drop the reference 442 * on the mark because it was in the idr. 443 */ 444static void inotify_remove_from_idr(struct fsnotify_group *group, 445 struct inotify_inode_mark *i_mark) 446{ 447 spinlock_t *idr_lock = &group->inotify_data.idr_lock; 448 struct inotify_inode_mark *found_i_mark = NULL; 449 int wd; 450 451 spin_lock(idr_lock); 452 wd = i_mark->wd; 453 454 /* 455 * does this i_mark think it is in the idr? we shouldn't get called 456 * if it wasn't.... 457 */ 458 if (wd == -1) { 459 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" 460 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, 461 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); 462 goto out; 463 } 464 465 /* Lets look in the idr to see if we find it */ 466 found_i_mark = inotify_idr_find_locked(group, wd); 467 if (unlikely(!found_i_mark)) { 468 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" 469 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, 470 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); 471 goto out; 472 } 473 474 /* 475 * We found an mark in the idr at the right wd, but it's 476 * not the mark we were told to remove. eparis seriously 477 * fucked up somewhere. 478 */ 479 if (unlikely(found_i_mark != i_mark)) { 480 WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p " 481 "mark->inode=%p found_i_mark=%p found_i_mark->wd=%d " 482 "found_i_mark->group=%p found_i_mark->inode=%p\n", 483 __func__, i_mark, i_mark->wd, i_mark->fsn_mark.group, 484 i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd, 485 found_i_mark->fsn_mark.group, 486 found_i_mark->fsn_mark.i.inode); 487 goto out; 488 } 489 490 /* 491 * One ref for being in the idr 492 * one ref held by the caller trying to kill us 493 * one ref grabbed by inotify_idr_find 494 */ 495 if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) { 496 printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p" 497 " i_mark->inode=%p\n", __func__, i_mark, i_mark->wd, 498 i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode); 499 /* we can't really recover with bad ref cnting.. */ 500 BUG(); 501 } 502 503 do_inotify_remove_from_idr(group, i_mark); 504out: 505 /* match the ref taken by inotify_idr_find_locked() */ 506 if (found_i_mark) 507 fsnotify_put_mark(&found_i_mark->fsn_mark); 508 i_mark->wd = -1; 509 spin_unlock(idr_lock); 510} 511 512/* 513 * Send IN_IGNORED for this wd, remove this wd from the idr. 514 */ 515void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark, 516 struct fsnotify_group *group) 517{ 518 struct inotify_inode_mark *i_mark; 519 struct fsnotify_event *ignored_event, *notify_event; 520 struct inotify_event_private_data *event_priv; 521 struct fsnotify_event_private_data *fsn_event_priv; 522 int ret; 523 524 ignored_event = fsnotify_create_event(NULL, FS_IN_IGNORED, NULL, 525 FSNOTIFY_EVENT_NONE, NULL, 0, 526 GFP_NOFS); 527 if (!ignored_event) 528 return; 529 530 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); 531 532 event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS); 533 if (unlikely(!event_priv)) 534 goto skip_send_ignore; 535 536 fsn_event_priv = &event_priv->fsnotify_event_priv_data; 537 538 fsn_event_priv->group = group; 539 event_priv->wd = i_mark->wd; 540 541 notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL); 542 if (notify_event) { 543 if (IS_ERR(notify_event)) 544 ret = PTR_ERR(notify_event); 545 else 546 fsnotify_put_event(notify_event); 547 inotify_free_event_priv(fsn_event_priv); 548 } 549 550skip_send_ignore: 551 552 /* matches the reference taken when the event was created */ 553 fsnotify_put_event(ignored_event); 554 555 /* remove this mark from the idr */ 556 inotify_remove_from_idr(group, i_mark); 557 558 atomic_dec(&group->inotify_data.user->inotify_watches); 559} 560 561/* ding dong the mark is dead */ 562static void inotify_free_mark(struct fsnotify_mark *fsn_mark) 563{ 564 struct inotify_inode_mark *i_mark; 565 566 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); 567 568 kmem_cache_free(inotify_inode_mark_cachep, i_mark); 569} 570 571static int inotify_update_existing_watch(struct fsnotify_group *group, 572 struct inode *inode, 573 u32 arg) 574{ 575 struct fsnotify_mark *fsn_mark; 576 struct inotify_inode_mark *i_mark; 577 __u32 old_mask, new_mask; 578 __u32 mask; 579 int add = (arg & IN_MASK_ADD); 580 int ret; 581 582 /* don't allow invalid bits: we don't want flags set */ 583 mask = inotify_arg_to_mask(arg); 584 if (unlikely(!(mask & IN_ALL_EVENTS))) 585 return -EINVAL; 586 587 fsn_mark = fsnotify_find_inode_mark(group, inode); 588 if (!fsn_mark) 589 return -ENOENT; 590 591 i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark); 592 593 spin_lock(&fsn_mark->lock); 594 595 old_mask = fsn_mark->mask; 596 if (add) 597 fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask)); 598 else 599 fsnotify_set_mark_mask_locked(fsn_mark, mask); 600 new_mask = fsn_mark->mask; 601 602 spin_unlock(&fsn_mark->lock); 603 604 if (old_mask != new_mask) { 605 /* more bits in old than in new? */ 606 int dropped = (old_mask & ~new_mask); 607 /* more bits in this fsn_mark than the inode's mask? */ 608 int do_inode = (new_mask & ~inode->i_fsnotify_mask); 609 610 /* update the inode with this new fsn_mark */ 611 if (dropped || do_inode) 612 fsnotify_recalc_inode_mask(inode); 613 614 } 615 616 /* return the wd */ 617 ret = i_mark->wd; 618 619 /* match the get from fsnotify_find_mark() */ 620 fsnotify_put_mark(fsn_mark); 621 622 return ret; 623} 624 625static int inotify_new_watch(struct fsnotify_group *group, 626 struct inode *inode, 627 u32 arg) 628{ 629 struct inotify_inode_mark *tmp_i_mark; 630 __u32 mask; 631 int ret; 632 struct idr *idr = &group->inotify_data.idr; 633 spinlock_t *idr_lock = &group->inotify_data.idr_lock; 634 635 /* don't allow invalid bits: we don't want flags set */ 636 mask = inotify_arg_to_mask(arg); 637 if (unlikely(!(mask & IN_ALL_EVENTS))) 638 return -EINVAL; 639 640 tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL); 641 if (unlikely(!tmp_i_mark)) 642 return -ENOMEM; 643 644 fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark); 645 tmp_i_mark->fsn_mark.mask = mask; 646 tmp_i_mark->wd = -1; 647 648 ret = -ENOSPC; 649 if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches) 650 goto out_err; 651 652 ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd, 653 tmp_i_mark); 654 if (ret) 655 goto out_err; 656 657 /* we are on the idr, now get on the inode */ 658 ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0); 659 if (ret) { 660 /* we failed to get on the inode, get off the idr */ 661 inotify_remove_from_idr(group, tmp_i_mark); 662 goto out_err; 663 } 664 665 /* increment the number of watches the user has */ 666 atomic_inc(&group->inotify_data.user->inotify_watches); 667 668 /* return the watch descriptor for this new mark */ 669 ret = tmp_i_mark->wd; 670 671out_err: 672 /* match the ref from fsnotify_init_mark() */ 673 fsnotify_put_mark(&tmp_i_mark->fsn_mark); 674 675 return ret; 676} 677 678static int inotify_update_watch(struct fsnotify_group *group, struct inode *inode, u32 arg) 679{ 680 int ret = 0; 681 682retry: 683 /* try to update and existing watch with the new arg */ 684 ret = inotify_update_existing_watch(group, inode, arg); 685 /* no mark present, try to add a new one */ 686 if (ret == -ENOENT) 687 ret = inotify_new_watch(group, inode, arg); 688 /* 689 * inotify_new_watch could race with another thread which did an 690 * inotify_new_watch between the update_existing and the add watch 691 * here, go back and try to update an existing mark again. 692 */ 693 if (ret == -EEXIST) 694 goto retry; 695 696 return ret; 697} 698 699static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events) 700{ 701 struct fsnotify_group *group; 702 703 group = fsnotify_alloc_group(&inotify_fsnotify_ops); 704 if (IS_ERR(group)) 705 return group; 706 707 group->max_events = max_events; 708 709 spin_lock_init(&group->inotify_data.idr_lock); 710 idr_init(&group->inotify_data.idr); 711 group->inotify_data.last_wd = 0; 712 group->inotify_data.user = user; 713 group->inotify_data.fa = NULL; 714 715 return group; 716} 717 718 719/* inotify syscalls */ 720SYSCALL_DEFINE1(inotify_init1, int, flags) 721{ 722 struct fsnotify_group *group; 723 struct user_struct *user; 724 int ret; 725 726 /* Check the IN_* constants for consistency. */ 727 BUILD_BUG_ON(IN_CLOEXEC != O_CLOEXEC); 728 BUILD_BUG_ON(IN_NONBLOCK != O_NONBLOCK); 729 730 if (flags & ~(IN_CLOEXEC | IN_NONBLOCK)) 731 return -EINVAL; 732 733 user = get_current_user(); 734 if (unlikely(atomic_read(&user->inotify_devs) >= 735 inotify_max_user_instances)) { 736 ret = -EMFILE; 737 goto out_free_uid; 738 } 739 740 /* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */ 741 group = inotify_new_group(user, inotify_max_queued_events); 742 if (IS_ERR(group)) { 743 ret = PTR_ERR(group); 744 goto out_free_uid; 745 } 746 747 atomic_inc(&user->inotify_devs); 748 749 ret = anon_inode_getfd("inotify", &inotify_fops, group, 750 O_RDONLY | flags); 751 if (ret >= 0) 752 return ret; 753 754 fsnotify_put_group(group); 755 atomic_dec(&user->inotify_devs); 756out_free_uid: 757 free_uid(user); 758 return ret; 759} 760 761SYSCALL_DEFINE0(inotify_init) 762{ 763 return sys_inotify_init1(0); 764} 765 766SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname, 767 u32, mask) 768{ 769 struct fsnotify_group *group; 770 struct inode *inode; 771 struct path path; 772 struct file *filp; 773 int ret, fput_needed; 774 unsigned flags = 0; 775 776 filp = fget_light(fd, &fput_needed); 777 if (unlikely(!filp)) 778 return -EBADF; 779 780 /* verify that this is indeed an inotify instance */ 781 if (unlikely(filp->f_op != &inotify_fops)) { 782 ret = -EINVAL; 783 goto fput_and_out; 784 } 785 786 if (!(mask & IN_DONT_FOLLOW)) 787 flags |= LOOKUP_FOLLOW; 788 if (mask & IN_ONLYDIR) 789 flags |= LOOKUP_DIRECTORY; 790 791 ret = inotify_find_inode(pathname, &path, flags); 792 if (ret) 793 goto fput_and_out; 794 795 /* inode held in place by reference to path; group by fget on fd */ 796 inode = path.dentry->d_inode; 797 group = filp->private_data; 798 799 /* create/update an inode mark */ 800 ret = inotify_update_watch(group, inode, mask); 801 path_put(&path); 802fput_and_out: 803 fput_light(filp, fput_needed); 804 return ret; 805} 806 807SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd) 808{ 809 struct fsnotify_group *group; 810 struct inotify_inode_mark *i_mark; 811 struct file *filp; 812 int ret = 0, fput_needed; 813 814 filp = fget_light(fd, &fput_needed); 815 if (unlikely(!filp)) 816 return -EBADF; 817 818 /* verify that this is indeed an inotify instance */ 819 ret = -EINVAL; 820 if (unlikely(filp->f_op != &inotify_fops)) 821 goto out; 822 823 group = filp->private_data; 824 825 ret = -EINVAL; 826 i_mark = inotify_idr_find(group, wd); 827 if (unlikely(!i_mark)) 828 goto out; 829 830 ret = 0; 831 832 fsnotify_destroy_mark(&i_mark->fsn_mark); 833 834 /* match ref taken by inotify_idr_find */ 835 fsnotify_put_mark(&i_mark->fsn_mark); 836 837out: 838 fput_light(filp, fput_needed); 839 return ret; 840} 841 842/* 843 * inotify_user_setup - Our initialization function. Note that we cannnot return 844 * error because we have compiled-in VFS hooks. So an (unlikely) failure here 845 * must result in panic(). 846 */ 847static int __init inotify_user_setup(void) 848{ 849 BUILD_BUG_ON(IN_ACCESS != FS_ACCESS); 850 BUILD_BUG_ON(IN_MODIFY != FS_MODIFY); 851 BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB); 852 BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE); 853 BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE); 854 BUILD_BUG_ON(IN_OPEN != FS_OPEN); 855 BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM); 856 BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO); 857 BUILD_BUG_ON(IN_CREATE != FS_CREATE); 858 BUILD_BUG_ON(IN_DELETE != FS_DELETE); 859 BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF); 860 BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF); 861 BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT); 862 BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW); 863 BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED); 864 BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK); 865 BUILD_BUG_ON(IN_ISDIR != FS_IN_ISDIR); 866 BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT); 867 868 BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21); 869 870 inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC); 871 event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC); 872 873 inotify_max_queued_events = 16384; 874 inotify_max_user_instances = 128; 875 inotify_max_user_watches = 8192; 876 877 return 0; 878} 879module_init(inotify_user_setup); 880