aboutsummaryrefslogtreecommitdiffstats
path: root/fs/notify/dnotify/dnotify.c
diff options
context:
space:
mode:
authorEric Paris <eparis@redhat.com>2009-05-21 17:01:33 -0400
committerEric Paris <eparis@redhat.com>2009-06-11 14:57:53 -0400
commit3c5119c05d624f95f4967d16b38c9624b816bdb9 (patch)
tree0b5f66106aea38e52adf62958762b0a975607322 /fs/notify/dnotify/dnotify.c
parentc28f7e56e9d95fb531dc3be8df2e7f52bee76d21 (diff)
dnotify: reimplement dnotify using fsnotify
Reimplement dnotify using fsnotify. Signed-off-by: Eric Paris <eparis@redhat.com> Acked-by: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@lst.de>
Diffstat (limited to 'fs/notify/dnotify/dnotify.c')
-rw-r--r--fs/notify/dnotify/dnotify.c469
1 files changed, 362 insertions, 107 deletions
diff --git a/fs/notify/dnotify/dnotify.c b/fs/notify/dnotify/dnotify.c
index b0aa2cde80bd..d9d80f502c6f 100644
--- a/fs/notify/dnotify/dnotify.c
+++ b/fs/notify/dnotify/dnotify.c
@@ -3,6 +3,9 @@
3 * 3 *
4 * Copyright (C) 2000,2001,2002 Stephen Rothwell 4 * Copyright (C) 2000,2001,2002 Stephen Rothwell
5 * 5 *
6 * Copyright (C) 2009 Eric Paris <Red Hat Inc>
7 * dnotify was largly rewritten to use the new fsnotify infrastructure
8 *
6 * This program is free software; you can redistribute it and/or modify it 9 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the 10 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; either version 2, or (at your option) any 11 * Free Software Foundation; either version 2, or (at your option) any
@@ -21,24 +24,178 @@
21#include <linux/spinlock.h> 24#include <linux/spinlock.h>
22#include <linux/slab.h> 25#include <linux/slab.h>
23#include <linux/fdtable.h> 26#include <linux/fdtable.h>
27#include <linux/fsnotify_backend.h>
24 28
25int dir_notify_enable __read_mostly = 1; 29int dir_notify_enable __read_mostly = 1;
26 30
27static struct kmem_cache *dn_cache __read_mostly; 31static struct kmem_cache *dnotify_struct_cache __read_mostly;
32static struct kmem_cache *dnotify_mark_entry_cache __read_mostly;
33static struct fsnotify_group *dnotify_group __read_mostly;
34static 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 */
41struct dnotify_mark_entry {
42 struct fsnotify_mark_entry fsn_entry;
43 struct dnotify_struct *dn;
44};
28 45
29static void redo_inode_mask(struct inode *inode) 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 */
54static void dnotify_recalc_inode_mask(struct fsnotify_mark_entry *entry)
30{ 55{
31 unsigned long new_mask; 56 __u32 new_mask, old_mask;
32 struct dnotify_struct *dn; 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);
33 63
64 old_mask = entry->mask;
34 new_mask = 0; 65 new_mask = 0;
35 for (dn = inode->i_dnotify; dn != NULL; dn = dn->dn_next) 66 for (dn = dnentry->dn; dn != NULL; dn = dn->dn_next)
36 new_mask |= dn->dn_mask & ~DN_MULTISHOT; 67 new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
37 inode->i_dnotify_mask = new_mask; 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);
38} 75}
39 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 */
85static 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 */
134static 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 spin_lock(&entry->lock);
157 send = (mask & entry->mask) ? true : false;
158 spin_unlock(&entry->lock);
159 fsnotify_put_mark(entry); /* matches fsnotify_find_mark_entry */
160
161 return send;
162}
163
164static void dnotify_freeing_mark(struct fsnotify_mark_entry *entry,
165 struct fsnotify_group *group)
166{
167 /* dnotify doesn't care than an inode is on the way out */
168}
169
170static void dnotify_free_mark(struct fsnotify_mark_entry *entry)
171{
172 struct dnotify_mark_entry *dnentry = container_of(entry,
173 struct dnotify_mark_entry,
174 fsn_entry);
175
176 BUG_ON(dnentry->dn);
177
178 kmem_cache_free(dnotify_mark_entry_cache, dnentry);
179}
180
181static struct fsnotify_ops dnotify_fsnotify_ops = {
182 .handle_event = dnotify_handle_event,
183 .should_send_event = dnotify_should_send_event,
184 .free_group_priv = NULL,
185 .freeing_mark = dnotify_freeing_mark,
186};
187
188/*
189 * Called every time a file is closed. Looks first for a dnotify mark on the
190 * inode. If one is found run all of the ->dn entries attached to that
191 * mark for one relevant to this process closing the file and remove that
192 * dnotify_struct. If that was the last dnotify_struct also remove the
193 * fsnotify_mark_entry.
194 */
40void dnotify_flush(struct file *filp, fl_owner_t id) 195void dnotify_flush(struct file *filp, fl_owner_t id)
41{ 196{
197 struct fsnotify_mark_entry *entry;
198 struct dnotify_mark_entry *dnentry;
42 struct dnotify_struct *dn; 199 struct dnotify_struct *dn;
43 struct dnotify_struct **prev; 200 struct dnotify_struct **prev;
44 struct inode *inode; 201 struct inode *inode;
@@ -46,145 +203,243 @@ void dnotify_flush(struct file *filp, fl_owner_t id)
46 inode = filp->f_path.dentry->d_inode; 203 inode = filp->f_path.dentry->d_inode;
47 if (!S_ISDIR(inode->i_mode)) 204 if (!S_ISDIR(inode->i_mode))
48 return; 205 return;
206
49 spin_lock(&inode->i_lock); 207 spin_lock(&inode->i_lock);
50 prev = &inode->i_dnotify; 208 entry = fsnotify_find_mark_entry(dnotify_group, inode);
209 spin_unlock(&inode->i_lock);
210 if (!entry)
211 return;
212 dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
213
214 mutex_lock(&dnotify_mark_mutex);
215
216 spin_lock(&entry->lock);
217 prev = &dnentry->dn;
51 while ((dn = *prev) != NULL) { 218 while ((dn = *prev) != NULL) {
52 if ((dn->dn_owner == id) && (dn->dn_filp == filp)) { 219 if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
53 *prev = dn->dn_next; 220 *prev = dn->dn_next;
54 redo_inode_mask(inode); 221 kmem_cache_free(dnotify_struct_cache, dn);
55 kmem_cache_free(dn_cache, dn); 222 dnotify_recalc_inode_mask(entry);
56 break; 223 break;
57 } 224 }
58 prev = &dn->dn_next; 225 prev = &dn->dn_next;
59 } 226 }
60 spin_unlock(&inode->i_lock); 227
228 spin_unlock(&entry->lock);
229
230 /* nothing else could have found us thanks to the dnotify_mark_mutex */
231 if (dnentry->dn == NULL)
232 fsnotify_destroy_mark_by_entry(entry);
233
234 fsnotify_recalc_group_mask(dnotify_group);
235
236 mutex_unlock(&dnotify_mark_mutex);
237
238 fsnotify_put_mark(entry);
239}
240
241/* this conversion is done only at watch creation */
242static __u32 convert_arg(unsigned long arg)
243{
244 __u32 new_mask = FS_EVENT_ON_CHILD;
245
246 if (arg & DN_MULTISHOT)
247 new_mask |= FS_DN_MULTISHOT;
248 if (arg & DN_DELETE)
249 new_mask |= (FS_DELETE | FS_MOVED_FROM);
250 if (arg & DN_MODIFY)
251 new_mask |= FS_MODIFY;
252 if (arg & DN_ACCESS)
253 new_mask |= FS_ACCESS;
254 if (arg & DN_ATTRIB)
255 new_mask |= FS_ATTRIB;
256 if (arg & DN_RENAME)
257 new_mask |= FS_DN_RENAME;
258 if (arg & DN_CREATE)
259 new_mask |= (FS_CREATE | FS_MOVED_TO);
260
261 return new_mask;
61} 262}
62 263
264/*
265 * If multiple processes watch the same inode with dnotify there is only one
266 * dnotify mark in inode->i_fsnotify_mark_entries but we chain a dnotify_struct
267 * onto that mark. This function either attaches the new dnotify_struct onto
268 * that list, or it |= the mask onto an existing dnofiy_struct.
269 */
270static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnentry,
271 fl_owner_t id, int fd, struct file *filp, __u32 mask)
272{
273 struct dnotify_struct *odn;
274
275 odn = dnentry->dn;
276 while (odn != NULL) {
277 /* adding more events to existing dnofiy_struct? */
278 if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
279 odn->dn_fd = fd;
280 odn->dn_mask |= mask;
281 return -EEXIST;
282 }
283 odn = odn->dn_next;
284 }
285
286 dn->dn_mask = mask;
287 dn->dn_fd = fd;
288 dn->dn_filp = filp;
289 dn->dn_owner = id;
290 dn->dn_next = dnentry->dn;
291 dnentry->dn = dn;
292
293 return 0;
294}
295
296/*
297 * When a process calls fcntl to attach a dnotify watch to a directory it ends
298 * up here. Allocate both a mark for fsnotify to add and a dnotify_struct to be
299 * attached to the fsnotify_mark.
300 */
63int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg) 301int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
64{ 302{
303 struct dnotify_mark_entry *new_dnentry, *dnentry;
304 struct fsnotify_mark_entry *new_entry, *entry;
65 struct dnotify_struct *dn; 305 struct dnotify_struct *dn;
66 struct dnotify_struct *odn;
67 struct dnotify_struct **prev;
68 struct inode *inode; 306 struct inode *inode;
69 fl_owner_t id = current->files; 307 fl_owner_t id = current->files;
70 struct file *f; 308 struct file *f;
71 int error = 0; 309 int destroy = 0, error = 0;
310 __u32 mask;
311
312 /* we use these to tell if we need to kfree */
313 new_entry = NULL;
314 dn = NULL;
72 315
316 if (!dir_notify_enable) {
317 error = -EINVAL;
318 goto out_err;
319 }
320
321 /* a 0 mask means we are explicitly removing the watch */
73 if ((arg & ~DN_MULTISHOT) == 0) { 322 if ((arg & ~DN_MULTISHOT) == 0) {
74 dnotify_flush(filp, id); 323 dnotify_flush(filp, id);
75 return 0; 324 error = 0;
325 goto out_err;
76 } 326 }
77 if (!dir_notify_enable) 327
78 return -EINVAL; 328 /* dnotify only works on directories */
79 inode = filp->f_path.dentry->d_inode; 329 inode = filp->f_path.dentry->d_inode;
80 if (!S_ISDIR(inode->i_mode)) 330 if (!S_ISDIR(inode->i_mode)) {
81 return -ENOTDIR; 331 error = -ENOTDIR;
82 dn = kmem_cache_alloc(dn_cache, GFP_KERNEL); 332 goto out_err;
83 if (dn == NULL)
84 return -ENOMEM;
85 spin_lock(&inode->i_lock);
86 prev = &inode->i_dnotify;
87 while ((odn = *prev) != NULL) {
88 if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
89 odn->dn_fd = fd;
90 odn->dn_mask |= arg;
91 inode->i_dnotify_mask |= arg & ~DN_MULTISHOT;
92 goto out_free;
93 }
94 prev = &odn->dn_next;
95 } 333 }
96 334
97 rcu_read_lock(); 335 /* expect most fcntl to add new rather than augment old */
98 f = fcheck(fd); 336 dn = kmem_cache_alloc(dnotify_struct_cache, GFP_KERNEL);
99 rcu_read_unlock(); 337 if (!dn) {
100 /* we'd lost the race with close(), sod off silently */ 338 error = -ENOMEM;
101 /* note that inode->i_lock prevents reordering problems 339 goto out_err;
102 * between accesses to descriptor table and ->i_dnotify */ 340 }
103 if (f != filp)
104 goto out_free;
105 341
106 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0); 342 /* new fsnotify mark, we expect most fcntl calls to add a new mark */
107 if (error) 343 new_dnentry = kmem_cache_alloc(dnotify_mark_entry_cache, GFP_KERNEL);
108 goto out_free; 344 if (!new_dnentry) {
345 error = -ENOMEM;
346 goto out_err;
347 }
109 348
110 dn->dn_mask = arg; 349 /* convert the userspace DN_* "arg" to the internal FS_* defines in fsnotify */
111 dn->dn_fd = fd; 350 mask = convert_arg(arg);
112 dn->dn_filp = filp;
113 dn->dn_owner = id;
114 inode->i_dnotify_mask |= arg & ~DN_MULTISHOT;
115 dn->dn_next = inode->i_dnotify;
116 inode->i_dnotify = dn;
117 spin_unlock(&inode->i_lock);
118 return 0;
119 351
120out_free: 352 /* set up the new_entry and new_dnentry */
121 spin_unlock(&inode->i_lock); 353 new_entry = &new_dnentry->fsn_entry;
122 kmem_cache_free(dn_cache, dn); 354 fsnotify_init_mark(new_entry, dnotify_free_mark);
123 return error; 355 new_entry->mask = mask;
124} 356 new_dnentry->dn = NULL;
125 357
126void __inode_dir_notify(struct inode *inode, unsigned long event) 358 /* this is needed to prevent the fcntl/close race described below */
127{ 359 mutex_lock(&dnotify_mark_mutex);
128 struct dnotify_struct * dn;
129 struct dnotify_struct **prev;
130 struct fown_struct * fown;
131 int changed = 0;
132 360
361 /* add the new_entry or find an old one. */
133 spin_lock(&inode->i_lock); 362 spin_lock(&inode->i_lock);
134 prev = &inode->i_dnotify; 363 entry = fsnotify_find_mark_entry(dnotify_group, inode);
135 while ((dn = *prev) != NULL) {
136 if ((dn->dn_mask & event) == 0) {
137 prev = &dn->dn_next;
138 continue;
139 }
140 fown = &dn->dn_filp->f_owner;
141 send_sigio(fown, dn->dn_fd, POLL_MSG);
142 if (dn->dn_mask & DN_MULTISHOT)
143 prev = &dn->dn_next;
144 else {
145 *prev = dn->dn_next;
146 changed = 1;
147 kmem_cache_free(dn_cache, dn);
148 }
149 }
150 if (changed)
151 redo_inode_mask(inode);
152 spin_unlock(&inode->i_lock); 364 spin_unlock(&inode->i_lock);
153} 365 if (entry) {
154 366 dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
155EXPORT_SYMBOL(__inode_dir_notify); 367 spin_lock(&entry->lock);
368 } else {
369 fsnotify_add_mark(new_entry, dnotify_group, inode);
370 spin_lock(&new_entry->lock);
371 entry = new_entry;
372 dnentry = new_dnentry;
373 /* we used new_entry, so don't free it */
374 new_entry = NULL;
375 }
156 376
157/* 377 rcu_read_lock();
158 * This is hopelessly wrong, but unfixable without API changes. At 378 f = fcheck(fd);
159 * least it doesn't oops the kernel... 379 rcu_read_unlock();
160 *
161 * To safely access ->d_parent we need to keep d_move away from it. Use the
162 * dentry's d_lock for this.
163 */
164void dnotify_parent(struct dentry *dentry, unsigned long event)
165{
166 struct dentry *parent;
167 380
168 if (!dir_notify_enable) 381 /* if (f != filp) means that we lost a race and another task/thread
169 return; 382 * actually closed the fd we are still playing with before we grabbed
383 * the dnotify_mark_mutex and entry->lock. Since closing the fd is the
384 * only time we clean up the mark entries we need to get our mark off
385 * the list. */
386 if (f != filp) {
387 /* if we added ourselves, shoot ourselves, it's possible that
388 * the flush actually did shoot this entry. That's fine too
389 * since multiple calls to destroy_mark is perfectly safe, if
390 * we found a dnentry already attached to the inode, just sod
391 * off silently as the flush at close time dealt with it.
392 */
393 if (dnentry == new_dnentry)
394 destroy = 1;
395 goto out;
396 }
170 397
171 spin_lock(&dentry->d_lock); 398 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
172 parent = dentry->d_parent; 399 if (error) {
173 if (parent->d_inode->i_dnotify_mask & event) { 400 /* if we added, we must shoot */
174 dget(parent); 401 if (dnentry == new_dnentry)
175 spin_unlock(&dentry->d_lock); 402 destroy = 1;
176 __inode_dir_notify(parent->d_inode, event); 403 goto out;
177 dput(parent);
178 } else {
179 spin_unlock(&dentry->d_lock);
180 } 404 }
405
406 error = attach_dn(dn, dnentry, id, fd, filp, mask);
407 /* !error means that we attached the dn to the dnentry, so don't free it */
408 if (!error)
409 dn = NULL;
410 /* -EEXIST means that we didn't add this new dn and used an old one.
411 * that isn't an error (and the unused dn should be freed) */
412 else if (error == -EEXIST)
413 error = 0;
414
415 dnotify_recalc_inode_mask(entry);
416out:
417 spin_unlock(&entry->lock);
418
419 if (destroy)
420 fsnotify_destroy_mark_by_entry(entry);
421
422 fsnotify_recalc_group_mask(dnotify_group);
423
424 mutex_unlock(&dnotify_mark_mutex);
425 fsnotify_put_mark(entry);
426out_err:
427 if (new_entry)
428 fsnotify_put_mark(new_entry);
429 if (dn)
430 kmem_cache_free(dnotify_struct_cache, dn);
431 return error;
181} 432}
182EXPORT_SYMBOL_GPL(dnotify_parent);
183 433
184static int __init dnotify_init(void) 434static int __init dnotify_init(void)
185{ 435{
186 dn_cache = kmem_cache_create("dnotify_cache", 436 dnotify_struct_cache = KMEM_CACHE(dnotify_struct, SLAB_PANIC);
187 sizeof(struct dnotify_struct), 0, SLAB_PANIC, NULL); 437 dnotify_mark_entry_cache = KMEM_CACHE(dnotify_mark_entry, SLAB_PANIC);
438
439 dnotify_group = fsnotify_obtain_group(DNOTIFY_GROUP_NUM,
440 0, &dnotify_fsnotify_ops);
441 if (IS_ERR(dnotify_group))
442 panic("unable to allocate fsnotify group for dnotify\n");
188 return 0; 443 return 0;
189} 444}
190 445