fs/notify/dnotify/dnotify.c

   1 /*
   2  * Directory notifications for Linux.
   3  *
   4  * Copyright (C) 2000,2001,2002 Stephen Rothwell
   5  *
   6  * Copyright (C) 2009 Eric Paris <Red Hat Inc>
   7  * dnotify was largly rewritten to use the new fsnotify infrastructure
   8  *
   9  * This program is free software; you can redistribute it and/or modify it
  10  * under the terms of the GNU General Public License as published by the
  11  * Free Software Foundation; either version 2, or (at your option) any
  12  * later version.
  13  *
  14  * This program is distributed in the hope that it will be useful, but
  15  * WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * General Public License for more details.
  18  */
  19 #include <linux/fs.h>
  20 #include <linux/module.h>
  21 #include <linux/sched.h>
  22 #include <linux/dnotify.h>
  23 #include <linux/init.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/slab.h>
  26 #include <linux/fdtable.h>
  27 #include <linux/fsnotify_backend.h>
  28
  29 int dir_notify_enable __read_mostly = 1;
  30
  31 static struct kmem_cache *dnotify_struct_cache __read_mostly;
  32 static struct kmem_cache *dnotify_mark_entry_cache __read_mostly;
  33 static struct fsnotify_group *dnotify_group __read_mostly;
  34 static DEFINE_MUTEX(dnotify_mark_mutex);
  35
  36 /*
  37  * dnotify will attach one of these to each inode (i_fsnotify_mark_entries) which
  38  * is being watched by dnotify.  If multiple userspace applications are watching
  39  * the same directory with dnotify their information is chained in dn
  40  */
  41 struct dnotify_mark_entry {
  42         struct fsnotify_mark_entry fsn_entry;
  43         struct dnotify_struct *dn;
  44 };
  45
  46 /*
  47  * When a process starts or stops watching an inode the set of events which
  48  * dnotify cares about for that inode may change.  This function runs the
  49  * list of everything receiving dnotify events about this directory and calculates
  50  * the set of all those events.  After it updates what dnotify is interested in
  51  * it calls the fsnotify function so it can update the set of all events relevant
  52  * to this inode.
  53  */
  54 static void dnotify_recalc_inode_mask(struct fsnotify_mark_entry *entry)
  55 {
  56         __u32 new_mask, old_mask;
  57         struct dnotify_struct *dn;
  58         struct dnotify_mark_entry *dnentry  = container_of(entry,
  59                                                            struct dnotify_mark_entry,
  60                                                            fsn_entry);
  61
  62         assert_spin_locked(&entry->lock);
  63
  64         old_mask = entry->mask;
  65         new_mask = 0;
  66         for (dn = dnentry->dn; dn != NULL; dn = dn->dn_next)
  67                 new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
  68         entry->mask = new_mask;
  69
  70         if (old_mask == new_mask)
  71                 return;
  72
  73         if (entry->inode)
  74                 fsnotify_recalc_inode_mask(entry->inode);
  75 }
  76
  77 /*
  78  * Mains fsnotify call where events are delivered to dnotify.
  79  * Find the dnotify mark on the relevant inode, run the list of dnotify structs
  80  * on that mark and determine which of them has expressed interest in receiving
  81  * events of this type.  When found send the correct process and signal and
  82  * destroy the dnotify struct if it was not registered to receive multiple
  83  * events.
  84  */
  85 static int dnotify_handle_event(struct fsnotify_group *group,
  86                                 struct fsnotify_event *event)
  87 {
  88         struct fsnotify_mark_entry *entry = NULL;
  89         struct dnotify_mark_entry *dnentry;
  90         struct inode *to_tell;
  91         struct dnotify_struct *dn;
  92         struct dnotify_struct **prev;
  93         struct fown_struct *fown;
  94
  95         to_tell = event->to_tell;
  96
  97         spin_lock(&to_tell->i_lock);
  98         entry = fsnotify_find_mark_entry(group, to_tell);
  99         spin_unlock(&to_tell->i_lock);
 100
 101         /* unlikely since we alreay passed dnotify_should_send_event() */
 102         if (unlikely(!entry))
 103                 return 0;
 104         dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 105
 106         spin_lock(&entry->lock);
 107         prev = &dnentry->dn;
 108         while ((dn = *prev) != NULL) {
 109                 if ((dn->dn_mask & event->mask) == 0) {
 110                         prev = &dn->dn_next;
 111                         continue;
 112                 }
 113                 fown = &dn->dn_filp->f_owner;
 114                 send_sigio(fown, dn->dn_fd, POLL_MSG);
 115                 if (dn->dn_mask & FS_DN_MULTISHOT)
 116                         prev = &dn->dn_next;
 117                 else {
 118                         *prev = dn->dn_next;
 119                         kmem_cache_free(dnotify_struct_cache, dn);
 120                         dnotify_recalc_inode_mask(entry);
 121                 }
 122         }
 123
 124         spin_unlock(&entry->lock);
 125         fsnotify_put_mark(entry);
 126
 127         return 0;
 128 }
 129
 130 /*
 131  * Given an inode and mask determine if dnotify would be interested in sending
 132  * userspace notification for that pair.
 133  */
 134 static bool dnotify_should_send_event(struct fsnotify_group *group,
 135                                       struct inode *inode, __u32 mask)
 136 {
 137         struct fsnotify_mark_entry *entry;
 138         bool send;
 139
 140         /* !dir_notify_enable should never get here, don't waste time checking
 141         if (!dir_notify_enable)
 142                 return 0; */
 143
 144         /* not a dir, dnotify doesn't care */
 145         if (!S_ISDIR(inode->i_mode))
 146                 return false;
 147
 148         spin_lock(&inode->i_lock);
 149         entry = fsnotify_find_mark_entry(group, inode);
 150         spin_unlock(&inode->i_lock);
 151
 152         /* no mark means no dnotify watch */
 153         if (!entry)
 154                 return false;
 155
 156         mask = (mask & ~FS_EVENT_ON_CHILD);
 157         send = (mask & entry->mask);
 158
 159         fsnotify_put_mark(entry); /* matches fsnotify_find_mark_entry */
 160
 161         return send;
 162 }
 163
 164 static void dnotify_free_mark(struct fsnotify_mark_entry *entry)
 165 {
 166         struct dnotify_mark_entry *dnentry = container_of(entry,
 167                                                           struct dnotify_mark_entry,
 168                                                           fsn_entry);
 169
 170         BUG_ON(dnentry->dn);
 171
 172         kmem_cache_free(dnotify_mark_entry_cache, dnentry);
 173 }
 174
 175 static struct fsnotify_ops dnotify_fsnotify_ops = {
 176         .handle_event = dnotify_handle_event,
 177         .should_send_event = dnotify_should_send_event,
 178         .free_group_priv = NULL,
 179         .freeing_mark = NULL,
 180         .free_event_priv = NULL,
 181 };
 182
 183 /*
 184  * Called every time a file is closed.  Looks first for a dnotify mark on the
 185  * inode.  If one is found run all of the ->dn entries attached to that
 186  * mark for one relevant to this process closing the file and remove that
 187  * dnotify_struct.  If that was the last dnotify_struct also remove the
 188  * fsnotify_mark_entry.
 189  */
 190 void dnotify_flush(struct file *filp, fl_owner_t id)
 191 {
 192         struct fsnotify_mark_entry *entry;
 193         struct dnotify_mark_entry *dnentry;
 194         struct dnotify_struct *dn;
 195         struct dnotify_struct **prev;
 196         struct inode *inode;
 197
 198         inode = filp->f_path.dentry->d_inode;
 199         if (!S_ISDIR(inode->i_mode))
 200                 return;
 201
 202         spin_lock(&inode->i_lock);
 203         entry = fsnotify_find_mark_entry(dnotify_group, inode);
 204         spin_unlock(&inode->i_lock);
 205         if (!entry)
 206                 return;
 207         dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 208
 209         mutex_lock(&dnotify_mark_mutex);
 210
 211         spin_lock(&entry->lock);
 212         prev = &dnentry->dn;
 213         while ((dn = *prev) != NULL) {
 214                 if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
 215                         *prev = dn->dn_next;
 216                         kmem_cache_free(dnotify_struct_cache, dn);
 217                         dnotify_recalc_inode_mask(entry);
 218                         break;
 219                 }
 220                 prev = &dn->dn_next;
 221         }
 222
 223         spin_unlock(&entry->lock);
 224
 225         /* nothing else could have found us thanks to the dnotify_mark_mutex */
 226         if (dnentry->dn == NULL)
 227                 fsnotify_destroy_mark_by_entry(entry);
 228
 229         fsnotify_recalc_group_mask(dnotify_group);
 230
 231         mutex_unlock(&dnotify_mark_mutex);
 232
 233         fsnotify_put_mark(entry);
 234 }
 235
 236 /* this conversion is done only at watch creation */
 237 static __u32 convert_arg(unsigned long arg)
 238 {
 239         __u32 new_mask = FS_EVENT_ON_CHILD;
 240
 241         if (arg & DN_MULTISHOT)
 242                 new_mask |= FS_DN_MULTISHOT;
 243         if (arg & DN_DELETE)
 244                 new_mask |= (FS_DELETE | FS_MOVED_FROM);
 245         if (arg & DN_MODIFY)
 246                 new_mask |= FS_MODIFY;
 247         if (arg & DN_ACCESS)
 248                 new_mask |= FS_ACCESS;
 249         if (arg & DN_ATTRIB)
 250                 new_mask |= FS_ATTRIB;
 251         if (arg & DN_RENAME)
 252                 new_mask |= FS_DN_RENAME;
 253         if (arg & DN_CREATE)
 254                 new_mask |= (FS_CREATE | FS_MOVED_TO);
 255
 256         return new_mask;
 257 }
 258
 259 /*
 260  * If multiple processes watch the same inode with dnotify there is only one
 261  * dnotify mark in inode->i_fsnotify_mark_entries but we chain a dnotify_struct
 262  * onto that mark.  This function either attaches the new dnotify_struct onto
 263  * that list, or it |= the mask onto an existing dnofiy_struct.
 264  */
 265 static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnentry,
 266                      fl_owner_t id, int fd, struct file *filp, __u32 mask)
 267 {
 268         struct dnotify_struct *odn;
 269
 270         odn = dnentry->dn;
 271         while (odn != NULL) {
 272                 /* adding more events to existing dnofiy_struct? */
 273                 if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
 274                         odn->dn_fd = fd;
 275                         odn->dn_mask |= mask;
 276                         return -EEXIST;
 277                 }
 278                 odn = odn->dn_next;
 279         }
 280
 281         dn->dn_mask = mask;
 282         dn->dn_fd = fd;
 283         dn->dn_filp = filp;
 284         dn->dn_owner = id;
 285         dn->dn_next = dnentry->dn;
 286         dnentry->dn = dn;
 287
 288         return 0;
 289 }
 290
 291 /*
 292  * When a process calls fcntl to attach a dnotify watch to a directory it ends
 293  * up here.  Allocate both a mark for fsnotify to add and a dnotify_struct to be
 294  * attached to the fsnotify_mark.
 295  */
 296 int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 297 {
 298         struct dnotify_mark_entry *new_dnentry, *dnentry;
 299         struct fsnotify_mark_entry *new_entry, *entry;
 300         struct dnotify_struct *dn;
 301         struct inode *inode;
 302         fl_owner_t id = current->files;
 303         struct file *f;
 304         int destroy = 0, error = 0;
 305         __u32 mask;
 306
 307         /* we use these to tell if we need to kfree */
 308         new_entry = NULL;
 309         dn = NULL;
 310
 311         if (!dir_notify_enable) {
 312                 error = -EINVAL;
 313                 goto out_err;
 314         }
 315
 316         /* a 0 mask means we are explicitly removing the watch */
 317         if ((arg & ~DN_MULTISHOT) == 0) {
 318                 dnotify_flush(filp, id);
 319                 error = 0;
 320                 goto out_err;
 321         }
 322
 323         /* dnotify only works on directories */
 324         inode = filp->f_path.dentry->d_inode;
 325         if (!S_ISDIR(inode->i_mode)) {
 326                 error = -ENOTDIR;
 327                 goto out_err;
 328         }
 329
 330         /* expect most fcntl to add new rather than augment old */
 331         dn = kmem_cache_alloc(dnotify_struct_cache, GFP_KERNEL);
 332         if (!dn) {
 333                 error = -ENOMEM;
 334                 goto out_err;
 335         }
 336
 337         /* new fsnotify mark, we expect most fcntl calls to add a new mark */
 338         new_dnentry = kmem_cache_alloc(dnotify_mark_entry_cache, GFP_KERNEL);
 339         if (!new_dnentry) {
 340                 error = -ENOMEM;
 341                 goto out_err;
 342         }
 343
 344         /* convert the userspace DN_* "arg" to the internal FS_* defines in fsnotify */
 345         mask = convert_arg(arg);
 346
 347         /* set up the new_entry and new_dnentry */
 348         new_entry = &new_dnentry->fsn_entry;
 349         fsnotify_init_mark(new_entry, dnotify_free_mark);
 350         new_entry->mask = mask;
 351         new_dnentry->dn = NULL;
 352
 353         /* this is needed to prevent the fcntl/close race described below */
 354         mutex_lock(&dnotify_mark_mutex);
 355
 356         /* add the new_entry or find an old one. */
 357         spin_lock(&inode->i_lock);
 358         entry = fsnotify_find_mark_entry(dnotify_group, inode);
 359         spin_unlock(&inode->i_lock);
 360         if (entry) {
 361                 dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
 362                 spin_lock(&entry->lock);
 363         } else {
 364                 fsnotify_add_mark(new_entry, dnotify_group, inode);
 365                 spin_lock(&new_entry->lock);
 366                 entry = new_entry;
 367                 dnentry = new_dnentry;
 368                 /* we used new_entry, so don't free it */
 369                 new_entry = NULL;
 370         }
 371
 372         rcu_read_lock();
 373         f = fcheck(fd);
 374         rcu_read_unlock();
 375
 376         /* if (f != filp) means that we lost a race and another task/thread
 377          * actually closed the fd we are still playing with before we grabbed
 378          * the dnotify_mark_mutex and entry->lock.  Since closing the fd is the
 379          * only time we clean up the mark entries we need to get our mark off
 380          * the list. */
 381         if (f != filp) {
 382                 /* if we added ourselves, shoot ourselves, it's possible that
 383                  * the flush actually did shoot this entry.  That's fine too
 384                  * since multiple calls to destroy_mark is perfectly safe, if
 385                  * we found a dnentry already attached to the inode, just sod
 386                  * off silently as the flush at close time dealt with it.
 387                  */
 388                 if (dnentry == new_dnentry)
 389                         destroy = 1;
 390                 goto out;
 391         }
 392
 393         error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
 394         if (error) {
 395                 /* if we added, we must shoot */
 396                 if (dnentry == new_dnentry)
 397                         destroy = 1;
 398                 goto out;
 399         }
 400
 401         error = attach_dn(dn, dnentry, id, fd, filp, mask);
 402         /* !error means that we attached the dn to the dnentry, so don't free it */
 403         if (!error)
 404                 dn = NULL;
 405         /* -EEXIST means that we didn't add this new dn and used an old one.
 406          * that isn't an error (and the unused dn should be freed) */
 407         else if (error == -EEXIST)
 408                 error = 0;
 409
 410         dnotify_recalc_inode_mask(entry);
 411 out:
 412         spin_unlock(&entry->lock);
 413
 414         if (destroy)
 415                 fsnotify_destroy_mark_by_entry(entry);
 416
 417         fsnotify_recalc_group_mask(dnotify_group);
 418
 419         mutex_unlock(&dnotify_mark_mutex);
 420         fsnotify_put_mark(entry);
 421 out_err:
 422         if (new_entry)
 423                 fsnotify_put_mark(new_entry);
 424         if (dn)
 425                 kmem_cache_free(dnotify_struct_cache, dn);
 426         return error;
 427 }
 428
 429 static int __init dnotify_init(void)
 430 {
 431         dnotify_struct_cache = KMEM_CACHE(dnotify_struct, SLAB_PANIC);
 432         dnotify_mark_entry_cache = KMEM_CACHE(dnotify_mark_entry, SLAB_PANIC);
 433
 434         dnotify_group = fsnotify_obtain_group(DNOTIFY_GROUP_NUM,
 435                                               0, &dnotify_fsnotify_ops);
 436         if (IS_ERR(dnotify_group))
 437                 panic("unable to allocate fsnotify group for dnotify\n");
 438         return 0;
 439 }
 440
 441 module_init(dnotify_init)