]>
Commit | Line | Data |
---|---|---|
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 | #define pr_fmt(fmt) "%s: " fmt, __func__ | |
12 | ||
13 | #include <linux/kernel.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/time.h> | |
17 | #include <linux/aio_abi.h> | |
18 | #include <linux/export.h> | |
19 | #include <linux/syscalls.h> | |
20 | #include <linux/backing-dev.h> | |
21 | #include <linux/uio.h> | |
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 | #include "internal.h" | |
43 | ||
44 | #define AIO_RING_MAGIC 0xa10a10a1 | |
45 | #define AIO_RING_COMPAT_FEATURES 1 | |
46 | #define AIO_RING_INCOMPAT_FEATURES 0 | |
47 | struct aio_ring { | |
48 | unsigned id; /* kernel internal index number */ | |
49 | unsigned nr; /* number of io_events */ | |
50 | unsigned head; | |
51 | unsigned tail; | |
52 | ||
53 | unsigned magic; | |
54 | unsigned compat_features; | |
55 | unsigned incompat_features; | |
56 | unsigned header_length; /* size of aio_ring */ | |
57 | ||
58 | ||
59 | struct io_event io_events[0]; | |
60 | }; /* 128 bytes + ring size */ | |
61 | ||
62 | #define AIO_RING_PAGES 8 | |
63 | ||
64 | struct kioctx { | |
65 | atomic_t users; | |
66 | atomic_t dead; | |
67 | ||
68 | /* This needs improving */ | |
69 | unsigned long user_id; | |
70 | struct hlist_node list; | |
71 | ||
72 | /* | |
73 | * This is what userspace passed to io_setup(), it's not used for | |
74 | * anything but counting against the global max_reqs quota. | |
75 | * | |
76 | * The real limit is nr_events - 1, which will be larger (see | |
77 | * aio_setup_ring()) | |
78 | */ | |
79 | unsigned max_reqs; | |
80 | ||
81 | /* Size of ringbuffer, in units of struct io_event */ | |
82 | unsigned nr_events; | |
83 | ||
84 | unsigned long mmap_base; | |
85 | unsigned long mmap_size; | |
86 | ||
87 | struct page **ring_pages; | |
88 | long nr_pages; | |
89 | ||
90 | struct rcu_head rcu_head; | |
91 | struct work_struct rcu_work; | |
92 | ||
93 | struct { | |
94 | atomic_t reqs_active; | |
95 | } ____cacheline_aligned_in_smp; | |
96 | ||
97 | struct { | |
98 | spinlock_t ctx_lock; | |
99 | struct list_head active_reqs; /* used for cancellation */ | |
100 | } ____cacheline_aligned_in_smp; | |
101 | ||
102 | struct { | |
103 | struct mutex ring_lock; | |
104 | wait_queue_head_t wait; | |
105 | } ____cacheline_aligned_in_smp; | |
106 | ||
107 | struct { | |
108 | unsigned tail; | |
109 | spinlock_t completion_lock; | |
110 | } ____cacheline_aligned_in_smp; | |
111 | ||
112 | struct page *internal_pages[AIO_RING_PAGES]; | |
113 | }; | |
114 | ||
115 | /*------ sysctl variables----*/ | |
116 | static DEFINE_SPINLOCK(aio_nr_lock); | |
117 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
118 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
119 | /*----end sysctl variables---*/ | |
120 | ||
121 | static struct kmem_cache *kiocb_cachep; | |
122 | static struct kmem_cache *kioctx_cachep; | |
123 | ||
124 | /* aio_setup | |
125 | * Creates the slab caches used by the aio routines, panic on | |
126 | * failure as this is done early during the boot sequence. | |
127 | */ | |
128 | static int __init aio_setup(void) | |
129 | { | |
130 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
131 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
132 | ||
133 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); | |
134 | ||
135 | return 0; | |
136 | } | |
137 | __initcall(aio_setup); | |
138 | ||
139 | static void aio_free_ring(struct kioctx *ctx) | |
140 | { | |
141 | long i; | |
142 | ||
143 | for (i = 0; i < ctx->nr_pages; i++) | |
144 | put_page(ctx->ring_pages[i]); | |
145 | ||
146 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) | |
147 | kfree(ctx->ring_pages); | |
148 | } | |
149 | ||
150 | static int aio_setup_ring(struct kioctx *ctx) | |
151 | { | |
152 | struct aio_ring *ring; | |
153 | unsigned nr_events = ctx->max_reqs; | |
154 | struct mm_struct *mm = current->mm; | |
155 | unsigned long size, populate; | |
156 | int nr_pages; | |
157 | ||
158 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
159 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
160 | ||
161 | size = sizeof(struct aio_ring); | |
162 | size += sizeof(struct io_event) * nr_events; | |
163 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; | |
164 | ||
165 | if (nr_pages < 0) | |
166 | return -EINVAL; | |
167 | ||
168 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); | |
169 | ||
170 | ctx->nr_events = 0; | |
171 | ctx->ring_pages = ctx->internal_pages; | |
172 | if (nr_pages > AIO_RING_PAGES) { | |
173 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), | |
174 | GFP_KERNEL); | |
175 | if (!ctx->ring_pages) | |
176 | return -ENOMEM; | |
177 | } | |
178 | ||
179 | ctx->mmap_size = nr_pages * PAGE_SIZE; | |
180 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
181 | down_write(&mm->mmap_sem); | |
182 | ctx->mmap_base = do_mmap_pgoff(NULL, 0, ctx->mmap_size, | |
183 | PROT_READ|PROT_WRITE, | |
184 | MAP_ANONYMOUS|MAP_PRIVATE, 0, &populate); | |
185 | if (IS_ERR((void *)ctx->mmap_base)) { | |
186 | up_write(&mm->mmap_sem); | |
187 | ctx->mmap_size = 0; | |
188 | aio_free_ring(ctx); | |
189 | return -EAGAIN; | |
190 | } | |
191 | ||
192 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); | |
193 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, | |
194 | 1, 0, ctx->ring_pages, NULL); | |
195 | up_write(&mm->mmap_sem); | |
196 | ||
197 | if (unlikely(ctx->nr_pages != nr_pages)) { | |
198 | aio_free_ring(ctx); | |
199 | return -EAGAIN; | |
200 | } | |
201 | if (populate) | |
202 | mm_populate(ctx->mmap_base, populate); | |
203 | ||
204 | ctx->user_id = ctx->mmap_base; | |
205 | ctx->nr_events = nr_events; /* trusted copy */ | |
206 | ||
207 | ring = kmap_atomic(ctx->ring_pages[0]); | |
208 | ring->nr = nr_events; /* user copy */ | |
209 | ring->id = ctx->user_id; | |
210 | ring->head = ring->tail = 0; | |
211 | ring->magic = AIO_RING_MAGIC; | |
212 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
213 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
214 | ring->header_length = sizeof(struct aio_ring); | |
215 | kunmap_atomic(ring); | |
216 | flush_dcache_page(ctx->ring_pages[0]); | |
217 | ||
218 | return 0; | |
219 | } | |
220 | ||
221 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) | |
222 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
223 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
224 | ||
225 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) | |
226 | { | |
227 | struct kioctx *ctx = req->ki_ctx; | |
228 | unsigned long flags; | |
229 | ||
230 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
231 | ||
232 | if (!req->ki_list.next) | |
233 | list_add(&req->ki_list, &ctx->active_reqs); | |
234 | ||
235 | req->ki_cancel = cancel; | |
236 | ||
237 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
238 | } | |
239 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
240 | ||
241 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, | |
242 | struct io_event *res) | |
243 | { | |
244 | kiocb_cancel_fn *old, *cancel; | |
245 | int ret = -EINVAL; | |
246 | ||
247 | /* | |
248 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
249 | * actually has a cancel function, hence the cmpxchg() | |
250 | */ | |
251 | ||
252 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
253 | do { | |
254 | if (!cancel || cancel == KIOCB_CANCELLED) | |
255 | return ret; | |
256 | ||
257 | old = cancel; | |
258 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
259 | } while (cancel != old); | |
260 | ||
261 | atomic_inc(&kiocb->ki_users); | |
262 | spin_unlock_irq(&ctx->ctx_lock); | |
263 | ||
264 | memset(res, 0, sizeof(*res)); | |
265 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
266 | res->data = kiocb->ki_user_data; | |
267 | ret = cancel(kiocb, res); | |
268 | ||
269 | spin_lock_irq(&ctx->ctx_lock); | |
270 | ||
271 | return ret; | |
272 | } | |
273 | ||
274 | static void free_ioctx_rcu(struct rcu_head *head) | |
275 | { | |
276 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
277 | kmem_cache_free(kioctx_cachep, ctx); | |
278 | } | |
279 | ||
280 | /* | |
281 | * When this function runs, the kioctx has been removed from the "hash table" | |
282 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
283 | * now it's safe to cancel any that need to be. | |
284 | */ | |
285 | static void free_ioctx(struct kioctx *ctx) | |
286 | { | |
287 | struct aio_ring *ring; | |
288 | struct io_event res; | |
289 | struct kiocb *req; | |
290 | unsigned head, avail; | |
291 | ||
292 | spin_lock_irq(&ctx->ctx_lock); | |
293 | ||
294 | while (!list_empty(&ctx->active_reqs)) { | |
295 | req = list_first_entry(&ctx->active_reqs, | |
296 | struct kiocb, ki_list); | |
297 | ||
298 | list_del_init(&req->ki_list); | |
299 | kiocb_cancel(ctx, req, &res); | |
300 | } | |
301 | ||
302 | spin_unlock_irq(&ctx->ctx_lock); | |
303 | ||
304 | ring = kmap_atomic(ctx->ring_pages[0]); | |
305 | head = ring->head; | |
306 | kunmap_atomic(ring); | |
307 | ||
308 | while (atomic_read(&ctx->reqs_active) > 0) { | |
309 | wait_event(ctx->wait, | |
310 | head != ctx->tail || | |
311 | atomic_read(&ctx->reqs_active) <= 0); | |
312 | ||
313 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; | |
314 | ||
315 | atomic_sub(avail, &ctx->reqs_active); | |
316 | head += avail; | |
317 | head %= ctx->nr_events; | |
318 | } | |
319 | ||
320 | WARN_ON(atomic_read(&ctx->reqs_active) < 0); | |
321 | ||
322 | aio_free_ring(ctx); | |
323 | ||
324 | pr_debug("freeing %p\n", ctx); | |
325 | ||
326 | /* | |
327 | * Here the call_rcu() is between the wait_event() for reqs_active to | |
328 | * hit 0, and freeing the ioctx. | |
329 | * | |
330 | * aio_complete() decrements reqs_active, but it has to touch the ioctx | |
331 | * after to issue a wakeup so we use rcu. | |
332 | */ | |
333 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); | |
334 | } | |
335 | ||
336 | static void put_ioctx(struct kioctx *ctx) | |
337 | { | |
338 | if (unlikely(atomic_dec_and_test(&ctx->users))) | |
339 | free_ioctx(ctx); | |
340 | } | |
341 | ||
342 | /* ioctx_alloc | |
343 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
344 | */ | |
345 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
346 | { | |
347 | struct mm_struct *mm = current->mm; | |
348 | struct kioctx *ctx; | |
349 | int err = -ENOMEM; | |
350 | ||
351 | /* Prevent overflows */ | |
352 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
353 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
354 | pr_debug("ENOMEM: nr_events too high\n"); | |
355 | return ERR_PTR(-EINVAL); | |
356 | } | |
357 | ||
358 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) | |
359 | return ERR_PTR(-EAGAIN); | |
360 | ||
361 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); | |
362 | if (!ctx) | |
363 | return ERR_PTR(-ENOMEM); | |
364 | ||
365 | ctx->max_reqs = nr_events; | |
366 | ||
367 | atomic_set(&ctx->users, 2); | |
368 | atomic_set(&ctx->dead, 0); | |
369 | spin_lock_init(&ctx->ctx_lock); | |
370 | spin_lock_init(&ctx->completion_lock); | |
371 | mutex_init(&ctx->ring_lock); | |
372 | init_waitqueue_head(&ctx->wait); | |
373 | ||
374 | INIT_LIST_HEAD(&ctx->active_reqs); | |
375 | ||
376 | if (aio_setup_ring(ctx) < 0) | |
377 | goto out_freectx; | |
378 | ||
379 | /* limit the number of system wide aios */ | |
380 | spin_lock(&aio_nr_lock); | |
381 | if (aio_nr + nr_events > aio_max_nr || | |
382 | aio_nr + nr_events < aio_nr) { | |
383 | spin_unlock(&aio_nr_lock); | |
384 | goto out_cleanup; | |
385 | } | |
386 | aio_nr += ctx->max_reqs; | |
387 | spin_unlock(&aio_nr_lock); | |
388 | ||
389 | /* now link into global list. */ | |
390 | spin_lock(&mm->ioctx_lock); | |
391 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
392 | spin_unlock(&mm->ioctx_lock); | |
393 | ||
394 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", | |
395 | ctx, ctx->user_id, mm, ctx->nr_events); | |
396 | return ctx; | |
397 | ||
398 | out_cleanup: | |
399 | err = -EAGAIN; | |
400 | aio_free_ring(ctx); | |
401 | out_freectx: | |
402 | kmem_cache_free(kioctx_cachep, ctx); | |
403 | pr_debug("error allocating ioctx %d\n", err); | |
404 | return ERR_PTR(err); | |
405 | } | |
406 | ||
407 | static void kill_ioctx_work(struct work_struct *work) | |
408 | { | |
409 | struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); | |
410 | ||
411 | wake_up_all(&ctx->wait); | |
412 | put_ioctx(ctx); | |
413 | } | |
414 | ||
415 | static void kill_ioctx_rcu(struct rcu_head *head) | |
416 | { | |
417 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
418 | ||
419 | INIT_WORK(&ctx->rcu_work, kill_ioctx_work); | |
420 | schedule_work(&ctx->rcu_work); | |
421 | } | |
422 | ||
423 | /* kill_ioctx | |
424 | * Cancels all outstanding aio requests on an aio context. Used | |
425 | * when the processes owning a context have all exited to encourage | |
426 | * the rapid destruction of the kioctx. | |
427 | */ | |
428 | static void kill_ioctx(struct kioctx *ctx) | |
429 | { | |
430 | if (!atomic_xchg(&ctx->dead, 1)) { | |
431 | hlist_del_rcu(&ctx->list); | |
432 | ||
433 | /* | |
434 | * It'd be more correct to do this in free_ioctx(), after all | |
435 | * the outstanding kiocbs have finished - but by then io_destroy | |
436 | * has already returned, so io_setup() could potentially return | |
437 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
438 | * could tell). | |
439 | */ | |
440 | spin_lock(&aio_nr_lock); | |
441 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); | |
442 | aio_nr -= ctx->max_reqs; | |
443 | spin_unlock(&aio_nr_lock); | |
444 | ||
445 | if (ctx->mmap_size) | |
446 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
447 | ||
448 | /* Between hlist_del_rcu() and dropping the initial ref */ | |
449 | call_rcu(&ctx->rcu_head, kill_ioctx_rcu); | |
450 | } | |
451 | } | |
452 | ||
453 | /* wait_on_sync_kiocb: | |
454 | * Waits on the given sync kiocb to complete. | |
455 | */ | |
456 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) | |
457 | { | |
458 | while (atomic_read(&iocb->ki_users)) { | |
459 | set_current_state(TASK_UNINTERRUPTIBLE); | |
460 | if (!atomic_read(&iocb->ki_users)) | |
461 | break; | |
462 | io_schedule(); | |
463 | } | |
464 | __set_current_state(TASK_RUNNING); | |
465 | return iocb->ki_user_data; | |
466 | } | |
467 | EXPORT_SYMBOL(wait_on_sync_kiocb); | |
468 | ||
469 | /* | |
470 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
471 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
472 | * called on the context. | |
473 | * | |
474 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
475 | * them. | |
476 | */ | |
477 | void exit_aio(struct mm_struct *mm) | |
478 | { | |
479 | struct kioctx *ctx; | |
480 | struct hlist_node *n; | |
481 | ||
482 | hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { | |
483 | if (1 != atomic_read(&ctx->users)) | |
484 | printk(KERN_DEBUG | |
485 | "exit_aio:ioctx still alive: %d %d %d\n", | |
486 | atomic_read(&ctx->users), | |
487 | atomic_read(&ctx->dead), | |
488 | atomic_read(&ctx->reqs_active)); | |
489 | /* | |
490 | * We don't need to bother with munmap() here - | |
491 | * exit_mmap(mm) is coming and it'll unmap everything. | |
492 | * Since aio_free_ring() uses non-zero ->mmap_size | |
493 | * as indicator that it needs to unmap the area, | |
494 | * just set it to 0; aio_free_ring() is the only | |
495 | * place that uses ->mmap_size, so it's safe. | |
496 | */ | |
497 | ctx->mmap_size = 0; | |
498 | ||
499 | kill_ioctx(ctx); | |
500 | } | |
501 | } | |
502 | ||
503 | /* aio_get_req | |
504 | * Allocate a slot for an aio request. Increments the ki_users count | |
505 | * of the kioctx so that the kioctx stays around until all requests are | |
506 | * complete. Returns NULL if no requests are free. | |
507 | * | |
508 | * Returns with kiocb->ki_users set to 2. The io submit code path holds | |
509 | * an extra reference while submitting the i/o. | |
510 | * This prevents races between the aio code path referencing the | |
511 | * req (after submitting it) and aio_complete() freeing the req. | |
512 | */ | |
513 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) | |
514 | { | |
515 | struct kiocb *req; | |
516 | ||
517 | if (atomic_read(&ctx->reqs_active) >= ctx->nr_events) | |
518 | return NULL; | |
519 | ||
520 | if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1) | |
521 | goto out_put; | |
522 | ||
523 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); | |
524 | if (unlikely(!req)) | |
525 | goto out_put; | |
526 | ||
527 | atomic_set(&req->ki_users, 2); | |
528 | req->ki_ctx = ctx; | |
529 | ||
530 | return req; | |
531 | out_put: | |
532 | atomic_dec(&ctx->reqs_active); | |
533 | return NULL; | |
534 | } | |
535 | ||
536 | static void kiocb_free(struct kiocb *req) | |
537 | { | |
538 | if (req->ki_filp) | |
539 | fput(req->ki_filp); | |
540 | if (req->ki_eventfd != NULL) | |
541 | eventfd_ctx_put(req->ki_eventfd); | |
542 | if (req->ki_dtor) | |
543 | req->ki_dtor(req); | |
544 | if (req->ki_iovec != &req->ki_inline_vec) | |
545 | kfree(req->ki_iovec); | |
546 | kmem_cache_free(kiocb_cachep, req); | |
547 | } | |
548 | ||
549 | void aio_put_req(struct kiocb *req) | |
550 | { | |
551 | if (atomic_dec_and_test(&req->ki_users)) | |
552 | kiocb_free(req); | |
553 | } | |
554 | EXPORT_SYMBOL(aio_put_req); | |
555 | ||
556 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) | |
557 | { | |
558 | struct mm_struct *mm = current->mm; | |
559 | struct kioctx *ctx, *ret = NULL; | |
560 | ||
561 | rcu_read_lock(); | |
562 | ||
563 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { | |
564 | if (ctx->user_id == ctx_id) { | |
565 | atomic_inc(&ctx->users); | |
566 | ret = ctx; | |
567 | break; | |
568 | } | |
569 | } | |
570 | ||
571 | rcu_read_unlock(); | |
572 | return ret; | |
573 | } | |
574 | ||
575 | /* aio_complete | |
576 | * Called when the io request on the given iocb is complete. | |
577 | */ | |
578 | void aio_complete(struct kiocb *iocb, long res, long res2) | |
579 | { | |
580 | struct kioctx *ctx = iocb->ki_ctx; | |
581 | struct aio_ring *ring; | |
582 | struct io_event *ev_page, *event; | |
583 | unsigned long flags; | |
584 | unsigned tail, pos; | |
585 | ||
586 | /* | |
587 | * Special case handling for sync iocbs: | |
588 | * - events go directly into the iocb for fast handling | |
589 | * - the sync task with the iocb in its stack holds the single iocb | |
590 | * ref, no other paths have a way to get another ref | |
591 | * - the sync task helpfully left a reference to itself in the iocb | |
592 | */ | |
593 | if (is_sync_kiocb(iocb)) { | |
594 | BUG_ON(atomic_read(&iocb->ki_users) != 1); | |
595 | iocb->ki_user_data = res; | |
596 | atomic_set(&iocb->ki_users, 0); | |
597 | wake_up_process(iocb->ki_obj.tsk); | |
598 | return; | |
599 | } | |
600 | ||
601 | /* | |
602 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we | |
603 | * need to issue a wakeup after decrementing reqs_active. | |
604 | */ | |
605 | rcu_read_lock(); | |
606 | ||
607 | if (iocb->ki_list.next) { | |
608 | unsigned long flags; | |
609 | ||
610 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
611 | list_del(&iocb->ki_list); | |
612 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
613 | } | |
614 | ||
615 | /* | |
616 | * cancelled requests don't get events, userland was given one | |
617 | * when the event got cancelled. | |
618 | */ | |
619 | if (unlikely(xchg(&iocb->ki_cancel, | |
620 | KIOCB_CANCELLED) == KIOCB_CANCELLED)) { | |
621 | atomic_dec(&ctx->reqs_active); | |
622 | /* Still need the wake_up in case free_ioctx is waiting */ | |
623 | goto put_rq; | |
624 | } | |
625 | ||
626 | /* | |
627 | * Add a completion event to the ring buffer. Must be done holding | |
628 | * ctx->ctx_lock to prevent other code from messing with the tail | |
629 | * pointer since we might be called from irq context. | |
630 | */ | |
631 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
632 | ||
633 | tail = ctx->tail; | |
634 | pos = tail + AIO_EVENTS_OFFSET; | |
635 | ||
636 | if (++tail >= ctx->nr_events) | |
637 | tail = 0; | |
638 | ||
639 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); | |
640 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; | |
641 | ||
642 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; | |
643 | event->data = iocb->ki_user_data; | |
644 | event->res = res; | |
645 | event->res2 = res2; | |
646 | ||
647 | kunmap_atomic(ev_page); | |
648 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); | |
649 | ||
650 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
651 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, | |
652 | res, res2); | |
653 | ||
654 | /* after flagging the request as done, we | |
655 | * must never even look at it again | |
656 | */ | |
657 | smp_wmb(); /* make event visible before updating tail */ | |
658 | ||
659 | ctx->tail = tail; | |
660 | ||
661 | ring = kmap_atomic(ctx->ring_pages[0]); | |
662 | ring->tail = tail; | |
663 | kunmap_atomic(ring); | |
664 | flush_dcache_page(ctx->ring_pages[0]); | |
665 | ||
666 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
667 | ||
668 | pr_debug("added to ring %p at [%u]\n", iocb, tail); | |
669 | ||
670 | /* | |
671 | * Check if the user asked us to deliver the result through an | |
672 | * eventfd. The eventfd_signal() function is safe to be called | |
673 | * from IRQ context. | |
674 | */ | |
675 | if (iocb->ki_eventfd != NULL) | |
676 | eventfd_signal(iocb->ki_eventfd, 1); | |
677 | ||
678 | put_rq: | |
679 | /* everything turned out well, dispose of the aiocb. */ | |
680 | aio_put_req(iocb); | |
681 | ||
682 | /* | |
683 | * We have to order our ring_info tail store above and test | |
684 | * of the wait list below outside the wait lock. This is | |
685 | * like in wake_up_bit() where clearing a bit has to be | |
686 | * ordered with the unlocked test. | |
687 | */ | |
688 | smp_mb(); | |
689 | ||
690 | if (waitqueue_active(&ctx->wait)) | |
691 | wake_up(&ctx->wait); | |
692 | ||
693 | rcu_read_unlock(); | |
694 | } | |
695 | EXPORT_SYMBOL(aio_complete); | |
696 | ||
697 | /* aio_read_events | |
698 | * Pull an event off of the ioctx's event ring. Returns the number of | |
699 | * events fetched | |
700 | */ | |
701 | static long aio_read_events_ring(struct kioctx *ctx, | |
702 | struct io_event __user *event, long nr) | |
703 | { | |
704 | struct aio_ring *ring; | |
705 | unsigned head, pos; | |
706 | long ret = 0; | |
707 | int copy_ret; | |
708 | ||
709 | mutex_lock(&ctx->ring_lock); | |
710 | ||
711 | ring = kmap_atomic(ctx->ring_pages[0]); | |
712 | head = ring->head; | |
713 | kunmap_atomic(ring); | |
714 | ||
715 | pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events); | |
716 | ||
717 | if (head == ctx->tail) | |
718 | goto out; | |
719 | ||
720 | while (ret < nr) { | |
721 | long avail; | |
722 | struct io_event *ev; | |
723 | struct page *page; | |
724 | ||
725 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; | |
726 | if (head == ctx->tail) | |
727 | break; | |
728 | ||
729 | avail = min(avail, nr - ret); | |
730 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
731 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
732 | ||
733 | pos = head + AIO_EVENTS_OFFSET; | |
734 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; | |
735 | pos %= AIO_EVENTS_PER_PAGE; | |
736 | ||
737 | ev = kmap(page); | |
738 | copy_ret = copy_to_user(event + ret, ev + pos, | |
739 | sizeof(*ev) * avail); | |
740 | kunmap(page); | |
741 | ||
742 | if (unlikely(copy_ret)) { | |
743 | ret = -EFAULT; | |
744 | goto out; | |
745 | } | |
746 | ||
747 | ret += avail; | |
748 | head += avail; | |
749 | head %= ctx->nr_events; | |
750 | } | |
751 | ||
752 | ring = kmap_atomic(ctx->ring_pages[0]); | |
753 | ring->head = head; | |
754 | kunmap_atomic(ring); | |
755 | flush_dcache_page(ctx->ring_pages[0]); | |
756 | ||
757 | pr_debug("%li h%u t%u\n", ret, head, ctx->tail); | |
758 | ||
759 | atomic_sub(ret, &ctx->reqs_active); | |
760 | out: | |
761 | mutex_unlock(&ctx->ring_lock); | |
762 | ||
763 | return ret; | |
764 | } | |
765 | ||
766 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, | |
767 | struct io_event __user *event, long *i) | |
768 | { | |
769 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); | |
770 | ||
771 | if (ret > 0) | |
772 | *i += ret; | |
773 | ||
774 | if (unlikely(atomic_read(&ctx->dead))) | |
775 | ret = -EINVAL; | |
776 | ||
777 | if (!*i) | |
778 | *i = ret; | |
779 | ||
780 | return ret < 0 || *i >= min_nr; | |
781 | } | |
782 | ||
783 | static long read_events(struct kioctx *ctx, long min_nr, long nr, | |
784 | struct io_event __user *event, | |
785 | struct timespec __user *timeout) | |
786 | { | |
787 | ktime_t until = { .tv64 = KTIME_MAX }; | |
788 | long ret = 0; | |
789 | ||
790 | if (timeout) { | |
791 | struct timespec ts; | |
792 | ||
793 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) | |
794 | return -EFAULT; | |
795 | ||
796 | until = timespec_to_ktime(ts); | |
797 | } | |
798 | ||
799 | /* | |
800 | * Note that aio_read_events() is being called as the conditional - i.e. | |
801 | * we're calling it after prepare_to_wait() has set task state to | |
802 | * TASK_INTERRUPTIBLE. | |
803 | * | |
804 | * But aio_read_events() can block, and if it blocks it's going to flip | |
805 | * the task state back to TASK_RUNNING. | |
806 | * | |
807 | * This should be ok, provided it doesn't flip the state back to | |
808 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
809 | * will only happen if the mutex_lock() call blocks, and we then find | |
810 | * the ringbuffer empty. So in practice we should be ok, but it's | |
811 | * something to be aware of when touching this code. | |
812 | */ | |
813 | wait_event_interruptible_hrtimeout(ctx->wait, | |
814 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
815 | ||
816 | if (!ret && signal_pending(current)) | |
817 | ret = -EINTR; | |
818 | ||
819 | return ret; | |
820 | } | |
821 | ||
822 | /* sys_io_setup: | |
823 | * Create an aio_context capable of receiving at least nr_events. | |
824 | * ctxp must not point to an aio_context that already exists, and | |
825 | * must be initialized to 0 prior to the call. On successful | |
826 | * creation of the aio_context, *ctxp is filled in with the resulting | |
827 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
828 | * if the specified nr_events exceeds internal limits. May fail | |
829 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
830 | * of available events. May fail with -ENOMEM if insufficient kernel | |
831 | * resources are available. May fail with -EFAULT if an invalid | |
832 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
833 | * implemented. | |
834 | */ | |
835 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) | |
836 | { | |
837 | struct kioctx *ioctx = NULL; | |
838 | unsigned long ctx; | |
839 | long ret; | |
840 | ||
841 | ret = get_user(ctx, ctxp); | |
842 | if (unlikely(ret)) | |
843 | goto out; | |
844 | ||
845 | ret = -EINVAL; | |
846 | if (unlikely(ctx || nr_events == 0)) { | |
847 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
848 | ctx, nr_events); | |
849 | goto out; | |
850 | } | |
851 | ||
852 | ioctx = ioctx_alloc(nr_events); | |
853 | ret = PTR_ERR(ioctx); | |
854 | if (!IS_ERR(ioctx)) { | |
855 | ret = put_user(ioctx->user_id, ctxp); | |
856 | if (ret) | |
857 | kill_ioctx(ioctx); | |
858 | put_ioctx(ioctx); | |
859 | } | |
860 | ||
861 | out: | |
862 | return ret; | |
863 | } | |
864 | ||
865 | /* sys_io_destroy: | |
866 | * Destroy the aio_context specified. May cancel any outstanding | |
867 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
868 | * implemented. May fail with -EINVAL if the context pointed to | |
869 | * is invalid. | |
870 | */ | |
871 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) | |
872 | { | |
873 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
874 | if (likely(NULL != ioctx)) { | |
875 | kill_ioctx(ioctx); | |
876 | put_ioctx(ioctx); | |
877 | return 0; | |
878 | } | |
879 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
880 | return -EINVAL; | |
881 | } | |
882 | ||
883 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) | |
884 | { | |
885 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; | |
886 | ||
887 | BUG_ON(ret <= 0); | |
888 | ||
889 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { | |
890 | ssize_t this = min((ssize_t)iov->iov_len, ret); | |
891 | iov->iov_base += this; | |
892 | iov->iov_len -= this; | |
893 | iocb->ki_left -= this; | |
894 | ret -= this; | |
895 | if (iov->iov_len == 0) { | |
896 | iocb->ki_cur_seg++; | |
897 | iov++; | |
898 | } | |
899 | } | |
900 | ||
901 | /* the caller should not have done more io than what fit in | |
902 | * the remaining iovecs */ | |
903 | BUG_ON(ret > 0 && iocb->ki_left == 0); | |
904 | } | |
905 | ||
906 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, | |
907 | unsigned long, loff_t); | |
908 | ||
909 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb, int rw, aio_rw_op *rw_op) | |
910 | { | |
911 | struct file *file = iocb->ki_filp; | |
912 | struct address_space *mapping = file->f_mapping; | |
913 | struct inode *inode = mapping->host; | |
914 | ssize_t ret = 0; | |
915 | ||
916 | /* This matches the pread()/pwrite() logic */ | |
917 | if (iocb->ki_pos < 0) | |
918 | return -EINVAL; | |
919 | ||
920 | if (rw == WRITE) | |
921 | file_start_write(file); | |
922 | do { | |
923 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], | |
924 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
925 | iocb->ki_pos); | |
926 | if (ret > 0) | |
927 | aio_advance_iovec(iocb, ret); | |
928 | ||
929 | /* retry all partial writes. retry partial reads as long as its a | |
930 | * regular file. */ | |
931 | } while (ret > 0 && iocb->ki_left > 0 && | |
932 | (rw == WRITE || | |
933 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); | |
934 | if (rw == WRITE) | |
935 | file_end_write(file); | |
936 | ||
937 | /* This means we must have transferred all that we could */ | |
938 | /* No need to retry anymore */ | |
939 | if ((ret == 0) || (iocb->ki_left == 0)) | |
940 | ret = iocb->ki_nbytes - iocb->ki_left; | |
941 | ||
942 | /* If we managed to write some out we return that, rather than | |
943 | * the eventual error. */ | |
944 | if (rw == WRITE | |
945 | && ret < 0 && ret != -EIOCBQUEUED | |
946 | && iocb->ki_nbytes - iocb->ki_left) | |
947 | ret = iocb->ki_nbytes - iocb->ki_left; | |
948 | ||
949 | return ret; | |
950 | } | |
951 | ||
952 | static ssize_t aio_setup_vectored_rw(int rw, struct kiocb *kiocb, bool compat) | |
953 | { | |
954 | ssize_t ret; | |
955 | ||
956 | kiocb->ki_nr_segs = kiocb->ki_nbytes; | |
957 | ||
958 | #ifdef CONFIG_COMPAT | |
959 | if (compat) | |
960 | ret = compat_rw_copy_check_uvector(rw, | |
961 | (struct compat_iovec __user *)kiocb->ki_buf, | |
962 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, | |
963 | &kiocb->ki_iovec); | |
964 | else | |
965 | #endif | |
966 | ret = rw_copy_check_uvector(rw, | |
967 | (struct iovec __user *)kiocb->ki_buf, | |
968 | kiocb->ki_nr_segs, 1, &kiocb->ki_inline_vec, | |
969 | &kiocb->ki_iovec); | |
970 | if (ret < 0) | |
971 | return ret; | |
972 | ||
973 | /* ki_nbytes now reflect bytes instead of segs */ | |
974 | kiocb->ki_nbytes = ret; | |
975 | return 0; | |
976 | } | |
977 | ||
978 | static ssize_t aio_setup_single_vector(int rw, struct kiocb *kiocb) | |
979 | { | |
980 | if (unlikely(!access_ok(!rw, kiocb->ki_buf, kiocb->ki_nbytes))) | |
981 | return -EFAULT; | |
982 | ||
983 | kiocb->ki_iovec = &kiocb->ki_inline_vec; | |
984 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
985 | kiocb->ki_iovec->iov_len = kiocb->ki_nbytes; | |
986 | kiocb->ki_nr_segs = 1; | |
987 | return 0; | |
988 | } | |
989 | ||
990 | /* | |
991 | * aio_setup_iocb: | |
992 | * Performs the initial checks and aio retry method | |
993 | * setup for the kiocb at the time of io submission. | |
994 | */ | |
995 | static ssize_t aio_run_iocb(struct kiocb *req, bool compat) | |
996 | { | |
997 | struct file *file = req->ki_filp; | |
998 | ssize_t ret; | |
999 | int rw; | |
1000 | fmode_t mode; | |
1001 | aio_rw_op *rw_op; | |
1002 | ||
1003 | switch (req->ki_opcode) { | |
1004 | case IOCB_CMD_PREAD: | |
1005 | case IOCB_CMD_PREADV: | |
1006 | mode = FMODE_READ; | |
1007 | rw = READ; | |
1008 | rw_op = file->f_op->aio_read; | |
1009 | goto rw_common; | |
1010 | ||
1011 | case IOCB_CMD_PWRITE: | |
1012 | case IOCB_CMD_PWRITEV: | |
1013 | mode = FMODE_WRITE; | |
1014 | rw = WRITE; | |
1015 | rw_op = file->f_op->aio_write; | |
1016 | goto rw_common; | |
1017 | rw_common: | |
1018 | if (unlikely(!(file->f_mode & mode))) | |
1019 | return -EBADF; | |
1020 | ||
1021 | if (!rw_op) | |
1022 | return -EINVAL; | |
1023 | ||
1024 | ret = (req->ki_opcode == IOCB_CMD_PREADV || | |
1025 | req->ki_opcode == IOCB_CMD_PWRITEV) | |
1026 | ? aio_setup_vectored_rw(rw, req, compat) | |
1027 | : aio_setup_single_vector(rw, req); | |
1028 | if (ret) | |
1029 | return ret; | |
1030 | ||
1031 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
1032 | if (ret < 0) | |
1033 | return ret; | |
1034 | ||
1035 | req->ki_nbytes = ret; | |
1036 | req->ki_left = ret; | |
1037 | ||
1038 | ret = aio_rw_vect_retry(req, rw, rw_op); | |
1039 | break; | |
1040 | ||
1041 | case IOCB_CMD_FDSYNC: | |
1042 | if (!file->f_op->aio_fsync) | |
1043 | return -EINVAL; | |
1044 | ||
1045 | ret = file->f_op->aio_fsync(req, 1); | |
1046 | break; | |
1047 | ||
1048 | case IOCB_CMD_FSYNC: | |
1049 | if (!file->f_op->aio_fsync) | |
1050 | return -EINVAL; | |
1051 | ||
1052 | ret = file->f_op->aio_fsync(req, 0); | |
1053 | break; | |
1054 | ||
1055 | default: | |
1056 | pr_debug("EINVAL: no operation provided\n"); | |
1057 | return -EINVAL; | |
1058 | } | |
1059 | ||
1060 | if (ret != -EIOCBQUEUED) { | |
1061 | /* | |
1062 | * There's no easy way to restart the syscall since other AIO's | |
1063 | * may be already running. Just fail this IO with EINTR. | |
1064 | */ | |
1065 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1066 | ret == -ERESTARTNOHAND || | |
1067 | ret == -ERESTART_RESTARTBLOCK)) | |
1068 | ret = -EINTR; | |
1069 | aio_complete(req, ret, 0); | |
1070 | } | |
1071 | ||
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, | |
1076 | struct iocb *iocb, bool compat) | |
1077 | { | |
1078 | struct kiocb *req; | |
1079 | ssize_t ret; | |
1080 | ||
1081 | /* enforce forwards compatibility on users */ | |
1082 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { | |
1083 | pr_debug("EINVAL: reserve field set\n"); | |
1084 | return -EINVAL; | |
1085 | } | |
1086 | ||
1087 | /* prevent overflows */ | |
1088 | if (unlikely( | |
1089 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1090 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1091 | ((ssize_t)iocb->aio_nbytes < 0) | |
1092 | )) { | |
1093 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1094 | return -EINVAL; | |
1095 | } | |
1096 | ||
1097 | req = aio_get_req(ctx); | |
1098 | if (unlikely(!req)) | |
1099 | return -EAGAIN; | |
1100 | ||
1101 | req->ki_filp = fget(iocb->aio_fildes); | |
1102 | if (unlikely(!req->ki_filp)) { | |
1103 | ret = -EBADF; | |
1104 | goto out_put_req; | |
1105 | } | |
1106 | ||
1107 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { | |
1108 | /* | |
1109 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1110 | * instance of the file* now. The file descriptor must be | |
1111 | * an eventfd() fd, and will be signaled for each completed | |
1112 | * event using the eventfd_signal() function. | |
1113 | */ | |
1114 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); | |
1115 | if (IS_ERR(req->ki_eventfd)) { | |
1116 | ret = PTR_ERR(req->ki_eventfd); | |
1117 | req->ki_eventfd = NULL; | |
1118 | goto out_put_req; | |
1119 | } | |
1120 | } | |
1121 | ||
1122 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); | |
1123 | if (unlikely(ret)) { | |
1124 | pr_debug("EFAULT: aio_key\n"); | |
1125 | goto out_put_req; | |
1126 | } | |
1127 | ||
1128 | req->ki_obj.user = user_iocb; | |
1129 | req->ki_user_data = iocb->aio_data; | |
1130 | req->ki_pos = iocb->aio_offset; | |
1131 | ||
1132 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1133 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1134 | req->ki_opcode = iocb->aio_lio_opcode; | |
1135 | ||
1136 | ret = aio_run_iocb(req, compat); | |
1137 | if (ret) | |
1138 | goto out_put_req; | |
1139 | ||
1140 | aio_put_req(req); /* drop extra ref to req */ | |
1141 | return 0; | |
1142 | out_put_req: | |
1143 | atomic_dec(&ctx->reqs_active); | |
1144 | aio_put_req(req); /* drop extra ref to req */ | |
1145 | aio_put_req(req); /* drop i/o ref to req */ | |
1146 | return ret; | |
1147 | } | |
1148 | ||
1149 | long do_io_submit(aio_context_t ctx_id, long nr, | |
1150 | struct iocb __user *__user *iocbpp, bool compat) | |
1151 | { | |
1152 | struct kioctx *ctx; | |
1153 | long ret = 0; | |
1154 | int i = 0; | |
1155 | struct blk_plug plug; | |
1156 | ||
1157 | if (unlikely(nr < 0)) | |
1158 | return -EINVAL; | |
1159 | ||
1160 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) | |
1161 | nr = LONG_MAX/sizeof(*iocbpp); | |
1162 | ||
1163 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) | |
1164 | return -EFAULT; | |
1165 | ||
1166 | ctx = lookup_ioctx(ctx_id); | |
1167 | if (unlikely(!ctx)) { | |
1168 | pr_debug("EINVAL: invalid context id\n"); | |
1169 | return -EINVAL; | |
1170 | } | |
1171 | ||
1172 | blk_start_plug(&plug); | |
1173 | ||
1174 | /* | |
1175 | * AKPM: should this return a partial result if some of the IOs were | |
1176 | * successfully submitted? | |
1177 | */ | |
1178 | for (i=0; i<nr; i++) { | |
1179 | struct iocb __user *user_iocb; | |
1180 | struct iocb tmp; | |
1181 | ||
1182 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1183 | ret = -EFAULT; | |
1184 | break; | |
1185 | } | |
1186 | ||
1187 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1188 | ret = -EFAULT; | |
1189 | break; | |
1190 | } | |
1191 | ||
1192 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); | |
1193 | if (ret) | |
1194 | break; | |
1195 | } | |
1196 | blk_finish_plug(&plug); | |
1197 | ||
1198 | put_ioctx(ctx); | |
1199 | return i ? i : ret; | |
1200 | } | |
1201 | ||
1202 | /* sys_io_submit: | |
1203 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1204 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1205 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1206 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1207 | * is invalid for the file descriptor in the iocb. May fail with | |
1208 | * -EFAULT if any of the data structures point to invalid data. May | |
1209 | * fail with -EBADF if the file descriptor specified in the first | |
1210 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1211 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1212 | * fail with -ENOSYS if not implemented. | |
1213 | */ | |
1214 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1215 | struct iocb __user * __user *, iocbpp) | |
1216 | { | |
1217 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1218 | } | |
1219 | ||
1220 | /* lookup_kiocb | |
1221 | * Finds a given iocb for cancellation. | |
1222 | */ | |
1223 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, | |
1224 | u32 key) | |
1225 | { | |
1226 | struct list_head *pos; | |
1227 | ||
1228 | assert_spin_locked(&ctx->ctx_lock); | |
1229 | ||
1230 | if (key != KIOCB_KEY) | |
1231 | return NULL; | |
1232 | ||
1233 | /* TODO: use a hash or array, this sucks. */ | |
1234 | list_for_each(pos, &ctx->active_reqs) { | |
1235 | struct kiocb *kiocb = list_kiocb(pos); | |
1236 | if (kiocb->ki_obj.user == iocb) | |
1237 | return kiocb; | |
1238 | } | |
1239 | return NULL; | |
1240 | } | |
1241 | ||
1242 | /* sys_io_cancel: | |
1243 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1244 | * the operation is successfully cancelled, the resulting event is | |
1245 | * copied into the memory pointed to by result without being placed | |
1246 | * into the completion queue and 0 is returned. May fail with | |
1247 | * -EFAULT if any of the data structures pointed to are invalid. | |
1248 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1249 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1250 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1251 | */ | |
1252 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, | |
1253 | struct io_event __user *, result) | |
1254 | { | |
1255 | struct io_event res; | |
1256 | struct kioctx *ctx; | |
1257 | struct kiocb *kiocb; | |
1258 | u32 key; | |
1259 | int ret; | |
1260 | ||
1261 | ret = get_user(key, &iocb->aio_key); | |
1262 | if (unlikely(ret)) | |
1263 | return -EFAULT; | |
1264 | ||
1265 | ctx = lookup_ioctx(ctx_id); | |
1266 | if (unlikely(!ctx)) | |
1267 | return -EINVAL; | |
1268 | ||
1269 | spin_lock_irq(&ctx->ctx_lock); | |
1270 | ||
1271 | kiocb = lookup_kiocb(ctx, iocb, key); | |
1272 | if (kiocb) | |
1273 | ret = kiocb_cancel(ctx, kiocb, &res); | |
1274 | else | |
1275 | ret = -EINVAL; | |
1276 | ||
1277 | spin_unlock_irq(&ctx->ctx_lock); | |
1278 | ||
1279 | if (!ret) { | |
1280 | /* Cancellation succeeded -- copy the result | |
1281 | * into the user's buffer. | |
1282 | */ | |
1283 | if (copy_to_user(result, &res, sizeof(res))) | |
1284 | ret = -EFAULT; | |
1285 | } | |
1286 | ||
1287 | put_ioctx(ctx); | |
1288 | ||
1289 | return ret; | |
1290 | } | |
1291 | ||
1292 | /* io_getevents: | |
1293 | * Attempts to read at least min_nr events and up to nr events from | |
1294 | * the completion queue for the aio_context specified by ctx_id. If | |
1295 | * it succeeds, the number of read events is returned. May fail with | |
1296 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1297 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1298 | * if any of the memory specified is invalid. May return 0 or | |
1299 | * < min_nr if the timeout specified by timeout has elapsed | |
1300 | * before sufficient events are available, where timeout == NULL | |
1301 | * specifies an infinite timeout. Note that the timeout pointed to by | |
1302 | * timeout is relative. Will fail with -ENOSYS if not implemented. | |
1303 | */ | |
1304 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, | |
1305 | long, min_nr, | |
1306 | long, nr, | |
1307 | struct io_event __user *, events, | |
1308 | struct timespec __user *, timeout) | |
1309 | { | |
1310 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1311 | long ret = -EINVAL; | |
1312 | ||
1313 | if (likely(ioctx)) { | |
1314 | if (likely(min_nr <= nr && min_nr >= 0)) | |
1315 | ret = read_events(ioctx, min_nr, nr, events, timeout); | |
1316 | put_ioctx(ioctx); | |
1317 | } | |
1318 | return ret; | |
1319 | } |