Root/
Source at commit b386be689295730688885552666ea40b2e639b14 created 11 years 11 months ago. By Maarten ter Huurne, Revert "MIPS: JZ4740: reset: Initialize hibernate wakeup counters." | |
---|---|
1 | /* |
2 | * An async IO implementation for Linux |
3 | * Written by Benjamin LaHaise <bcrl@kvack.org> |
4 | * |
5 | * Implements an efficient asynchronous io interface. |
6 | * |
7 | * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. |
8 | * |
9 | * See ../COPYING for licensing terms. |
10 | */ |
11 | #include <linux/kernel.h> |
12 | #include <linux/init.h> |
13 | #include <linux/errno.h> |
14 | #include <linux/time.h> |
15 | #include <linux/aio_abi.h> |
16 | #include <linux/module.h> |
17 | #include <linux/syscalls.h> |
18 | #include <linux/backing-dev.h> |
19 | #include <linux/uio.h> |
20 | |
21 | #define DEBUG 0 |
22 | |
23 | #include <linux/sched.h> |
24 | #include <linux/fs.h> |
25 | #include <linux/file.h> |
26 | #include <linux/mm.h> |
27 | #include <linux/mman.h> |
28 | #include <linux/mmu_context.h> |
29 | #include <linux/slab.h> |
30 | #include <linux/timer.h> |
31 | #include <linux/aio.h> |
32 | #include <linux/highmem.h> |
33 | #include <linux/workqueue.h> |
34 | #include <linux/security.h> |
35 | #include <linux/eventfd.h> |
36 | #include <linux/blkdev.h> |
37 | #include <linux/compat.h> |
38 | |
39 | #include <asm/kmap_types.h> |
40 | #include <asm/uaccess.h> |
41 | |
42 | #if DEBUG > 1 |
43 | #define dprintk printk |
44 | #else |
45 | #define dprintk(x...) do { ; } while (0) |
46 | #endif |
47 | |
48 | /*------ sysctl variables----*/ |
49 | static DEFINE_SPINLOCK(aio_nr_lock); |
50 | unsigned long aio_nr; /* current system wide number of aio requests */ |
51 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ |
52 | /*----end sysctl variables---*/ |
53 | |
54 | static struct kmem_cache *kiocb_cachep; |
55 | static struct kmem_cache *kioctx_cachep; |
56 | |
57 | static struct workqueue_struct *aio_wq; |
58 | |
59 | /* Used for rare fput completion. */ |
60 | static void aio_fput_routine(struct work_struct *); |
61 | static DECLARE_WORK(fput_work, aio_fput_routine); |
62 | |
63 | static DEFINE_SPINLOCK(fput_lock); |
64 | static LIST_HEAD(fput_head); |
65 | |
66 | static void aio_kick_handler(struct work_struct *); |
67 | static void aio_queue_work(struct kioctx *); |
68 | |
69 | /* aio_setup |
70 | * Creates the slab caches used by the aio routines, panic on |
71 | * failure as this is done early during the boot sequence. |
72 | */ |
73 | static int __init aio_setup(void) |
74 | { |
75 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
76 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
77 | |
78 | aio_wq = alloc_workqueue("aio", 0, 1); /* used to limit concurrency */ |
79 | BUG_ON(!aio_wq); |
80 | |
81 | pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); |
82 | |
83 | return 0; |
84 | } |
85 | __initcall(aio_setup); |
86 | |
87 | static void aio_free_ring(struct kioctx *ctx) |
88 | { |
89 | struct aio_ring_info *info = &ctx->ring_info; |
90 | long i; |
91 | |
92 | for (i=0; i<info->nr_pages; i++) |
93 | put_page(info->ring_pages[i]); |
94 | |
95 | if (info->mmap_size) { |
96 | down_write(&ctx->mm->mmap_sem); |
97 | do_munmap(ctx->mm, info->mmap_base, info->mmap_size); |
98 | up_write(&ctx->mm->mmap_sem); |
99 | } |
100 | |
101 | if (info->ring_pages && info->ring_pages != info->internal_pages) |
102 | kfree(info->ring_pages); |
103 | info->ring_pages = NULL; |
104 | info->nr = 0; |
105 | } |
106 | |
107 | static int aio_setup_ring(struct kioctx *ctx) |
108 | { |
109 | struct aio_ring *ring; |
110 | struct aio_ring_info *info = &ctx->ring_info; |
111 | unsigned nr_events = ctx->max_reqs; |
112 | unsigned long size; |
113 | int nr_pages; |
114 | |
115 | /* Compensate for the ring buffer's head/tail overlap entry */ |
116 | nr_events += 2; /* 1 is required, 2 for good luck */ |
117 | |
118 | size = sizeof(struct aio_ring); |
119 | size += sizeof(struct io_event) * nr_events; |
120 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; |
121 | |
122 | if (nr_pages < 0) |
123 | return -EINVAL; |
124 | |
125 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); |
126 | |
127 | info->nr = 0; |
128 | info->ring_pages = info->internal_pages; |
129 | if (nr_pages > AIO_RING_PAGES) { |
130 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
131 | if (!info->ring_pages) |
132 | return -ENOMEM; |
133 | } |
134 | |
135 | info->mmap_size = nr_pages * PAGE_SIZE; |
136 | dprintk("attempting mmap of %lu bytes\n", info->mmap_size); |
137 | down_write(&ctx->mm->mmap_sem); |
138 | info->mmap_base = do_mmap(NULL, 0, info->mmap_size, |
139 | PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, |
140 | 0); |
141 | if (IS_ERR((void *)info->mmap_base)) { |
142 | up_write(&ctx->mm->mmap_sem); |
143 | info->mmap_size = 0; |
144 | aio_free_ring(ctx); |
145 | return -EAGAIN; |
146 | } |
147 | |
148 | dprintk("mmap address: 0x%08lx\n", info->mmap_base); |
149 | info->nr_pages = get_user_pages(current, ctx->mm, |
150 | info->mmap_base, nr_pages, |
151 | 1, 0, info->ring_pages, NULL); |
152 | up_write(&ctx->mm->mmap_sem); |
153 | |
154 | if (unlikely(info->nr_pages != nr_pages)) { |
155 | aio_free_ring(ctx); |
156 | return -EAGAIN; |
157 | } |
158 | |
159 | ctx->user_id = info->mmap_base; |
160 | |
161 | info->nr = nr_events; /* trusted copy */ |
162 | |
163 | ring = kmap_atomic(info->ring_pages[0], KM_USER0); |
164 | ring->nr = nr_events; /* user copy */ |
165 | ring->id = ctx->user_id; |
166 | ring->head = ring->tail = 0; |
167 | ring->magic = AIO_RING_MAGIC; |
168 | ring->compat_features = AIO_RING_COMPAT_FEATURES; |
169 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; |
170 | ring->header_length = sizeof(struct aio_ring); |
171 | kunmap_atomic(ring, KM_USER0); |
172 | |
173 | return 0; |
174 | } |
175 | |
176 | |
177 | /* aio_ring_event: returns a pointer to the event at the given index from |
178 | * kmap_atomic(, km). Release the pointer with put_aio_ring_event(); |
179 | */ |
180 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
181 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) |
182 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) |
183 | |
184 | #define aio_ring_event(info, nr, km) ({ \ |
185 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
186 | struct io_event *__event; \ |
187 | __event = kmap_atomic( \ |
188 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \ |
189 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
190 | __event; \ |
191 | }) |
192 | |
193 | #define put_aio_ring_event(event, km) do { \ |
194 | struct io_event *__event = (event); \ |
195 | (void)__event; \ |
196 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \ |
197 | } while(0) |
198 | |
199 | static void ctx_rcu_free(struct rcu_head *head) |
200 | { |
201 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); |
202 | unsigned nr_events = ctx->max_reqs; |
203 | |
204 | kmem_cache_free(kioctx_cachep, ctx); |
205 | |
206 | if (nr_events) { |
207 | spin_lock(&aio_nr_lock); |
208 | BUG_ON(aio_nr - nr_events > aio_nr); |
209 | aio_nr -= nr_events; |
210 | spin_unlock(&aio_nr_lock); |
211 | } |
212 | } |
213 | |
214 | /* __put_ioctx |
215 | * Called when the last user of an aio context has gone away, |
216 | * and the struct needs to be freed. |
217 | */ |
218 | static void __put_ioctx(struct kioctx *ctx) |
219 | { |
220 | BUG_ON(ctx->reqs_active); |
221 | |
222 | cancel_delayed_work(&ctx->wq); |
223 | cancel_work_sync(&ctx->wq.work); |
224 | aio_free_ring(ctx); |
225 | mmdrop(ctx->mm); |
226 | ctx->mm = NULL; |
227 | pr_debug("__put_ioctx: freeing %p\n", ctx); |
228 | call_rcu(&ctx->rcu_head, ctx_rcu_free); |
229 | } |
230 | |
231 | static inline int try_get_ioctx(struct kioctx *kioctx) |
232 | { |
233 | return atomic_inc_not_zero(&kioctx->users); |
234 | } |
235 | |
236 | static inline void put_ioctx(struct kioctx *kioctx) |
237 | { |
238 | BUG_ON(atomic_read(&kioctx->users) <= 0); |
239 | if (unlikely(atomic_dec_and_test(&kioctx->users))) |
240 | __put_ioctx(kioctx); |
241 | } |
242 | |
243 | /* ioctx_alloc |
244 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. |
245 | */ |
246 | static struct kioctx *ioctx_alloc(unsigned nr_events) |
247 | { |
248 | struct mm_struct *mm; |
249 | struct kioctx *ctx; |
250 | int did_sync = 0; |
251 | |
252 | /* Prevent overflows */ |
253 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || |
254 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { |
255 | pr_debug("ENOMEM: nr_events too high\n"); |
256 | return ERR_PTR(-EINVAL); |
257 | } |
258 | |
259 | if ((unsigned long)nr_events > aio_max_nr) |
260 | return ERR_PTR(-EAGAIN); |
261 | |
262 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
263 | if (!ctx) |
264 | return ERR_PTR(-ENOMEM); |
265 | |
266 | ctx->max_reqs = nr_events; |
267 | mm = ctx->mm = current->mm; |
268 | atomic_inc(&mm->mm_count); |
269 | |
270 | atomic_set(&ctx->users, 2); |
271 | spin_lock_init(&ctx->ctx_lock); |
272 | spin_lock_init(&ctx->ring_info.ring_lock); |
273 | init_waitqueue_head(&ctx->wait); |
274 | |
275 | INIT_LIST_HEAD(&ctx->active_reqs); |
276 | INIT_LIST_HEAD(&ctx->run_list); |
277 | INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler); |
278 | |
279 | if (aio_setup_ring(ctx) < 0) |
280 | goto out_freectx; |
281 | |
282 | /* limit the number of system wide aios */ |
283 | do { |
284 | spin_lock_bh(&aio_nr_lock); |
285 | if (aio_nr + nr_events > aio_max_nr || |
286 | aio_nr + nr_events < aio_nr) |
287 | ctx->max_reqs = 0; |
288 | else |
289 | aio_nr += ctx->max_reqs; |
290 | spin_unlock_bh(&aio_nr_lock); |
291 | if (ctx->max_reqs || did_sync) |
292 | break; |
293 | |
294 | /* wait for rcu callbacks to have completed before giving up */ |
295 | synchronize_rcu(); |
296 | did_sync = 1; |
297 | ctx->max_reqs = nr_events; |
298 | } while (1); |
299 | |
300 | if (ctx->max_reqs == 0) |
301 | goto out_cleanup; |
302 | |
303 | /* now link into global list. */ |
304 | spin_lock(&mm->ioctx_lock); |
305 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); |
306 | spin_unlock(&mm->ioctx_lock); |
307 | |
308 | dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
309 | ctx, ctx->user_id, current->mm, ctx->ring_info.nr); |
310 | return ctx; |
311 | |
312 | out_cleanup: |
313 | __put_ioctx(ctx); |
314 | return ERR_PTR(-EAGAIN); |
315 | |
316 | out_freectx: |
317 | mmdrop(mm); |
318 | kmem_cache_free(kioctx_cachep, ctx); |
319 | ctx = ERR_PTR(-ENOMEM); |
320 | |
321 | dprintk("aio: error allocating ioctx %p\n", ctx); |
322 | return ctx; |
323 | } |
324 | |
325 | /* aio_cancel_all |
326 | * Cancels all outstanding aio requests on an aio context. Used |
327 | * when the processes owning a context have all exited to encourage |
328 | * the rapid destruction of the kioctx. |
329 | */ |
330 | static void aio_cancel_all(struct kioctx *ctx) |
331 | { |
332 | int (*cancel)(struct kiocb *, struct io_event *); |
333 | struct io_event res; |
334 | spin_lock_irq(&ctx->ctx_lock); |
335 | ctx->dead = 1; |
336 | while (!list_empty(&ctx->active_reqs)) { |
337 | struct list_head *pos = ctx->active_reqs.next; |
338 | struct kiocb *iocb = list_kiocb(pos); |
339 | list_del_init(&iocb->ki_list); |
340 | cancel = iocb->ki_cancel; |
341 | kiocbSetCancelled(iocb); |
342 | if (cancel) { |
343 | iocb->ki_users++; |
344 | spin_unlock_irq(&ctx->ctx_lock); |
345 | cancel(iocb, &res); |
346 | spin_lock_irq(&ctx->ctx_lock); |
347 | } |
348 | } |
349 | spin_unlock_irq(&ctx->ctx_lock); |
350 | } |
351 | |
352 | static void wait_for_all_aios(struct kioctx *ctx) |
353 | { |
354 | struct task_struct *tsk = current; |
355 | DECLARE_WAITQUEUE(wait, tsk); |
356 | |
357 | spin_lock_irq(&ctx->ctx_lock); |
358 | if (!ctx->reqs_active) |
359 | goto out; |
360 | |
361 | add_wait_queue(&ctx->wait, &wait); |
362 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
363 | while (ctx->reqs_active) { |
364 | spin_unlock_irq(&ctx->ctx_lock); |
365 | io_schedule(); |
366 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
367 | spin_lock_irq(&ctx->ctx_lock); |
368 | } |
369 | __set_task_state(tsk, TASK_RUNNING); |
370 | remove_wait_queue(&ctx->wait, &wait); |
371 | |
372 | out: |
373 | spin_unlock_irq(&ctx->ctx_lock); |
374 | } |
375 | |
376 | /* wait_on_sync_kiocb: |
377 | * Waits on the given sync kiocb to complete. |
378 | */ |
379 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
380 | { |
381 | while (iocb->ki_users) { |
382 | set_current_state(TASK_UNINTERRUPTIBLE); |
383 | if (!iocb->ki_users) |
384 | break; |
385 | io_schedule(); |
386 | } |
387 | __set_current_state(TASK_RUNNING); |
388 | return iocb->ki_user_data; |
389 | } |
390 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
391 | |
392 | /* exit_aio: called when the last user of mm goes away. At this point, |
393 | * there is no way for any new requests to be submited or any of the |
394 | * io_* syscalls to be called on the context. However, there may be |
395 | * outstanding requests which hold references to the context; as they |
396 | * go away, they will call put_ioctx and release any pinned memory |
397 | * associated with the request (held via struct page * references). |
398 | */ |
399 | void exit_aio(struct mm_struct *mm) |
400 | { |
401 | struct kioctx *ctx; |
402 | |
403 | while (!hlist_empty(&mm->ioctx_list)) { |
404 | ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); |
405 | hlist_del_rcu(&ctx->list); |
406 | |
407 | aio_cancel_all(ctx); |
408 | |
409 | wait_for_all_aios(ctx); |
410 | /* |
411 | * Ensure we don't leave the ctx on the aio_wq |
412 | */ |
413 | cancel_work_sync(&ctx->wq.work); |
414 | |
415 | if (1 != atomic_read(&ctx->users)) |
416 | printk(KERN_DEBUG |
417 | "exit_aio:ioctx still alive: %d %d %d\n", |
418 | atomic_read(&ctx->users), ctx->dead, |
419 | ctx->reqs_active); |
420 | put_ioctx(ctx); |
421 | } |
422 | } |
423 | |
424 | /* aio_get_req |
425 | * Allocate a slot for an aio request. Increments the users count |
426 | * of the kioctx so that the kioctx stays around until all requests are |
427 | * complete. Returns NULL if no requests are free. |
428 | * |
429 | * Returns with kiocb->users set to 2. The io submit code path holds |
430 | * an extra reference while submitting the i/o. |
431 | * This prevents races between the aio code path referencing the |
432 | * req (after submitting it) and aio_complete() freeing the req. |
433 | */ |
434 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
435 | { |
436 | struct kiocb *req = NULL; |
437 | |
438 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); |
439 | if (unlikely(!req)) |
440 | return NULL; |
441 | |
442 | req->ki_flags = 0; |
443 | req->ki_users = 2; |
444 | req->ki_key = 0; |
445 | req->ki_ctx = ctx; |
446 | req->ki_cancel = NULL; |
447 | req->ki_retry = NULL; |
448 | req->ki_dtor = NULL; |
449 | req->private = NULL; |
450 | req->ki_iovec = NULL; |
451 | INIT_LIST_HEAD(&req->ki_run_list); |
452 | req->ki_eventfd = NULL; |
453 | |
454 | return req; |
455 | } |
456 | |
457 | /* |
458 | * struct kiocb's are allocated in batches to reduce the number of |
459 | * times the ctx lock is acquired and released. |
460 | */ |
461 | #define KIOCB_BATCH_SIZE 32L |
462 | struct kiocb_batch { |
463 | struct list_head head; |
464 | long count; /* number of requests left to allocate */ |
465 | }; |
466 | |
467 | static void kiocb_batch_init(struct kiocb_batch *batch, long total) |
468 | { |
469 | INIT_LIST_HEAD(&batch->head); |
470 | batch->count = total; |
471 | } |
472 | |
473 | static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) |
474 | { |
475 | struct kiocb *req, *n; |
476 | |
477 | if (list_empty(&batch->head)) |
478 | return; |
479 | |
480 | spin_lock_irq(&ctx->ctx_lock); |
481 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
482 | list_del(&req->ki_batch); |
483 | list_del(&req->ki_list); |
484 | kmem_cache_free(kiocb_cachep, req); |
485 | ctx->reqs_active--; |
486 | } |
487 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
488 | wake_up_all(&ctx->wait); |
489 | spin_unlock_irq(&ctx->ctx_lock); |
490 | } |
491 | |
492 | /* |
493 | * Allocate a batch of kiocbs. This avoids taking and dropping the |
494 | * context lock a lot during setup. |
495 | */ |
496 | static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) |
497 | { |
498 | unsigned short allocated, to_alloc; |
499 | long avail; |
500 | bool called_fput = false; |
501 | struct kiocb *req, *n; |
502 | struct aio_ring *ring; |
503 | |
504 | to_alloc = min(batch->count, KIOCB_BATCH_SIZE); |
505 | for (allocated = 0; allocated < to_alloc; allocated++) { |
506 | req = __aio_get_req(ctx); |
507 | if (!req) |
508 | /* allocation failed, go with what we've got */ |
509 | break; |
510 | list_add(&req->ki_batch, &batch->head); |
511 | } |
512 | |
513 | if (allocated == 0) |
514 | goto out; |
515 | |
516 | retry: |
517 | spin_lock_irq(&ctx->ctx_lock); |
518 | ring = kmap_atomic(ctx->ring_info.ring_pages[0]); |
519 | |
520 | avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active; |
521 | BUG_ON(avail < 0); |
522 | if (avail == 0 && !called_fput) { |
523 | /* |
524 | * Handle a potential starvation case. It is possible that |
525 | * we hold the last reference on a struct file, causing us |
526 | * to delay the final fput to non-irq context. In this case, |
527 | * ctx->reqs_active is artificially high. Calling the fput |
528 | * routine here may free up a slot in the event completion |
529 | * ring, allowing this allocation to succeed. |
530 | */ |
531 | kunmap_atomic(ring); |
532 | spin_unlock_irq(&ctx->ctx_lock); |
533 | aio_fput_routine(NULL); |
534 | called_fput = true; |
535 | goto retry; |
536 | } |
537 | |
538 | if (avail < allocated) { |
539 | /* Trim back the number of requests. */ |
540 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
541 | list_del(&req->ki_batch); |
542 | kmem_cache_free(kiocb_cachep, req); |
543 | if (--allocated <= avail) |
544 | break; |
545 | } |
546 | } |
547 | |
548 | batch->count -= allocated; |
549 | list_for_each_entry(req, &batch->head, ki_batch) { |
550 | list_add(&req->ki_list, &ctx->active_reqs); |
551 | ctx->reqs_active++; |
552 | } |
553 | |
554 | kunmap_atomic(ring); |
555 | spin_unlock_irq(&ctx->ctx_lock); |
556 | |
557 | out: |
558 | return allocated; |
559 | } |
560 | |
561 | static inline struct kiocb *aio_get_req(struct kioctx *ctx, |
562 | struct kiocb_batch *batch) |
563 | { |
564 | struct kiocb *req; |
565 | |
566 | if (list_empty(&batch->head)) |
567 | if (kiocb_batch_refill(ctx, batch) == 0) |
568 | return NULL; |
569 | req = list_first_entry(&batch->head, struct kiocb, ki_batch); |
570 | list_del(&req->ki_batch); |
571 | return req; |
572 | } |
573 | |
574 | static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) |
575 | { |
576 | assert_spin_locked(&ctx->ctx_lock); |
577 | |
578 | if (req->ki_eventfd != NULL) |
579 | eventfd_ctx_put(req->ki_eventfd); |
580 | if (req->ki_dtor) |
581 | req->ki_dtor(req); |
582 | if (req->ki_iovec != &req->ki_inline_vec) |
583 | kfree(req->ki_iovec); |
584 | kmem_cache_free(kiocb_cachep, req); |
585 | ctx->reqs_active--; |
586 | |
587 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
588 | wake_up_all(&ctx->wait); |
589 | } |
590 | |
591 | static void aio_fput_routine(struct work_struct *data) |
592 | { |
593 | spin_lock_irq(&fput_lock); |
594 | while (likely(!list_empty(&fput_head))) { |
595 | struct kiocb *req = list_kiocb(fput_head.next); |
596 | struct kioctx *ctx = req->ki_ctx; |
597 | |
598 | list_del(&req->ki_list); |
599 | spin_unlock_irq(&fput_lock); |
600 | |
601 | /* Complete the fput(s) */ |
602 | if (req->ki_filp != NULL) |
603 | fput(req->ki_filp); |
604 | |
605 | /* Link the iocb into the context's free list */ |
606 | rcu_read_lock(); |
607 | spin_lock_irq(&ctx->ctx_lock); |
608 | really_put_req(ctx, req); |
609 | /* |
610 | * at that point ctx might've been killed, but actual |
611 | * freeing is RCU'd |
612 | */ |
613 | spin_unlock_irq(&ctx->ctx_lock); |
614 | rcu_read_unlock(); |
615 | |
616 | spin_lock_irq(&fput_lock); |
617 | } |
618 | spin_unlock_irq(&fput_lock); |
619 | } |
620 | |
621 | /* __aio_put_req |
622 | * Returns true if this put was the last user of the request. |
623 | */ |
624 | static int __aio_put_req(struct kioctx *ctx, struct kiocb *req) |
625 | { |
626 | dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n", |
627 | req, atomic_long_read(&req->ki_filp->f_count)); |
628 | |
629 | assert_spin_locked(&ctx->ctx_lock); |
630 | |
631 | req->ki_users--; |
632 | BUG_ON(req->ki_users < 0); |
633 | if (likely(req->ki_users)) |
634 | return 0; |
635 | list_del(&req->ki_list); /* remove from active_reqs */ |
636 | req->ki_cancel = NULL; |
637 | req->ki_retry = NULL; |
638 | |
639 | /* |
640 | * Try to optimize the aio and eventfd file* puts, by avoiding to |
641 | * schedule work in case it is not final fput() time. In normal cases, |
642 | * we would not be holding the last reference to the file*, so |
643 | * this function will be executed w/out any aio kthread wakeup. |
644 | */ |
645 | if (unlikely(!fput_atomic(req->ki_filp))) { |
646 | spin_lock(&fput_lock); |
647 | list_add(&req->ki_list, &fput_head); |
648 | spin_unlock(&fput_lock); |
649 | schedule_work(&fput_work); |
650 | } else { |
651 | req->ki_filp = NULL; |
652 | really_put_req(ctx, req); |
653 | } |
654 | return 1; |
655 | } |
656 | |
657 | /* aio_put_req |
658 | * Returns true if this put was the last user of the kiocb, |
659 | * false if the request is still in use. |
660 | */ |
661 | int aio_put_req(struct kiocb *req) |
662 | { |
663 | struct kioctx *ctx = req->ki_ctx; |
664 | int ret; |
665 | spin_lock_irq(&ctx->ctx_lock); |
666 | ret = __aio_put_req(ctx, req); |
667 | spin_unlock_irq(&ctx->ctx_lock); |
668 | return ret; |
669 | } |
670 | EXPORT_SYMBOL(aio_put_req); |
671 | |
672 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
673 | { |
674 | struct mm_struct *mm = current->mm; |
675 | struct kioctx *ctx, *ret = NULL; |
676 | struct hlist_node *n; |
677 | |
678 | rcu_read_lock(); |
679 | |
680 | hlist_for_each_entry_rcu(ctx, n, &mm->ioctx_list, list) { |
681 | /* |
682 | * RCU protects us against accessing freed memory but |
683 | * we have to be careful not to get a reference when the |
684 | * reference count already dropped to 0 (ctx->dead test |
685 | * is unreliable because of races). |
686 | */ |
687 | if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){ |
688 | ret = ctx; |
689 | break; |
690 | } |
691 | } |
692 | |
693 | rcu_read_unlock(); |
694 | return ret; |
695 | } |
696 | |
697 | /* |
698 | * Queue up a kiocb to be retried. Assumes that the kiocb |
699 | * has already been marked as kicked, and places it on |
700 | * the retry run list for the corresponding ioctx, if it |
701 | * isn't already queued. Returns 1 if it actually queued |
702 | * the kiocb (to tell the caller to activate the work |
703 | * queue to process it), or 0, if it found that it was |
704 | * already queued. |
705 | */ |
706 | static inline int __queue_kicked_iocb(struct kiocb *iocb) |
707 | { |
708 | struct kioctx *ctx = iocb->ki_ctx; |
709 | |
710 | assert_spin_locked(&ctx->ctx_lock); |
711 | |
712 | if (list_empty(&iocb->ki_run_list)) { |
713 | list_add_tail(&iocb->ki_run_list, |
714 | &ctx->run_list); |
715 | return 1; |
716 | } |
717 | return 0; |
718 | } |
719 | |
720 | /* aio_run_iocb |
721 | * This is the core aio execution routine. It is |
722 | * invoked both for initial i/o submission and |
723 | * subsequent retries via the aio_kick_handler. |
724 | * Expects to be invoked with iocb->ki_ctx->lock |
725 | * already held. The lock is released and reacquired |
726 | * as needed during processing. |
727 | * |
728 | * Calls the iocb retry method (already setup for the |
729 | * iocb on initial submission) for operation specific |
730 | * handling, but takes care of most of common retry |
731 | * execution details for a given iocb. The retry method |
732 | * needs to be non-blocking as far as possible, to avoid |
733 | * holding up other iocbs waiting to be serviced by the |
734 | * retry kernel thread. |
735 | * |
736 | * The trickier parts in this code have to do with |
737 | * ensuring that only one retry instance is in progress |
738 | * for a given iocb at any time. Providing that guarantee |
739 | * simplifies the coding of individual aio operations as |
740 | * it avoids various potential races. |
741 | */ |
742 | static ssize_t aio_run_iocb(struct kiocb *iocb) |
743 | { |
744 | struct kioctx *ctx = iocb->ki_ctx; |
745 | ssize_t (*retry)(struct kiocb *); |
746 | ssize_t ret; |
747 | |
748 | if (!(retry = iocb->ki_retry)) { |
749 | printk("aio_run_iocb: iocb->ki_retry = NULL\n"); |
750 | return 0; |
751 | } |
752 | |
753 | /* |
754 | * We don't want the next retry iteration for this |
755 | * operation to start until this one has returned and |
756 | * updated the iocb state. However, wait_queue functions |
757 | * can trigger a kick_iocb from interrupt context in the |
758 | * meantime, indicating that data is available for the next |
759 | * iteration. We want to remember that and enable the |
760 | * next retry iteration _after_ we are through with |
761 | * this one. |
762 | * |
763 | * So, in order to be able to register a "kick", but |
764 | * prevent it from being queued now, we clear the kick |
765 | * flag, but make the kick code *think* that the iocb is |
766 | * still on the run list until we are actually done. |
767 | * When we are done with this iteration, we check if |
768 | * the iocb was kicked in the meantime and if so, queue |
769 | * it up afresh. |
770 | */ |
771 | |
772 | kiocbClearKicked(iocb); |
773 | |
774 | /* |
775 | * This is so that aio_complete knows it doesn't need to |
776 | * pull the iocb off the run list (We can't just call |
777 | * INIT_LIST_HEAD because we don't want a kick_iocb to |
778 | * queue this on the run list yet) |
779 | */ |
780 | iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL; |
781 | spin_unlock_irq(&ctx->ctx_lock); |
782 | |
783 | /* Quit retrying if the i/o has been cancelled */ |
784 | if (kiocbIsCancelled(iocb)) { |
785 | ret = -EINTR; |
786 | aio_complete(iocb, ret, 0); |
787 | /* must not access the iocb after this */ |
788 | goto out; |
789 | } |
790 | |
791 | /* |
792 | * Now we are all set to call the retry method in async |
793 | * context. |
794 | */ |
795 | ret = retry(iocb); |
796 | |
797 | if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) { |
798 | /* |
799 | * There's no easy way to restart the syscall since other AIO's |
800 | * may be already running. Just fail this IO with EINTR. |
801 | */ |
802 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || |
803 | ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK)) |
804 | ret = -EINTR; |
805 | aio_complete(iocb, ret, 0); |
806 | } |
807 | out: |
808 | spin_lock_irq(&ctx->ctx_lock); |
809 | |
810 | if (-EIOCBRETRY == ret) { |
811 | /* |
812 | * OK, now that we are done with this iteration |
813 | * and know that there is more left to go, |
814 | * this is where we let go so that a subsequent |
815 | * "kick" can start the next iteration |
816 | */ |
817 | |
818 | /* will make __queue_kicked_iocb succeed from here on */ |
819 | INIT_LIST_HEAD(&iocb->ki_run_list); |
820 | /* we must queue the next iteration ourselves, if it |
821 | * has already been kicked */ |
822 | if (kiocbIsKicked(iocb)) { |
823 | __queue_kicked_iocb(iocb); |
824 | |
825 | /* |
826 | * __queue_kicked_iocb will always return 1 here, because |
827 | * iocb->ki_run_list is empty at this point so it should |
828 | * be safe to unconditionally queue the context into the |
829 | * work queue. |
830 | */ |
831 | aio_queue_work(ctx); |
832 | } |
833 | } |
834 | return ret; |
835 | } |
836 | |
837 | /* |
838 | * __aio_run_iocbs: |
839 | * Process all pending retries queued on the ioctx |
840 | * run list. |
841 | * Assumes it is operating within the aio issuer's mm |
842 | * context. |
843 | */ |
844 | static int __aio_run_iocbs(struct kioctx *ctx) |
845 | { |
846 | struct kiocb *iocb; |
847 | struct list_head run_list; |
848 | |
849 | assert_spin_locked(&ctx->ctx_lock); |
850 | |
851 | list_replace_init(&ctx->run_list, &run_list); |
852 | while (!list_empty(&run_list)) { |
853 | iocb = list_entry(run_list.next, struct kiocb, |
854 | ki_run_list); |
855 | list_del(&iocb->ki_run_list); |
856 | /* |
857 | * Hold an extra reference while retrying i/o. |
858 | */ |
859 | iocb->ki_users++; /* grab extra reference */ |
860 | aio_run_iocb(iocb); |
861 | __aio_put_req(ctx, iocb); |
862 | } |
863 | if (!list_empty(&ctx->run_list)) |
864 | return 1; |
865 | return 0; |
866 | } |
867 | |
868 | static void aio_queue_work(struct kioctx * ctx) |
869 | { |
870 | unsigned long timeout; |
871 | /* |
872 | * if someone is waiting, get the work started right |
873 | * away, otherwise, use a longer delay |
874 | */ |
875 | smp_mb(); |
876 | if (waitqueue_active(&ctx->wait)) |
877 | timeout = 1; |
878 | else |
879 | timeout = HZ/10; |
880 | queue_delayed_work(aio_wq, &ctx->wq, timeout); |
881 | } |
882 | |
883 | /* |
884 | * aio_run_all_iocbs: |
885 | * Process all pending retries queued on the ioctx |
886 | * run list, and keep running them until the list |
887 | * stays empty. |
888 | * Assumes it is operating within the aio issuer's mm context. |
889 | */ |
890 | static inline void aio_run_all_iocbs(struct kioctx *ctx) |
891 | { |
892 | spin_lock_irq(&ctx->ctx_lock); |
893 | while (__aio_run_iocbs(ctx)) |
894 | ; |
895 | spin_unlock_irq(&ctx->ctx_lock); |
896 | } |
897 | |
898 | /* |
899 | * aio_kick_handler: |
900 | * Work queue handler triggered to process pending |
901 | * retries on an ioctx. Takes on the aio issuer's |
902 | * mm context before running the iocbs, so that |
903 | * copy_xxx_user operates on the issuer's address |
904 | * space. |
905 | * Run on aiod's context. |
906 | */ |
907 | static void aio_kick_handler(struct work_struct *work) |
908 | { |
909 | struct kioctx *ctx = container_of(work, struct kioctx, wq.work); |
910 | mm_segment_t oldfs = get_fs(); |
911 | struct mm_struct *mm; |
912 | int requeue; |
913 | |
914 | set_fs(USER_DS); |
915 | use_mm(ctx->mm); |
916 | spin_lock_irq(&ctx->ctx_lock); |
917 | requeue =__aio_run_iocbs(ctx); |
918 | mm = ctx->mm; |
919 | spin_unlock_irq(&ctx->ctx_lock); |
920 | unuse_mm(mm); |
921 | set_fs(oldfs); |
922 | /* |
923 | * we're in a worker thread already, don't use queue_delayed_work, |
924 | */ |
925 | if (requeue) |
926 | queue_delayed_work(aio_wq, &ctx->wq, 0); |
927 | } |
928 | |
929 | |
930 | /* |
931 | * Called by kick_iocb to queue the kiocb for retry |
932 | * and if required activate the aio work queue to process |
933 | * it |
934 | */ |
935 | static void try_queue_kicked_iocb(struct kiocb *iocb) |
936 | { |
937 | struct kioctx *ctx = iocb->ki_ctx; |
938 | unsigned long flags; |
939 | int run = 0; |
940 | |
941 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
942 | /* set this inside the lock so that we can't race with aio_run_iocb() |
943 | * testing it and putting the iocb on the run list under the lock */ |
944 | if (!kiocbTryKick(iocb)) |
945 | run = __queue_kicked_iocb(iocb); |
946 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
947 | if (run) |
948 | aio_queue_work(ctx); |
949 | } |
950 | |
951 | /* |
952 | * kick_iocb: |
953 | * Called typically from a wait queue callback context |
954 | * to trigger a retry of the iocb. |
955 | * The retry is usually executed by aio workqueue |
956 | * threads (See aio_kick_handler). |
957 | */ |
958 | void kick_iocb(struct kiocb *iocb) |
959 | { |
960 | /* sync iocbs are easy: they can only ever be executing from a |
961 | * single context. */ |
962 | if (is_sync_kiocb(iocb)) { |
963 | kiocbSetKicked(iocb); |
964 | wake_up_process(iocb->ki_obj.tsk); |
965 | return; |
966 | } |
967 | |
968 | try_queue_kicked_iocb(iocb); |
969 | } |
970 | EXPORT_SYMBOL(kick_iocb); |
971 | |
972 | /* aio_complete |
973 | * Called when the io request on the given iocb is complete. |
974 | * Returns true if this is the last user of the request. The |
975 | * only other user of the request can be the cancellation code. |
976 | */ |
977 | int aio_complete(struct kiocb *iocb, long res, long res2) |
978 | { |
979 | struct kioctx *ctx = iocb->ki_ctx; |
980 | struct aio_ring_info *info; |
981 | struct aio_ring *ring; |
982 | struct io_event *event; |
983 | unsigned long flags; |
984 | unsigned long tail; |
985 | int ret; |
986 | |
987 | /* |
988 | * Special case handling for sync iocbs: |
989 | * - events go directly into the iocb for fast handling |
990 | * - the sync task with the iocb in its stack holds the single iocb |
991 | * ref, no other paths have a way to get another ref |
992 | * - the sync task helpfully left a reference to itself in the iocb |
993 | */ |
994 | if (is_sync_kiocb(iocb)) { |
995 | BUG_ON(iocb->ki_users != 1); |
996 | iocb->ki_user_data = res; |
997 | iocb->ki_users = 0; |
998 | wake_up_process(iocb->ki_obj.tsk); |
999 | return 1; |
1000 | } |
1001 | |
1002 | info = &ctx->ring_info; |
1003 | |
1004 | /* add a completion event to the ring buffer. |
1005 | * must be done holding ctx->ctx_lock to prevent |
1006 | * other code from messing with the tail |
1007 | * pointer since we might be called from irq |
1008 | * context. |
1009 | */ |
1010 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
1011 | |
1012 | if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list)) |
1013 | list_del_init(&iocb->ki_run_list); |
1014 | |
1015 | /* |
1016 | * cancelled requests don't get events, userland was given one |
1017 | * when the event got cancelled. |
1018 | */ |
1019 | if (kiocbIsCancelled(iocb)) |
1020 | goto put_rq; |
1021 | |
1022 | ring = kmap_atomic(info->ring_pages[0], KM_IRQ1); |
1023 | |
1024 | tail = info->tail; |
1025 | event = aio_ring_event(info, tail, KM_IRQ0); |
1026 | if (++tail >= info->nr) |
1027 | tail = 0; |
1028 | |
1029 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
1030 | event->data = iocb->ki_user_data; |
1031 | event->res = res; |
1032 | event->res2 = res2; |
1033 | |
1034 | dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n", |
1035 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
1036 | res, res2); |
1037 | |
1038 | /* after flagging the request as done, we |
1039 | * must never even look at it again |
1040 | */ |
1041 | smp_wmb(); /* make event visible before updating tail */ |
1042 | |
1043 | info->tail = tail; |
1044 | ring->tail = tail; |
1045 | |
1046 | put_aio_ring_event(event, KM_IRQ0); |
1047 | kunmap_atomic(ring, KM_IRQ1); |
1048 | |
1049 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); |
1050 | |
1051 | /* |
1052 | * Check if the user asked us to deliver the result through an |
1053 | * eventfd. The eventfd_signal() function is safe to be called |
1054 | * from IRQ context. |
1055 | */ |
1056 | if (iocb->ki_eventfd != NULL) |
1057 | eventfd_signal(iocb->ki_eventfd, 1); |
1058 | |
1059 | put_rq: |
1060 | /* everything turned out well, dispose of the aiocb. */ |
1061 | ret = __aio_put_req(ctx, iocb); |
1062 | |
1063 | /* |
1064 | * We have to order our ring_info tail store above and test |
1065 | * of the wait list below outside the wait lock. This is |
1066 | * like in wake_up_bit() where clearing a bit has to be |
1067 | * ordered with the unlocked test. |
1068 | */ |
1069 | smp_mb(); |
1070 | |
1071 | if (waitqueue_active(&ctx->wait)) |
1072 | wake_up(&ctx->wait); |
1073 | |
1074 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
1075 | return ret; |
1076 | } |
1077 | EXPORT_SYMBOL(aio_complete); |
1078 | |
1079 | /* aio_read_evt |
1080 | * Pull an event off of the ioctx's event ring. Returns the number of |
1081 | * events fetched (0 or 1 ;-) |
1082 | * FIXME: make this use cmpxchg. |
1083 | * TODO: make the ringbuffer user mmap()able (requires FIXME). |
1084 | */ |
1085 | static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) |
1086 | { |
1087 | struct aio_ring_info *info = &ioctx->ring_info; |
1088 | struct aio_ring *ring; |
1089 | unsigned long head; |
1090 | int ret = 0; |
1091 | |
1092 | ring = kmap_atomic(info->ring_pages[0], KM_USER0); |
1093 | dprintk("in aio_read_evt h%lu t%lu m%lu\n", |
1094 | (unsigned long)ring->head, (unsigned long)ring->tail, |
1095 | (unsigned long)ring->nr); |
1096 | |
1097 | if (ring->head == ring->tail) |
1098 | goto out; |
1099 | |
1100 | spin_lock(&info->ring_lock); |
1101 | |
1102 | head = ring->head % info->nr; |
1103 | if (head != ring->tail) { |
1104 | struct io_event *evp = aio_ring_event(info, head, KM_USER1); |
1105 | *ent = *evp; |
1106 | head = (head + 1) % info->nr; |
1107 | smp_mb(); /* finish reading the event before updatng the head */ |
1108 | ring->head = head; |
1109 | ret = 1; |
1110 | put_aio_ring_event(evp, KM_USER1); |
1111 | } |
1112 | spin_unlock(&info->ring_lock); |
1113 | |
1114 | out: |
1115 | kunmap_atomic(ring, KM_USER0); |
1116 | dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret, |
1117 | (unsigned long)ring->head, (unsigned long)ring->tail); |
1118 | return ret; |
1119 | } |
1120 | |
1121 | struct aio_timeout { |
1122 | struct timer_list timer; |
1123 | int timed_out; |
1124 | struct task_struct *p; |
1125 | }; |
1126 | |
1127 | static void timeout_func(unsigned long data) |
1128 | { |
1129 | struct aio_timeout *to = (struct aio_timeout *)data; |
1130 | |
1131 | to->timed_out = 1; |
1132 | wake_up_process(to->p); |
1133 | } |
1134 | |
1135 | static inline void init_timeout(struct aio_timeout *to) |
1136 | { |
1137 | setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to); |
1138 | to->timed_out = 0; |
1139 | to->p = current; |
1140 | } |
1141 | |
1142 | static inline void set_timeout(long start_jiffies, struct aio_timeout *to, |
1143 | const struct timespec *ts) |
1144 | { |
1145 | to->timer.expires = start_jiffies + timespec_to_jiffies(ts); |
1146 | if (time_after(to->timer.expires, jiffies)) |
1147 | add_timer(&to->timer); |
1148 | else |
1149 | to->timed_out = 1; |
1150 | } |
1151 | |
1152 | static inline void clear_timeout(struct aio_timeout *to) |
1153 | { |
1154 | del_singleshot_timer_sync(&to->timer); |
1155 | } |
1156 | |
1157 | static int read_events(struct kioctx *ctx, |
1158 | long min_nr, long nr, |
1159 | struct io_event __user *event, |
1160 | struct timespec __user *timeout) |
1161 | { |
1162 | long start_jiffies = jiffies; |
1163 | struct task_struct *tsk = current; |
1164 | DECLARE_WAITQUEUE(wait, tsk); |
1165 | int ret; |
1166 | int i = 0; |
1167 | struct io_event ent; |
1168 | struct aio_timeout to; |
1169 | int retry = 0; |
1170 | |
1171 | /* needed to zero any padding within an entry (there shouldn't be |
1172 | * any, but C is fun! |
1173 | */ |
1174 | memset(&ent, 0, sizeof(ent)); |
1175 | retry: |
1176 | ret = 0; |
1177 | while (likely(i < nr)) { |
1178 | ret = aio_read_evt(ctx, &ent); |
1179 | if (unlikely(ret <= 0)) |
1180 | break; |
1181 | |
1182 | dprintk("read event: %Lx %Lx %Lx %Lx\n", |
1183 | ent.data, ent.obj, ent.res, ent.res2); |
1184 | |
1185 | /* Could we split the check in two? */ |
1186 | ret = -EFAULT; |
1187 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
1188 | dprintk("aio: lost an event due to EFAULT.\n"); |
1189 | break; |
1190 | } |
1191 | ret = 0; |
1192 | |
1193 | /* Good, event copied to userland, update counts. */ |
1194 | event ++; |
1195 | i ++; |
1196 | } |
1197 | |
1198 | if (min_nr <= i) |
1199 | return i; |
1200 | if (ret) |
1201 | return ret; |
1202 | |
1203 | /* End fast path */ |
1204 | |
1205 | /* racey check, but it gets redone */ |
1206 | if (!retry && unlikely(!list_empty(&ctx->run_list))) { |
1207 | retry = 1; |
1208 | aio_run_all_iocbs(ctx); |
1209 | goto retry; |
1210 | } |
1211 | |
1212 | init_timeout(&to); |
1213 | if (timeout) { |
1214 | struct timespec ts; |
1215 | ret = -EFAULT; |
1216 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
1217 | goto out; |
1218 | |
1219 | set_timeout(start_jiffies, &to, &ts); |
1220 | } |
1221 | |
1222 | while (likely(i < nr)) { |
1223 | add_wait_queue_exclusive(&ctx->wait, &wait); |
1224 | do { |
1225 | set_task_state(tsk, TASK_INTERRUPTIBLE); |
1226 | ret = aio_read_evt(ctx, &ent); |
1227 | if (ret) |
1228 | break; |
1229 | if (min_nr <= i) |
1230 | break; |
1231 | if (unlikely(ctx->dead)) { |
1232 | ret = -EINVAL; |
1233 | break; |
1234 | } |
1235 | if (to.timed_out) /* Only check after read evt */ |
1236 | break; |
1237 | /* Try to only show up in io wait if there are ops |
1238 | * in flight */ |
1239 | if (ctx->reqs_active) |
1240 | io_schedule(); |
1241 | else |
1242 | schedule(); |
1243 | if (signal_pending(tsk)) { |
1244 | ret = -EINTR; |
1245 | break; |
1246 | } |
1247 | /*ret = aio_read_evt(ctx, &ent);*/ |
1248 | } while (1) ; |
1249 | |
1250 | set_task_state(tsk, TASK_RUNNING); |
1251 | remove_wait_queue(&ctx->wait, &wait); |
1252 | |
1253 | if (unlikely(ret <= 0)) |
1254 | break; |
1255 | |
1256 | ret = -EFAULT; |
1257 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { |
1258 | dprintk("aio: lost an event due to EFAULT.\n"); |
1259 | break; |
1260 | } |
1261 | |
1262 | /* Good, event copied to userland, update counts. */ |
1263 | event ++; |
1264 | i ++; |
1265 | } |
1266 | |
1267 | if (timeout) |
1268 | clear_timeout(&to); |
1269 | out: |
1270 | destroy_timer_on_stack(&to.timer); |
1271 | return i ? i : ret; |
1272 | } |
1273 | |
1274 | /* Take an ioctx and remove it from the list of ioctx's. Protects |
1275 | * against races with itself via ->dead. |
1276 | */ |
1277 | static void io_destroy(struct kioctx *ioctx) |
1278 | { |
1279 | struct mm_struct *mm = current->mm; |
1280 | int was_dead; |
1281 | |
1282 | /* delete the entry from the list is someone else hasn't already */ |
1283 | spin_lock(&mm->ioctx_lock); |
1284 | was_dead = ioctx->dead; |
1285 | ioctx->dead = 1; |
1286 | hlist_del_rcu(&ioctx->list); |
1287 | spin_unlock(&mm->ioctx_lock); |
1288 | |
1289 | dprintk("aio_release(%p)\n", ioctx); |
1290 | if (likely(!was_dead)) |
1291 | put_ioctx(ioctx); /* twice for the list */ |
1292 | |
1293 | aio_cancel_all(ioctx); |
1294 | wait_for_all_aios(ioctx); |
1295 | |
1296 | /* |
1297 | * Wake up any waiters. The setting of ctx->dead must be seen |
1298 | * by other CPUs at this point. Right now, we rely on the |
1299 | * locking done by the above calls to ensure this consistency. |
1300 | */ |
1301 | wake_up_all(&ioctx->wait); |
1302 | put_ioctx(ioctx); /* once for the lookup */ |
1303 | } |
1304 | |
1305 | /* sys_io_setup: |
1306 | * Create an aio_context capable of receiving at least nr_events. |
1307 | * ctxp must not point to an aio_context that already exists, and |
1308 | * must be initialized to 0 prior to the call. On successful |
1309 | * creation of the aio_context, *ctxp is filled in with the resulting |
1310 | * handle. May fail with -EINVAL if *ctxp is not initialized, |
1311 | * if the specified nr_events exceeds internal limits. May fail |
1312 | * with -EAGAIN if the specified nr_events exceeds the user's limit |
1313 | * of available events. May fail with -ENOMEM if insufficient kernel |
1314 | * resources are available. May fail with -EFAULT if an invalid |
1315 | * pointer is passed for ctxp. Will fail with -ENOSYS if not |
1316 | * implemented. |
1317 | */ |
1318 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1319 | { |
1320 | struct kioctx *ioctx = NULL; |
1321 | unsigned long ctx; |
1322 | long ret; |
1323 | |
1324 | ret = get_user(ctx, ctxp); |
1325 | if (unlikely(ret)) |
1326 | goto out; |
1327 | |
1328 | ret = -EINVAL; |
1329 | if (unlikely(ctx || nr_events == 0)) { |
1330 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", |
1331 | ctx, nr_events); |
1332 | goto out; |
1333 | } |
1334 | |
1335 | ioctx = ioctx_alloc(nr_events); |
1336 | ret = PTR_ERR(ioctx); |
1337 | if (!IS_ERR(ioctx)) { |
1338 | ret = put_user(ioctx->user_id, ctxp); |
1339 | if (!ret) { |
1340 | put_ioctx(ioctx); |
1341 | return 0; |
1342 | } |
1343 | io_destroy(ioctx); |
1344 | } |
1345 | |
1346 | out: |
1347 | return ret; |
1348 | } |
1349 | |
1350 | /* sys_io_destroy: |
1351 | * Destroy the aio_context specified. May cancel any outstanding |
1352 | * AIOs and block on completion. Will fail with -ENOSYS if not |
1353 | * implemented. May fail with -EINVAL if the context pointed to |
1354 | * is invalid. |
1355 | */ |
1356 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1357 | { |
1358 | struct kioctx *ioctx = lookup_ioctx(ctx); |
1359 | if (likely(NULL != ioctx)) { |
1360 | io_destroy(ioctx); |
1361 | return 0; |
1362 | } |
1363 | pr_debug("EINVAL: io_destroy: invalid context id\n"); |
1364 | return -EINVAL; |
1365 | } |
1366 | |
1367 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1368 | { |
1369 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
1370 | |
1371 | BUG_ON(ret <= 0); |
1372 | |
1373 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { |
1374 | ssize_t this = min((ssize_t)iov->iov_len, ret); |
1375 | iov->iov_base += this; |
1376 | iov->iov_len -= this; |
1377 | iocb->ki_left -= this; |
1378 | ret -= this; |
1379 | if (iov->iov_len == 0) { |
1380 | iocb->ki_cur_seg++; |
1381 | iov++; |
1382 | } |
1383 | } |
1384 | |
1385 | /* the caller should not have done more io than what fit in |
1386 | * the remaining iovecs */ |
1387 | BUG_ON(ret > 0 && iocb->ki_left == 0); |
1388 | } |
1389 | |
1390 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
1391 | { |
1392 | struct file *file = iocb->ki_filp; |
1393 | struct address_space *mapping = file->f_mapping; |
1394 | struct inode *inode = mapping->host; |
1395 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, |
1396 | unsigned long, loff_t); |
1397 | ssize_t ret = 0; |
1398 | unsigned short opcode; |
1399 | |
1400 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || |
1401 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { |
1402 | rw_op = file->f_op->aio_read; |
1403 | opcode = IOCB_CMD_PREADV; |
1404 | } else { |
1405 | rw_op = file->f_op->aio_write; |
1406 | opcode = IOCB_CMD_PWRITEV; |
1407 | } |
1408 | |
1409 | /* This matches the pread()/pwrite() logic */ |
1410 | if (iocb->ki_pos < 0) |
1411 | return -EINVAL; |
1412 | |
1413 | do { |
1414 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
1415 | iocb->ki_nr_segs - iocb->ki_cur_seg, |
1416 | iocb->ki_pos); |
1417 | if (ret > 0) |
1418 | aio_advance_iovec(iocb, ret); |
1419 | |
1420 | /* retry all partial writes. retry partial reads as long as its a |
1421 | * regular file. */ |
1422 | } while (ret > 0 && iocb->ki_left > 0 && |
1423 | (opcode == IOCB_CMD_PWRITEV || |
1424 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); |
1425 | |
1426 | /* This means we must have transferred all that we could */ |
1427 | /* No need to retry anymore */ |
1428 | if ((ret == 0) || (iocb->ki_left == 0)) |
1429 | ret = iocb->ki_nbytes - iocb->ki_left; |
1430 | |
1431 | /* If we managed to write some out we return that, rather than |
1432 | * the eventual error. */ |
1433 | if (opcode == IOCB_CMD_PWRITEV |
1434 | && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY |
1435 | && iocb->ki_nbytes - iocb->ki_left) |
1436 | ret = iocb->ki_nbytes - iocb->ki_left; |
1437 | |
1438 | return ret; |
1439 | } |
1440 | |
1441 | static ssize_t aio_fdsync(struct kiocb *iocb) |
1442 | { |
1443 | struct file *file = iocb->ki_filp; |
1444 | ssize_t ret = -EINVAL; |
1445 | |
1446 | if (file->f_op->aio_fsync) |
1447 | ret = file->f_op->aio_fsync(iocb, 1); |
1448 | return ret; |
1449 | } |
1450 | |
1451 | static ssize_t aio_fsync(struct kiocb *iocb) |
1452 | { |
1453 | struct file *file = iocb->ki_filp; |
1454 | ssize_t ret = -EINVAL; |
1455 | |
1456 | if (file->f_op->aio_fsync) |
1457 | ret = file->f_op->aio_fsync(iocb, 0); |
1458 | return ret; |
1459 | } |
1460 | |
1461 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
1462 | { |
1463 | ssize_t ret; |
1464 | |
1465 | #ifdef CONFIG_COMPAT |
1466 | if (compat) |
1467 | ret = compat_rw_copy_check_uvector(type, |
1468 | (struct compat_iovec __user *)kiocb->ki_buf, |
1469 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
1470 | &kiocb->ki_iovec, 1); |
1471 | else |
1472 | #endif |
1473 | ret = rw_copy_check_uvector(type, |
1474 | (struct iovec __user *)kiocb->ki_buf, |
1475 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, |
1476 | &kiocb->ki_iovec, 1); |
1477 | if (ret < 0) |
1478 | goto out; |
1479 | |
1480 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1481 | kiocb->ki_cur_seg = 0; |
1482 | /* ki_nbytes/left now reflect bytes instead of segs */ |
1483 | kiocb->ki_nbytes = ret; |
1484 | kiocb->ki_left = ret; |
1485 | |
1486 | ret = 0; |
1487 | out: |
1488 | return ret; |
1489 | } |
1490 | |
1491 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb) |
1492 | { |
1493 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1494 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; |
1495 | kiocb->ki_iovec->iov_len = kiocb->ki_left; |
1496 | kiocb->ki_nr_segs = 1; |
1497 | kiocb->ki_cur_seg = 0; |
1498 | return 0; |
1499 | } |
1500 | |
1501 | /* |
1502 | * aio_setup_iocb: |
1503 | * Performs the initial checks and aio retry method |
1504 | * setup for the kiocb at the time of io submission. |
1505 | */ |
1506 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
1507 | { |
1508 | struct file *file = kiocb->ki_filp; |
1509 | ssize_t ret = 0; |
1510 | |
1511 | switch (kiocb->ki_opcode) { |
1512 | case IOCB_CMD_PREAD: |
1513 | ret = -EBADF; |
1514 | if (unlikely(!(file->f_mode & FMODE_READ))) |
1515 | break; |
1516 | ret = -EFAULT; |
1517 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, |
1518 | kiocb->ki_left))) |
1519 | break; |
1520 | ret = security_file_permission(file, MAY_READ); |
1521 | if (unlikely(ret)) |
1522 | break; |
1523 | ret = aio_setup_single_vector(kiocb); |
1524 | if (ret) |
1525 | break; |
1526 | ret = -EINVAL; |
1527 | if (file->f_op->aio_read) |
1528 | kiocb->ki_retry = aio_rw_vect_retry; |
1529 | break; |
1530 | case IOCB_CMD_PWRITE: |
1531 | ret = -EBADF; |
1532 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
1533 | break; |
1534 | ret = -EFAULT; |
1535 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, |
1536 | kiocb->ki_left))) |
1537 | break; |
1538 | ret = security_file_permission(file, MAY_WRITE); |
1539 | if (unlikely(ret)) |
1540 | break; |
1541 | ret = aio_setup_single_vector(kiocb); |
1542 | if (ret) |
1543 | break; |
1544 | ret = -EINVAL; |
1545 | if (file->f_op->aio_write) |
1546 | kiocb->ki_retry = aio_rw_vect_retry; |
1547 | break; |
1548 | case IOCB_CMD_PREADV: |
1549 | ret = -EBADF; |
1550 | if (unlikely(!(file->f_mode & FMODE_READ))) |
1551 | break; |
1552 | ret = security_file_permission(file, MAY_READ); |
1553 | if (unlikely(ret)) |
1554 | break; |
1555 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
1556 | if (ret) |
1557 | break; |
1558 | ret = -EINVAL; |
1559 | if (file->f_op->aio_read) |
1560 | kiocb->ki_retry = aio_rw_vect_retry; |
1561 | break; |
1562 | case IOCB_CMD_PWRITEV: |
1563 | ret = -EBADF; |
1564 | if (unlikely(!(file->f_mode & FMODE_WRITE))) |
1565 | break; |
1566 | ret = security_file_permission(file, MAY_WRITE); |
1567 | if (unlikely(ret)) |
1568 | break; |
1569 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
1570 | if (ret) |
1571 | break; |
1572 | ret = -EINVAL; |
1573 | if (file->f_op->aio_write) |
1574 | kiocb->ki_retry = aio_rw_vect_retry; |
1575 | break; |
1576 | case IOCB_CMD_FDSYNC: |
1577 | ret = -EINVAL; |
1578 | if (file->f_op->aio_fsync) |
1579 | kiocb->ki_retry = aio_fdsync; |
1580 | break; |
1581 | case IOCB_CMD_FSYNC: |
1582 | ret = -EINVAL; |
1583 | if (file->f_op->aio_fsync) |
1584 | kiocb->ki_retry = aio_fsync; |
1585 | break; |
1586 | default: |
1587 | dprintk("EINVAL: io_submit: no operation provided\n"); |
1588 | ret = -EINVAL; |
1589 | } |
1590 | |
1591 | if (!kiocb->ki_retry) |
1592 | return ret; |
1593 | |
1594 | return 0; |
1595 | } |
1596 | |
1597 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
1598 | struct iocb *iocb, struct kiocb_batch *batch, |
1599 | bool compat) |
1600 | { |
1601 | struct kiocb *req; |
1602 | struct file *file; |
1603 | ssize_t ret; |
1604 | |
1605 | /* enforce forwards compatibility on users */ |
1606 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
1607 | pr_debug("EINVAL: io_submit: reserve field set\n"); |
1608 | return -EINVAL; |
1609 | } |
1610 | |
1611 | /* prevent overflows */ |
1612 | if (unlikely( |
1613 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || |
1614 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || |
1615 | ((ssize_t)iocb->aio_nbytes < 0) |
1616 | )) { |
1617 | pr_debug("EINVAL: io_submit: overflow check\n"); |
1618 | return -EINVAL; |
1619 | } |
1620 | |
1621 | file = fget(iocb->aio_fildes); |
1622 | if (unlikely(!file)) |
1623 | return -EBADF; |
1624 | |
1625 | req = aio_get_req(ctx, batch); /* returns with 2 references to req */ |
1626 | if (unlikely(!req)) { |
1627 | fput(file); |
1628 | return -EAGAIN; |
1629 | } |
1630 | req->ki_filp = file; |
1631 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1632 | /* |
1633 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an |
1634 | * instance of the file* now. The file descriptor must be |
1635 | * an eventfd() fd, and will be signaled for each completed |
1636 | * event using the eventfd_signal() function. |
1637 | */ |
1638 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
1639 | if (IS_ERR(req->ki_eventfd)) { |
1640 | ret = PTR_ERR(req->ki_eventfd); |
1641 | req->ki_eventfd = NULL; |
1642 | goto out_put_req; |
1643 | } |
1644 | } |
1645 | |
1646 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
1647 | if (unlikely(ret)) { |
1648 | dprintk("EFAULT: aio_key\n"); |
1649 | goto out_put_req; |
1650 | } |
1651 | |
1652 | req->ki_obj.user = user_iocb; |
1653 | req->ki_user_data = iocb->aio_data; |
1654 | req->ki_pos = iocb->aio_offset; |
1655 | |
1656 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; |
1657 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; |
1658 | req->ki_opcode = iocb->aio_lio_opcode; |
1659 | |
1660 | ret = aio_setup_iocb(req, compat); |
1661 | |
1662 | if (ret) |
1663 | goto out_put_req; |
1664 | |
1665 | spin_lock_irq(&ctx->ctx_lock); |
1666 | /* |
1667 | * We could have raced with io_destroy() and are currently holding a |
1668 | * reference to ctx which should be destroyed. We cannot submit IO |
1669 | * since ctx gets freed as soon as io_submit() puts its reference. The |
1670 | * check here is reliable: io_destroy() sets ctx->dead before waiting |
1671 | * for outstanding IO and the barrier between these two is realized by |
1672 | * unlock of mm->ioctx_lock and lock of ctx->ctx_lock. Analogously we |
1673 | * increment ctx->reqs_active before checking for ctx->dead and the |
1674 | * barrier is realized by unlock and lock of ctx->ctx_lock. Thus if we |
1675 | * don't see ctx->dead set here, io_destroy() waits for our IO to |
1676 | * finish. |
1677 | */ |
1678 | if (ctx->dead) { |
1679 | spin_unlock_irq(&ctx->ctx_lock); |
1680 | ret = -EINVAL; |
1681 | goto out_put_req; |
1682 | } |
1683 | aio_run_iocb(req); |
1684 | if (!list_empty(&ctx->run_list)) { |
1685 | /* drain the run list */ |
1686 | while (__aio_run_iocbs(ctx)) |
1687 | ; |
1688 | } |
1689 | spin_unlock_irq(&ctx->ctx_lock); |
1690 | |
1691 | aio_put_req(req); /* drop extra ref to req */ |
1692 | return 0; |
1693 | |
1694 | out_put_req: |
1695 | aio_put_req(req); /* drop extra ref to req */ |
1696 | aio_put_req(req); /* drop i/o ref to req */ |
1697 | return ret; |
1698 | } |
1699 | |
1700 | long do_io_submit(aio_context_t ctx_id, long nr, |
1701 | struct iocb __user *__user *iocbpp, bool compat) |
1702 | { |
1703 | struct kioctx *ctx; |
1704 | long ret = 0; |
1705 | int i = 0; |
1706 | struct blk_plug plug; |
1707 | struct kiocb_batch batch; |
1708 | |
1709 | if (unlikely(nr < 0)) |
1710 | return -EINVAL; |
1711 | |
1712 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1713 | nr = LONG_MAX/sizeof(*iocbpp); |
1714 | |
1715 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1716 | return -EFAULT; |
1717 | |
1718 | ctx = lookup_ioctx(ctx_id); |
1719 | if (unlikely(!ctx)) { |
1720 | pr_debug("EINVAL: io_submit: invalid context id\n"); |
1721 | return -EINVAL; |
1722 | } |
1723 | |
1724 | kiocb_batch_init(&batch, nr); |
1725 | |
1726 | blk_start_plug(&plug); |
1727 | |
1728 | /* |
1729 | * AKPM: should this return a partial result if some of the IOs were |
1730 | * successfully submitted? |
1731 | */ |
1732 | for (i=0; i<nr; i++) { |
1733 | struct iocb __user *user_iocb; |
1734 | struct iocb tmp; |
1735 | |
1736 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { |
1737 | ret = -EFAULT; |
1738 | break; |
1739 | } |
1740 | |
1741 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { |
1742 | ret = -EFAULT; |
1743 | break; |
1744 | } |
1745 | |
1746 | ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat); |
1747 | if (ret) |
1748 | break; |
1749 | } |
1750 | blk_finish_plug(&plug); |
1751 | |
1752 | kiocb_batch_free(ctx, &batch); |
1753 | put_ioctx(ctx); |
1754 | return i ? i : ret; |
1755 | } |
1756 | |
1757 | /* sys_io_submit: |
1758 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns |
1759 | * the number of iocbs queued. May return -EINVAL if the aio_context |
1760 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at |
1761 | * *iocbpp[0] is not properly initialized, if the operation specified |
1762 | * is invalid for the file descriptor in the iocb. May fail with |
1763 | * -EFAULT if any of the data structures point to invalid data. May |
1764 | * fail with -EBADF if the file descriptor specified in the first |
1765 | * iocb is invalid. May fail with -EAGAIN if insufficient resources |
1766 | * are available to queue any iocbs. Will return 0 if nr is 0. Will |
1767 | * fail with -ENOSYS if not implemented. |
1768 | */ |
1769 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, |
1770 | struct iocb __user * __user *, iocbpp) |
1771 | { |
1772 | return do_io_submit(ctx_id, nr, iocbpp, 0); |
1773 | } |
1774 | |
1775 | /* lookup_kiocb |
1776 | * Finds a given iocb for cancellation. |
1777 | */ |
1778 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1779 | u32 key) |
1780 | { |
1781 | struct list_head *pos; |
1782 | |
1783 | assert_spin_locked(&ctx->ctx_lock); |
1784 | |
1785 | /* TODO: use a hash or array, this sucks. */ |
1786 | list_for_each(pos, &ctx->active_reqs) { |
1787 | struct kiocb *kiocb = list_kiocb(pos); |
1788 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) |
1789 | return kiocb; |
1790 | } |
1791 | return NULL; |
1792 | } |
1793 | |
1794 | /* sys_io_cancel: |
1795 | * Attempts to cancel an iocb previously passed to io_submit. If |
1796 | * the operation is successfully cancelled, the resulting event is |
1797 | * copied into the memory pointed to by result without being placed |
1798 | * into the completion queue and 0 is returned. May fail with |
1799 | * -EFAULT if any of the data structures pointed to are invalid. |
1800 | * May fail with -EINVAL if aio_context specified by ctx_id is |
1801 | * invalid. May fail with -EAGAIN if the iocb specified was not |
1802 | * cancelled. Will fail with -ENOSYS if not implemented. |
1803 | */ |
1804 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1805 | struct io_event __user *, result) |
1806 | { |
1807 | int (*cancel)(struct kiocb *iocb, struct io_event *res); |
1808 | struct kioctx *ctx; |
1809 | struct kiocb *kiocb; |
1810 | u32 key; |
1811 | int ret; |
1812 | |
1813 | ret = get_user(key, &iocb->aio_key); |
1814 | if (unlikely(ret)) |
1815 | return -EFAULT; |
1816 | |
1817 | ctx = lookup_ioctx(ctx_id); |
1818 | if (unlikely(!ctx)) |
1819 | return -EINVAL; |
1820 | |
1821 | spin_lock_irq(&ctx->ctx_lock); |
1822 | ret = -EAGAIN; |
1823 | kiocb = lookup_kiocb(ctx, iocb, key); |
1824 | if (kiocb && kiocb->ki_cancel) { |
1825 | cancel = kiocb->ki_cancel; |
1826 | kiocb->ki_users ++; |
1827 | kiocbSetCancelled(kiocb); |
1828 | } else |
1829 | cancel = NULL; |
1830 | spin_unlock_irq(&ctx->ctx_lock); |
1831 | |
1832 | if (NULL != cancel) { |
1833 | struct io_event tmp; |
1834 | pr_debug("calling cancel\n"); |
1835 | memset(&tmp, 0, sizeof(tmp)); |
1836 | tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user; |
1837 | tmp.data = kiocb->ki_user_data; |
1838 | ret = cancel(kiocb, &tmp); |
1839 | if (!ret) { |
1840 | /* Cancellation succeeded -- copy the result |
1841 | * into the user's buffer. |
1842 | */ |
1843 | if (copy_to_user(result, &tmp, sizeof(tmp))) |
1844 | ret = -EFAULT; |
1845 | } |
1846 | } else |
1847 | ret = -EINVAL; |
1848 | |
1849 | put_ioctx(ctx); |
1850 | |
1851 | return ret; |
1852 | } |
1853 | |
1854 | /* io_getevents: |
1855 | * Attempts to read at least min_nr events and up to nr events from |
1856 | * the completion queue for the aio_context specified by ctx_id. If |
1857 | * it succeeds, the number of read events is returned. May fail with |
1858 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is |
1859 | * out of range, if timeout is out of range. May fail with -EFAULT |
1860 | * if any of the memory specified is invalid. May return 0 or |
1861 | * < min_nr if the timeout specified by timeout has elapsed |
1862 | * before sufficient events are available, where timeout == NULL |
1863 | * specifies an infinite timeout. Note that the timeout pointed to by |
1864 | * timeout is relative and will be updated if not NULL and the |
1865 | * operation blocks. Will fail with -ENOSYS if not implemented. |
1866 | */ |
1867 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1868 | long, min_nr, |
1869 | long, nr, |
1870 | struct io_event __user *, events, |
1871 | struct timespec __user *, timeout) |
1872 | { |
1873 | struct kioctx *ioctx = lookup_ioctx(ctx_id); |
1874 | long ret = -EINVAL; |
1875 | |
1876 | if (likely(ioctx)) { |
1877 | if (likely(min_nr <= nr && min_nr >= 0)) |
1878 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1879 | put_ioctx(ioctx); |
1880 | } |
1881 | |
1882 | asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout); |
1883 | return ret; |
1884 | } |
1885 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9