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) {
 164                 /* we are notifying a parent so come up with the new mask which
 165                  * specifies these are events which came from a child. */
 166                 mask |= FS_EVENT_ON_CHILD;
 167
 168                 if (path)
 169                         ret = fsnotify(p_inode, mask, path, FSNOTIFY_EVENT_PATH,
 170                                        dentry->d_name.name, 0);
 171                 else
 172                         ret = fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
 173                                        dentry->d_name.name, 0);
 174         }
 175
 176         dput(parent);
 177
 178         return ret;
 179 }
 180 EXPORT_SYMBOL_GPL(__fsnotify_parent);
 181
 182 static int send_to_group(struct inode *to_tell,
 183                          struct fsnotify_mark *inode_mark,
 184                          struct fsnotify_mark *vfsmount_mark,
 185                          __u32 mask, const void *data,
 186                          int data_is, u32 cookie,
 187                          const unsigned char *file_name,
 188                          struct fsnotify_iter_info *iter_info)
 189 {
 190         struct fsnotify_group *group = NULL;
 191         __u32 inode_test_mask = 0;
 192         __u32 vfsmount_test_mask = 0;
 193
 194         if (unlikely(!inode_mark && !vfsmount_mark)) {
 195                 BUG();
 196                 return 0;
 197         }
 198
 199         /* clear ignored on inode modification */
 200         if (mask & FS_MODIFY) {
 201                 if (inode_mark &&
 202                     !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
 203                         inode_mark->ignored_mask = 0;
 204                 if (vfsmount_mark &&
 205                     !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
 206                         vfsmount_mark->ignored_mask = 0;
 207         }
 208
 209         /* does the inode mark tell us to do something? */
 210         if (inode_mark) {
 211                 group = inode_mark->group;
 212                 inode_test_mask = (mask & ~FS_EVENT_ON_CHILD);
 213                 inode_test_mask &= inode_mark->mask;
 214                 inode_test_mask &= ~inode_mark->ignored_mask;
 215         }
 216
 217         /* does the vfsmount_mark tell us to do something? */
 218         if (vfsmount_mark) {
 219                 vfsmount_test_mask = (mask & ~FS_EVENT_ON_CHILD);
 220                 group = vfsmount_mark->group;
 221                 vfsmount_test_mask &= vfsmount_mark->mask;
 222                 vfsmount_test_mask &= ~vfsmount_mark->ignored_mask;
 223                 if (inode_mark)
 224                         vfsmount_test_mask &= ~inode_mark->ignored_mask;
 225         }
 226
 227         pr_debug("%s: group=%p to_tell=%p mask=%x inode_mark=%p"
 228                  " inode_test_mask=%x vfsmount_mark=%p vfsmount_test_mask=%x"
 229                  " data=%p data_is=%d cookie=%d\n",
 230                  __func__, group, to_tell, mask, inode_mark,
 231                  inode_test_mask, vfsmount_mark, vfsmount_test_mask, data,
 232                  data_is, cookie);
 233
 234         if (!inode_test_mask && !vfsmount_test_mask)
 235                 return 0;
 236
 237         return group->ops->handle_event(group, to_tell, inode_mark,
 238                                         vfsmount_mark, mask, data, data_is,
 239                                         file_name, cookie, iter_info);
 240 }
 241
 242 /*
 243  * This is the main call to fsnotify.  The VFS calls into hook specific functions
 244  * in linux/fsnotify.h.  Those functions then in turn call here.  Here will call
 245  * out to all of the registered fsnotify_group.  Those groups can then use the
 246  * notification event in whatever means they feel necessary.
 247  */
 248 int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
 249              const unsigned char *file_name, u32 cookie)
 250 {
 251         struct hlist_node *inode_node = NULL, *vfsmount_node = NULL;
 252         struct fsnotify_mark *inode_mark = NULL, *vfsmount_mark = NULL;
 253         struct fsnotify_group *inode_group, *vfsmount_group;
 254         struct fsnotify_mark_connector *inode_conn, *vfsmount_conn;
 255         struct fsnotify_iter_info iter_info;
 256         struct mount *mnt;
 257         int ret = 0;
 258         /* global tests shouldn't care about events on child only the specific event */
 259         __u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
 260
 261         if (data_is == FSNOTIFY_EVENT_PATH)
 262                 mnt = real_mount(((const struct path *)data)->mnt);
 263         else
 264                 mnt = NULL;
 265
 266         /*
 267          * Optimization: srcu_read_lock() has a memory barrier which can
 268          * be expensive.  It protects walking the *_fsnotify_marks lists.
 269          * However, if we do not walk the lists, we do not have to do
 270          * SRCU because we have no references to any objects and do not
 271          * need SRCU to keep them "alive".
 272          */
 273         if (!to_tell->i_fsnotify_marks &&
 274             (!mnt || !mnt->mnt_fsnotify_marks))
 275                 return 0;
 276         /*
 277          * if this is a modify event we may need to clear the ignored masks
 278          * otherwise return if neither the inode nor the vfsmount care about
 279          * this type of event.
 280          */
 281         if (!(mask & FS_MODIFY) &&
 282             !(test_mask & to_tell->i_fsnotify_mask) &&
 283             !(mnt && test_mask & mnt->mnt_fsnotify_mask))
 284                 return 0;
 285
 286         iter_info.srcu_idx = srcu_read_lock(&fsnotify_mark_srcu);
 287
 288         if ((mask & FS_MODIFY) ||
 289             (test_mask & to_tell->i_fsnotify_mask)) {
 290                 inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
 291                                               &fsnotify_mark_srcu);
 292                 if (inode_conn)
 293                         inode_node = srcu_dereference(inode_conn->list.first,
 294                                                       &fsnotify_mark_srcu);
 295         }
 296
 297         if (mnt && ((mask & FS_MODIFY) ||
 298                     (test_mask & mnt->mnt_fsnotify_mask))) {
 299                 inode_conn = srcu_dereference(to_tell->i_fsnotify_marks,
 300                                               &fsnotify_mark_srcu);
 301                 if (inode_conn)
 302                         inode_node = srcu_dereference(inode_conn->list.first,
 303                                                       &fsnotify_mark_srcu);
 304                 vfsmount_conn = srcu_dereference(mnt->mnt_fsnotify_marks,
 305                                                  &fsnotify_mark_srcu);
 306                 if (vfsmount_conn)
 307                         vfsmount_node = srcu_dereference(
 308                                                 vfsmount_conn->list.first,
 309                                                 &fsnotify_mark_srcu);
 310         }
 311
 312         /*
 313          * We need to merge inode & vfsmount mark lists so that inode mark
 314          * ignore masks are properly reflected for mount mark notifications.
 315          * That's why this traversal is so complicated...
 316          */
 317         while (inode_node || vfsmount_node) {
 318                 inode_group = NULL;
 319                 inode_mark = NULL;
 320                 vfsmount_group = NULL;
 321                 vfsmount_mark = NULL;
 322
 323                 if (inode_node) {
 324                         inode_mark = hlist_entry(srcu_dereference(inode_node, &fsnotify_mark_srcu),
 325                                                  struct fsnotify_mark, obj_list);
 326                         inode_group = inode_mark->group;
 327                 }
 328
 329                 if (vfsmount_node) {
 330                         vfsmount_mark = hlist_entry(srcu_dereference(vfsmount_node, &fsnotify_mark_srcu),
 331                                                     struct fsnotify_mark, obj_list);
 332                         vfsmount_group = vfsmount_mark->group;
 333                 }
 334
 335                 if (inode_group && vfsmount_group) {
 336                         int cmp = fsnotify_compare_groups(inode_group,
 337                                                           vfsmount_group);
 338                         if (cmp > 0) {
 339                                 inode_group = NULL;
 340                                 inode_mark = NULL;
 341                         } else if (cmp < 0) {
 342                                 vfsmount_group = NULL;
 343                                 vfsmount_mark = NULL;
 344                         }
 345                 }
 346
 347                 iter_info.inode_mark = inode_mark;
 348                 iter_info.vfsmount_mark = vfsmount_mark;
 349
 350                 ret = send_to_group(to_tell, inode_mark, vfsmount_mark, mask,
 351                                     data, data_is, cookie, file_name,
 352                                     &iter_info);
 353
 354                 if (ret && (mask & ALL_FSNOTIFY_PERM_EVENTS))
 355                         goto out;
 356
 357                 if (inode_group)
 358                         inode_node = srcu_dereference(inode_node->next,
 359                                                       &fsnotify_mark_srcu);
 360                 if (vfsmount_group)
 361                         vfsmount_node = srcu_dereference(vfsmount_node->next,
 362                                                          &fsnotify_mark_srcu);
 363         }
 364         ret = 0;
 365 out:
 366         srcu_read_unlock(&fsnotify_mark_srcu, iter_info.srcu_idx);
 367
 368         return ret;
 369 }
 370 EXPORT_SYMBOL_GPL(fsnotify);
 371
 372 extern struct kmem_cache *fsnotify_mark_connector_cachep;
 373
 374 static __init int fsnotify_init(void)
 375 {
 376         int ret;
 377
 378         BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23);
 379
 380         ret = init_srcu_struct(&fsnotify_mark_srcu);
 381         if (ret)
 382                 panic("initializing fsnotify_mark_srcu");
 383
 384         fsnotify_mark_connector_cachep = KMEM_CACHE(fsnotify_mark_connector,
 385                                                     SLAB_PANIC);
 386
 387         return 0;
 388 }
 389 core_initcall(fsnotify_init);