1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com> 4 */ 5 6 #include <linux/dcache.h> 7 #include <linux/fs.h> 8 #include <linux/gfp.h> 9 #include <linux/init.h> 10 #include <linux/module.h> 11 #include <linux/mount.h> 12 #include <linux/srcu.h> 13 14 #include <linux/fsnotify_backend.h> 15 #include "fsnotify.h" 16 17 /* 18 * Clear all of the marks on an inode when it is being evicted from core 19 */ 20 void __fsnotify_inode_delete(struct inode *inode) 21 { 22 fsnotify_clear_marks_by_inode(inode); 23 } 24 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete); 25 26 void __fsnotify_vfsmount_delete(struct vfsmount *mnt) 27 { 28 fsnotify_clear_marks_by_mount(mnt); 29 } 30 31 /** 32 * fsnotify_unmount_inodes - an sb is unmounting. handle any watched inodes. 33 * @sb: superblock being unmounted. 34 * 35 * Called during unmount with no locks held, so needs to be safe against 36 * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block. 37 */ 38 static void fsnotify_unmount_inodes(struct super_block *sb) 39 { 40 struct inode *inode, *iput_inode = NULL; 41 42 spin_lock(&sb->s_inode_list_lock); 43 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 44 /* 45 * We cannot __iget() an inode in state I_FREEING, 46 * I_WILL_FREE, or I_NEW which is fine because by that point 47 * the inode cannot have any associated watches. 48 */ 49 spin_lock(&inode->i_lock); 50 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { 51 spin_unlock(&inode->i_lock); 52 continue; 53 } 54 55 /* 56 * If i_count is zero, the inode cannot have any watches and 57 * doing an __iget/iput with SB_ACTIVE clear would actually 58 * evict all inodes with zero i_count from icache which is 59 * unnecessarily violent and may in fact be illegal to do. 60 * However, we should have been called /after/ evict_inodes 61 * removed all zero refcount inodes, in any case. Test to 62 * be sure. 63 */ 64 if (!atomic_read(&inode->i_count)) { 65 spin_unlock(&inode->i_lock); 66 continue; 67 } 68 69 __iget(inode); 70 spin_unlock(&inode->i_lock); 71 spin_unlock(&sb->s_inode_list_lock); 72 73 if (iput_inode) 74 iput(iput_inode); 75 76 /* for each watch, send FS_UNMOUNT and then remove it */ 77 fsnotify_inode(inode, FS_UNMOUNT); 78 79 fsnotify_inode_delete(inode); 80 81 iput_inode = inode; 82 83 cond_resched(); 84 spin_lock(&sb->s_inode_list_lock); 85 } 86 spin_unlock(&sb->s_inode_list_lock); 87 88 if (iput_inode) 89 iput(iput_inode); 90 /* Wait for outstanding inode references from connectors */ 91 wait_var_event(&sb->s_fsnotify_inode_refs, 92 !atomic_long_read(&sb->s_fsnotify_inode_refs)); 93 } 94 95 void fsnotify_sb_delete(struct super_block *sb) 96 { 97 fsnotify_unmount_inodes(sb); 98 fsnotify_clear_marks_by_sb(sb); 99 } 100 101 /* 102 * Given an inode, first check if we care what happens to our children. Inotify 103 * and dnotify both tell their parents about events. If we care about any event 104 * on a child we run all of our children and set a dentry flag saying that the 105 * parent cares. Thus when an event happens on a child it can quickly tell if 106 * if there is a need to find a parent and send the event to the parent. 107 */ 108 void __fsnotify_update_child_dentry_flags(struct inode *inode) 109 { 110 struct dentry *alias; 111 int watched; 112 113 if (!S_ISDIR(inode->i_mode)) 114 return; 115 116 /* determine if the children should tell inode about their events */ 117 watched = fsnotify_inode_watches_children(inode); 118 119 spin_lock(&inode->i_lock); 120 /* run all of the dentries associated with this inode. Since this is a 121 * directory, there damn well better only be one item on this list */ 122 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 123 struct dentry *child; 124 125 /* run all of the children of the original inode and fix their 126 * d_flags to indicate parental interest (their parent is the 127 * original inode) */ 128 spin_lock(&alias->d_lock); 129 list_for_each_entry(child, &alias->d_subdirs, d_child) { 130 if (!child->d_inode) 131 continue; 132 133 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED); 134 if (watched) 135 child->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED; 136 else 137 child->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED; 138 spin_unlock(&child->d_lock); 139 } 140 spin_unlock(&alias->d_lock); 141 } 142 spin_unlock(&inode->i_lock); 143 } 144 145 /* Are inode/sb/mount interested in parent and name info with this event? */ 146 static bool fsnotify_event_needs_parent(struct inode *inode, struct mount *mnt, 147 __u32 mask) 148 { 149 __u32 marks_mask = 0; 150 151 /* We only send parent/name to inode/sb/mount for events on non-dir */ 152 if (mask & FS_ISDIR) 153 return false; 154 155 /* 156 * All events that are possible on child can also may be reported with 157 * parent/name info to inode/sb/mount. Otherwise, a watching parent 158 * could result in events reported with unexpected name info to sb/mount. 159 */ 160 BUILD_BUG_ON(FS_EVENTS_POSS_ON_CHILD & ~FS_EVENTS_POSS_TO_PARENT); 161 162 /* Did either inode/sb/mount subscribe for events with parent/name? */ 163 marks_mask |= fsnotify_parent_needed_mask(inode->i_fsnotify_mask); 164 marks_mask |= fsnotify_parent_needed_mask(inode->i_sb->s_fsnotify_mask); 165 if (mnt) 166 marks_mask |= fsnotify_parent_needed_mask(mnt->mnt_fsnotify_mask); 167 168 /* Did they subscribe for this event with parent/name info? */ 169 return mask & marks_mask; 170 } 171 172 /* 173 * Notify this dentry's parent about a child's events with child name info 174 * if parent is watching or if inode/sb/mount are interested in events with 175 * parent and name info. 176 * 177 * Notify only the child without name info if parent is not watching and 178 * inode/sb/mount are not interested in events with parent and name info. 179 */ 180 int __fsnotify_parent(struct dentry *dentry, __u32 mask, const void *data, 181 int data_type) 182 { 183 const struct path *path = fsnotify_data_path(data, data_type); 184 struct mount *mnt = path ? real_mount(path->mnt) : NULL; 185 struct inode *inode = d_inode(dentry); 186 struct dentry *parent; 187 bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED; 188 bool parent_needed, parent_interested; 189 __u32 p_mask; 190 struct inode *p_inode = NULL; 191 struct name_snapshot name; 192 struct qstr *file_name = NULL; 193 int ret = 0; 194 195 /* 196 * Do inode/sb/mount care about parent and name info on non-dir? 197 * Do they care about any event at all? 198 */ 199 if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks && 200 (!mnt || !mnt->mnt_fsnotify_marks) && !parent_watched) 201 return 0; 202 203 parent = NULL; 204 parent_needed = fsnotify_event_needs_parent(inode, mnt, mask); 205 if (!parent_watched && !parent_needed) 206 goto notify; 207 208 /* Does parent inode care about events on children? */ 209 parent = dget_parent(dentry); 210 p_inode = parent->d_inode; 211 p_mask = fsnotify_inode_watches_children(p_inode); 212 if (unlikely(parent_watched && !p_mask)) 213 __fsnotify_update_child_dentry_flags(p_inode); 214 215 /* 216 * Include parent/name in notification either if some notification 217 * groups require parent info or the parent is interested in this event. 218 */ 219 parent_interested = mask & p_mask & ALL_FSNOTIFY_EVENTS; 220 if (parent_needed || parent_interested) { 221 /* When notifying parent, child should be passed as data */ 222 WARN_ON_ONCE(inode != fsnotify_data_inode(data, data_type)); 223 224 /* Notify both parent and child with child name info */ 225 take_dentry_name_snapshot(&name, dentry); 226 file_name = &name.name; 227 if (parent_interested) 228 mask |= FS_EVENT_ON_CHILD; 229 } 230 231 notify: 232 ret = fsnotify(mask, data, data_type, p_inode, file_name, inode, 0); 233 234 if (file_name) 235 release_dentry_name_snapshot(&name); 236 dput(parent); 237 238 return ret; 239 } 240 EXPORT_SYMBOL_GPL(__fsnotify_parent); 241 242 static int fsnotify_handle_inode_event(struct fsnotify_group *group, 243 struct fsnotify_mark *inode_mark, 244 u32 mask, const void *data, int data_type, 245 struct inode *dir, const struct qstr *name, 246 u32 cookie) 247 { 248 const struct path *path = fsnotify_data_path(data, data_type); 249 struct inode *inode = fsnotify_data_inode(data, data_type); 250 const struct fsnotify_ops *ops = group->ops; 251 252 if (WARN_ON_ONCE(!ops->handle_inode_event)) 253 return 0; 254 255 if ((inode_mark->mask & FS_EXCL_UNLINK) && 256 path && d_unlinked(path->dentry)) 257 return 0; 258 259 /* Check interest of this mark in case event was sent with two marks */ 260 if (!(mask & inode_mark->mask & ALL_FSNOTIFY_EVENTS)) 261 return 0; 262 263 return ops->handle_inode_event(inode_mark, mask, inode, dir, name, cookie); 264 } 265 266 static int fsnotify_handle_event(struct fsnotify_group *group, __u32 mask, 267 const void *data, int data_type, 268 struct inode *dir, const struct qstr *name, 269 u32 cookie, struct fsnotify_iter_info *iter_info) 270 { 271 struct fsnotify_mark *inode_mark = fsnotify_iter_inode_mark(iter_info); 272 struct fsnotify_mark *parent_mark = fsnotify_iter_parent_mark(iter_info); 273 int ret; 274 275 if (WARN_ON_ONCE(fsnotify_iter_sb_mark(iter_info)) || 276 WARN_ON_ONCE(fsnotify_iter_vfsmount_mark(iter_info))) 277 return 0; 278 279 if (parent_mark) { 280 /* 281 * parent_mark indicates that the parent inode is watching 282 * children and interested in this event, which is an event 283 * possible on child. But is *this mark* watching children and 284 * interested in this event? 285 */ 286 if (parent_mark->mask & FS_EVENT_ON_CHILD) { 287 ret = fsnotify_handle_inode_event(group, parent_mark, mask, 288 data, data_type, dir, name, 0); 289 if (ret) 290 return ret; 291 } 292 if (!inode_mark) 293 return 0; 294 } 295 296 if (mask & FS_EVENT_ON_CHILD) { 297 /* 298 * Some events can be sent on both parent dir and child marks 299 * (e.g. FS_ATTRIB). If both parent dir and child are 300 * watching, report the event once to parent dir with name (if 301 * interested) and once to child without name (if interested). 302 * The child watcher is expecting an event without a file name 303 * and without the FS_EVENT_ON_CHILD flag. 304 */ 305 mask &= ~FS_EVENT_ON_CHILD; 306 dir = NULL; 307 name = NULL; 308 } 309 310 return fsnotify_handle_inode_event(group, inode_mark, mask, data, data_type, 311 dir, name, cookie); 312 } 313 314 static int send_to_group(__u32 mask, const void *data, int data_type, 315 struct inode *dir, const struct qstr *file_name, 316 u32 cookie, struct fsnotify_iter_info *iter_info) 317 { 318 struct fsnotify_group *group = NULL; 319 __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS); 320 __u32 marks_mask = 0; 321 __u32 marks_ignored_mask = 0; 322 struct fsnotify_mark *mark; 323 int type; 324 325 if (WARN_ON(!iter_info->report_mask)) 326 return 0; 327 328 /* clear ignored on inode modification */ 329 if (mask & FS_MODIFY) { 330 fsnotify_foreach_obj_type(type) { 331 if (!fsnotify_iter_should_report_type(iter_info, type)) 332 continue; 333 mark = iter_info->marks[type]; 334 if (mark && 335 !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY)) 336 mark->ignored_mask = 0; 337 } 338 } 339 340 fsnotify_foreach_obj_type(type) { 341 if (!fsnotify_iter_should_report_type(iter_info, type)) 342 continue; 343 mark = iter_info->marks[type]; 344 /* does the object mark tell us to do something? */ 345 if (mark) { 346 group = mark->group; 347 marks_mask |= mark->mask; 348 marks_ignored_mask |= mark->ignored_mask; 349 } 350 } 351 352 pr_debug("%s: group=%p mask=%x marks_mask=%x marks_ignored_mask=%x data=%p data_type=%d dir=%p cookie=%d\n", 353 __func__, group, mask, marks_mask, marks_ignored_mask, 354 data, data_type, dir, cookie); 355 356 if (!(test_mask & marks_mask & ~marks_ignored_mask)) 357 return 0; 358 359 if (group->ops->handle_event) { 360 return group->ops->handle_event(group, mask, data, data_type, dir, 361 file_name, cookie, iter_info); 362 } 363 364 return fsnotify_handle_event(group, mask, data, data_type, dir, 365 file_name, cookie, iter_info); 366 } 367 368 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp) 369 { 370 struct fsnotify_mark_connector *conn; 371 struct hlist_node *node = NULL; 372 373 conn = srcu_dereference(*connp, &fsnotify_mark_srcu); 374 if (conn) 375 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu); 376 377 return hlist_entry_safe(node, struct fsnotify_mark, obj_list); 378 } 379 380 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark) 381 { 382 struct hlist_node *node = NULL; 383 384 if (mark) 385 node = srcu_dereference(mark->obj_list.next, 386 &fsnotify_mark_srcu); 387 388 return hlist_entry_safe(node, struct fsnotify_mark, obj_list); 389 } 390 391 /* 392 * iter_info is a multi head priority queue of marks. 393 * Pick a subset of marks from queue heads, all with the 394 * same group and set the report_mask for selected subset. 395 * Returns the report_mask of the selected subset. 396 */ 397 static unsigned int fsnotify_iter_select_report_types( 398 struct fsnotify_iter_info *iter_info) 399 { 400 struct fsnotify_group *max_prio_group = NULL; 401 struct fsnotify_mark *mark; 402 int type; 403 404 /* Choose max prio group among groups of all queue heads */ 405 fsnotify_foreach_obj_type(type) { 406 mark = iter_info->marks[type]; 407 if (mark && 408 fsnotify_compare_groups(max_prio_group, mark->group) > 0) 409 max_prio_group = mark->group; 410 } 411 412 if (!max_prio_group) 413 return 0; 414 415 /* Set the report mask for marks from same group as max prio group */ 416 iter_info->report_mask = 0; 417 fsnotify_foreach_obj_type(type) { 418 mark = iter_info->marks[type]; 419 if (mark && 420 fsnotify_compare_groups(max_prio_group, mark->group) == 0) 421 fsnotify_iter_set_report_type(iter_info, type); 422 } 423 424 return iter_info->report_mask; 425 } 426 427 /* 428 * Pop from iter_info multi head queue, the marks that were iterated in the 429 * current iteration step. 430 */ 431 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info) 432 { 433 int type; 434 435 fsnotify_foreach_obj_type(type) { 436 if (fsnotify_iter_should_report_type(iter_info, type)) 437 iter_info->marks[type] = 438 fsnotify_next_mark(iter_info->marks[type]); 439 } 440 } 441 442 /* 443 * fsnotify - This is the main call to fsnotify. 444 * 445 * The VFS calls into hook specific functions in linux/fsnotify.h. 446 * Those functions then in turn call here. Here will call out to all of the 447 * registered fsnotify_group. Those groups can then use the notification event 448 * in whatever means they feel necessary. 449 * 450 * @mask: event type and flags 451 * @data: object that event happened on 452 * @data_type: type of object for fanotify_data_XXX() accessors 453 * @dir: optional directory associated with event - 454 * if @file_name is not NULL, this is the directory that 455 * @file_name is relative to 456 * @file_name: optional file name associated with event 457 * @inode: optional inode associated with event - 458 * either @dir or @inode must be non-NULL. 459 * if both are non-NULL event may be reported to both. 460 * @cookie: inotify rename cookie 461 */ 462 int fsnotify(__u32 mask, const void *data, int data_type, struct inode *dir, 463 const struct qstr *file_name, struct inode *inode, u32 cookie) 464 { 465 const struct path *path = fsnotify_data_path(data, data_type); 466 struct fsnotify_iter_info iter_info = {}; 467 struct super_block *sb; 468 struct mount *mnt = NULL; 469 struct inode *parent = NULL; 470 int ret = 0; 471 __u32 test_mask, marks_mask; 472 473 if (path) 474 mnt = real_mount(path->mnt); 475 476 if (!inode) { 477 /* Dirent event - report on TYPE_INODE to dir */ 478 inode = dir; 479 } else if (mask & FS_EVENT_ON_CHILD) { 480 /* 481 * Event on child - report on TYPE_PARENT to dir if it is 482 * watching children and on TYPE_INODE to child. 483 */ 484 parent = dir; 485 } 486 sb = inode->i_sb; 487 488 /* 489 * Optimization: srcu_read_lock() has a memory barrier which can 490 * be expensive. It protects walking the *_fsnotify_marks lists. 491 * However, if we do not walk the lists, we do not have to do 492 * SRCU because we have no references to any objects and do not 493 * need SRCU to keep them "alive". 494 */ 495 if (!sb->s_fsnotify_marks && 496 (!mnt || !mnt->mnt_fsnotify_marks) && 497 (!inode || !inode->i_fsnotify_marks) && 498 (!parent || !parent->i_fsnotify_marks)) 499 return 0; 500 501 marks_mask = sb->s_fsnotify_mask; 502 if (mnt) 503 marks_mask |= mnt->mnt_fsnotify_mask; 504 if (inode) 505 marks_mask |= inode->i_fsnotify_mask; 506 if (parent) 507 marks_mask |= parent->i_fsnotify_mask; 508 509 510 /* 511 * if this is a modify event we may need to clear the ignored masks 512 * otherwise return if none of the marks care about this type of event. 513 */ 514 test_mask = (mask & ALL_FSNOTIFY_EVENTS); 515 if (!(mask & FS_MODIFY) && !(test_mask & marks_mask)) 516 return 0; 517 518 iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu); 519 520 iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] = 521 fsnotify_first_mark(&sb->s_fsnotify_marks); 522 if (mnt) { 523 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] = 524 fsnotify_first_mark(&mnt->mnt_fsnotify_marks); 525 } 526 if (inode) { 527 iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] = 528 fsnotify_first_mark(&inode->i_fsnotify_marks); 529 } 530 if (parent) { 531 iter_info.marks[FSNOTIFY_OBJ_TYPE_PARENT] = 532 fsnotify_first_mark(&parent->i_fsnotify_marks); 533 } 534 535 /* 536 * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark 537 * ignore masks are properly reflected for mount/sb mark notifications. 538 * That's why this traversal is so complicated... 539 */ 540 while (fsnotify_iter_select_report_types(&iter_info)) { 541 ret = send_to_group(mask, data, data_type, dir, file_name, 542 cookie, &iter_info); 543 544 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS)) 545 goto out; 546 547 fsnotify_iter_next(&iter_info); 548 } 549 ret = 0; 550 out: 551 srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx); 552 553 return ret; 554 } 555 EXPORT_SYMBOL_GPL(fsnotify); 556 557 static __init int fsnotify_init(void) 558 { 559 int ret; 560 561 BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25); 562 563 ret = init_srcu_struct(&fsnotify_mark_srcu); 564 if (ret) 565 panic("initializing fsnotify_mark_srcu"); 566 567 fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector, 568 SLAB_PANIC); 569 570 return 0; 571 } 572 core_initcall(fsnotify_init); 573