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
29int dir_notify_enable __read_mostly = 1;
30
31static struct kmem_cache *dnotify_struct_cache __read_mostly;
32static struct kmem_cache *dnotify_mark_cache __read_mostly;
33static struct fsnotify_group *dnotify_group __read_mostly;
34
35/*
36 * dnotify will attach one of these to each inode (i_fsnotify_marks) which
37 * is being watched by dnotify.  If multiple userspace applications are watching
38 * the same directory with dnotify their information is chained in dn
39 */
40struct dnotify_mark {
41	struct fsnotify_mark fsn_mark;
42	struct dnotify_struct *dn;
43};
44
45/*
46 * When a process starts or stops watching an inode the set of events which
47 * dnotify cares about for that inode may change.  This function runs the
48 * list of everything receiving dnotify events about this directory and calculates
49 * the set of all those events.  After it updates what dnotify is interested in
50 * it calls the fsnotify function so it can update the set of all events relevant
51 * to this inode.
52 */
53static void dnotify_recalc_inode_mask(struct fsnotify_mark *fsn_mark)
54{
55	__u32 new_mask, old_mask;
56	struct dnotify_struct *dn;
57	struct dnotify_mark *dn_mark  = container_of(fsn_mark,
58						     struct dnotify_mark,
59						     fsn_mark);
60
61	assert_spin_locked(&fsn_mark->lock);
62
63	old_mask = fsn_mark->mask;
64	new_mask = 0;
65	for (dn = dn_mark->dn; dn != NULL; dn = dn->dn_next)
66		new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
67	fsnotify_set_mark_mask_locked(fsn_mark, new_mask);
68
69	if (old_mask == new_mask)
70		return;
71
72	if (fsn_mark->inode)
73		fsnotify_recalc_inode_mask(fsn_mark->inode);
74}
75
76/*
77 * Mains fsnotify call where events are delivered to dnotify.
78 * Find the dnotify mark on the relevant inode, run the list of dnotify structs
79 * on that mark and determine which of them has expressed interest in receiving
80 * events of this type.  When found send the correct process and signal and
81 * destroy the dnotify struct if it was not registered to receive multiple
82 * events.
83 */
84static int dnotify_handle_event(struct fsnotify_group *group,
85				struct inode *inode,
86				struct fsnotify_mark *inode_mark,
87				struct fsnotify_mark *vfsmount_mark,
88				u32 mask, void *data, int data_type,
89				const unsigned char *file_name, u32 cookie)
90{
91	struct dnotify_mark *dn_mark;
92	struct dnotify_struct *dn;
93	struct dnotify_struct **prev;
94	struct fown_struct *fown;
95	__u32 test_mask = mask & ~FS_EVENT_ON_CHILD;
96
97	/* not a dir, dnotify doesn't care */
98	if (!S_ISDIR(inode->i_mode))
99		return 0;
100
101	BUG_ON(vfsmount_mark);
102
103	dn_mark = container_of(inode_mark, struct dnotify_mark, fsn_mark);
104
105	spin_lock(&inode_mark->lock);
106	prev = &dn_mark->dn;
107	while ((dn = *prev) != NULL) {
108		if ((dn->dn_mask & test_mask) == 0) {
109			prev = &dn->dn_next;
110			continue;
111		}
112		fown = &dn->dn_filp->f_owner;
113		send_sigio(fown, dn->dn_fd, POLL_MSG);
114		if (dn->dn_mask & FS_DN_MULTISHOT)
115			prev = &dn->dn_next;
116		else {
117			*prev = dn->dn_next;
118			kmem_cache_free(dnotify_struct_cache, dn);
119			dnotify_recalc_inode_mask(inode_mark);
120		}
121	}
122
123	spin_unlock(&inode_mark->lock);
124
125	return 0;
126}
127
128static void dnotify_free_mark(struct fsnotify_mark *fsn_mark)
129{
130	struct dnotify_mark *dn_mark = container_of(fsn_mark,
131						    struct dnotify_mark,
132						    fsn_mark);
133
134	BUG_ON(dn_mark->dn);
135
136	kmem_cache_free(dnotify_mark_cache, dn_mark);
137}
138
139static struct fsnotify_ops dnotify_fsnotify_ops = {
140	.handle_event = dnotify_handle_event,
141};
142
143/*
144 * Called every time a file is closed.  Looks first for a dnotify mark on the
145 * inode.  If one is found run all of the ->dn structures attached to that
146 * mark for one relevant to this process closing the file and remove that
147 * dnotify_struct.  If that was the last dnotify_struct also remove the
148 * fsnotify_mark.
149 */
150void dnotify_flush(struct file *filp, fl_owner_t id)
151{
152	struct fsnotify_mark *fsn_mark;
153	struct dnotify_mark *dn_mark;
154	struct dnotify_struct *dn;
155	struct dnotify_struct **prev;
156	struct inode *inode;
157
158	inode = file_inode(filp);
159	if (!S_ISDIR(inode->i_mode))
160		return;
161
162	fsn_mark = fsnotify_find_inode_mark(dnotify_group, inode);
163	if (!fsn_mark)
164		return;
165	dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
166
167	mutex_lock(&dnotify_group->mark_mutex);
168
169	spin_lock(&fsn_mark->lock);
170	prev = &dn_mark->dn;
171	while ((dn = *prev) != NULL) {
172		if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
173			*prev = dn->dn_next;
174			kmem_cache_free(dnotify_struct_cache, dn);
175			dnotify_recalc_inode_mask(fsn_mark);
176			break;
177		}
178		prev = &dn->dn_next;
179	}
180
181	spin_unlock(&fsn_mark->lock);
182
183	/* nothing else could have found us thanks to the dnotify_groups
184	   mark_mutex */
185	if (dn_mark->dn == NULL)
186		fsnotify_destroy_mark_locked(fsn_mark, dnotify_group);
187
188	mutex_unlock(&dnotify_group->mark_mutex);
189
190	fsnotify_put_mark(fsn_mark);
191}
192
193/* this conversion is done only at watch creation */
194static __u32 convert_arg(unsigned long arg)
195{
196	__u32 new_mask = FS_EVENT_ON_CHILD;
197
198	if (arg & DN_MULTISHOT)
199		new_mask |= FS_DN_MULTISHOT;
200	if (arg & DN_DELETE)
201		new_mask |= (FS_DELETE | FS_MOVED_FROM);
202	if (arg & DN_MODIFY)
203		new_mask |= FS_MODIFY;
204	if (arg & DN_ACCESS)
205		new_mask |= FS_ACCESS;
206	if (arg & DN_ATTRIB)
207		new_mask |= FS_ATTRIB;
208	if (arg & DN_RENAME)
209		new_mask |= FS_DN_RENAME;
210	if (arg & DN_CREATE)
211		new_mask |= (FS_CREATE | FS_MOVED_TO);
212
213	return new_mask;
214}
215
216/*
217 * If multiple processes watch the same inode with dnotify there is only one
218 * dnotify mark in inode->i_fsnotify_marks but we chain a dnotify_struct
219 * onto that mark.  This function either attaches the new dnotify_struct onto
220 * that list, or it |= the mask onto an existing dnofiy_struct.
221 */
222static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark *dn_mark,
223		     fl_owner_t id, int fd, struct file *filp, __u32 mask)
224{
225	struct dnotify_struct *odn;
226
227	odn = dn_mark->dn;
228	while (odn != NULL) {
229		/* adding more events to existing dnofiy_struct? */
230		if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
231			odn->dn_fd = fd;
232			odn->dn_mask |= mask;
233			return -EEXIST;
234		}
235		odn = odn->dn_next;
236	}
237
238	dn->dn_mask = mask;
239	dn->dn_fd = fd;
240	dn->dn_filp = filp;
241	dn->dn_owner = id;
242	dn->dn_next = dn_mark->dn;
243	dn_mark->dn = dn;
244
245	return 0;
246}
247
248/*
249 * When a process calls fcntl to attach a dnotify watch to a directory it ends
250 * up here.  Allocate both a mark for fsnotify to add and a dnotify_struct to be
251 * attached to the fsnotify_mark.
252 */
253int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
254{
255	struct dnotify_mark *new_dn_mark, *dn_mark;
256	struct fsnotify_mark *new_fsn_mark, *fsn_mark;
257	struct dnotify_struct *dn;
258	struct inode *inode;
259	fl_owner_t id = current->files;
260	struct file *f;
261	int destroy = 0, error = 0;
262	__u32 mask;
263
264	/* we use these to tell if we need to kfree */
265	new_fsn_mark = NULL;
266	dn = NULL;
267
268	if (!dir_notify_enable) {
269		error = -EINVAL;
270		goto out_err;
271	}
272
273	/* a 0 mask means we are explicitly removing the watch */
274	if ((arg & ~DN_MULTISHOT) == 0) {
275		dnotify_flush(filp, id);
276		error = 0;
277		goto out_err;
278	}
279
280	/* dnotify only works on directories */
281	inode = file_inode(filp);
282	if (!S_ISDIR(inode->i_mode)) {
283		error = -ENOTDIR;
284		goto out_err;
285	}
286
287	/* expect most fcntl to add new rather than augment old */
288	dn = kmem_cache_alloc(dnotify_struct_cache, GFP_KERNEL);
289	if (!dn) {
290		error = -ENOMEM;
291		goto out_err;
292	}
293
294	/* new fsnotify mark, we expect most fcntl calls to add a new mark */
295	new_dn_mark = kmem_cache_alloc(dnotify_mark_cache, GFP_KERNEL);
296	if (!new_dn_mark) {
297		error = -ENOMEM;
298		goto out_err;
299	}
300
301	/* convert the userspace DN_* "arg" to the internal FS_* defines in fsnotify */
302	mask = convert_arg(arg);
303
304	/* set up the new_fsn_mark and new_dn_mark */
305	new_fsn_mark = &new_dn_mark->fsn_mark;
306	fsnotify_init_mark(new_fsn_mark, dnotify_free_mark);
307	new_fsn_mark->mask = mask;
308	new_dn_mark->dn = NULL;
309
310	/* this is needed to prevent the fcntl/close race described below */
311	mutex_lock(&dnotify_group->mark_mutex);
312
313	/* add the new_fsn_mark or find an old one. */
314	fsn_mark = fsnotify_find_inode_mark(dnotify_group, inode);
315	if (fsn_mark) {
316		dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
317		spin_lock(&fsn_mark->lock);
318	} else {
319		fsnotify_add_mark_locked(new_fsn_mark, dnotify_group, inode,
320					 NULL, 0);
321		spin_lock(&new_fsn_mark->lock);
322		fsn_mark = new_fsn_mark;
323		dn_mark = new_dn_mark;
324		/* we used new_fsn_mark, so don't free it */
325		new_fsn_mark = NULL;
326	}
327
328	rcu_read_lock();
329	f = fcheck(fd);
330	rcu_read_unlock();
331
332	/* if (f != filp) means that we lost a race and another task/thread
333	 * actually closed the fd we are still playing with before we grabbed
334	 * the dnotify_groups mark_mutex and fsn_mark->lock.  Since closing the
335	 * fd is the only time we clean up the marks we need to get our mark
336	 * off the list. */
337	if (f != filp) {
338		/* if we added ourselves, shoot ourselves, it's possible that
339		 * the flush actually did shoot this fsn_mark.  That's fine too
340		 * since multiple calls to destroy_mark is perfectly safe, if
341		 * we found a dn_mark already attached to the inode, just sod
342		 * off silently as the flush at close time dealt with it.
343		 */
344		if (dn_mark == new_dn_mark)
345			destroy = 1;
346		goto out;
347	}
348
349	__f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
350
351	error = attach_dn(dn, dn_mark, id, fd, filp, mask);
352	/* !error means that we attached the dn to the dn_mark, so don't free it */
353	if (!error)
354		dn = NULL;
355	/* -EEXIST means that we didn't add this new dn and used an old one.
356	 * that isn't an error (and the unused dn should be freed) */
357	else if (error == -EEXIST)
358		error = 0;
359
360	dnotify_recalc_inode_mask(fsn_mark);
361out:
362	spin_unlock(&fsn_mark->lock);
363
364	if (destroy)
365		fsnotify_destroy_mark_locked(fsn_mark, dnotify_group);
366
367	mutex_unlock(&dnotify_group->mark_mutex);
368	fsnotify_put_mark(fsn_mark);
369out_err:
370	if (new_fsn_mark)
371		fsnotify_put_mark(new_fsn_mark);
372	if (dn)
373		kmem_cache_free(dnotify_struct_cache, dn);
374	return error;
375}
376
377static int __init dnotify_init(void)
378{
379	dnotify_struct_cache = KMEM_CACHE(dnotify_struct, SLAB_PANIC);
380	dnotify_mark_cache = KMEM_CACHE(dnotify_mark, SLAB_PANIC);
381
382	dnotify_group = fsnotify_alloc_group(&dnotify_fsnotify_ops);
383	if (IS_ERR(dnotify_group))
384		panic("unable to allocate fsnotify group for dnotify\n");
385	return 0;
386}
387
388module_init(dnotify_init)
389