Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | */ | |
caf4167a KO |
11 | #define pr_fmt(fmt) "%s: " fmt, __func__ |
12 | ||
1da177e4 LT |
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> | |
630d9c47 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/syscalls.h> |
b9d128f1 | 20 | #include <linux/backing-dev.h> |
027445c3 | 21 | #include <linux/uio.h> |
1da177e4 | 22 | |
1da177e4 LT |
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> | |
3d2d827f | 28 | #include <linux/mmu_context.h> |
1da177e4 LT |
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> | |
9c3060be | 35 | #include <linux/eventfd.h> |
cfb1e33e | 36 | #include <linux/blkdev.h> |
9d85cba7 | 37 | #include <linux/compat.h> |
1da177e4 LT |
38 | |
39 | #include <asm/kmap_types.h> | |
40 | #include <asm/uaccess.h> | |
1da177e4 | 41 | |
4e179bca KO |
42 | #define AIO_RING_MAGIC 0xa10a10a1 |
43 | #define AIO_RING_COMPAT_FEATURES 1 | |
44 | #define AIO_RING_INCOMPAT_FEATURES 0 | |
45 | struct aio_ring { | |
46 | unsigned id; /* kernel internal index number */ | |
47 | unsigned nr; /* number of io_events */ | |
48 | unsigned head; | |
49 | unsigned tail; | |
50 | ||
51 | unsigned magic; | |
52 | unsigned compat_features; | |
53 | unsigned incompat_features; | |
54 | unsigned header_length; /* size of aio_ring */ | |
55 | ||
56 | ||
57 | struct io_event io_events[0]; | |
58 | }; /* 128 bytes + ring size */ | |
59 | ||
60 | #define AIO_RING_PAGES 8 | |
61 | struct aio_ring_info { | |
62 | unsigned long mmap_base; | |
63 | unsigned long mmap_size; | |
64 | ||
65 | struct page **ring_pages; | |
66 | spinlock_t ring_lock; | |
67 | long nr_pages; | |
68 | ||
69 | unsigned nr, tail; | |
70 | ||
71 | struct page *internal_pages[AIO_RING_PAGES]; | |
72 | }; | |
73 | ||
74 | static inline unsigned aio_ring_avail(struct aio_ring_info *info, | |
75 | struct aio_ring *ring) | |
76 | { | |
77 | return (ring->head + info->nr - 1 - ring->tail) % info->nr; | |
78 | } | |
79 | ||
80 | struct kioctx { | |
81 | atomic_t users; | |
82 | int dead; | |
83 | ||
84 | /* This needs improving */ | |
85 | unsigned long user_id; | |
86 | struct hlist_node list; | |
87 | ||
88 | wait_queue_head_t wait; | |
89 | ||
90 | spinlock_t ctx_lock; | |
91 | ||
92 | int reqs_active; | |
93 | struct list_head active_reqs; /* used for cancellation */ | |
94 | ||
95 | /* sys_io_setup currently limits this to an unsigned int */ | |
96 | unsigned max_reqs; | |
97 | ||
98 | struct aio_ring_info ring_info; | |
99 | ||
100 | struct rcu_head rcu_head; | |
101 | }; | |
102 | ||
1da177e4 | 103 | /*------ sysctl variables----*/ |
d55b5fda ZB |
104 | static DEFINE_SPINLOCK(aio_nr_lock); |
105 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
106 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
107 | /*----end sysctl variables---*/ |
108 | ||
e18b890b CL |
109 | static struct kmem_cache *kiocb_cachep; |
110 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 111 | |
1da177e4 LT |
112 | /* aio_setup |
113 | * Creates the slab caches used by the aio routines, panic on | |
114 | * failure as this is done early during the boot sequence. | |
115 | */ | |
116 | static int __init aio_setup(void) | |
117 | { | |
0a31bd5f CL |
118 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
119 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 120 | |
caf4167a | 121 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
122 | |
123 | return 0; | |
124 | } | |
385773e0 | 125 | __initcall(aio_setup); |
1da177e4 LT |
126 | |
127 | static void aio_free_ring(struct kioctx *ctx) | |
128 | { | |
129 | struct aio_ring_info *info = &ctx->ring_info; | |
130 | long i; | |
131 | ||
132 | for (i=0; i<info->nr_pages; i++) | |
133 | put_page(info->ring_pages[i]); | |
134 | ||
936af157 | 135 | if (info->mmap_size) { |
bfce281c | 136 | vm_munmap(info->mmap_base, info->mmap_size); |
936af157 | 137 | } |
1da177e4 LT |
138 | |
139 | if (info->ring_pages && info->ring_pages != info->internal_pages) | |
140 | kfree(info->ring_pages); | |
141 | info->ring_pages = NULL; | |
142 | info->nr = 0; | |
143 | } | |
144 | ||
145 | static int aio_setup_ring(struct kioctx *ctx) | |
146 | { | |
147 | struct aio_ring *ring; | |
148 | struct aio_ring_info *info = &ctx->ring_info; | |
149 | unsigned nr_events = ctx->max_reqs; | |
41003a7b | 150 | struct mm_struct *mm = current->mm; |
41badc15 | 151 | unsigned long size, populate; |
1da177e4 LT |
152 | int nr_pages; |
153 | ||
154 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
155 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
156 | ||
157 | size = sizeof(struct aio_ring); | |
158 | size += sizeof(struct io_event) * nr_events; | |
159 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; | |
160 | ||
161 | if (nr_pages < 0) | |
162 | return -EINVAL; | |
163 | ||
164 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); | |
165 | ||
166 | info->nr = 0; | |
167 | info->ring_pages = info->internal_pages; | |
168 | if (nr_pages > AIO_RING_PAGES) { | |
11b0b5ab | 169 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
170 | if (!info->ring_pages) |
171 | return -ENOMEM; | |
1da177e4 LT |
172 | } |
173 | ||
174 | info->mmap_size = nr_pages * PAGE_SIZE; | |
caf4167a | 175 | pr_debug("attempting mmap of %lu bytes\n", info->mmap_size); |
41003a7b | 176 | down_write(&mm->mmap_sem); |
e3fc629d AV |
177 | info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size, |
178 | PROT_READ|PROT_WRITE, | |
bebeb3d6 ML |
179 | MAP_ANONYMOUS|MAP_PRIVATE, 0, |
180 | &populate); | |
1da177e4 | 181 | if (IS_ERR((void *)info->mmap_base)) { |
41003a7b | 182 | up_write(&mm->mmap_sem); |
1da177e4 LT |
183 | info->mmap_size = 0; |
184 | aio_free_ring(ctx); | |
185 | return -EAGAIN; | |
186 | } | |
187 | ||
caf4167a | 188 | pr_debug("mmap address: 0x%08lx\n", info->mmap_base); |
41003a7b | 189 | info->nr_pages = get_user_pages(current, mm, info->mmap_base, nr_pages, |
1da177e4 | 190 | 1, 0, info->ring_pages, NULL); |
41003a7b | 191 | up_write(&mm->mmap_sem); |
1da177e4 LT |
192 | |
193 | if (unlikely(info->nr_pages != nr_pages)) { | |
194 | aio_free_ring(ctx); | |
195 | return -EAGAIN; | |
196 | } | |
bebeb3d6 | 197 | if (populate) |
41badc15 | 198 | mm_populate(info->mmap_base, populate); |
1da177e4 LT |
199 | |
200 | ctx->user_id = info->mmap_base; | |
201 | ||
202 | info->nr = nr_events; /* trusted copy */ | |
203 | ||
e8e3c3d6 | 204 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
205 | ring->nr = nr_events; /* user copy */ |
206 | ring->id = ctx->user_id; | |
207 | ring->head = ring->tail = 0; | |
208 | ring->magic = AIO_RING_MAGIC; | |
209 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
210 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
211 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 212 | kunmap_atomic(ring); |
1da177e4 LT |
213 | |
214 | return 0; | |
215 | } | |
216 | ||
217 | ||
218 | /* aio_ring_event: returns a pointer to the event at the given index from | |
e8e3c3d6 | 219 | * kmap_atomic(). Release the pointer with put_aio_ring_event(); |
1da177e4 LT |
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 | ||
e8e3c3d6 | 225 | #define aio_ring_event(info, nr) ({ \ |
1da177e4 LT |
226 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
227 | struct io_event *__event; \ | |
228 | __event = kmap_atomic( \ | |
e8e3c3d6 | 229 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \ |
1da177e4 LT |
230 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
231 | __event; \ | |
232 | }) | |
233 | ||
e8e3c3d6 | 234 | #define put_aio_ring_event(event) do { \ |
1da177e4 LT |
235 | struct io_event *__event = (event); \ |
236 | (void)__event; \ | |
e8e3c3d6 | 237 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \ |
1da177e4 LT |
238 | } while(0) |
239 | ||
abf137dd JA |
240 | static void ctx_rcu_free(struct rcu_head *head) |
241 | { | |
242 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
abf137dd | 243 | kmem_cache_free(kioctx_cachep, ctx); |
abf137dd | 244 | } |
d5470b59 AB |
245 | |
246 | /* __put_ioctx | |
247 | * Called when the last user of an aio context has gone away, | |
248 | * and the struct needs to be freed. | |
249 | */ | |
250 | static void __put_ioctx(struct kioctx *ctx) | |
251 | { | |
2dd542b7 | 252 | unsigned nr_events = ctx->max_reqs; |
d5470b59 AB |
253 | BUG_ON(ctx->reqs_active); |
254 | ||
d5470b59 | 255 | aio_free_ring(ctx); |
2dd542b7 AV |
256 | if (nr_events) { |
257 | spin_lock(&aio_nr_lock); | |
258 | BUG_ON(aio_nr - nr_events > aio_nr); | |
259 | aio_nr -= nr_events; | |
260 | spin_unlock(&aio_nr_lock); | |
261 | } | |
caf4167a | 262 | pr_debug("freeing %p\n", ctx); |
abf137dd | 263 | call_rcu(&ctx->rcu_head, ctx_rcu_free); |
d5470b59 AB |
264 | } |
265 | ||
3bd9a5d7 NP |
266 | static inline int try_get_ioctx(struct kioctx *kioctx) |
267 | { | |
268 | return atomic_inc_not_zero(&kioctx->users); | |
269 | } | |
270 | ||
271 | static inline void put_ioctx(struct kioctx *kioctx) | |
272 | { | |
273 | BUG_ON(atomic_read(&kioctx->users) <= 0); | |
274 | if (unlikely(atomic_dec_and_test(&kioctx->users))) | |
275 | __put_ioctx(kioctx); | |
276 | } | |
d5470b59 | 277 | |
906b973c KO |
278 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
279 | struct io_event *res) | |
280 | { | |
281 | int (*cancel)(struct kiocb *, struct io_event *); | |
282 | int ret = -EINVAL; | |
283 | ||
284 | cancel = kiocb->ki_cancel; | |
285 | kiocbSetCancelled(kiocb); | |
286 | if (cancel) { | |
287 | kiocb->ki_users++; | |
288 | spin_unlock_irq(&ctx->ctx_lock); | |
289 | ||
290 | memset(res, 0, sizeof(*res)); | |
291 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
292 | res->data = kiocb->ki_user_data; | |
293 | ret = cancel(kiocb, res); | |
294 | ||
295 | spin_lock_irq(&ctx->ctx_lock); | |
296 | } | |
297 | ||
298 | return ret; | |
299 | } | |
300 | ||
1da177e4 LT |
301 | /* ioctx_alloc |
302 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
303 | */ | |
304 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
305 | { | |
41003a7b | 306 | struct mm_struct *mm = current->mm; |
1da177e4 | 307 | struct kioctx *ctx; |
e23754f8 | 308 | int err = -ENOMEM; |
1da177e4 LT |
309 | |
310 | /* Prevent overflows */ | |
311 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
312 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
313 | pr_debug("ENOMEM: nr_events too high\n"); | |
314 | return ERR_PTR(-EINVAL); | |
315 | } | |
316 | ||
2dd542b7 | 317 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
1da177e4 LT |
318 | return ERR_PTR(-EAGAIN); |
319 | ||
c3762229 | 320 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
321 | if (!ctx) |
322 | return ERR_PTR(-ENOMEM); | |
323 | ||
1da177e4 | 324 | ctx->max_reqs = nr_events; |
1da177e4 | 325 | |
86b62a2c | 326 | atomic_set(&ctx->users, 2); |
1da177e4 LT |
327 | spin_lock_init(&ctx->ctx_lock); |
328 | spin_lock_init(&ctx->ring_info.ring_lock); | |
329 | init_waitqueue_head(&ctx->wait); | |
330 | ||
331 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 LT |
332 | |
333 | if (aio_setup_ring(ctx) < 0) | |
334 | goto out_freectx; | |
335 | ||
336 | /* limit the number of system wide aios */ | |
9fa1cb39 | 337 | spin_lock(&aio_nr_lock); |
2dd542b7 AV |
338 | if (aio_nr + nr_events > aio_max_nr || |
339 | aio_nr + nr_events < aio_nr) { | |
9fa1cb39 | 340 | spin_unlock(&aio_nr_lock); |
1da177e4 | 341 | goto out_cleanup; |
2dd542b7 AV |
342 | } |
343 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 344 | spin_unlock(&aio_nr_lock); |
1da177e4 | 345 | |
39fa0031 | 346 | /* now link into global list. */ |
abf137dd JA |
347 | spin_lock(&mm->ioctx_lock); |
348 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
349 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 | 350 | |
caf4167a | 351 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
41003a7b | 352 | ctx, ctx->user_id, mm, ctx->ring_info.nr); |
1da177e4 LT |
353 | return ctx; |
354 | ||
355 | out_cleanup: | |
e23754f8 AV |
356 | err = -EAGAIN; |
357 | aio_free_ring(ctx); | |
1da177e4 | 358 | out_freectx: |
1da177e4 | 359 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 360 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 361 | return ERR_PTR(err); |
1da177e4 LT |
362 | } |
363 | ||
06af121e | 364 | /* kill_ctx |
1da177e4 LT |
365 | * Cancels all outstanding aio requests on an aio context. Used |
366 | * when the processes owning a context have all exited to encourage | |
367 | * the rapid destruction of the kioctx. | |
368 | */ | |
06af121e | 369 | static void kill_ctx(struct kioctx *ctx) |
1da177e4 | 370 | { |
06af121e AV |
371 | struct task_struct *tsk = current; |
372 | DECLARE_WAITQUEUE(wait, tsk); | |
1da177e4 | 373 | struct io_event res; |
906b973c | 374 | struct kiocb *req; |
06af121e | 375 | |
1da177e4 LT |
376 | spin_lock_irq(&ctx->ctx_lock); |
377 | ctx->dead = 1; | |
378 | while (!list_empty(&ctx->active_reqs)) { | |
906b973c KO |
379 | req = list_first_entry(&ctx->active_reqs, |
380 | struct kiocb, ki_list); | |
381 | ||
382 | list_del_init(&req->ki_list); | |
383 | kiocb_cancel(ctx, req, &res); | |
1da177e4 | 384 | } |
1da177e4 | 385 | |
1da177e4 | 386 | if (!ctx->reqs_active) |
dee11c23 | 387 | goto out; |
1da177e4 LT |
388 | |
389 | add_wait_queue(&ctx->wait, &wait); | |
390 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
391 | while (ctx->reqs_active) { | |
dee11c23 | 392 | spin_unlock_irq(&ctx->ctx_lock); |
41d10da3 | 393 | io_schedule(); |
1da177e4 | 394 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
dee11c23 | 395 | spin_lock_irq(&ctx->ctx_lock); |
1da177e4 LT |
396 | } |
397 | __set_task_state(tsk, TASK_RUNNING); | |
398 | remove_wait_queue(&ctx->wait, &wait); | |
dee11c23 KC |
399 | |
400 | out: | |
401 | spin_unlock_irq(&ctx->ctx_lock); | |
1da177e4 LT |
402 | } |
403 | ||
404 | /* wait_on_sync_kiocb: | |
405 | * Waits on the given sync kiocb to complete. | |
406 | */ | |
fc9b52cd | 407 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
1da177e4 LT |
408 | { |
409 | while (iocb->ki_users) { | |
410 | set_current_state(TASK_UNINTERRUPTIBLE); | |
411 | if (!iocb->ki_users) | |
412 | break; | |
41d10da3 | 413 | io_schedule(); |
1da177e4 LT |
414 | } |
415 | __set_current_state(TASK_RUNNING); | |
416 | return iocb->ki_user_data; | |
417 | } | |
385773e0 | 418 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 LT |
419 | |
420 | /* exit_aio: called when the last user of mm goes away. At this point, | |
421 | * there is no way for any new requests to be submited or any of the | |
422 | * io_* syscalls to be called on the context. However, there may be | |
423 | * outstanding requests which hold references to the context; as they | |
424 | * go away, they will call put_ioctx and release any pinned memory | |
425 | * associated with the request (held via struct page * references). | |
426 | */ | |
fc9b52cd | 427 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 428 | { |
abf137dd JA |
429 | struct kioctx *ctx; |
430 | ||
431 | while (!hlist_empty(&mm->ioctx_list)) { | |
432 | ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); | |
433 | hlist_del_rcu(&ctx->list); | |
434 | ||
06af121e | 435 | kill_ctx(ctx); |
1da177e4 LT |
436 | |
437 | if (1 != atomic_read(&ctx->users)) | |
438 | printk(KERN_DEBUG | |
439 | "exit_aio:ioctx still alive: %d %d %d\n", | |
440 | atomic_read(&ctx->users), ctx->dead, | |
441 | ctx->reqs_active); | |
936af157 AV |
442 | /* |
443 | * We don't need to bother with munmap() here - | |
444 | * exit_mmap(mm) is coming and it'll unmap everything. | |
445 | * Since aio_free_ring() uses non-zero ->mmap_size | |
446 | * as indicator that it needs to unmap the area, | |
447 | * just set it to 0; aio_free_ring() is the only | |
448 | * place that uses ->mmap_size, so it's safe. | |
936af157 AV |
449 | */ |
450 | ctx->ring_info.mmap_size = 0; | |
1da177e4 | 451 | put_ioctx(ctx); |
1da177e4 LT |
452 | } |
453 | } | |
454 | ||
1da177e4 LT |
455 | /* aio_get_req |
456 | * Allocate a slot for an aio request. Increments the users count | |
457 | * of the kioctx so that the kioctx stays around until all requests are | |
458 | * complete. Returns NULL if no requests are free. | |
459 | * | |
460 | * Returns with kiocb->users set to 2. The io submit code path holds | |
461 | * an extra reference while submitting the i/o. | |
462 | * This prevents races between the aio code path referencing the | |
463 | * req (after submitting it) and aio_complete() freeing the req. | |
464 | */ | |
fc9b52cd | 465 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
1da177e4 LT |
466 | { |
467 | struct kiocb *req = NULL; | |
1da177e4 LT |
468 | |
469 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); | |
470 | if (unlikely(!req)) | |
471 | return NULL; | |
472 | ||
4faa5285 | 473 | req->ki_flags = 0; |
1da177e4 LT |
474 | req->ki_users = 2; |
475 | req->ki_key = 0; | |
476 | req->ki_ctx = ctx; | |
477 | req->ki_cancel = NULL; | |
478 | req->ki_retry = NULL; | |
1da177e4 LT |
479 | req->ki_dtor = NULL; |
480 | req->private = NULL; | |
eed4e51f | 481 | req->ki_iovec = NULL; |
87c3a86e | 482 | req->ki_eventfd = NULL; |
1da177e4 | 483 | |
080d676d JM |
484 | return req; |
485 | } | |
486 | ||
487 | /* | |
488 | * struct kiocb's are allocated in batches to reduce the number of | |
489 | * times the ctx lock is acquired and released. | |
490 | */ | |
491 | #define KIOCB_BATCH_SIZE 32L | |
492 | struct kiocb_batch { | |
493 | struct list_head head; | |
494 | long count; /* number of requests left to allocate */ | |
495 | }; | |
496 | ||
497 | static void kiocb_batch_init(struct kiocb_batch *batch, long total) | |
498 | { | |
499 | INIT_LIST_HEAD(&batch->head); | |
500 | batch->count = total; | |
501 | } | |
502 | ||
69e4747e | 503 | static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) |
080d676d JM |
504 | { |
505 | struct kiocb *req, *n; | |
506 | ||
69e4747e GN |
507 | if (list_empty(&batch->head)) |
508 | return; | |
509 | ||
510 | spin_lock_irq(&ctx->ctx_lock); | |
080d676d JM |
511 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
512 | list_del(&req->ki_batch); | |
69e4747e | 513 | list_del(&req->ki_list); |
080d676d | 514 | kmem_cache_free(kiocb_cachep, req); |
69e4747e | 515 | ctx->reqs_active--; |
080d676d | 516 | } |
880641bb JM |
517 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
518 | wake_up_all(&ctx->wait); | |
69e4747e | 519 | spin_unlock_irq(&ctx->ctx_lock); |
080d676d JM |
520 | } |
521 | ||
522 | /* | |
523 | * Allocate a batch of kiocbs. This avoids taking and dropping the | |
524 | * context lock a lot during setup. | |
525 | */ | |
526 | static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) | |
527 | { | |
528 | unsigned short allocated, to_alloc; | |
529 | long avail; | |
080d676d JM |
530 | struct kiocb *req, *n; |
531 | struct aio_ring *ring; | |
532 | ||
533 | to_alloc = min(batch->count, KIOCB_BATCH_SIZE); | |
534 | for (allocated = 0; allocated < to_alloc; allocated++) { | |
535 | req = __aio_get_req(ctx); | |
536 | if (!req) | |
537 | /* allocation failed, go with what we've got */ | |
538 | break; | |
539 | list_add(&req->ki_batch, &batch->head); | |
540 | } | |
541 | ||
542 | if (allocated == 0) | |
543 | goto out; | |
544 | ||
1da177e4 | 545 | spin_lock_irq(&ctx->ctx_lock); |
080d676d JM |
546 | ring = kmap_atomic(ctx->ring_info.ring_pages[0]); |
547 | ||
548 | avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active; | |
549 | BUG_ON(avail < 0); | |
080d676d JM |
550 | if (avail < allocated) { |
551 | /* Trim back the number of requests. */ | |
552 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { | |
553 | list_del(&req->ki_batch); | |
554 | kmem_cache_free(kiocb_cachep, req); | |
555 | if (--allocated <= avail) | |
556 | break; | |
557 | } | |
558 | } | |
559 | ||
560 | batch->count -= allocated; | |
561 | list_for_each_entry(req, &batch->head, ki_batch) { | |
1da177e4 | 562 | list_add(&req->ki_list, &ctx->active_reqs); |
1da177e4 | 563 | ctx->reqs_active++; |
1da177e4 | 564 | } |
1da177e4 | 565 | |
080d676d JM |
566 | kunmap_atomic(ring); |
567 | spin_unlock_irq(&ctx->ctx_lock); | |
1da177e4 | 568 | |
080d676d JM |
569 | out: |
570 | return allocated; | |
1da177e4 LT |
571 | } |
572 | ||
080d676d JM |
573 | static inline struct kiocb *aio_get_req(struct kioctx *ctx, |
574 | struct kiocb_batch *batch) | |
1da177e4 LT |
575 | { |
576 | struct kiocb *req; | |
080d676d JM |
577 | |
578 | if (list_empty(&batch->head)) | |
579 | if (kiocb_batch_refill(ctx, batch) == 0) | |
580 | return NULL; | |
581 | req = list_first_entry(&batch->head, struct kiocb, ki_batch); | |
582 | list_del(&req->ki_batch); | |
1da177e4 LT |
583 | return req; |
584 | } | |
585 | ||
586 | static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) | |
587 | { | |
d00689af ZB |
588 | assert_spin_locked(&ctx->ctx_lock); |
589 | ||
1d98ebfc KO |
590 | if (req->ki_filp) |
591 | fput(req->ki_filp); | |
13389010 DL |
592 | if (req->ki_eventfd != NULL) |
593 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 LT |
594 | if (req->ki_dtor) |
595 | req->ki_dtor(req); | |
eed4e51f BP |
596 | if (req->ki_iovec != &req->ki_inline_vec) |
597 | kfree(req->ki_iovec); | |
1da177e4 LT |
598 | kmem_cache_free(kiocb_cachep, req); |
599 | ctx->reqs_active--; | |
600 | ||
601 | if (unlikely(!ctx->reqs_active && ctx->dead)) | |
e91f90bb | 602 | wake_up_all(&ctx->wait); |
1da177e4 LT |
603 | } |
604 | ||
1da177e4 LT |
605 | /* __aio_put_req |
606 | * Returns true if this put was the last user of the request. | |
607 | */ | |
2d68449e | 608 | static void __aio_put_req(struct kioctx *ctx, struct kiocb *req) |
1da177e4 | 609 | { |
d00689af ZB |
610 | assert_spin_locked(&ctx->ctx_lock); |
611 | ||
87c3a86e | 612 | req->ki_users--; |
93e06b41 | 613 | BUG_ON(req->ki_users < 0); |
1da177e4 | 614 | if (likely(req->ki_users)) |
2d68449e | 615 | return; |
1da177e4 LT |
616 | list_del(&req->ki_list); /* remove from active_reqs */ |
617 | req->ki_cancel = NULL; | |
618 | req->ki_retry = NULL; | |
619 | ||
3ffa3c0e | 620 | really_put_req(ctx, req); |
1da177e4 LT |
621 | } |
622 | ||
623 | /* aio_put_req | |
624 | * Returns true if this put was the last user of the kiocb, | |
625 | * false if the request is still in use. | |
626 | */ | |
2d68449e | 627 | void aio_put_req(struct kiocb *req) |
1da177e4 LT |
628 | { |
629 | struct kioctx *ctx = req->ki_ctx; | |
1da177e4 | 630 | spin_lock_irq(&ctx->ctx_lock); |
2d68449e | 631 | __aio_put_req(ctx, req); |
1da177e4 | 632 | spin_unlock_irq(&ctx->ctx_lock); |
1da177e4 | 633 | } |
385773e0 | 634 | EXPORT_SYMBOL(aio_put_req); |
1da177e4 | 635 | |
d5470b59 | 636 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 637 | { |
abf137dd | 638 | struct mm_struct *mm = current->mm; |
65c24491 | 639 | struct kioctx *ctx, *ret = NULL; |
1da177e4 | 640 | |
abf137dd JA |
641 | rcu_read_lock(); |
642 | ||
b67bfe0d | 643 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
3bd9a5d7 NP |
644 | /* |
645 | * RCU protects us against accessing freed memory but | |
646 | * we have to be careful not to get a reference when the | |
647 | * reference count already dropped to 0 (ctx->dead test | |
648 | * is unreliable because of races). | |
649 | */ | |
650 | if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){ | |
65c24491 | 651 | ret = ctx; |
1da177e4 LT |
652 | break; |
653 | } | |
abf137dd | 654 | } |
1da177e4 | 655 | |
abf137dd | 656 | rcu_read_unlock(); |
65c24491 | 657 | return ret; |
1da177e4 LT |
658 | } |
659 | ||
1da177e4 LT |
660 | /* aio_complete |
661 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 662 | */ |
2d68449e | 663 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
664 | { |
665 | struct kioctx *ctx = iocb->ki_ctx; | |
666 | struct aio_ring_info *info; | |
667 | struct aio_ring *ring; | |
668 | struct io_event *event; | |
669 | unsigned long flags; | |
670 | unsigned long tail; | |
1da177e4 | 671 | |
20dcae32 ZB |
672 | /* |
673 | * Special case handling for sync iocbs: | |
674 | * - events go directly into the iocb for fast handling | |
675 | * - the sync task with the iocb in its stack holds the single iocb | |
676 | * ref, no other paths have a way to get another ref | |
677 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
678 | */ |
679 | if (is_sync_kiocb(iocb)) { | |
20dcae32 | 680 | BUG_ON(iocb->ki_users != 1); |
1da177e4 | 681 | iocb->ki_user_data = res; |
20dcae32 | 682 | iocb->ki_users = 0; |
1da177e4 | 683 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 684 | return; |
1da177e4 LT |
685 | } |
686 | ||
687 | info = &ctx->ring_info; | |
688 | ||
689 | /* add a completion event to the ring buffer. | |
690 | * must be done holding ctx->ctx_lock to prevent | |
691 | * other code from messing with the tail | |
692 | * pointer since we might be called from irq | |
693 | * context. | |
694 | */ | |
695 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
696 | ||
1da177e4 LT |
697 | /* |
698 | * cancelled requests don't get events, userland was given one | |
699 | * when the event got cancelled. | |
700 | */ | |
701 | if (kiocbIsCancelled(iocb)) | |
702 | goto put_rq; | |
703 | ||
e8e3c3d6 | 704 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
705 | |
706 | tail = info->tail; | |
e8e3c3d6 | 707 | event = aio_ring_event(info, tail); |
4bf69b2a KC |
708 | if (++tail >= info->nr) |
709 | tail = 0; | |
1da177e4 LT |
710 | |
711 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; | |
712 | event->data = iocb->ki_user_data; | |
713 | event->res = res; | |
714 | event->res2 = res2; | |
715 | ||
caf4167a KO |
716 | pr_debug("%p[%lu]: %p: %p %Lx %lx %lx\n", |
717 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, | |
718 | res, res2); | |
1da177e4 LT |
719 | |
720 | /* after flagging the request as done, we | |
721 | * must never even look at it again | |
722 | */ | |
723 | smp_wmb(); /* make event visible before updating tail */ | |
724 | ||
725 | info->tail = tail; | |
726 | ring->tail = tail; | |
727 | ||
e8e3c3d6 CW |
728 | put_aio_ring_event(event); |
729 | kunmap_atomic(ring); | |
1da177e4 LT |
730 | |
731 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); | |
8d1c98b0 DL |
732 | |
733 | /* | |
734 | * Check if the user asked us to deliver the result through an | |
735 | * eventfd. The eventfd_signal() function is safe to be called | |
736 | * from IRQ context. | |
737 | */ | |
87c3a86e | 738 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
739 | eventfd_signal(iocb->ki_eventfd, 1); |
740 | ||
1da177e4 LT |
741 | put_rq: |
742 | /* everything turned out well, dispose of the aiocb. */ | |
2d68449e | 743 | __aio_put_req(ctx, iocb); |
1da177e4 | 744 | |
6cb2a210 QB |
745 | /* |
746 | * We have to order our ring_info tail store above and test | |
747 | * of the wait list below outside the wait lock. This is | |
748 | * like in wake_up_bit() where clearing a bit has to be | |
749 | * ordered with the unlocked test. | |
750 | */ | |
751 | smp_mb(); | |
752 | ||
1da177e4 LT |
753 | if (waitqueue_active(&ctx->wait)) |
754 | wake_up(&ctx->wait); | |
755 | ||
dee11c23 | 756 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
1da177e4 | 757 | } |
385773e0 | 758 | EXPORT_SYMBOL(aio_complete); |
1da177e4 LT |
759 | |
760 | /* aio_read_evt | |
761 | * Pull an event off of the ioctx's event ring. Returns the number of | |
762 | * events fetched (0 or 1 ;-) | |
763 | * FIXME: make this use cmpxchg. | |
764 | * TODO: make the ringbuffer user mmap()able (requires FIXME). | |
765 | */ | |
766 | static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) | |
767 | { | |
768 | struct aio_ring_info *info = &ioctx->ring_info; | |
769 | struct aio_ring *ring; | |
770 | unsigned long head; | |
771 | int ret = 0; | |
772 | ||
e8e3c3d6 | 773 | ring = kmap_atomic(info->ring_pages[0]); |
caf4167a | 774 | pr_debug("h%u t%u m%u\n", ring->head, ring->tail, ring->nr); |
1da177e4 LT |
775 | |
776 | if (ring->head == ring->tail) | |
777 | goto out; | |
778 | ||
779 | spin_lock(&info->ring_lock); | |
780 | ||
781 | head = ring->head % info->nr; | |
782 | if (head != ring->tail) { | |
e8e3c3d6 | 783 | struct io_event *evp = aio_ring_event(info, head); |
1da177e4 LT |
784 | *ent = *evp; |
785 | head = (head + 1) % info->nr; | |
786 | smp_mb(); /* finish reading the event before updatng the head */ | |
787 | ring->head = head; | |
788 | ret = 1; | |
e8e3c3d6 | 789 | put_aio_ring_event(evp); |
1da177e4 LT |
790 | } |
791 | spin_unlock(&info->ring_lock); | |
792 | ||
793 | out: | |
91d80a84 | 794 | kunmap_atomic(ring); |
caf4167a | 795 | pr_debug("%d h%u t%u\n", ret, ring->head, ring->tail); |
1da177e4 LT |
796 | return ret; |
797 | } | |
798 | ||
799 | struct aio_timeout { | |
800 | struct timer_list timer; | |
801 | int timed_out; | |
802 | struct task_struct *p; | |
803 | }; | |
804 | ||
805 | static void timeout_func(unsigned long data) | |
806 | { | |
807 | struct aio_timeout *to = (struct aio_timeout *)data; | |
808 | ||
809 | to->timed_out = 1; | |
810 | wake_up_process(to->p); | |
811 | } | |
812 | ||
813 | static inline void init_timeout(struct aio_timeout *to) | |
814 | { | |
c6f3a97f | 815 | setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to); |
1da177e4 LT |
816 | to->timed_out = 0; |
817 | to->p = current; | |
818 | } | |
819 | ||
820 | static inline void set_timeout(long start_jiffies, struct aio_timeout *to, | |
821 | const struct timespec *ts) | |
822 | { | |
823 | to->timer.expires = start_jiffies + timespec_to_jiffies(ts); | |
824 | if (time_after(to->timer.expires, jiffies)) | |
825 | add_timer(&to->timer); | |
826 | else | |
827 | to->timed_out = 1; | |
828 | } | |
829 | ||
830 | static inline void clear_timeout(struct aio_timeout *to) | |
831 | { | |
832 | del_singleshot_timer_sync(&to->timer); | |
833 | } | |
834 | ||
835 | static int read_events(struct kioctx *ctx, | |
836 | long min_nr, long nr, | |
837 | struct io_event __user *event, | |
838 | struct timespec __user *timeout) | |
839 | { | |
840 | long start_jiffies = jiffies; | |
841 | struct task_struct *tsk = current; | |
842 | DECLARE_WAITQUEUE(wait, tsk); | |
843 | int ret; | |
844 | int i = 0; | |
845 | struct io_event ent; | |
846 | struct aio_timeout to; | |
1da177e4 LT |
847 | |
848 | /* needed to zero any padding within an entry (there shouldn't be | |
849 | * any, but C is fun! | |
850 | */ | |
851 | memset(&ent, 0, sizeof(ent)); | |
1da177e4 LT |
852 | ret = 0; |
853 | while (likely(i < nr)) { | |
854 | ret = aio_read_evt(ctx, &ent); | |
855 | if (unlikely(ret <= 0)) | |
856 | break; | |
857 | ||
caf4167a KO |
858 | pr_debug("%Lx %Lx %Lx %Lx\n", |
859 | ent.data, ent.obj, ent.res, ent.res2); | |
1da177e4 LT |
860 | |
861 | /* Could we split the check in two? */ | |
862 | ret = -EFAULT; | |
863 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { | |
caf4167a | 864 | pr_debug("lost an event due to EFAULT.\n"); |
1da177e4 LT |
865 | break; |
866 | } | |
867 | ret = 0; | |
868 | ||
869 | /* Good, event copied to userland, update counts. */ | |
870 | event ++; | |
871 | i ++; | |
872 | } | |
873 | ||
874 | if (min_nr <= i) | |
875 | return i; | |
876 | if (ret) | |
877 | return ret; | |
878 | ||
879 | /* End fast path */ | |
880 | ||
1da177e4 LT |
881 | init_timeout(&to); |
882 | if (timeout) { | |
883 | struct timespec ts; | |
884 | ret = -EFAULT; | |
885 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) | |
886 | goto out; | |
887 | ||
888 | set_timeout(start_jiffies, &to, &ts); | |
889 | } | |
890 | ||
891 | while (likely(i < nr)) { | |
892 | add_wait_queue_exclusive(&ctx->wait, &wait); | |
893 | do { | |
894 | set_task_state(tsk, TASK_INTERRUPTIBLE); | |
895 | ret = aio_read_evt(ctx, &ent); | |
896 | if (ret) | |
897 | break; | |
898 | if (min_nr <= i) | |
899 | break; | |
e92adcba JM |
900 | if (unlikely(ctx->dead)) { |
901 | ret = -EINVAL; | |
902 | break; | |
903 | } | |
1da177e4 LT |
904 | if (to.timed_out) /* Only check after read evt */ |
905 | break; | |
e00ba3da JM |
906 | /* Try to only show up in io wait if there are ops |
907 | * in flight */ | |
908 | if (ctx->reqs_active) | |
909 | io_schedule(); | |
910 | else | |
911 | schedule(); | |
1da177e4 LT |
912 | if (signal_pending(tsk)) { |
913 | ret = -EINTR; | |
914 | break; | |
915 | } | |
916 | /*ret = aio_read_evt(ctx, &ent);*/ | |
917 | } while (1) ; | |
918 | ||
919 | set_task_state(tsk, TASK_RUNNING); | |
920 | remove_wait_queue(&ctx->wait, &wait); | |
921 | ||
922 | if (unlikely(ret <= 0)) | |
923 | break; | |
924 | ||
925 | ret = -EFAULT; | |
926 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { | |
caf4167a | 927 | pr_debug("lost an event due to EFAULT.\n"); |
1da177e4 LT |
928 | break; |
929 | } | |
930 | ||
931 | /* Good, event copied to userland, update counts. */ | |
932 | event ++; | |
933 | i ++; | |
934 | } | |
935 | ||
936 | if (timeout) | |
937 | clear_timeout(&to); | |
938 | out: | |
c6f3a97f | 939 | destroy_timer_on_stack(&to.timer); |
1da177e4 LT |
940 | return i ? i : ret; |
941 | } | |
942 | ||
943 | /* Take an ioctx and remove it from the list of ioctx's. Protects | |
944 | * against races with itself via ->dead. | |
945 | */ | |
946 | static void io_destroy(struct kioctx *ioctx) | |
947 | { | |
948 | struct mm_struct *mm = current->mm; | |
1da177e4 LT |
949 | int was_dead; |
950 | ||
951 | /* delete the entry from the list is someone else hasn't already */ | |
abf137dd | 952 | spin_lock(&mm->ioctx_lock); |
1da177e4 LT |
953 | was_dead = ioctx->dead; |
954 | ioctx->dead = 1; | |
abf137dd JA |
955 | hlist_del_rcu(&ioctx->list); |
956 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 | 957 | |
caf4167a | 958 | pr_debug("(%p)\n", ioctx); |
1da177e4 LT |
959 | if (likely(!was_dead)) |
960 | put_ioctx(ioctx); /* twice for the list */ | |
961 | ||
06af121e | 962 | kill_ctx(ioctx); |
e92adcba JM |
963 | |
964 | /* | |
965 | * Wake up any waiters. The setting of ctx->dead must be seen | |
966 | * by other CPUs at this point. Right now, we rely on the | |
967 | * locking done by the above calls to ensure this consistency. | |
968 | */ | |
e91f90bb | 969 | wake_up_all(&ioctx->wait); |
1da177e4 LT |
970 | } |
971 | ||
972 | /* sys_io_setup: | |
973 | * Create an aio_context capable of receiving at least nr_events. | |
974 | * ctxp must not point to an aio_context that already exists, and | |
975 | * must be initialized to 0 prior to the call. On successful | |
976 | * creation of the aio_context, *ctxp is filled in with the resulting | |
977 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
978 | * if the specified nr_events exceeds internal limits. May fail | |
979 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
980 | * of available events. May fail with -ENOMEM if insufficient kernel | |
981 | * resources are available. May fail with -EFAULT if an invalid | |
982 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
983 | * implemented. | |
984 | */ | |
002c8976 | 985 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
986 | { |
987 | struct kioctx *ioctx = NULL; | |
988 | unsigned long ctx; | |
989 | long ret; | |
990 | ||
991 | ret = get_user(ctx, ctxp); | |
992 | if (unlikely(ret)) | |
993 | goto out; | |
994 | ||
995 | ret = -EINVAL; | |
d55b5fda ZB |
996 | if (unlikely(ctx || nr_events == 0)) { |
997 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
998 | ctx, nr_events); | |
1da177e4 LT |
999 | goto out; |
1000 | } | |
1001 | ||
1002 | ioctx = ioctx_alloc(nr_events); | |
1003 | ret = PTR_ERR(ioctx); | |
1004 | if (!IS_ERR(ioctx)) { | |
1005 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a AV |
1006 | if (ret) |
1007 | io_destroy(ioctx); | |
1008 | put_ioctx(ioctx); | |
1da177e4 LT |
1009 | } |
1010 | ||
1011 | out: | |
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | /* sys_io_destroy: | |
1016 | * Destroy the aio_context specified. May cancel any outstanding | |
1017 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1018 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1019 | * is invalid. |
1020 | */ | |
002c8976 | 1021 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1022 | { |
1023 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1024 | if (likely(NULL != ioctx)) { | |
1025 | io_destroy(ioctx); | |
a2e1859a | 1026 | put_ioctx(ioctx); |
1da177e4 LT |
1027 | return 0; |
1028 | } | |
1029 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1030 | return -EINVAL; | |
1031 | } | |
1032 | ||
eed4e51f | 1033 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1da177e4 | 1034 | { |
eed4e51f BP |
1035 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
1036 | ||
1037 | BUG_ON(ret <= 0); | |
1038 | ||
1039 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { | |
1040 | ssize_t this = min((ssize_t)iov->iov_len, ret); | |
1041 | iov->iov_base += this; | |
1042 | iov->iov_len -= this; | |
1043 | iocb->ki_left -= this; | |
1044 | ret -= this; | |
1045 | if (iov->iov_len == 0) { | |
1046 | iocb->ki_cur_seg++; | |
1047 | iov++; | |
897f15fb | 1048 | } |
eed4e51f | 1049 | } |
1da177e4 | 1050 | |
eed4e51f BP |
1051 | /* the caller should not have done more io than what fit in |
1052 | * the remaining iovecs */ | |
1053 | BUG_ON(ret > 0 && iocb->ki_left == 0); | |
1da177e4 LT |
1054 | } |
1055 | ||
eed4e51f | 1056 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
1da177e4 LT |
1057 | { |
1058 | struct file *file = iocb->ki_filp; | |
eed4e51f BP |
1059 | struct address_space *mapping = file->f_mapping; |
1060 | struct inode *inode = mapping->host; | |
1061 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, | |
1062 | unsigned long, loff_t); | |
1da177e4 | 1063 | ssize_t ret = 0; |
eed4e51f BP |
1064 | unsigned short opcode; |
1065 | ||
1066 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || | |
1067 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { | |
1068 | rw_op = file->f_op->aio_read; | |
1069 | opcode = IOCB_CMD_PREADV; | |
1070 | } else { | |
1071 | rw_op = file->f_op->aio_write; | |
1072 | opcode = IOCB_CMD_PWRITEV; | |
1073 | } | |
1da177e4 | 1074 | |
c2ec6682 RR |
1075 | /* This matches the pread()/pwrite() logic */ |
1076 | if (iocb->ki_pos < 0) | |
1077 | return -EINVAL; | |
1078 | ||
8d71db4f AV |
1079 | if (opcode == IOCB_CMD_PWRITEV) |
1080 | file_start_write(file); | |
897f15fb | 1081 | do { |
eed4e51f BP |
1082 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
1083 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
1084 | iocb->ki_pos); | |
1085 | if (ret > 0) | |
1086 | aio_advance_iovec(iocb, ret); | |
1087 | ||
1088 | /* retry all partial writes. retry partial reads as long as its a | |
1089 | * regular file. */ | |
1090 | } while (ret > 0 && iocb->ki_left > 0 && | |
1091 | (opcode == IOCB_CMD_PWRITEV || | |
1092 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); | |
8d71db4f AV |
1093 | if (opcode == IOCB_CMD_PWRITEV) |
1094 | file_end_write(file); | |
1da177e4 | 1095 | |
eed4e51f BP |
1096 | /* This means we must have transferred all that we could */ |
1097 | /* No need to retry anymore */ | |
1da177e4 LT |
1098 | if ((ret == 0) || (iocb->ki_left == 0)) |
1099 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1100 | ||
7adfa2ff RR |
1101 | /* If we managed to write some out we return that, rather than |
1102 | * the eventual error. */ | |
1103 | if (opcode == IOCB_CMD_PWRITEV | |
41003a7b | 1104 | && ret < 0 && ret != -EIOCBQUEUED |
7adfa2ff RR |
1105 | && iocb->ki_nbytes - iocb->ki_left) |
1106 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1107 | ||
1da177e4 LT |
1108 | return ret; |
1109 | } | |
1110 | ||
1111 | static ssize_t aio_fdsync(struct kiocb *iocb) | |
1112 | { | |
1113 | struct file *file = iocb->ki_filp; | |
1114 | ssize_t ret = -EINVAL; | |
1115 | ||
1116 | if (file->f_op->aio_fsync) | |
1117 | ret = file->f_op->aio_fsync(iocb, 1); | |
1118 | return ret; | |
1119 | } | |
1120 | ||
1121 | static ssize_t aio_fsync(struct kiocb *iocb) | |
1122 | { | |
1123 | struct file *file = iocb->ki_filp; | |
1124 | ssize_t ret = -EINVAL; | |
1125 | ||
1126 | if (file->f_op->aio_fsync) | |
1127 | ret = file->f_op->aio_fsync(iocb, 0); | |
1128 | return ret; | |
1129 | } | |
1130 | ||
9d85cba7 | 1131 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
eed4e51f BP |
1132 | { |
1133 | ssize_t ret; | |
1134 | ||
9d85cba7 JM |
1135 | #ifdef CONFIG_COMPAT |
1136 | if (compat) | |
1137 | ret = compat_rw_copy_check_uvector(type, | |
1138 | (struct compat_iovec __user *)kiocb->ki_buf, | |
1139 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1140 | &kiocb->ki_iovec); |
9d85cba7 JM |
1141 | else |
1142 | #endif | |
1143 | ret = rw_copy_check_uvector(type, | |
1144 | (struct iovec __user *)kiocb->ki_buf, | |
1145 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1146 | &kiocb->ki_iovec); |
eed4e51f BP |
1147 | if (ret < 0) |
1148 | goto out; | |
1149 | ||
a70b52ec LT |
1150 | ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret); |
1151 | if (ret < 0) | |
1152 | goto out; | |
1153 | ||
eed4e51f BP |
1154 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1155 | kiocb->ki_cur_seg = 0; | |
1156 | /* ki_nbytes/left now reflect bytes instead of segs */ | |
1157 | kiocb->ki_nbytes = ret; | |
1158 | kiocb->ki_left = ret; | |
1159 | ||
1160 | ret = 0; | |
1161 | out: | |
1162 | return ret; | |
1163 | } | |
1164 | ||
a70b52ec | 1165 | static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb) |
eed4e51f | 1166 | { |
a70b52ec LT |
1167 | int bytes; |
1168 | ||
1169 | bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left); | |
1170 | if (bytes < 0) | |
1171 | return bytes; | |
1172 | ||
eed4e51f BP |
1173 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1174 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
a70b52ec | 1175 | kiocb->ki_iovec->iov_len = bytes; |
eed4e51f BP |
1176 | kiocb->ki_nr_segs = 1; |
1177 | kiocb->ki_cur_seg = 0; | |
eed4e51f BP |
1178 | return 0; |
1179 | } | |
1180 | ||
1da177e4 LT |
1181 | /* |
1182 | * aio_setup_iocb: | |
1183 | * Performs the initial checks and aio retry method | |
1184 | * setup for the kiocb at the time of io submission. | |
1185 | */ | |
9d85cba7 | 1186 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
1da177e4 LT |
1187 | { |
1188 | struct file *file = kiocb->ki_filp; | |
1189 | ssize_t ret = 0; | |
1190 | ||
1191 | switch (kiocb->ki_opcode) { | |
1192 | case IOCB_CMD_PREAD: | |
1193 | ret = -EBADF; | |
1194 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1195 | break; | |
1196 | ret = -EFAULT; | |
1197 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, | |
1198 | kiocb->ki_left))) | |
1199 | break; | |
a70b52ec | 1200 | ret = aio_setup_single_vector(READ, file, kiocb); |
eed4e51f BP |
1201 | if (ret) |
1202 | break; | |
1da177e4 LT |
1203 | ret = -EINVAL; |
1204 | if (file->f_op->aio_read) | |
eed4e51f | 1205 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1206 | break; |
1207 | case IOCB_CMD_PWRITE: | |
1208 | ret = -EBADF; | |
1209 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1210 | break; | |
1211 | ret = -EFAULT; | |
1212 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, | |
1213 | kiocb->ki_left))) | |
1214 | break; | |
a70b52ec | 1215 | ret = aio_setup_single_vector(WRITE, file, kiocb); |
eed4e51f BP |
1216 | if (ret) |
1217 | break; | |
1218 | ret = -EINVAL; | |
1219 | if (file->f_op->aio_write) | |
1220 | kiocb->ki_retry = aio_rw_vect_retry; | |
1221 | break; | |
1222 | case IOCB_CMD_PREADV: | |
1223 | ret = -EBADF; | |
1224 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1225 | break; | |
9d85cba7 | 1226 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
eed4e51f BP |
1227 | if (ret) |
1228 | break; | |
1229 | ret = -EINVAL; | |
1230 | if (file->f_op->aio_read) | |
1231 | kiocb->ki_retry = aio_rw_vect_retry; | |
1232 | break; | |
1233 | case IOCB_CMD_PWRITEV: | |
1234 | ret = -EBADF; | |
1235 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1236 | break; | |
9d85cba7 | 1237 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
eed4e51f BP |
1238 | if (ret) |
1239 | break; | |
1da177e4 LT |
1240 | ret = -EINVAL; |
1241 | if (file->f_op->aio_write) | |
eed4e51f | 1242 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1243 | break; |
1244 | case IOCB_CMD_FDSYNC: | |
1245 | ret = -EINVAL; | |
1246 | if (file->f_op->aio_fsync) | |
1247 | kiocb->ki_retry = aio_fdsync; | |
1248 | break; | |
1249 | case IOCB_CMD_FSYNC: | |
1250 | ret = -EINVAL; | |
1251 | if (file->f_op->aio_fsync) | |
1252 | kiocb->ki_retry = aio_fsync; | |
1253 | break; | |
1254 | default: | |
caf4167a | 1255 | pr_debug("EINVAL: no operation provided\n"); |
1da177e4 LT |
1256 | ret = -EINVAL; |
1257 | } | |
1258 | ||
1259 | if (!kiocb->ki_retry) | |
1260 | return ret; | |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | ||
d5470b59 | 1265 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
080d676d JM |
1266 | struct iocb *iocb, struct kiocb_batch *batch, |
1267 | bool compat) | |
1da177e4 LT |
1268 | { |
1269 | struct kiocb *req; | |
1da177e4 LT |
1270 | ssize_t ret; |
1271 | ||
1272 | /* enforce forwards compatibility on users */ | |
9c3060be | 1273 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1274 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1275 | return -EINVAL; |
1276 | } | |
1277 | ||
1278 | /* prevent overflows */ | |
1279 | if (unlikely( | |
1280 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1281 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1282 | ((ssize_t)iocb->aio_nbytes < 0) | |
1283 | )) { | |
1284 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1285 | return -EINVAL; | |
1286 | } | |
1287 | ||
080d676d | 1288 | req = aio_get_req(ctx, batch); /* returns with 2 references to req */ |
1d98ebfc | 1289 | if (unlikely(!req)) |
1da177e4 | 1290 | return -EAGAIN; |
1d98ebfc KO |
1291 | |
1292 | req->ki_filp = fget(iocb->aio_fildes); | |
1293 | if (unlikely(!req->ki_filp)) { | |
1294 | ret = -EBADF; | |
1295 | goto out_put_req; | |
1da177e4 | 1296 | } |
1d98ebfc | 1297 | |
9c3060be DL |
1298 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1299 | /* | |
1300 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1301 | * instance of the file* now. The file descriptor must be | |
1302 | * an eventfd() fd, and will be signaled for each completed | |
1303 | * event using the eventfd_signal() function. | |
1304 | */ | |
13389010 | 1305 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1306 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1307 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1308 | req->ki_eventfd = NULL; |
9c3060be DL |
1309 | goto out_put_req; |
1310 | } | |
1311 | } | |
1da177e4 | 1312 | |
212079cf | 1313 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
1da177e4 | 1314 | if (unlikely(ret)) { |
caf4167a | 1315 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1316 | goto out_put_req; |
1317 | } | |
1318 | ||
1319 | req->ki_obj.user = user_iocb; | |
1320 | req->ki_user_data = iocb->aio_data; | |
1321 | req->ki_pos = iocb->aio_offset; | |
1322 | ||
1323 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1324 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1325 | req->ki_opcode = iocb->aio_lio_opcode; | |
1da177e4 | 1326 | |
9d85cba7 | 1327 | ret = aio_setup_iocb(req, compat); |
1da177e4 LT |
1328 | |
1329 | if (ret) | |
1330 | goto out_put_req; | |
1331 | ||
1332 | spin_lock_irq(&ctx->ctx_lock); | |
7137c6bd JK |
1333 | /* |
1334 | * We could have raced with io_destroy() and are currently holding a | |
1335 | * reference to ctx which should be destroyed. We cannot submit IO | |
1336 | * since ctx gets freed as soon as io_submit() puts its reference. The | |
1337 | * check here is reliable: io_destroy() sets ctx->dead before waiting | |
1338 | * for outstanding IO and the barrier between these two is realized by | |
1339 | * unlock of mm->ioctx_lock and lock of ctx->ctx_lock. Analogously we | |
1340 | * increment ctx->reqs_active before checking for ctx->dead and the | |
1341 | * barrier is realized by unlock and lock of ctx->ctx_lock. Thus if we | |
1342 | * don't see ctx->dead set here, io_destroy() waits for our IO to | |
1343 | * finish. | |
1344 | */ | |
41003a7b | 1345 | if (ctx->dead) |
7137c6bd | 1346 | ret = -EINVAL; |
41003a7b ZB |
1347 | spin_unlock_irq(&ctx->ctx_lock); |
1348 | if (ret) | |
7137c6bd | 1349 | goto out_put_req; |
41003a7b ZB |
1350 | |
1351 | if (unlikely(kiocbIsCancelled(req))) | |
1352 | ret = -EINTR; | |
1353 | else | |
1354 | ret = req->ki_retry(req); | |
1355 | ||
1356 | if (ret != -EIOCBQUEUED) { | |
1357 | /* | |
1358 | * There's no easy way to restart the syscall since other AIO's | |
1359 | * may be already running. Just fail this IO with EINTR. | |
1360 | */ | |
1361 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1362 | ret == -ERESTARTNOHAND || | |
1363 | ret == -ERESTART_RESTARTBLOCK)) | |
1364 | ret = -EINTR; | |
1365 | aio_complete(req, ret, 0); | |
7137c6bd | 1366 | } |
cfb1e33e | 1367 | |
1da177e4 LT |
1368 | aio_put_req(req); /* drop extra ref to req */ |
1369 | return 0; | |
1370 | ||
1371 | out_put_req: | |
1372 | aio_put_req(req); /* drop extra ref to req */ | |
1373 | aio_put_req(req); /* drop i/o ref to req */ | |
1374 | return ret; | |
1375 | } | |
1376 | ||
9d85cba7 JM |
1377 | long do_io_submit(aio_context_t ctx_id, long nr, |
1378 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1379 | { |
1380 | struct kioctx *ctx; | |
1381 | long ret = 0; | |
080d676d | 1382 | int i = 0; |
9f5b9425 | 1383 | struct blk_plug plug; |
080d676d | 1384 | struct kiocb_batch batch; |
1da177e4 LT |
1385 | |
1386 | if (unlikely(nr < 0)) | |
1387 | return -EINVAL; | |
1388 | ||
75e1c70f JM |
1389 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1390 | nr = LONG_MAX/sizeof(*iocbpp); | |
1391 | ||
1da177e4 LT |
1392 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1393 | return -EFAULT; | |
1394 | ||
1395 | ctx = lookup_ioctx(ctx_id); | |
1396 | if (unlikely(!ctx)) { | |
caf4167a | 1397 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1398 | return -EINVAL; |
1399 | } | |
1400 | ||
080d676d JM |
1401 | kiocb_batch_init(&batch, nr); |
1402 | ||
9f5b9425 SL |
1403 | blk_start_plug(&plug); |
1404 | ||
1da177e4 LT |
1405 | /* |
1406 | * AKPM: should this return a partial result if some of the IOs were | |
1407 | * successfully submitted? | |
1408 | */ | |
1409 | for (i=0; i<nr; i++) { | |
1410 | struct iocb __user *user_iocb; | |
1411 | struct iocb tmp; | |
1412 | ||
1413 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1414 | ret = -EFAULT; | |
1415 | break; | |
1416 | } | |
1417 | ||
1418 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1419 | ret = -EFAULT; | |
1420 | break; | |
1421 | } | |
1422 | ||
080d676d | 1423 | ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat); |
1da177e4 LT |
1424 | if (ret) |
1425 | break; | |
1426 | } | |
9f5b9425 | 1427 | blk_finish_plug(&plug); |
1da177e4 | 1428 | |
69e4747e | 1429 | kiocb_batch_free(ctx, &batch); |
1da177e4 LT |
1430 | put_ioctx(ctx); |
1431 | return i ? i : ret; | |
1432 | } | |
1433 | ||
9d85cba7 JM |
1434 | /* sys_io_submit: |
1435 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1436 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1437 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1438 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1439 | * is invalid for the file descriptor in the iocb. May fail with | |
1440 | * -EFAULT if any of the data structures point to invalid data. May | |
1441 | * fail with -EBADF if the file descriptor specified in the first | |
1442 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1443 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1444 | * fail with -ENOSYS if not implemented. | |
1445 | */ | |
1446 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1447 | struct iocb __user * __user *, iocbpp) | |
1448 | { | |
1449 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1450 | } | |
1451 | ||
1da177e4 LT |
1452 | /* lookup_kiocb |
1453 | * Finds a given iocb for cancellation. | |
1da177e4 | 1454 | */ |
25ee7e38 AB |
1455 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1456 | u32 key) | |
1da177e4 LT |
1457 | { |
1458 | struct list_head *pos; | |
d00689af ZB |
1459 | |
1460 | assert_spin_locked(&ctx->ctx_lock); | |
1461 | ||
1da177e4 LT |
1462 | /* TODO: use a hash or array, this sucks. */ |
1463 | list_for_each(pos, &ctx->active_reqs) { | |
1464 | struct kiocb *kiocb = list_kiocb(pos); | |
1465 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) | |
1466 | return kiocb; | |
1467 | } | |
1468 | return NULL; | |
1469 | } | |
1470 | ||
1471 | /* sys_io_cancel: | |
1472 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1473 | * the operation is successfully cancelled, the resulting event is | |
1474 | * copied into the memory pointed to by result without being placed | |
1475 | * into the completion queue and 0 is returned. May fail with | |
1476 | * -EFAULT if any of the data structures pointed to are invalid. | |
1477 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1478 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1479 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1480 | */ | |
002c8976 HC |
1481 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1482 | struct io_event __user *, result) | |
1da177e4 | 1483 | { |
906b973c | 1484 | struct io_event res; |
1da177e4 LT |
1485 | struct kioctx *ctx; |
1486 | struct kiocb *kiocb; | |
1487 | u32 key; | |
1488 | int ret; | |
1489 | ||
1490 | ret = get_user(key, &iocb->aio_key); | |
1491 | if (unlikely(ret)) | |
1492 | return -EFAULT; | |
1493 | ||
1494 | ctx = lookup_ioctx(ctx_id); | |
1495 | if (unlikely(!ctx)) | |
1496 | return -EINVAL; | |
1497 | ||
1498 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1499 | |
1da177e4 | 1500 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c KO |
1501 | if (kiocb) |
1502 | ret = kiocb_cancel(ctx, kiocb, &res); | |
1503 | else | |
1504 | ret = -EINVAL; | |
1505 | ||
1da177e4 LT |
1506 | spin_unlock_irq(&ctx->ctx_lock); |
1507 | ||
906b973c KO |
1508 | if (!ret) { |
1509 | /* Cancellation succeeded -- copy the result | |
1510 | * into the user's buffer. | |
1511 | */ | |
1512 | if (copy_to_user(result, &res, sizeof(res))) | |
1513 | ret = -EFAULT; | |
1514 | } | |
1da177e4 LT |
1515 | |
1516 | put_ioctx(ctx); | |
1517 | ||
1518 | return ret; | |
1519 | } | |
1520 | ||
1521 | /* io_getevents: | |
1522 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1523 | * the completion queue for the aio_context specified by ctx_id. If |
1524 | * it succeeds, the number of read events is returned. May fail with | |
1525 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1526 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1527 | * if any of the memory specified is invalid. May return 0 or | |
1528 | * < min_nr if the timeout specified by timeout has elapsed | |
1529 | * before sufficient events are available, where timeout == NULL | |
1530 | * specifies an infinite timeout. Note that the timeout pointed to by | |
1531 | * timeout is relative and will be updated if not NULL and the | |
1532 | * operation blocks. Will fail with -ENOSYS if not implemented. | |
1da177e4 | 1533 | */ |
002c8976 HC |
1534 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1535 | long, min_nr, | |
1536 | long, nr, | |
1537 | struct io_event __user *, events, | |
1538 | struct timespec __user *, timeout) | |
1da177e4 LT |
1539 | { |
1540 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1541 | long ret = -EINVAL; | |
1542 | ||
1543 | if (likely(ioctx)) { | |
2e410255 | 1544 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 LT |
1545 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1546 | put_ioctx(ioctx); | |
1547 | } | |
1da177e4 LT |
1548 | return ret; |
1549 | } |