]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/eventpoll.c ( Efficent event polling implementation ) | |
3 | * Copyright (C) 2001,...,2003 Davide Libenzi | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * Davide Libenzi <davidel@xmailserver.org> | |
11 | * | |
12 | */ | |
13 | ||
14 | #include <linux/module.h> | |
15 | #include <linux/init.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/fs.h> | |
19 | #include <linux/file.h> | |
20 | #include <linux/signal.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/mm.h> | |
23 | #include <linux/slab.h> | |
24 | #include <linux/poll.h> | |
25 | #include <linux/smp_lock.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/list.h> | |
28 | #include <linux/hash.h> | |
29 | #include <linux/spinlock.h> | |
30 | #include <linux/syscalls.h> | |
31 | #include <linux/rwsem.h> | |
32 | #include <linux/rbtree.h> | |
33 | #include <linux/wait.h> | |
34 | #include <linux/eventpoll.h> | |
35 | #include <linux/mount.h> | |
36 | #include <linux/bitops.h> | |
37 | #include <asm/uaccess.h> | |
38 | #include <asm/system.h> | |
39 | #include <asm/io.h> | |
40 | #include <asm/mman.h> | |
41 | #include <asm/atomic.h> | |
42 | #include <asm/semaphore.h> | |
43 | ||
44 | ||
45 | /* | |
46 | * LOCKING: | |
47 | * There are three level of locking required by epoll : | |
48 | * | |
49 | * 1) epsem (semaphore) | |
50 | * 2) ep->sem (rw_semaphore) | |
51 | * 3) ep->lock (rw_lock) | |
52 | * | |
53 | * The acquire order is the one listed above, from 1 to 3. | |
54 | * We need a spinlock (ep->lock) because we manipulate objects | |
55 | * from inside the poll callback, that might be triggered from | |
56 | * a wake_up() that in turn might be called from IRQ context. | |
57 | * So we can't sleep inside the poll callback and hence we need | |
58 | * a spinlock. During the event transfer loop (from kernel to | |
59 | * user space) we could end up sleeping due a copy_to_user(), so | |
60 | * we need a lock that will allow us to sleep. This lock is a | |
61 | * read-write semaphore (ep->sem). It is acquired on read during | |
62 | * the event transfer loop and in write during epoll_ctl(EPOLL_CTL_DEL) | |
63 | * and during eventpoll_release_file(). Then we also need a global | |
64 | * semaphore to serialize eventpoll_release_file() and ep_free(). | |
65 | * This semaphore is acquired by ep_free() during the epoll file | |
66 | * cleanup path and it is also acquired by eventpoll_release_file() | |
67 | * if a file has been pushed inside an epoll set and it is then | |
68 | * close()d without a previous call toepoll_ctl(EPOLL_CTL_DEL). | |
69 | * It is possible to drop the "ep->sem" and to use the global | |
70 | * semaphore "epsem" (together with "ep->lock") to have it working, | |
71 | * but having "ep->sem" will make the interface more scalable. | |
72 | * Events that require holding "epsem" are very rare, while for | |
73 | * normal operations the epoll private "ep->sem" will guarantee | |
74 | * a greater scalability. | |
75 | */ | |
76 | ||
77 | ||
78 | #define EVENTPOLLFS_MAGIC 0x03111965 /* My birthday should work for this :) */ | |
79 | ||
80 | #define DEBUG_EPOLL 0 | |
81 | ||
82 | #if DEBUG_EPOLL > 0 | |
83 | #define DPRINTK(x) printk x | |
84 | #define DNPRINTK(n, x) do { if ((n) <= DEBUG_EPOLL) printk x; } while (0) | |
85 | #else /* #if DEBUG_EPOLL > 0 */ | |
86 | #define DPRINTK(x) (void) 0 | |
87 | #define DNPRINTK(n, x) (void) 0 | |
88 | #endif /* #if DEBUG_EPOLL > 0 */ | |
89 | ||
90 | #define DEBUG_EPI 0 | |
91 | ||
92 | #if DEBUG_EPI != 0 | |
93 | #define EPI_SLAB_DEBUG (SLAB_DEBUG_FREE | SLAB_RED_ZONE /* | SLAB_POISON */) | |
94 | #else /* #if DEBUG_EPI != 0 */ | |
95 | #define EPI_SLAB_DEBUG 0 | |
96 | #endif /* #if DEBUG_EPI != 0 */ | |
97 | ||
98 | /* Epoll private bits inside the event mask */ | |
99 | #define EP_PRIVATE_BITS (EPOLLONESHOT | EPOLLET) | |
100 | ||
101 | /* Maximum number of poll wake up nests we are allowing */ | |
102 | #define EP_MAX_POLLWAKE_NESTS 4 | |
103 | ||
104 | /* Macro to allocate a "struct epitem" from the slab cache */ | |
105 | #define EPI_MEM_ALLOC() (struct epitem *) kmem_cache_alloc(epi_cache, SLAB_KERNEL) | |
106 | ||
107 | /* Macro to free a "struct epitem" to the slab cache */ | |
108 | #define EPI_MEM_FREE(p) kmem_cache_free(epi_cache, p) | |
109 | ||
110 | /* Macro to allocate a "struct eppoll_entry" from the slab cache */ | |
111 | #define PWQ_MEM_ALLOC() (struct eppoll_entry *) kmem_cache_alloc(pwq_cache, SLAB_KERNEL) | |
112 | ||
113 | /* Macro to free a "struct eppoll_entry" to the slab cache */ | |
114 | #define PWQ_MEM_FREE(p) kmem_cache_free(pwq_cache, p) | |
115 | ||
116 | /* Fast test to see if the file is an evenpoll file */ | |
117 | #define IS_FILE_EPOLL(f) ((f)->f_op == &eventpoll_fops) | |
118 | ||
119 | /* Setup the structure that is used as key for the rb-tree */ | |
120 | #define EP_SET_FFD(p, f, d) do { (p)->file = (f); (p)->fd = (d); } while (0) | |
121 | ||
122 | /* Compare rb-tree keys */ | |
123 | #define EP_CMP_FFD(p1, p2) ((p1)->file > (p2)->file ? +1: \ | |
124 | ((p1)->file < (p2)->file ? -1: (p1)->fd - (p2)->fd)) | |
125 | ||
126 | /* Special initialization for the rb-tree node to detect linkage */ | |
127 | #define EP_RB_INITNODE(n) (n)->rb_parent = (n) | |
128 | ||
129 | /* Removes a node from the rb-tree and marks it for a fast is-linked check */ | |
130 | #define EP_RB_ERASE(n, r) do { rb_erase(n, r); (n)->rb_parent = (n); } while (0) | |
131 | ||
132 | /* Fast check to verify that the item is linked to the main rb-tree */ | |
133 | #define EP_RB_LINKED(n) ((n)->rb_parent != (n)) | |
134 | ||
135 | /* | |
136 | * Remove the item from the list and perform its initialization. | |
137 | * This is useful for us because we can test if the item is linked | |
138 | * using "EP_IS_LINKED(p)". | |
139 | */ | |
140 | #define EP_LIST_DEL(p) do { list_del(p); INIT_LIST_HEAD(p); } while (0) | |
141 | ||
142 | /* Tells us if the item is currently linked */ | |
143 | #define EP_IS_LINKED(p) (!list_empty(p)) | |
144 | ||
145 | /* Get the "struct epitem" from a wait queue pointer */ | |
146 | #define EP_ITEM_FROM_WAIT(p) ((struct epitem *) container_of(p, struct eppoll_entry, wait)->base) | |
147 | ||
148 | /* Get the "struct epitem" from an epoll queue wrapper */ | |
149 | #define EP_ITEM_FROM_EPQUEUE(p) (container_of(p, struct ep_pqueue, pt)->epi) | |
150 | ||
151 | /* Tells if the epoll_ctl(2) operation needs an event copy from userspace */ | |
152 | #define EP_OP_HASH_EVENT(op) ((op) != EPOLL_CTL_DEL) | |
153 | ||
154 | ||
155 | struct epoll_filefd { | |
156 | struct file *file; | |
157 | int fd; | |
158 | }; | |
159 | ||
160 | /* | |
161 | * Node that is linked into the "wake_task_list" member of the "struct poll_safewake". | |
162 | * It is used to keep track on all tasks that are currently inside the wake_up() code | |
163 | * to 1) short-circuit the one coming from the same task and same wait queue head | |
164 | * ( loop ) 2) allow a maximum number of epoll descriptors inclusion nesting | |
165 | * 3) let go the ones coming from other tasks. | |
166 | */ | |
167 | struct wake_task_node { | |
168 | struct list_head llink; | |
169 | task_t *task; | |
170 | wait_queue_head_t *wq; | |
171 | }; | |
172 | ||
173 | /* | |
174 | * This is used to implement the safe poll wake up avoiding to reenter | |
175 | * the poll callback from inside wake_up(). | |
176 | */ | |
177 | struct poll_safewake { | |
178 | struct list_head wake_task_list; | |
179 | spinlock_t lock; | |
180 | }; | |
181 | ||
182 | /* | |
183 | * This structure is stored inside the "private_data" member of the file | |
184 | * structure and rapresent the main data sructure for the eventpoll | |
185 | * interface. | |
186 | */ | |
187 | struct eventpoll { | |
188 | /* Protect the this structure access */ | |
189 | rwlock_t lock; | |
190 | ||
191 | /* | |
192 | * This semaphore is used to ensure that files are not removed | |
193 | * while epoll is using them. This is read-held during the event | |
194 | * collection loop and it is write-held during the file cleanup | |
195 | * path, the epoll file exit code and the ctl operations. | |
196 | */ | |
197 | struct rw_semaphore sem; | |
198 | ||
199 | /* Wait queue used by sys_epoll_wait() */ | |
200 | wait_queue_head_t wq; | |
201 | ||
202 | /* Wait queue used by file->poll() */ | |
203 | wait_queue_head_t poll_wait; | |
204 | ||
205 | /* List of ready file descriptors */ | |
206 | struct list_head rdllist; | |
207 | ||
208 | /* RB-Tree root used to store monitored fd structs */ | |
209 | struct rb_root rbr; | |
210 | }; | |
211 | ||
212 | /* Wait structure used by the poll hooks */ | |
213 | struct eppoll_entry { | |
214 | /* List header used to link this structure to the "struct epitem" */ | |
215 | struct list_head llink; | |
216 | ||
217 | /* The "base" pointer is set to the container "struct epitem" */ | |
218 | void *base; | |
219 | ||
220 | /* | |
221 | * Wait queue item that will be linked to the target file wait | |
222 | * queue head. | |
223 | */ | |
224 | wait_queue_t wait; | |
225 | ||
226 | /* The wait queue head that linked the "wait" wait queue item */ | |
227 | wait_queue_head_t *whead; | |
228 | }; | |
229 | ||
230 | /* | |
231 | * Each file descriptor added to the eventpoll interface will | |
232 | * have an entry of this type linked to the hash. | |
233 | */ | |
234 | struct epitem { | |
235 | /* RB-Tree node used to link this structure to the eventpoll rb-tree */ | |
236 | struct rb_node rbn; | |
237 | ||
238 | /* List header used to link this structure to the eventpoll ready list */ | |
239 | struct list_head rdllink; | |
240 | ||
241 | /* The file descriptor information this item refers to */ | |
242 | struct epoll_filefd ffd; | |
243 | ||
244 | /* Number of active wait queue attached to poll operations */ | |
245 | int nwait; | |
246 | ||
247 | /* List containing poll wait queues */ | |
248 | struct list_head pwqlist; | |
249 | ||
250 | /* The "container" of this item */ | |
251 | struct eventpoll *ep; | |
252 | ||
253 | /* The structure that describe the interested events and the source fd */ | |
254 | struct epoll_event event; | |
255 | ||
256 | /* | |
257 | * Used to keep track of the usage count of the structure. This avoids | |
258 | * that the structure will desappear from underneath our processing. | |
259 | */ | |
260 | atomic_t usecnt; | |
261 | ||
262 | /* List header used to link this item to the "struct file" items list */ | |
263 | struct list_head fllink; | |
264 | ||
265 | /* List header used to link the item to the transfer list */ | |
266 | struct list_head txlink; | |
267 | ||
268 | /* | |
269 | * This is used during the collection/transfer of events to userspace | |
270 | * to pin items empty events set. | |
271 | */ | |
272 | unsigned int revents; | |
273 | }; | |
274 | ||
275 | /* Wrapper struct used by poll queueing */ | |
276 | struct ep_pqueue { | |
277 | poll_table pt; | |
278 | struct epitem *epi; | |
279 | }; | |
280 | ||
281 | ||
282 | ||
283 | static void ep_poll_safewake_init(struct poll_safewake *psw); | |
284 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq); | |
285 | static int ep_getfd(int *efd, struct inode **einode, struct file **efile); | |
286 | static int ep_file_init(struct file *file); | |
287 | static void ep_free(struct eventpoll *ep); | |
288 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd); | |
289 | static void ep_use_epitem(struct epitem *epi); | |
290 | static void ep_release_epitem(struct epitem *epi); | |
291 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
292 | poll_table *pt); | |
293 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi); | |
294 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, | |
295 | struct file *tfile, int fd); | |
296 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, | |
297 | struct epoll_event *event); | |
298 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi); | |
299 | static int ep_unlink(struct eventpoll *ep, struct epitem *epi); | |
300 | static int ep_remove(struct eventpoll *ep, struct epitem *epi); | |
301 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key); | |
302 | static int ep_eventpoll_close(struct inode *inode, struct file *file); | |
303 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait); | |
304 | static int ep_collect_ready_items(struct eventpoll *ep, | |
305 | struct list_head *txlist, int maxevents); | |
306 | static int ep_send_events(struct eventpoll *ep, struct list_head *txlist, | |
307 | struct epoll_event __user *events); | |
308 | static void ep_reinject_items(struct eventpoll *ep, struct list_head *txlist); | |
309 | static int ep_events_transfer(struct eventpoll *ep, | |
310 | struct epoll_event __user *events, | |
311 | int maxevents); | |
312 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, | |
313 | int maxevents, long timeout); | |
314 | static int eventpollfs_delete_dentry(struct dentry *dentry); | |
315 | static struct inode *ep_eventpoll_inode(void); | |
316 | static struct super_block *eventpollfs_get_sb(struct file_system_type *fs_type, | |
317 | int flags, const char *dev_name, | |
318 | void *data); | |
319 | ||
320 | /* | |
321 | * This semaphore is used to serialize ep_free() and eventpoll_release_file(). | |
322 | */ | |
75c96f85 | 323 | static struct semaphore epsem; |
1da177e4 LT |
324 | |
325 | /* Safe wake up implementation */ | |
326 | static struct poll_safewake psw; | |
327 | ||
328 | /* Slab cache used to allocate "struct epitem" */ | |
329 | static kmem_cache_t *epi_cache; | |
330 | ||
331 | /* Slab cache used to allocate "struct eppoll_entry" */ | |
332 | static kmem_cache_t *pwq_cache; | |
333 | ||
334 | /* Virtual fs used to allocate inodes for eventpoll files */ | |
335 | static struct vfsmount *eventpoll_mnt; | |
336 | ||
337 | /* File callbacks that implement the eventpoll file behaviour */ | |
338 | static struct file_operations eventpoll_fops = { | |
339 | .release = ep_eventpoll_close, | |
340 | .poll = ep_eventpoll_poll | |
341 | }; | |
342 | ||
343 | /* | |
344 | * This is used to register the virtual file system from where | |
345 | * eventpoll inodes are allocated. | |
346 | */ | |
347 | static struct file_system_type eventpoll_fs_type = { | |
348 | .name = "eventpollfs", | |
349 | .get_sb = eventpollfs_get_sb, | |
350 | .kill_sb = kill_anon_super, | |
351 | }; | |
352 | ||
353 | /* Very basic directory entry operations for the eventpoll virtual file system */ | |
354 | static struct dentry_operations eventpollfs_dentry_operations = { | |
355 | .d_delete = eventpollfs_delete_dentry, | |
356 | }; | |
357 | ||
358 | ||
359 | ||
360 | /* Initialize the poll safe wake up structure */ | |
361 | static void ep_poll_safewake_init(struct poll_safewake *psw) | |
362 | { | |
363 | ||
364 | INIT_LIST_HEAD(&psw->wake_task_list); | |
365 | spin_lock_init(&psw->lock); | |
366 | } | |
367 | ||
368 | ||
369 | /* | |
370 | * Perform a safe wake up of the poll wait list. The problem is that | |
371 | * with the new callback'd wake up system, it is possible that the | |
372 | * poll callback is reentered from inside the call to wake_up() done | |
373 | * on the poll wait queue head. The rule is that we cannot reenter the | |
374 | * wake up code from the same task more than EP_MAX_POLLWAKE_NESTS times, | |
375 | * and we cannot reenter the same wait queue head at all. This will | |
376 | * enable to have a hierarchy of epoll file descriptor of no more than | |
377 | * EP_MAX_POLLWAKE_NESTS deep. We need the irq version of the spin lock | |
378 | * because this one gets called by the poll callback, that in turn is called | |
379 | * from inside a wake_up(), that might be called from irq context. | |
380 | */ | |
381 | static void ep_poll_safewake(struct poll_safewake *psw, wait_queue_head_t *wq) | |
382 | { | |
383 | int wake_nests = 0; | |
384 | unsigned long flags; | |
385 | task_t *this_task = current; | |
386 | struct list_head *lsthead = &psw->wake_task_list, *lnk; | |
387 | struct wake_task_node *tncur; | |
388 | struct wake_task_node tnode; | |
389 | ||
390 | spin_lock_irqsave(&psw->lock, flags); | |
391 | ||
392 | /* Try to see if the current task is already inside this wakeup call */ | |
393 | list_for_each(lnk, lsthead) { | |
394 | tncur = list_entry(lnk, struct wake_task_node, llink); | |
395 | ||
396 | if (tncur->wq == wq || | |
397 | (tncur->task == this_task && ++wake_nests > EP_MAX_POLLWAKE_NESTS)) { | |
398 | /* | |
399 | * Ops ... loop detected or maximum nest level reached. | |
400 | * We abort this wake by breaking the cycle itself. | |
401 | */ | |
402 | spin_unlock_irqrestore(&psw->lock, flags); | |
403 | return; | |
404 | } | |
405 | } | |
406 | ||
407 | /* Add the current task to the list */ | |
408 | tnode.task = this_task; | |
409 | tnode.wq = wq; | |
410 | list_add(&tnode.llink, lsthead); | |
411 | ||
412 | spin_unlock_irqrestore(&psw->lock, flags); | |
413 | ||
414 | /* Do really wake up now */ | |
415 | wake_up(wq); | |
416 | ||
417 | /* Remove the current task from the list */ | |
418 | spin_lock_irqsave(&psw->lock, flags); | |
419 | list_del(&tnode.llink); | |
420 | spin_unlock_irqrestore(&psw->lock, flags); | |
421 | } | |
422 | ||
423 | ||
424 | /* Used to initialize the epoll bits inside the "struct file" */ | |
425 | void eventpoll_init_file(struct file *file) | |
426 | { | |
427 | ||
428 | INIT_LIST_HEAD(&file->f_ep_links); | |
429 | spin_lock_init(&file->f_ep_lock); | |
430 | } | |
431 | ||
432 | ||
433 | /* | |
434 | * This is called from eventpoll_release() to unlink files from the eventpoll | |
435 | * interface. We need to have this facility to cleanup correctly files that are | |
436 | * closed without being removed from the eventpoll interface. | |
437 | */ | |
438 | void eventpoll_release_file(struct file *file) | |
439 | { | |
440 | struct list_head *lsthead = &file->f_ep_links; | |
441 | struct eventpoll *ep; | |
442 | struct epitem *epi; | |
443 | ||
444 | /* | |
445 | * We don't want to get "file->f_ep_lock" because it is not | |
446 | * necessary. It is not necessary because we're in the "struct file" | |
447 | * cleanup path, and this means that noone is using this file anymore. | |
448 | * The only hit might come from ep_free() but by holding the semaphore | |
449 | * will correctly serialize the operation. We do need to acquire | |
450 | * "ep->sem" after "epsem" because ep_remove() requires it when called | |
451 | * from anywhere but ep_free(). | |
452 | */ | |
453 | down(&epsem); | |
454 | ||
455 | while (!list_empty(lsthead)) { | |
456 | epi = list_entry(lsthead->next, struct epitem, fllink); | |
457 | ||
458 | ep = epi->ep; | |
459 | EP_LIST_DEL(&epi->fllink); | |
460 | down_write(&ep->sem); | |
461 | ep_remove(ep, epi); | |
462 | up_write(&ep->sem); | |
463 | } | |
464 | ||
465 | up(&epsem); | |
466 | } | |
467 | ||
468 | ||
469 | /* | |
470 | * It opens an eventpoll file descriptor by suggesting a storage of "size" | |
471 | * file descriptors. The size parameter is just an hint about how to size | |
472 | * data structures. It won't prevent the user to store more than "size" | |
473 | * file descriptors inside the epoll interface. It is the kernel part of | |
474 | * the userspace epoll_create(2). | |
475 | */ | |
476 | asmlinkage long sys_epoll_create(int size) | |
477 | { | |
478 | int error, fd; | |
479 | struct inode *inode; | |
480 | struct file *file; | |
481 | ||
482 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d)\n", | |
483 | current, size)); | |
484 | ||
485 | /* Sanity check on the size parameter */ | |
486 | error = -EINVAL; | |
487 | if (size <= 0) | |
488 | goto eexit_1; | |
489 | ||
490 | /* | |
491 | * Creates all the items needed to setup an eventpoll file. That is, | |
492 | * a file structure, and inode and a free file descriptor. | |
493 | */ | |
494 | error = ep_getfd(&fd, &inode, &file); | |
495 | if (error) | |
496 | goto eexit_1; | |
497 | ||
498 | /* Setup the file internal data structure ( "struct eventpoll" ) */ | |
499 | error = ep_file_init(file); | |
500 | if (error) | |
501 | goto eexit_2; | |
502 | ||
503 | ||
504 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", | |
505 | current, size, fd)); | |
506 | ||
507 | return fd; | |
508 | ||
509 | eexit_2: | |
510 | sys_close(fd); | |
511 | eexit_1: | |
512 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_create(%d) = %d\n", | |
513 | current, size, error)); | |
514 | return error; | |
515 | } | |
516 | ||
517 | ||
518 | /* | |
519 | * The following function implements the controller interface for | |
520 | * the eventpoll file that enables the insertion/removal/change of | |
521 | * file descriptors inside the interest set. It represents | |
522 | * the kernel part of the user space epoll_ctl(2). | |
523 | */ | |
524 | asmlinkage long | |
525 | sys_epoll_ctl(int epfd, int op, int fd, struct epoll_event __user *event) | |
526 | { | |
527 | int error; | |
528 | struct file *file, *tfile; | |
529 | struct eventpoll *ep; | |
530 | struct epitem *epi; | |
531 | struct epoll_event epds; | |
532 | ||
533 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p)\n", | |
534 | current, epfd, op, fd, event)); | |
535 | ||
536 | error = -EFAULT; | |
537 | if (EP_OP_HASH_EVENT(op) && | |
538 | copy_from_user(&epds, event, sizeof(struct epoll_event))) | |
539 | goto eexit_1; | |
540 | ||
541 | /* Get the "struct file *" for the eventpoll file */ | |
542 | error = -EBADF; | |
543 | file = fget(epfd); | |
544 | if (!file) | |
545 | goto eexit_1; | |
546 | ||
547 | /* Get the "struct file *" for the target file */ | |
548 | tfile = fget(fd); | |
549 | if (!tfile) | |
550 | goto eexit_2; | |
551 | ||
552 | /* The target file descriptor must support poll */ | |
553 | error = -EPERM; | |
554 | if (!tfile->f_op || !tfile->f_op->poll) | |
555 | goto eexit_3; | |
556 | ||
557 | /* | |
558 | * We have to check that the file structure underneath the file descriptor | |
559 | * the user passed to us _is_ an eventpoll file. And also we do not permit | |
560 | * adding an epoll file descriptor inside itself. | |
561 | */ | |
562 | error = -EINVAL; | |
563 | if (file == tfile || !IS_FILE_EPOLL(file)) | |
564 | goto eexit_3; | |
565 | ||
566 | /* | |
567 | * At this point it is safe to assume that the "private_data" contains | |
568 | * our own data structure. | |
569 | */ | |
570 | ep = file->private_data; | |
571 | ||
572 | down_write(&ep->sem); | |
573 | ||
574 | /* Try to lookup the file inside our hash table */ | |
575 | epi = ep_find(ep, tfile, fd); | |
576 | ||
577 | error = -EINVAL; | |
578 | switch (op) { | |
579 | case EPOLL_CTL_ADD: | |
580 | if (!epi) { | |
581 | epds.events |= POLLERR | POLLHUP; | |
582 | ||
583 | error = ep_insert(ep, &epds, tfile, fd); | |
584 | } else | |
585 | error = -EEXIST; | |
586 | break; | |
587 | case EPOLL_CTL_DEL: | |
588 | if (epi) | |
589 | error = ep_remove(ep, epi); | |
590 | else | |
591 | error = -ENOENT; | |
592 | break; | |
593 | case EPOLL_CTL_MOD: | |
594 | if (epi) { | |
595 | epds.events |= POLLERR | POLLHUP; | |
596 | error = ep_modify(ep, epi, &epds); | |
597 | } else | |
598 | error = -ENOENT; | |
599 | break; | |
600 | } | |
601 | ||
602 | /* | |
603 | * The function ep_find() increments the usage count of the structure | |
604 | * so, if this is not NULL, we need to release it. | |
605 | */ | |
606 | if (epi) | |
607 | ep_release_epitem(epi); | |
608 | ||
609 | up_write(&ep->sem); | |
610 | ||
611 | eexit_3: | |
612 | fput(tfile); | |
613 | eexit_2: | |
614 | fput(file); | |
615 | eexit_1: | |
616 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_ctl(%d, %d, %d, %p) = %d\n", | |
617 | current, epfd, op, fd, event, error)); | |
618 | ||
619 | return error; | |
620 | } | |
621 | ||
622 | #define MAX_EVENTS (INT_MAX / sizeof(struct epoll_event)) | |
623 | ||
624 | /* | |
625 | * Implement the event wait interface for the eventpoll file. It is the kernel | |
626 | * part of the user space epoll_wait(2). | |
627 | */ | |
628 | asmlinkage long sys_epoll_wait(int epfd, struct epoll_event __user *events, | |
629 | int maxevents, int timeout) | |
630 | { | |
631 | int error; | |
632 | struct file *file; | |
633 | struct eventpoll *ep; | |
634 | ||
635 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d)\n", | |
636 | current, epfd, events, maxevents, timeout)); | |
637 | ||
638 | /* The maximum number of event must be greater than zero */ | |
639 | if (maxevents <= 0 || maxevents > MAX_EVENTS) | |
640 | return -EINVAL; | |
641 | ||
642 | /* Verify that the area passed by the user is writeable */ | |
643 | if (!access_ok(VERIFY_WRITE, events, maxevents * sizeof(struct epoll_event))) { | |
644 | error = -EFAULT; | |
645 | goto eexit_1; | |
646 | } | |
647 | ||
648 | /* Get the "struct file *" for the eventpoll file */ | |
649 | error = -EBADF; | |
650 | file = fget(epfd); | |
651 | if (!file) | |
652 | goto eexit_1; | |
653 | ||
654 | /* | |
655 | * We have to check that the file structure underneath the fd | |
656 | * the user passed to us _is_ an eventpoll file. | |
657 | */ | |
658 | error = -EINVAL; | |
659 | if (!IS_FILE_EPOLL(file)) | |
660 | goto eexit_2; | |
661 | ||
662 | /* | |
663 | * At this point it is safe to assume that the "private_data" contains | |
664 | * our own data structure. | |
665 | */ | |
666 | ep = file->private_data; | |
667 | ||
668 | /* Time to fish for events ... */ | |
669 | error = ep_poll(ep, events, maxevents, timeout); | |
670 | ||
671 | eexit_2: | |
672 | fput(file); | |
673 | eexit_1: | |
674 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: sys_epoll_wait(%d, %p, %d, %d) = %d\n", | |
675 | current, epfd, events, maxevents, timeout, error)); | |
676 | ||
677 | return error; | |
678 | } | |
679 | ||
680 | ||
681 | /* | |
682 | * Creates the file descriptor to be used by the epoll interface. | |
683 | */ | |
684 | static int ep_getfd(int *efd, struct inode **einode, struct file **efile) | |
685 | { | |
686 | struct qstr this; | |
687 | char name[32]; | |
688 | struct dentry *dentry; | |
689 | struct inode *inode; | |
690 | struct file *file; | |
691 | int error, fd; | |
692 | ||
693 | /* Get an ready to use file */ | |
694 | error = -ENFILE; | |
695 | file = get_empty_filp(); | |
696 | if (!file) | |
697 | goto eexit_1; | |
698 | ||
699 | /* Allocates an inode from the eventpoll file system */ | |
700 | inode = ep_eventpoll_inode(); | |
701 | error = PTR_ERR(inode); | |
702 | if (IS_ERR(inode)) | |
703 | goto eexit_2; | |
704 | ||
705 | /* Allocates a free descriptor to plug the file onto */ | |
706 | error = get_unused_fd(); | |
707 | if (error < 0) | |
708 | goto eexit_3; | |
709 | fd = error; | |
710 | ||
711 | /* | |
712 | * Link the inode to a directory entry by creating a unique name | |
713 | * using the inode number. | |
714 | */ | |
715 | error = -ENOMEM; | |
716 | sprintf(name, "[%lu]", inode->i_ino); | |
717 | this.name = name; | |
718 | this.len = strlen(name); | |
719 | this.hash = inode->i_ino; | |
720 | dentry = d_alloc(eventpoll_mnt->mnt_sb->s_root, &this); | |
721 | if (!dentry) | |
722 | goto eexit_4; | |
723 | dentry->d_op = &eventpollfs_dentry_operations; | |
724 | d_add(dentry, inode); | |
725 | file->f_vfsmnt = mntget(eventpoll_mnt); | |
726 | file->f_dentry = dentry; | |
727 | file->f_mapping = inode->i_mapping; | |
728 | ||
729 | file->f_pos = 0; | |
730 | file->f_flags = O_RDONLY; | |
731 | file->f_op = &eventpoll_fops; | |
732 | file->f_mode = FMODE_READ; | |
733 | file->f_version = 0; | |
734 | file->private_data = NULL; | |
735 | ||
736 | /* Install the new setup file into the allocated fd. */ | |
737 | fd_install(fd, file); | |
738 | ||
739 | *efd = fd; | |
740 | *einode = inode; | |
741 | *efile = file; | |
742 | return 0; | |
743 | ||
744 | eexit_4: | |
745 | put_unused_fd(fd); | |
746 | eexit_3: | |
747 | iput(inode); | |
748 | eexit_2: | |
749 | put_filp(file); | |
750 | eexit_1: | |
751 | return error; | |
752 | } | |
753 | ||
754 | ||
755 | static int ep_file_init(struct file *file) | |
756 | { | |
757 | struct eventpoll *ep; | |
758 | ||
759 | if (!(ep = kmalloc(sizeof(struct eventpoll), GFP_KERNEL))) | |
760 | return -ENOMEM; | |
761 | ||
762 | memset(ep, 0, sizeof(*ep)); | |
763 | rwlock_init(&ep->lock); | |
764 | init_rwsem(&ep->sem); | |
765 | init_waitqueue_head(&ep->wq); | |
766 | init_waitqueue_head(&ep->poll_wait); | |
767 | INIT_LIST_HEAD(&ep->rdllist); | |
768 | ep->rbr = RB_ROOT; | |
769 | ||
770 | file->private_data = ep; | |
771 | ||
772 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_file_init() ep=%p\n", | |
773 | current, ep)); | |
774 | return 0; | |
775 | } | |
776 | ||
777 | ||
778 | static void ep_free(struct eventpoll *ep) | |
779 | { | |
780 | struct rb_node *rbp; | |
781 | struct epitem *epi; | |
782 | ||
783 | /* We need to release all tasks waiting for these file */ | |
784 | if (waitqueue_active(&ep->poll_wait)) | |
785 | ep_poll_safewake(&psw, &ep->poll_wait); | |
786 | ||
787 | /* | |
788 | * We need to lock this because we could be hit by | |
789 | * eventpoll_release_file() while we're freeing the "struct eventpoll". | |
790 | * We do not need to hold "ep->sem" here because the epoll file | |
791 | * is on the way to be removed and no one has references to it | |
792 | * anymore. The only hit might come from eventpoll_release_file() but | |
793 | * holding "epsem" is sufficent here. | |
794 | */ | |
795 | down(&epsem); | |
796 | ||
797 | /* | |
798 | * Walks through the whole tree by unregistering poll callbacks. | |
799 | */ | |
800 | for (rbp = rb_first(&ep->rbr); rbp; rbp = rb_next(rbp)) { | |
801 | epi = rb_entry(rbp, struct epitem, rbn); | |
802 | ||
803 | ep_unregister_pollwait(ep, epi); | |
804 | } | |
805 | ||
806 | /* | |
807 | * Walks through the whole hash by freeing each "struct epitem". At this | |
808 | * point we are sure no poll callbacks will be lingering around, and also by | |
809 | * write-holding "sem" we can be sure that no file cleanup code will hit | |
810 | * us during this operation. So we can avoid the lock on "ep->lock". | |
811 | */ | |
812 | while ((rbp = rb_first(&ep->rbr)) != 0) { | |
813 | epi = rb_entry(rbp, struct epitem, rbn); | |
814 | ep_remove(ep, epi); | |
815 | } | |
816 | ||
817 | up(&epsem); | |
818 | } | |
819 | ||
820 | ||
821 | /* | |
822 | * Search the file inside the eventpoll hash. It add usage count to | |
823 | * the returned item, so the caller must call ep_release_epitem() | |
824 | * after finished using the "struct epitem". | |
825 | */ | |
826 | static struct epitem *ep_find(struct eventpoll *ep, struct file *file, int fd) | |
827 | { | |
828 | int kcmp; | |
829 | unsigned long flags; | |
830 | struct rb_node *rbp; | |
831 | struct epitem *epi, *epir = NULL; | |
832 | struct epoll_filefd ffd; | |
833 | ||
834 | EP_SET_FFD(&ffd, file, fd); | |
835 | read_lock_irqsave(&ep->lock, flags); | |
836 | for (rbp = ep->rbr.rb_node; rbp; ) { | |
837 | epi = rb_entry(rbp, struct epitem, rbn); | |
838 | kcmp = EP_CMP_FFD(&ffd, &epi->ffd); | |
839 | if (kcmp > 0) | |
840 | rbp = rbp->rb_right; | |
841 | else if (kcmp < 0) | |
842 | rbp = rbp->rb_left; | |
843 | else { | |
844 | ep_use_epitem(epi); | |
845 | epir = epi; | |
846 | break; | |
847 | } | |
848 | } | |
849 | read_unlock_irqrestore(&ep->lock, flags); | |
850 | ||
851 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_find(%p) -> %p\n", | |
852 | current, file, epir)); | |
853 | ||
854 | return epir; | |
855 | } | |
856 | ||
857 | ||
858 | /* | |
859 | * Increment the usage count of the "struct epitem" making it sure | |
860 | * that the user will have a valid pointer to reference. | |
861 | */ | |
862 | static void ep_use_epitem(struct epitem *epi) | |
863 | { | |
864 | ||
865 | atomic_inc(&epi->usecnt); | |
866 | } | |
867 | ||
868 | ||
869 | /* | |
870 | * Decrement ( release ) the usage count by signaling that the user | |
871 | * has finished using the structure. It might lead to freeing the | |
872 | * structure itself if the count goes to zero. | |
873 | */ | |
874 | static void ep_release_epitem(struct epitem *epi) | |
875 | { | |
876 | ||
877 | if (atomic_dec_and_test(&epi->usecnt)) | |
878 | EPI_MEM_FREE(epi); | |
879 | } | |
880 | ||
881 | ||
882 | /* | |
883 | * This is the callback that is used to add our wait queue to the | |
884 | * target file wakeup lists. | |
885 | */ | |
886 | static void ep_ptable_queue_proc(struct file *file, wait_queue_head_t *whead, | |
887 | poll_table *pt) | |
888 | { | |
889 | struct epitem *epi = EP_ITEM_FROM_EPQUEUE(pt); | |
890 | struct eppoll_entry *pwq; | |
891 | ||
892 | if (epi->nwait >= 0 && (pwq = PWQ_MEM_ALLOC())) { | |
893 | init_waitqueue_func_entry(&pwq->wait, ep_poll_callback); | |
894 | pwq->whead = whead; | |
895 | pwq->base = epi; | |
896 | add_wait_queue(whead, &pwq->wait); | |
897 | list_add_tail(&pwq->llink, &epi->pwqlist); | |
898 | epi->nwait++; | |
899 | } else { | |
900 | /* We have to signal that an error occurred */ | |
901 | epi->nwait = -1; | |
902 | } | |
903 | } | |
904 | ||
905 | ||
906 | static void ep_rbtree_insert(struct eventpoll *ep, struct epitem *epi) | |
907 | { | |
908 | int kcmp; | |
909 | struct rb_node **p = &ep->rbr.rb_node, *parent = NULL; | |
910 | struct epitem *epic; | |
911 | ||
912 | while (*p) { | |
913 | parent = *p; | |
914 | epic = rb_entry(parent, struct epitem, rbn); | |
915 | kcmp = EP_CMP_FFD(&epi->ffd, &epic->ffd); | |
916 | if (kcmp > 0) | |
917 | p = &parent->rb_right; | |
918 | else | |
919 | p = &parent->rb_left; | |
920 | } | |
921 | rb_link_node(&epi->rbn, parent, p); | |
922 | rb_insert_color(&epi->rbn, &ep->rbr); | |
923 | } | |
924 | ||
925 | ||
926 | static int ep_insert(struct eventpoll *ep, struct epoll_event *event, | |
927 | struct file *tfile, int fd) | |
928 | { | |
929 | int error, revents, pwake = 0; | |
930 | unsigned long flags; | |
931 | struct epitem *epi; | |
932 | struct ep_pqueue epq; | |
933 | ||
934 | error = -ENOMEM; | |
935 | if (!(epi = EPI_MEM_ALLOC())) | |
936 | goto eexit_1; | |
937 | ||
938 | /* Item initialization follow here ... */ | |
939 | EP_RB_INITNODE(&epi->rbn); | |
940 | INIT_LIST_HEAD(&epi->rdllink); | |
941 | INIT_LIST_HEAD(&epi->fllink); | |
942 | INIT_LIST_HEAD(&epi->txlink); | |
943 | INIT_LIST_HEAD(&epi->pwqlist); | |
944 | epi->ep = ep; | |
945 | EP_SET_FFD(&epi->ffd, tfile, fd); | |
946 | epi->event = *event; | |
947 | atomic_set(&epi->usecnt, 1); | |
948 | epi->nwait = 0; | |
949 | ||
950 | /* Initialize the poll table using the queue callback */ | |
951 | epq.epi = epi; | |
952 | init_poll_funcptr(&epq.pt, ep_ptable_queue_proc); | |
953 | ||
954 | /* | |
955 | * Attach the item to the poll hooks and get current event bits. | |
956 | * We can safely use the file* here because its usage count has | |
957 | * been increased by the caller of this function. | |
958 | */ | |
959 | revents = tfile->f_op->poll(tfile, &epq.pt); | |
960 | ||
961 | /* | |
962 | * We have to check if something went wrong during the poll wait queue | |
963 | * install process. Namely an allocation for a wait queue failed due | |
964 | * high memory pressure. | |
965 | */ | |
966 | if (epi->nwait < 0) | |
967 | goto eexit_2; | |
968 | ||
969 | /* Add the current item to the list of active epoll hook for this file */ | |
970 | spin_lock(&tfile->f_ep_lock); | |
971 | list_add_tail(&epi->fllink, &tfile->f_ep_links); | |
972 | spin_unlock(&tfile->f_ep_lock); | |
973 | ||
974 | /* We have to drop the new item inside our item list to keep track of it */ | |
975 | write_lock_irqsave(&ep->lock, flags); | |
976 | ||
977 | /* Add the current item to the rb-tree */ | |
978 | ep_rbtree_insert(ep, epi); | |
979 | ||
980 | /* If the file is already "ready" we drop it inside the ready list */ | |
981 | if ((revents & event->events) && !EP_IS_LINKED(&epi->rdllink)) { | |
982 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
983 | ||
984 | /* Notify waiting tasks that events are available */ | |
985 | if (waitqueue_active(&ep->wq)) | |
986 | wake_up(&ep->wq); | |
987 | if (waitqueue_active(&ep->poll_wait)) | |
988 | pwake++; | |
989 | } | |
990 | ||
991 | write_unlock_irqrestore(&ep->lock, flags); | |
992 | ||
993 | /* We have to call this outside the lock */ | |
994 | if (pwake) | |
995 | ep_poll_safewake(&psw, &ep->poll_wait); | |
996 | ||
997 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_insert(%p, %p, %d)\n", | |
998 | current, ep, tfile, fd)); | |
999 | ||
1000 | return 0; | |
1001 | ||
1002 | eexit_2: | |
1003 | ep_unregister_pollwait(ep, epi); | |
1004 | ||
1005 | /* | |
1006 | * We need to do this because an event could have been arrived on some | |
1007 | * allocated wait queue. | |
1008 | */ | |
1009 | write_lock_irqsave(&ep->lock, flags); | |
1010 | if (EP_IS_LINKED(&epi->rdllink)) | |
1011 | EP_LIST_DEL(&epi->rdllink); | |
1012 | write_unlock_irqrestore(&ep->lock, flags); | |
1013 | ||
1014 | EPI_MEM_FREE(epi); | |
1015 | eexit_1: | |
1016 | return error; | |
1017 | } | |
1018 | ||
1019 | ||
1020 | /* | |
1021 | * Modify the interest event mask by dropping an event if the new mask | |
1022 | * has a match in the current file status. | |
1023 | */ | |
1024 | static int ep_modify(struct eventpoll *ep, struct epitem *epi, struct epoll_event *event) | |
1025 | { | |
1026 | int pwake = 0; | |
1027 | unsigned int revents; | |
1028 | unsigned long flags; | |
1029 | ||
1030 | /* | |
1031 | * Set the new event interest mask before calling f_op->poll(), otherwise | |
1032 | * a potential race might occur. In fact if we do this operation inside | |
1033 | * the lock, an event might happen between the f_op->poll() call and the | |
1034 | * new event set registering. | |
1035 | */ | |
1036 | epi->event.events = event->events; | |
1037 | ||
1038 | /* | |
1039 | * Get current event bits. We can safely use the file* here because | |
1040 | * its usage count has been increased by the caller of this function. | |
1041 | */ | |
1042 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1043 | ||
1044 | write_lock_irqsave(&ep->lock, flags); | |
1045 | ||
1046 | /* Copy the data member from inside the lock */ | |
1047 | epi->event.data = event->data; | |
1048 | ||
1049 | /* | |
1050 | * If the item is not linked to the hash it means that it's on its | |
1051 | * way toward the removal. Do nothing in this case. | |
1052 | */ | |
1053 | if (EP_RB_LINKED(&epi->rbn)) { | |
1054 | /* | |
1055 | * If the item is "hot" and it is not registered inside the ready | |
1056 | * list, push it inside. If the item is not "hot" and it is currently | |
1057 | * registered inside the ready list, unlink it. | |
1058 | */ | |
1059 | if (revents & event->events) { | |
1060 | if (!EP_IS_LINKED(&epi->rdllink)) { | |
1061 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1062 | ||
1063 | /* Notify waiting tasks that events are available */ | |
1064 | if (waitqueue_active(&ep->wq)) | |
1065 | wake_up(&ep->wq); | |
1066 | if (waitqueue_active(&ep->poll_wait)) | |
1067 | pwake++; | |
1068 | } | |
1069 | } | |
1070 | } | |
1071 | ||
1072 | write_unlock_irqrestore(&ep->lock, flags); | |
1073 | ||
1074 | /* We have to call this outside the lock */ | |
1075 | if (pwake) | |
1076 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1077 | ||
1078 | return 0; | |
1079 | } | |
1080 | ||
1081 | ||
1082 | /* | |
1083 | * This function unregister poll callbacks from the associated file descriptor. | |
1084 | * Since this must be called without holding "ep->lock" the atomic exchange trick | |
1085 | * will protect us from multiple unregister. | |
1086 | */ | |
1087 | static void ep_unregister_pollwait(struct eventpoll *ep, struct epitem *epi) | |
1088 | { | |
1089 | int nwait; | |
1090 | struct list_head *lsthead = &epi->pwqlist; | |
1091 | struct eppoll_entry *pwq; | |
1092 | ||
1093 | /* This is called without locks, so we need the atomic exchange */ | |
1094 | nwait = xchg(&epi->nwait, 0); | |
1095 | ||
1096 | if (nwait) { | |
1097 | while (!list_empty(lsthead)) { | |
1098 | pwq = list_entry(lsthead->next, struct eppoll_entry, llink); | |
1099 | ||
1100 | EP_LIST_DEL(&pwq->llink); | |
1101 | remove_wait_queue(pwq->whead, &pwq->wait); | |
1102 | PWQ_MEM_FREE(pwq); | |
1103 | } | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | ||
1108 | /* | |
1109 | * Unlink the "struct epitem" from all places it might have been hooked up. | |
1110 | * This function must be called with write IRQ lock on "ep->lock". | |
1111 | */ | |
1112 | static int ep_unlink(struct eventpoll *ep, struct epitem *epi) | |
1113 | { | |
1114 | int error; | |
1115 | ||
1116 | /* | |
1117 | * It can happen that this one is called for an item already unlinked. | |
1118 | * The check protect us from doing a double unlink ( crash ). | |
1119 | */ | |
1120 | error = -ENOENT; | |
1121 | if (!EP_RB_LINKED(&epi->rbn)) | |
1122 | goto eexit_1; | |
1123 | ||
1124 | /* | |
1125 | * Clear the event mask for the unlinked item. This will avoid item | |
1126 | * notifications to be sent after the unlink operation from inside | |
1127 | * the kernel->userspace event transfer loop. | |
1128 | */ | |
1129 | epi->event.events = 0; | |
1130 | ||
1131 | /* | |
1132 | * At this point is safe to do the job, unlink the item from our rb-tree. | |
1133 | * This operation togheter with the above check closes the door to | |
1134 | * double unlinks. | |
1135 | */ | |
1136 | EP_RB_ERASE(&epi->rbn, &ep->rbr); | |
1137 | ||
1138 | /* | |
1139 | * If the item we are going to remove is inside the ready file descriptors | |
1140 | * we want to remove it from this list to avoid stale events. | |
1141 | */ | |
1142 | if (EP_IS_LINKED(&epi->rdllink)) | |
1143 | EP_LIST_DEL(&epi->rdllink); | |
1144 | ||
1145 | error = 0; | |
1146 | eexit_1: | |
1147 | ||
1148 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_unlink(%p, %p) = %d\n", | |
1149 | current, ep, epi->file, error)); | |
1150 | ||
1151 | return error; | |
1152 | } | |
1153 | ||
1154 | ||
1155 | /* | |
1156 | * Removes a "struct epitem" from the eventpoll hash and deallocates | |
1157 | * all the associated resources. | |
1158 | */ | |
1159 | static int ep_remove(struct eventpoll *ep, struct epitem *epi) | |
1160 | { | |
1161 | int error; | |
1162 | unsigned long flags; | |
1163 | struct file *file = epi->ffd.file; | |
1164 | ||
1165 | /* | |
1166 | * Removes poll wait queue hooks. We _have_ to do this without holding | |
1167 | * the "ep->lock" otherwise a deadlock might occur. This because of the | |
1168 | * sequence of the lock acquisition. Here we do "ep->lock" then the wait | |
1169 | * queue head lock when unregistering the wait queue. The wakeup callback | |
1170 | * will run by holding the wait queue head lock and will call our callback | |
1171 | * that will try to get "ep->lock". | |
1172 | */ | |
1173 | ep_unregister_pollwait(ep, epi); | |
1174 | ||
1175 | /* Remove the current item from the list of epoll hooks */ | |
1176 | spin_lock(&file->f_ep_lock); | |
1177 | if (EP_IS_LINKED(&epi->fllink)) | |
1178 | EP_LIST_DEL(&epi->fllink); | |
1179 | spin_unlock(&file->f_ep_lock); | |
1180 | ||
1181 | /* We need to acquire the write IRQ lock before calling ep_unlink() */ | |
1182 | write_lock_irqsave(&ep->lock, flags); | |
1183 | ||
1184 | /* Really unlink the item from the hash */ | |
1185 | error = ep_unlink(ep, epi); | |
1186 | ||
1187 | write_unlock_irqrestore(&ep->lock, flags); | |
1188 | ||
1189 | if (error) | |
1190 | goto eexit_1; | |
1191 | ||
1192 | /* At this point it is safe to free the eventpoll item */ | |
1193 | ep_release_epitem(epi); | |
1194 | ||
1195 | error = 0; | |
1196 | eexit_1: | |
1197 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: ep_remove(%p, %p) = %d\n", | |
1198 | current, ep, file, error)); | |
1199 | ||
1200 | return error; | |
1201 | } | |
1202 | ||
1203 | ||
1204 | /* | |
1205 | * This is the callback that is passed to the wait queue wakeup | |
1206 | * machanism. It is called by the stored file descriptors when they | |
1207 | * have events to report. | |
1208 | */ | |
1209 | static int ep_poll_callback(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
1210 | { | |
1211 | int pwake = 0; | |
1212 | unsigned long flags; | |
1213 | struct epitem *epi = EP_ITEM_FROM_WAIT(wait); | |
1214 | struct eventpoll *ep = epi->ep; | |
1215 | ||
1216 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: poll_callback(%p) epi=%p ep=%p\n", | |
1217 | current, epi->file, epi, ep)); | |
1218 | ||
1219 | write_lock_irqsave(&ep->lock, flags); | |
1220 | ||
1221 | /* | |
1222 | * If the event mask does not contain any poll(2) event, we consider the | |
1223 | * descriptor to be disabled. This condition is likely the effect of the | |
1224 | * EPOLLONESHOT bit that disables the descriptor when an event is received, | |
1225 | * until the next EPOLL_CTL_MOD will be issued. | |
1226 | */ | |
1227 | if (!(epi->event.events & ~EP_PRIVATE_BITS)) | |
1228 | goto is_disabled; | |
1229 | ||
1230 | /* If this file is already in the ready list we exit soon */ | |
1231 | if (EP_IS_LINKED(&epi->rdllink)) | |
1232 | goto is_linked; | |
1233 | ||
1234 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1235 | ||
1236 | is_linked: | |
1237 | /* | |
1238 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1239 | * wait list. | |
1240 | */ | |
1241 | if (waitqueue_active(&ep->wq)) | |
1242 | wake_up(&ep->wq); | |
1243 | if (waitqueue_active(&ep->poll_wait)) | |
1244 | pwake++; | |
1245 | ||
1246 | is_disabled: | |
1247 | write_unlock_irqrestore(&ep->lock, flags); | |
1248 | ||
1249 | /* We have to call this outside the lock */ | |
1250 | if (pwake) | |
1251 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1252 | ||
1253 | return 1; | |
1254 | } | |
1255 | ||
1256 | ||
1257 | static int ep_eventpoll_close(struct inode *inode, struct file *file) | |
1258 | { | |
1259 | struct eventpoll *ep = file->private_data; | |
1260 | ||
1261 | if (ep) { | |
1262 | ep_free(ep); | |
1263 | kfree(ep); | |
1264 | } | |
1265 | ||
1266 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: close() ep=%p\n", current, ep)); | |
1267 | return 0; | |
1268 | } | |
1269 | ||
1270 | ||
1271 | static unsigned int ep_eventpoll_poll(struct file *file, poll_table *wait) | |
1272 | { | |
1273 | unsigned int pollflags = 0; | |
1274 | unsigned long flags; | |
1275 | struct eventpoll *ep = file->private_data; | |
1276 | ||
1277 | /* Insert inside our poll wait queue */ | |
1278 | poll_wait(file, &ep->poll_wait, wait); | |
1279 | ||
1280 | /* Check our condition */ | |
1281 | read_lock_irqsave(&ep->lock, flags); | |
1282 | if (!list_empty(&ep->rdllist)) | |
1283 | pollflags = POLLIN | POLLRDNORM; | |
1284 | read_unlock_irqrestore(&ep->lock, flags); | |
1285 | ||
1286 | return pollflags; | |
1287 | } | |
1288 | ||
1289 | ||
1290 | /* | |
1291 | * Since we have to release the lock during the __copy_to_user() operation and | |
1292 | * during the f_op->poll() call, we try to collect the maximum number of items | |
1293 | * by reducing the irqlock/irqunlock switching rate. | |
1294 | */ | |
1295 | static int ep_collect_ready_items(struct eventpoll *ep, struct list_head *txlist, int maxevents) | |
1296 | { | |
1297 | int nepi; | |
1298 | unsigned long flags; | |
1299 | struct list_head *lsthead = &ep->rdllist, *lnk; | |
1300 | struct epitem *epi; | |
1301 | ||
1302 | write_lock_irqsave(&ep->lock, flags); | |
1303 | ||
1304 | for (nepi = 0, lnk = lsthead->next; lnk != lsthead && nepi < maxevents;) { | |
1305 | epi = list_entry(lnk, struct epitem, rdllink); | |
1306 | ||
1307 | lnk = lnk->next; | |
1308 | ||
1309 | /* If this file is already in the ready list we exit soon */ | |
1310 | if (!EP_IS_LINKED(&epi->txlink)) { | |
1311 | /* | |
1312 | * This is initialized in this way so that the default | |
1313 | * behaviour of the reinjecting code will be to push back | |
1314 | * the item inside the ready list. | |
1315 | */ | |
1316 | epi->revents = epi->event.events; | |
1317 | ||
1318 | /* Link the ready item into the transfer list */ | |
1319 | list_add(&epi->txlink, txlist); | |
1320 | nepi++; | |
1321 | ||
1322 | /* | |
1323 | * Unlink the item from the ready list. | |
1324 | */ | |
1325 | EP_LIST_DEL(&epi->rdllink); | |
1326 | } | |
1327 | } | |
1328 | ||
1329 | write_unlock_irqrestore(&ep->lock, flags); | |
1330 | ||
1331 | return nepi; | |
1332 | } | |
1333 | ||
1334 | ||
1335 | /* | |
1336 | * This function is called without holding the "ep->lock" since the call to | |
1337 | * __copy_to_user() might sleep, and also f_op->poll() might reenable the IRQ | |
1338 | * because of the way poll() is traditionally implemented in Linux. | |
1339 | */ | |
1340 | static int ep_send_events(struct eventpoll *ep, struct list_head *txlist, | |
1341 | struct epoll_event __user *events) | |
1342 | { | |
1343 | int eventcnt = 0; | |
1344 | unsigned int revents; | |
1345 | struct list_head *lnk; | |
1346 | struct epitem *epi; | |
1347 | ||
1348 | /* | |
1349 | * We can loop without lock because this is a task private list. | |
1350 | * The test done during the collection loop will guarantee us that | |
1351 | * another task will not try to collect this file. Also, items | |
1352 | * cannot vanish during the loop because we are holding "sem". | |
1353 | */ | |
1354 | list_for_each(lnk, txlist) { | |
1355 | epi = list_entry(lnk, struct epitem, txlink); | |
1356 | ||
1357 | /* | |
1358 | * Get the ready file event set. We can safely use the file | |
1359 | * because we are holding the "sem" in read and this will | |
1360 | * guarantee that both the file and the item will not vanish. | |
1361 | */ | |
1362 | revents = epi->ffd.file->f_op->poll(epi->ffd.file, NULL); | |
1363 | ||
1364 | /* | |
1365 | * Set the return event set for the current file descriptor. | |
1366 | * Note that only the task task was successfully able to link | |
1367 | * the item to its "txlist" will write this field. | |
1368 | */ | |
1369 | epi->revents = revents & epi->event.events; | |
1370 | ||
1371 | if (epi->revents) { | |
1372 | if (__put_user(epi->revents, | |
1373 | &events[eventcnt].events) || | |
1374 | __put_user(epi->event.data, | |
1375 | &events[eventcnt].data)) | |
1376 | return -EFAULT; | |
1377 | if (epi->event.events & EPOLLONESHOT) | |
1378 | epi->event.events &= EP_PRIVATE_BITS; | |
1379 | eventcnt++; | |
1380 | } | |
1381 | } | |
1382 | return eventcnt; | |
1383 | } | |
1384 | ||
1385 | ||
1386 | /* | |
1387 | * Walk through the transfer list we collected with ep_collect_ready_items() | |
1388 | * and, if 1) the item is still "alive" 2) its event set is not empty 3) it's | |
1389 | * not already linked, links it to the ready list. Same as above, we are holding | |
1390 | * "sem" so items cannot vanish underneath our nose. | |
1391 | */ | |
1392 | static void ep_reinject_items(struct eventpoll *ep, struct list_head *txlist) | |
1393 | { | |
1394 | int ricnt = 0, pwake = 0; | |
1395 | unsigned long flags; | |
1396 | struct epitem *epi; | |
1397 | ||
1398 | write_lock_irqsave(&ep->lock, flags); | |
1399 | ||
1400 | while (!list_empty(txlist)) { | |
1401 | epi = list_entry(txlist->next, struct epitem, txlink); | |
1402 | ||
1403 | /* Unlink the current item from the transfer list */ | |
1404 | EP_LIST_DEL(&epi->txlink); | |
1405 | ||
1406 | /* | |
1407 | * If the item is no more linked to the interest set, we don't | |
1408 | * have to push it inside the ready list because the following | |
1409 | * ep_release_epitem() is going to drop it. Also, if the current | |
1410 | * item is set to have an Edge Triggered behaviour, we don't have | |
1411 | * to push it back either. | |
1412 | */ | |
1413 | if (EP_RB_LINKED(&epi->rbn) && !(epi->event.events & EPOLLET) && | |
1414 | (epi->revents & epi->event.events) && !EP_IS_LINKED(&epi->rdllink)) { | |
1415 | list_add_tail(&epi->rdllink, &ep->rdllist); | |
1416 | ricnt++; | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | if (ricnt) { | |
1421 | /* | |
1422 | * Wake up ( if active ) both the eventpoll wait list and the ->poll() | |
1423 | * wait list. | |
1424 | */ | |
1425 | if (waitqueue_active(&ep->wq)) | |
1426 | wake_up(&ep->wq); | |
1427 | if (waitqueue_active(&ep->poll_wait)) | |
1428 | pwake++; | |
1429 | } | |
1430 | ||
1431 | write_unlock_irqrestore(&ep->lock, flags); | |
1432 | ||
1433 | /* We have to call this outside the lock */ | |
1434 | if (pwake) | |
1435 | ep_poll_safewake(&psw, &ep->poll_wait); | |
1436 | } | |
1437 | ||
1438 | ||
1439 | /* | |
1440 | * Perform the transfer of events to user space. | |
1441 | */ | |
1442 | static int ep_events_transfer(struct eventpoll *ep, | |
1443 | struct epoll_event __user *events, int maxevents) | |
1444 | { | |
1445 | int eventcnt = 0; | |
1446 | struct list_head txlist; | |
1447 | ||
1448 | INIT_LIST_HEAD(&txlist); | |
1449 | ||
1450 | /* | |
1451 | * We need to lock this because we could be hit by | |
1452 | * eventpoll_release_file() and epoll_ctl(EPOLL_CTL_DEL). | |
1453 | */ | |
1454 | down_read(&ep->sem); | |
1455 | ||
1456 | /* Collect/extract ready items */ | |
1457 | if (ep_collect_ready_items(ep, &txlist, maxevents) > 0) { | |
1458 | /* Build result set in userspace */ | |
1459 | eventcnt = ep_send_events(ep, &txlist, events); | |
1460 | ||
1461 | /* Reinject ready items into the ready list */ | |
1462 | ep_reinject_items(ep, &txlist); | |
1463 | } | |
1464 | ||
1465 | up_read(&ep->sem); | |
1466 | ||
1467 | return eventcnt; | |
1468 | } | |
1469 | ||
1470 | ||
1471 | static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events, | |
1472 | int maxevents, long timeout) | |
1473 | { | |
1474 | int res, eavail; | |
1475 | unsigned long flags; | |
1476 | long jtimeout; | |
1477 | wait_queue_t wait; | |
1478 | ||
1479 | /* | |
1480 | * Calculate the timeout by checking for the "infinite" value ( -1 ) | |
1481 | * and the overflow condition. The passed timeout is in milliseconds, | |
1482 | * that why (t * HZ) / 1000. | |
1483 | */ | |
1484 | jtimeout = timeout == -1 || timeout > (MAX_SCHEDULE_TIMEOUT - 1000) / HZ ? | |
1485 | MAX_SCHEDULE_TIMEOUT: (timeout * HZ + 999) / 1000; | |
1486 | ||
1487 | retry: | |
1488 | write_lock_irqsave(&ep->lock, flags); | |
1489 | ||
1490 | res = 0; | |
1491 | if (list_empty(&ep->rdllist)) { | |
1492 | /* | |
1493 | * We don't have any available event to return to the caller. | |
1494 | * We need to sleep here, and we will be wake up by | |
1495 | * ep_poll_callback() when events will become available. | |
1496 | */ | |
1497 | init_waitqueue_entry(&wait, current); | |
1498 | add_wait_queue(&ep->wq, &wait); | |
1499 | ||
1500 | for (;;) { | |
1501 | /* | |
1502 | * We don't want to sleep if the ep_poll_callback() sends us | |
1503 | * a wakeup in between. That's why we set the task state | |
1504 | * to TASK_INTERRUPTIBLE before doing the checks. | |
1505 | */ | |
1506 | set_current_state(TASK_INTERRUPTIBLE); | |
1507 | if (!list_empty(&ep->rdllist) || !jtimeout) | |
1508 | break; | |
1509 | if (signal_pending(current)) { | |
1510 | res = -EINTR; | |
1511 | break; | |
1512 | } | |
1513 | ||
1514 | write_unlock_irqrestore(&ep->lock, flags); | |
1515 | jtimeout = schedule_timeout(jtimeout); | |
1516 | write_lock_irqsave(&ep->lock, flags); | |
1517 | } | |
1518 | remove_wait_queue(&ep->wq, &wait); | |
1519 | ||
1520 | set_current_state(TASK_RUNNING); | |
1521 | } | |
1522 | ||
1523 | /* Is it worth to try to dig for events ? */ | |
1524 | eavail = !list_empty(&ep->rdllist); | |
1525 | ||
1526 | write_unlock_irqrestore(&ep->lock, flags); | |
1527 | ||
1528 | /* | |
1529 | * Try to transfer events to user space. In case we get 0 events and | |
1530 | * there's still timeout left over, we go trying again in search of | |
1531 | * more luck. | |
1532 | */ | |
1533 | if (!res && eavail && | |
1534 | !(res = ep_events_transfer(ep, events, maxevents)) && jtimeout) | |
1535 | goto retry; | |
1536 | ||
1537 | return res; | |
1538 | } | |
1539 | ||
1540 | ||
1541 | static int eventpollfs_delete_dentry(struct dentry *dentry) | |
1542 | { | |
1543 | ||
1544 | return 1; | |
1545 | } | |
1546 | ||
1547 | ||
1548 | static struct inode *ep_eventpoll_inode(void) | |
1549 | { | |
1550 | int error = -ENOMEM; | |
1551 | struct inode *inode = new_inode(eventpoll_mnt->mnt_sb); | |
1552 | ||
1553 | if (!inode) | |
1554 | goto eexit_1; | |
1555 | ||
1556 | inode->i_fop = &eventpoll_fops; | |
1557 | ||
1558 | /* | |
1559 | * Mark the inode dirty from the very beginning, | |
1560 | * that way it will never be moved to the dirty | |
1561 | * list because mark_inode_dirty() will think | |
1562 | * that it already _is_ on the dirty list. | |
1563 | */ | |
1564 | inode->i_state = I_DIRTY; | |
1565 | inode->i_mode = S_IRUSR | S_IWUSR; | |
1566 | inode->i_uid = current->fsuid; | |
1567 | inode->i_gid = current->fsgid; | |
1568 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
1569 | inode->i_blksize = PAGE_SIZE; | |
1570 | return inode; | |
1571 | ||
1572 | eexit_1: | |
1573 | return ERR_PTR(error); | |
1574 | } | |
1575 | ||
1576 | ||
1577 | static struct super_block * | |
1578 | eventpollfs_get_sb(struct file_system_type *fs_type, int flags, | |
1579 | const char *dev_name, void *data) | |
1580 | { | |
1581 | return get_sb_pseudo(fs_type, "eventpoll:", NULL, EVENTPOLLFS_MAGIC); | |
1582 | } | |
1583 | ||
1584 | ||
1585 | static int __init eventpoll_init(void) | |
1586 | { | |
1587 | int error; | |
1588 | ||
1589 | init_MUTEX(&epsem); | |
1590 | ||
1591 | /* Initialize the structure used to perform safe poll wait head wake ups */ | |
1592 | ep_poll_safewake_init(&psw); | |
1593 | ||
1594 | /* Allocates slab cache used to allocate "struct epitem" items */ | |
1595 | epi_cache = kmem_cache_create("eventpoll_epi", sizeof(struct epitem), | |
1596 | 0, SLAB_HWCACHE_ALIGN|EPI_SLAB_DEBUG|SLAB_PANIC, | |
1597 | NULL, NULL); | |
1598 | ||
1599 | /* Allocates slab cache used to allocate "struct eppoll_entry" */ | |
1600 | pwq_cache = kmem_cache_create("eventpoll_pwq", | |
1601 | sizeof(struct eppoll_entry), 0, | |
1602 | EPI_SLAB_DEBUG|SLAB_PANIC, NULL, NULL); | |
1603 | ||
1604 | /* | |
1605 | * Register the virtual file system that will be the source of inodes | |
1606 | * for the eventpoll files | |
1607 | */ | |
1608 | error = register_filesystem(&eventpoll_fs_type); | |
1609 | if (error) | |
1610 | goto epanic; | |
1611 | ||
1612 | /* Mount the above commented virtual file system */ | |
1613 | eventpoll_mnt = kern_mount(&eventpoll_fs_type); | |
1614 | error = PTR_ERR(eventpoll_mnt); | |
1615 | if (IS_ERR(eventpoll_mnt)) | |
1616 | goto epanic; | |
1617 | ||
1618 | DNPRINTK(3, (KERN_INFO "[%p] eventpoll: successfully initialized.\n", | |
1619 | current)); | |
1620 | return 0; | |
1621 | ||
1622 | epanic: | |
1623 | panic("eventpoll_init() failed\n"); | |
1624 | } | |
1625 | ||
1626 | ||
1627 | static void __exit eventpoll_exit(void) | |
1628 | { | |
1629 | /* Undo all operations done inside eventpoll_init() */ | |
1630 | unregister_filesystem(&eventpoll_fs_type); | |
1631 | mntput(eventpoll_mnt); | |
1632 | kmem_cache_destroy(pwq_cache); | |
1633 | kmem_cache_destroy(epi_cache); | |
1634 | } | |
1635 | ||
1636 | module_init(eventpoll_init); | |
1637 | module_exit(eventpoll_exit); | |
1638 | ||
1639 | MODULE_LICENSE("GPL"); |