fs/notify/fsnotify.c

   1 /*
   2  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
   3  *
   4  *  This program is free software; you can redistribute it and/or modify
   5  *  it under the terms of the GNU General Public License as published by
   6  *  the Free Software Foundation; either version 2, or (at your option)
   7  *  any later version.
   8  *
   9  *  This program is distributed in the hope that it will be useful,
  10  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  *  GNU General Public License for more details.
  13  *
  14  *  You should have received a copy of the GNU General Public License
  15  *  along with this program; see the file COPYING.  If not, write to
  16  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  17  */
  18
  19 #include <linux/dcache.h>
  20 #include <linux/fs.h>
  21 #include <linux/gfp.h>
  22 #include <linux/init.h>
  23 #include <linux/module.h>
  24 #include <linux/mount.h>
  25 #include <linux/srcu.h>
  26
  27 #include <linux/fsnotify_backend.h>
  28 #include "fsnotify.h"
  29
  30 /*
  31  * Clear all of the marks on an inode when it is being evicted from core
  32  */
  33 void __fsnotify_inode_delete(struct inode *inode)
  34 {
  35         fsnotify_clear_marks_by_inode(inode);
  36 }
  37 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
  38
  39 void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
  40 {
  41         fsnotify_clear_marks_by_mount(mnt);
  42 }
  43
  44 /**
  45  * fsnotify_unmount_inodes - an sb is unmounting.  handle any watched inodes.
  46  * @sb: superblock being unmounted.
  47  *
  48  * Called during unmount with no locks held, so needs to be safe against
  49  * concurrent modifiers. We temporarily drop sb->s_inode_list_lock and CAN block.
  50  */
  51 void fsnotify_unmount_inodes(struct super_block *sb)
  52 {
  53         struct inode *inode, *iput_inode = NULL;
  54
  55         spin_lock(&sb->s_inode_list_lock);
  56         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
  57                 /*
  58                  * We cannot __iget() an inode in state I_FREEING,
  59                  * I_WILL_FREE, or I_NEW which is fine because by that point
  60                  * the inode cannot have any associated watches.
  61                  */
  62                 spin_lock(&inode->i_lock);
  63                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
  64                         spin_unlock(&inode->i_lock);
  65                         continue;
  66                 }
  67
  68                 /*
  69                  * If i_count is zero, the inode cannot have any watches and
  70                  * doing an __iget/iput with MS_ACTIVE clear would actually
  71                  * evict all inodes with zero i_count from icache which is
  72                  * unnecessarily violent and may in fact be illegal to do.
  73                  */
  74                 if (!atomic_read(&inode->i_count)) {
  75                         spin_unlock(&inode->i_lock);
  76                         continue;
  77                 }
  78
  79                 __iget(inode);
  80                 spin_unlock(&inode->i_lock);
  81                 spin_unlock(&sb->s_inode_list_lock);
  82
  83                 if (iput_inode)
  84                         iput(iput_inode);
  85
  86                 /* for each watch, send FS_UNMOUNT and then remove it */
  87                 fsnotify(inode, FS_UNMOUNT, inode, FSNOTIFY_EVENT_INODE, NULL, 0);
  88
  89                 fsnotify_inode_delete(inode);
  90
  91                 iput_inode = inode;
  92
  93                 spin_lock(&sb->s_inode_list_lock);
  94         }
  95         spin_unlock(&sb->s_inode_list_lock);
  96
  97         if (iput_inode)
  98                 iput(iput_inode);
  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 & ~FS_EVENT_ON_CHILD) {
 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                          struct fsnotify_mark *inode_mark,
 188                          struct fsnotify_mark *vfsmount_mark,
 189                          __u32 mask, const void *data,
 190                          int data_is, u32 cookie,
 191                          const unsigned char *file_name,
 192                          struct fsnotify_iter_info *iter_info)
 193 {
 194         struct fsnotify_group *group = NULL;
 195         __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
 196         __u32 marks_mask = 0;
 197         __u32 marks_ignored_mask = 0;
 198
 199         if (unlikely(!inode_mark && !vfsmount_mark)) {
 200                 BUG();
 201                 return 0;
 202         }
 203
 204         /* clear ignored on inode modification */
 205         if (mask & FS_MODIFY) {
 206                 if (inode_mark &&
 207                     !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
 208                         inode_mark->ignored_mask = 0;
 209                 if (vfsmount_mark &&
 210                     !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
 211                         vfsmount_mark->ignored_mask = 0;
 212         }
 213
 214         /* does the inode mark tell us to do something? */
 215         if (inode_mark) {
 216                 group = inode_mark->group;
 217                 marks_mask |= inode_mark->mask;
 218                 marks_ignored_mask |= inode_mark->ignored_mask;
 219         }
 220
 221         /* does the vfsmount_mark tell us to do something? */
 222         if (vfsmount_mark) {
 223                 group = vfsmount_mark->group;
 224                 marks_mask |= vfsmount_mark->mask;
 225                 marks_ignored_mask |= vfsmount_mark->ignored_mask;
 226         }
 227
 228         pr_debug("%s: group=%p to_tell=%p mask=%x inode_mark=%p"
 229                  " vfsmount_mark=%p marks_mask=%x marks_ignored_mask=%x"
 230                  " data=%p data_is=%d cookie=%d\n",
 231                  __func__, group, to_tell, mask, inode_mark, vfsmount_mark,
 232                  marks_mask, marks_ignored_mask, data,
 233                  data_is, cookie);
 234
 235         if (!(test_mask & marks_mask & ~marks_ignored_mask))
 236                 return 0;
 237
 238         return group->ops->handle_event(group, to_tell, inode_mark,
 239                                         vfsmount_mark, mask, data, data_is,
 240                                         file_name, cookie, iter_info);
 241 }
 242
 243 /*
 244  * This is the main call to fsnotify.  The VFS calls into hook specific functions
 245  * in linux/fsnotify.h.  Those functions then in turn call here.  Here will call
 246  * out to all of the registered fsnotify_group.  Those groups can then use the
 247  * notification event in whatever means they feel necessary.
 248  */
 249 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
 250              const unsigned char *file_name, u32 cookie)
 251 {
 252         struct hlist_node *inode_node = NULL, *vfsmount_node = NULL;
 253         struct fsnotify_mark *inode_mark = NULL, *vfsmount_mark = NULL;
 254         struct fsnotify_group *inode_group, *vfsmount_group;
 255         struct fsnotify_mark_connector *inode_conn, *vfsmount_conn;
 256         struct fsnotify_iter_info iter_info;
 257         struct mount *mnt;
 258         int ret = 0;
 259         /* global tests shouldn't care about events on child only the specific event */
 260         __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
 261
 262         if (data_is == FSNOTIFY_EVENT_PATH)
 263                 mnt = real_mount(((const struct path *)data)->mnt);
 264         else
 265                 mnt = NULL;
 266
 267         /* An event "on child" is not intended for a mount mark */
 268         if (mask & FS_EVENT_ON_CHILD)
 269                 mnt = NULL;
 270
 271         /*
 272          * Optimization: srcu_read_lock() has a memory barrier which can
 273          * be expensive.  It protects walking the *_fsnotify_marks lists.
 274          * However, if we do not walk the lists, we do not have to do
 275          * SRCU because we have no references to any objects and do not
 276          * need SRCU to keep them "alive".
 277          */
 278         if (!to_tell->i_fsnotify_marks &&
 279             (!mnt || !mnt->mnt_fsnotify_marks))
 280                 return 0;
 281         /*
 282          * if this is a modify event we may need to clear the ignored masks
 283          * otherwise return if neither the inode nor the vfsmount care about
 284          * this type of event.
 285          */
 286         if (!(mask & FS_MODIFY) &&
 287             !(test_mask & to_tell->i_fsnotify_mask) &&
 288             !(mnt && test_mask & mnt->mnt_fsnotify_mask))
 289                 return 0;
 290
 291         iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
 292
 293         inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
 294                                       &fsnotify_mark_srcu);
 295         if (inode_conn)
 296                 inode_node = srcu_dereference(inode_conn->list.first,
 297                                               &fsnotify_mark_srcu);
 298
 299         if (mnt) {
 300                 inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
 301                                               &fsnotify_mark_srcu);
 302                 if (inode_conn)
 303                         inode_node = srcu_dereference(inode_conn->list.first,
 304                                                       &fsnotify_mark_srcu);
 305                 vfsmount_conn = srcu_dereference(mnt->mnt_fsnotify_marks,
 306                                                  &fsnotify_mark_srcu);
 307                 if (vfsmount_conn)
 308                         vfsmount_node = srcu_dereference(
 309                                                 vfsmount_conn->list.first,
 310                                                 &fsnotify_mark_srcu);
 311         }
 312
 313         /*
 314          * We need to merge inode & vfsmount mark lists so that inode mark
 315          * ignore masks are properly reflected for mount mark notifications.
 316          * That's why this traversal is so complicated...
 317          */
 318         while (inode_node || vfsmount_node) {
 319                 inode_group = NULL;
 320                 inode_mark = NULL;
 321                 vfsmount_group = NULL;
 322                 vfsmount_mark = NULL;
 323
 324                 if (inode_node) {
 325                         inode_mark = hlist_entry(srcu_dereference(inode_node, &fsnotify_mark_srcu),
 326                                                  struct fsnotify_mark, obj_list);
 327                         inode_group = inode_mark->group;
 328                 }
 329
 330                 if (vfsmount_node) {
 331                         vfsmount_mark = hlist_entry(srcu_dereference(vfsmount_node, &fsnotify_mark_srcu),
 332                                                     struct fsnotify_mark, obj_list);
 333                         vfsmount_group = vfsmount_mark->group;
 334                 }
 335                 /*
 336                  * Need to protect both marks against freeing so that we can
 337                  * continue iteration from this place, regardless of which mark
 338                  * we actually happen to send an event for.
 339                  */
 340                 iter_info.inode_mark = inode_mark;
 341                 iter_info.vfsmount_mark = vfsmount_mark;
 342
 343                 if (inode_group && vfsmount_group) {
 344                         int cmp = fsnotify_compare_groups(inode_group,
 345                                                           vfsmount_group);
 346                         if (cmp > 0) {
 347                                 inode_group = NULL;
 348                                 inode_mark = NULL;
 349                         } else if (cmp < 0) {
 350                                 vfsmount_group = NULL;
 351                                 vfsmount_mark = NULL;
 352                         }
 353                 }
 354
 355                 ret = send_to_group(to_tell, inode_mark, vfsmount_mark, mask,
 356                                     data, data_is, cookie, file_name,
 357                                     &iter_info);
 358
 359                 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
 360                         goto out;
 361
 362                 if (inode_group)
 363                         inode_node = srcu_dereference(inode_node->next,
 364                                                       &fsnotify_mark_srcu);
 365                 if (vfsmount_group)
 366                         vfsmount_node = srcu_dereference(vfsmount_node->next,
 367                                                          &fsnotify_mark_srcu);
 368         }
 369         ret = 0;
 370 out:
 371         srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
 372
 373         return ret;
 374 }
 375 EXPORT_SYMBOL_GPL(fsnotify);
 376
 377 extern struct kmem_cache *fsnotify_mark_connector_cachep;
 378
 379 static __init int fsnotify_init(void)
 380 {
 381         int ret;
 382
 383         BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23);
 384
 385         ret = init_srcu_struct(&fsnotify_mark_srcu);
 386         if (ret)
 387                 panic("initializing fsnotify_mark_srcu");
 388
 389         fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
 390                                                     SLAB_PANIC);
 391
 392         return 0;
 393 }
 394 core_initcall(fsnotify_init);