root/fs/notify/fsnotify.c

DEFINITIONS

1. __fsnotify_inode_delete
2. __fsnotify_vfsmount_delete
3. fsnotify_unmount_inodes
4. fsnotify_sb_delete
5. __fsnotify_update_child_dentry_flags
6. __fsnotify_parent
7. send_to_group
8. fsnotify_first_mark
9. fsnotify_next_mark
10. fsnotify_iter_select_report_types
11. fsnotify_iter_next
12. fsnotify
13. fsnotify_init

   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, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
  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 /* Notify this dentry's parent about a child's events. */
 146 int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
 147 {
 148         struct dentry *parent;
 149         struct inode *p_inode;
 150         int ret = 0;
 151 
 152         if (!dentry)
 153                 dentry = path->dentry;
 154 
 155         if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
 156                 return 0;
 157 
 158         parent = dget_parent(dentry);
 159         p_inode = parent->d_inode;
 160 
 161         if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
 162                 __fsnotify_update_child_dentry_flags(p_inode);
 163         } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
 164                 struct name_snapshot name;
 165 
 166                 /* we are notifying a parent so come up with the new mask which
 167                  * specifies these are events which came from a child. */
 168                 mask |= FS_EVENT_ON_CHILD;
 169 
 170                 take_dentry_name_snapshot(&name, dentry);
 171                 if (path)
 172                         ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
 173                                        &name.name, 0);
 174                 else
 175                         ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
 176                                        &name.name, 0);
 177                 release_dentry_name_snapshot(&name);
 178         }
 179 
 180         dput(parent);
 181 
 182         return ret;
 183 }
 184 EXPORT_SYMBOL_GPL(__fsnotify_parent);
 185 
 186 static int send_to_group(struct inode *to_tell,
 187                          __u32 mask, const void *data,
 188                          int data_is, u32 cookie,
 189                          const struct qstr *file_name,
 190                          struct fsnotify_iter_info *iter_info)
 191 {
 192         struct fsnotify_group *group = NULL;
 193         __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
 194         __u32 marks_mask = 0;
 195         __u32 marks_ignored_mask = 0;
 196         struct fsnotify_mark *mark;
 197         int type;
 198 
 199         if (WARN_ON(!iter_info->report_mask))
 200                 return 0;
 201 
 202         /* clear ignored on inode modification */
 203         if (mask & FS_MODIFY) {
 204                 fsnotify_foreach_obj_type(type) {
 205                         if (!fsnotify_iter_should_report_type(iter_info, type))
 206                                 continue;
 207                         mark = iter_info->marks[type];
 208                         if (mark &&
 209                             !(mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
 210                                 mark->ignored_mask = 0;
 211                 }
 212         }
 213 
 214         fsnotify_foreach_obj_type(type) {
 215                 if (!fsnotify_iter_should_report_type(iter_info, type))
 216                         continue;
 217                 mark = iter_info->marks[type];
 218                 /* does the object mark tell us to do something? */
 219                 if (mark) {
 220                         group = mark->group;
 221                         marks_mask |= mark->mask;
 222                         marks_ignored_mask |= mark->ignored_mask;
 223                 }
 224         }
 225 
 226         pr_debug("%s: group=%p to_tell=%p mask=%x marks_mask=%x marks_ignored_mask=%x"
 227                  " data=%p data_is=%d cookie=%d\n",
 228                  __func__, group, to_tell, mask, marks_mask, marks_ignored_mask,
 229                  data, data_is, cookie);
 230 
 231         if (!(test_mask & marks_mask & ~marks_ignored_mask))
 232                 return 0;
 233 
 234         return group->ops->handle_event(group, to_tell, mask, data, data_is,
 235                                         file_name, cookie, iter_info);
 236 }
 237 
 238 static struct fsnotify_mark *fsnotify_first_mark(struct fsnotify_mark_connector **connp)
 239 {
 240         struct fsnotify_mark_connector *conn;
 241         struct hlist_node *node = NULL;
 242 
 243         conn = srcu_dereference(*connp, &fsnotify_mark_srcu);
 244         if (conn)
 245                 node = srcu_dereference(conn->list.first, &fsnotify_mark_srcu);
 246 
 247         return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
 248 }
 249 
 250 static struct fsnotify_mark *fsnotify_next_mark(struct fsnotify_mark *mark)
 251 {
 252         struct hlist_node *node = NULL;
 253 
 254         if (mark)
 255                 node = srcu_dereference(mark->obj_list.next,
 256                                         &fsnotify_mark_srcu);
 257 
 258         return hlist_entry_safe(node, struct fsnotify_mark, obj_list);
 259 }
 260 
 261 /*
 262  * iter_info is a multi head priority queue of marks.
 263  * Pick a subset of marks from queue heads, all with the
 264  * same group and set the report_mask for selected subset.
 265  * Returns the report_mask of the selected subset.
 266  */
 267 static unsigned int fsnotify_iter_select_report_types(
 268                 struct fsnotify_iter_info *iter_info)
 269 {
 270         struct fsnotify_group *max_prio_group = NULL;
 271         struct fsnotify_mark *mark;
 272         int type;
 273 
 274         /* Choose max prio group among groups of all queue heads */
 275         fsnotify_foreach_obj_type(type) {
 276                 mark = iter_info->marks[type];
 277                 if (mark &&
 278                     fsnotify_compare_groups(max_prio_group, mark->group) > 0)
 279                         max_prio_group = mark->group;
 280         }
 281 
 282         if (!max_prio_group)
 283                 return 0;
 284 
 285         /* Set the report mask for marks from same group as max prio group */
 286         iter_info->report_mask = 0;
 287         fsnotify_foreach_obj_type(type) {
 288                 mark = iter_info->marks[type];
 289                 if (mark &&
 290                     fsnotify_compare_groups(max_prio_group, mark->group) == 0)
 291                         fsnotify_iter_set_report_type(iter_info, type);
 292         }
 293 
 294         return iter_info->report_mask;
 295 }
 296 
 297 /*
 298  * Pop from iter_info multi head queue, the marks that were iterated in the
 299  * current iteration step.
 300  */
 301 static void fsnotify_iter_next(struct fsnotify_iter_info *iter_info)
 302 {
 303         int type;
 304 
 305         fsnotify_foreach_obj_type(type) {
 306                 if (fsnotify_iter_should_report_type(iter_info, type))
 307                         iter_info->marks[type] =
 308                                 fsnotify_next_mark(iter_info->marks[type]);
 309         }
 310 }
 311 
 312 /*
 313  * This is the main call to fsnotify.  The VFS calls into hook specific functions
 314  * in linux/fsnotify.h.  Those functions then in turn call here.  Here will call
 315  * out to all of the registered fsnotify_group.  Those groups can then use the
 316  * notification event in whatever means they feel necessary.
 317  */
 318 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
 319              const struct qstr *file_name, u32 cookie)
 320 {
 321         struct fsnotify_iter_info iter_info = {};
 322         struct super_block *sb = to_tell->i_sb;
 323         struct mount *mnt = NULL;
 324         __u32 mnt_or_sb_mask = sb->s_fsnotify_mask;
 325         int ret = 0;
 326         __u32 test_mask = (mask & ALL_FSNOTIFY_EVENTS);
 327 
 328         if (data_is == FSNOTIFY_EVENT_PATH) {
 329                 mnt = real_mount(((const struct path *)data)->mnt);
 330                 mnt_or_sb_mask |= mnt->mnt_fsnotify_mask;
 331         }
 332         /* An event "on child" is not intended for a mount/sb mark */
 333         if (mask & FS_EVENT_ON_CHILD)
 334                 mnt_or_sb_mask = 0;
 335 
 336         /*
 337          * Optimization: srcu_read_lock() has a memory barrier which can
 338          * be expensive.  It protects walking the *_fsnotify_marks lists.
 339          * However, if we do not walk the lists, we do not have to do
 340          * SRCU because we have no references to any objects and do not
 341          * need SRCU to keep them "alive".
 342          */
 343         if (!to_tell->i_fsnotify_marks && !sb->s_fsnotify_marks &&
 344             (!mnt || !mnt->mnt_fsnotify_marks))
 345                 return 0;
 346         /*
 347          * if this is a modify event we may need to clear the ignored masks
 348          * otherwise return if neither the inode nor the vfsmount/sb care about
 349          * this type of event.
 350          */
 351         if (!(mask & FS_MODIFY) &&
 352             !(test_mask & (to_tell->i_fsnotify_mask | mnt_or_sb_mask)))
 353                 return 0;
 354 
 355         iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
 356 
 357         iter_info.marks[FSNOTIFY_OBJ_TYPE_INODE] =
 358                 fsnotify_first_mark(&to_tell->i_fsnotify_marks);
 359         iter_info.marks[FSNOTIFY_OBJ_TYPE_SB] =
 360                 fsnotify_first_mark(&sb->s_fsnotify_marks);
 361         if (mnt) {
 362                 iter_info.marks[FSNOTIFY_OBJ_TYPE_VFSMOUNT] =
 363                         fsnotify_first_mark(&mnt->mnt_fsnotify_marks);
 364         }
 365 
 366         /*
 367          * We need to merge inode/vfsmount/sb mark lists so that e.g. inode mark
 368          * ignore masks are properly reflected for mount/sb mark notifications.
 369          * That's why this traversal is so complicated...
 370          */
 371         while (fsnotify_iter_select_report_types(&iter_info)) {
 372                 ret = send_to_group(to_tell, mask, data, data_is, cookie,
 373                                     file_name, &iter_info);
 374 
 375                 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
 376                         goto out;
 377 
 378                 fsnotify_iter_next(&iter_info);
 379         }
 380         ret = 0;
 381 out:
 382         srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
 383 
 384         return ret;
 385 }
 386 EXPORT_SYMBOL_GPL(fsnotify);
 387 
 388 extern struct kmem_cache *fsnotify_mark_connector_cachep;
 389 
 390 static __init int fsnotify_init(void)
 391 {
 392         int ret;
 393 
 394         BUILD_BUG_ON(HWEIGHT32(ALL_FSNOTIFY_BITS) != 25);
 395 
 396         ret = init_srcu_struct(&fsnotify_mark_srcu);
 397         if (ret)
 398                 panic("initializing fsnotify_mark_srcu");
 399 
 400         fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
 401                                                     SLAB_PANIC);
 402 
 403         return 0;
 404 }
 405 core_initcall(fsnotify_init);