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 | |
4e179bca | 61 | |
4e179bca KO |
62 | struct kioctx { |
63 | atomic_t users; | |
36f55889 | 64 | atomic_t dead; |
4e179bca KO |
65 | |
66 | /* This needs improving */ | |
67 | unsigned long user_id; | |
68 | struct hlist_node list; | |
69 | ||
3e845ce0 KO |
70 | /* |
71 | * This is what userspace passed to io_setup(), it's not used for | |
72 | * anything but counting against the global max_reqs quota. | |
73 | * | |
58c85dc2 | 74 | * The real limit is nr_events - 1, which will be larger (see |
3e845ce0 KO |
75 | * aio_setup_ring()) |
76 | */ | |
4e179bca KO |
77 | unsigned max_reqs; |
78 | ||
58c85dc2 KO |
79 | /* Size of ringbuffer, in units of struct io_event */ |
80 | unsigned nr_events; | |
4e179bca | 81 | |
58c85dc2 KO |
82 | unsigned long mmap_base; |
83 | unsigned long mmap_size; | |
84 | ||
85 | struct page **ring_pages; | |
86 | long nr_pages; | |
87 | ||
4e23bcae KO |
88 | struct rcu_head rcu_head; |
89 | struct work_struct rcu_work; | |
90 | ||
91 | struct { | |
92 | atomic_t reqs_active; | |
93 | } ____cacheline_aligned_in_smp; | |
94 | ||
95 | struct { | |
96 | spinlock_t ctx_lock; | |
97 | struct list_head active_reqs; /* used for cancellation */ | |
98 | } ____cacheline_aligned_in_smp; | |
99 | ||
58c85dc2 KO |
100 | struct { |
101 | struct mutex ring_lock; | |
4e23bcae KO |
102 | wait_queue_head_t wait; |
103 | } ____cacheline_aligned_in_smp; | |
58c85dc2 KO |
104 | |
105 | struct { | |
106 | unsigned tail; | |
107 | spinlock_t completion_lock; | |
4e23bcae | 108 | } ____cacheline_aligned_in_smp; |
58c85dc2 KO |
109 | |
110 | struct page *internal_pages[AIO_RING_PAGES]; | |
4e179bca KO |
111 | }; |
112 | ||
1da177e4 | 113 | /*------ sysctl variables----*/ |
d55b5fda ZB |
114 | static DEFINE_SPINLOCK(aio_nr_lock); |
115 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
116 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
117 | /*----end sysctl variables---*/ |
118 | ||
e18b890b CL |
119 | static struct kmem_cache *kiocb_cachep; |
120 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 121 | |
1da177e4 LT |
122 | /* aio_setup |
123 | * Creates the slab caches used by the aio routines, panic on | |
124 | * failure as this is done early during the boot sequence. | |
125 | */ | |
126 | static int __init aio_setup(void) | |
127 | { | |
0a31bd5f CL |
128 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
129 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 130 | |
caf4167a | 131 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
132 | |
133 | return 0; | |
134 | } | |
385773e0 | 135 | __initcall(aio_setup); |
1da177e4 LT |
136 | |
137 | static void aio_free_ring(struct kioctx *ctx) | |
138 | { | |
1da177e4 LT |
139 | long i; |
140 | ||
58c85dc2 KO |
141 | for (i = 0; i < ctx->nr_pages; i++) |
142 | put_page(ctx->ring_pages[i]); | |
1da177e4 | 143 | |
58c85dc2 KO |
144 | if (ctx->mmap_size) |
145 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
1da177e4 | 146 | |
58c85dc2 KO |
147 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
148 | kfree(ctx->ring_pages); | |
1da177e4 LT |
149 | } |
150 | ||
151 | static int aio_setup_ring(struct kioctx *ctx) | |
152 | { | |
153 | struct aio_ring *ring; | |
1da177e4 | 154 | unsigned nr_events = ctx->max_reqs; |
41003a7b | 155 | struct mm_struct *mm = current->mm; |
41badc15 | 156 | unsigned long size, populate; |
1da177e4 LT |
157 | int nr_pages; |
158 | ||
159 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
160 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
161 | ||
162 | size = sizeof(struct aio_ring); | |
163 | size += sizeof(struct io_event) * nr_events; | |
164 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; | |
165 | ||
166 | if (nr_pages < 0) | |
167 | return -EINVAL; | |
168 | ||
169 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); | |
170 | ||
58c85dc2 KO |
171 | ctx->nr_events = 0; |
172 | ctx->ring_pages = ctx->internal_pages; | |
1da177e4 | 173 | if (nr_pages > AIO_RING_PAGES) { |
58c85dc2 KO |
174 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
175 | GFP_KERNEL); | |
176 | if (!ctx->ring_pages) | |
1da177e4 | 177 | return -ENOMEM; |
1da177e4 LT |
178 | } |
179 | ||
58c85dc2 KO |
180 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
181 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
41003a7b | 182 | down_write(&mm->mmap_sem); |
58c85dc2 KO |
183 | ctx->mmap_base = do_mmap_pgoff(NULL, 0, ctx->mmap_size, |
184 | PROT_READ|PROT_WRITE, | |
185 | MAP_ANONYMOUS|MAP_PRIVATE, 0, &populate); | |
186 | if (IS_ERR((void *)ctx->mmap_base)) { | |
41003a7b | 187 | up_write(&mm->mmap_sem); |
58c85dc2 | 188 | ctx->mmap_size = 0; |
1da177e4 LT |
189 | aio_free_ring(ctx); |
190 | return -EAGAIN; | |
191 | } | |
192 | ||
58c85dc2 KO |
193 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
194 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, | |
195 | 1, 0, ctx->ring_pages, NULL); | |
41003a7b | 196 | up_write(&mm->mmap_sem); |
1da177e4 | 197 | |
58c85dc2 | 198 | if (unlikely(ctx->nr_pages != nr_pages)) { |
1da177e4 LT |
199 | aio_free_ring(ctx); |
200 | return -EAGAIN; | |
201 | } | |
bebeb3d6 | 202 | if (populate) |
58c85dc2 | 203 | mm_populate(ctx->mmap_base, populate); |
1da177e4 | 204 | |
58c85dc2 KO |
205 | ctx->user_id = ctx->mmap_base; |
206 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 207 | |
58c85dc2 | 208 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 LT |
209 | ring->nr = nr_events; /* user copy */ |
210 | ring->id = ctx->user_id; | |
211 | ring->head = ring->tail = 0; | |
212 | ring->magic = AIO_RING_MAGIC; | |
213 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
214 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
215 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 216 | kunmap_atomic(ring); |
58c85dc2 | 217 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
218 | |
219 | return 0; | |
220 | } | |
221 | ||
1da177e4 LT |
222 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
223 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
224 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
225 | ||
0460fef2 KO |
226 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
227 | { | |
228 | struct kioctx *ctx = req->ki_ctx; | |
229 | unsigned long flags; | |
230 | ||
231 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
232 | ||
233 | if (!req->ki_list.next) | |
234 | list_add(&req->ki_list, &ctx->active_reqs); | |
235 | ||
236 | req->ki_cancel = cancel; | |
237 | ||
238 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
239 | } | |
240 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
241 | ||
906b973c KO |
242 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
243 | struct io_event *res) | |
244 | { | |
0460fef2 | 245 | kiocb_cancel_fn *old, *cancel; |
906b973c KO |
246 | int ret = -EINVAL; |
247 | ||
0460fef2 KO |
248 | /* |
249 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
250 | * actually has a cancel function, hence the cmpxchg() | |
251 | */ | |
252 | ||
253 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
254 | do { | |
255 | if (!cancel || cancel == KIOCB_CANCELLED) | |
256 | return ret; | |
906b973c | 257 | |
0460fef2 KO |
258 | old = cancel; |
259 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
260 | } while (cancel != old); | |
906b973c | 261 | |
0460fef2 KO |
262 | atomic_inc(&kiocb->ki_users); |
263 | spin_unlock_irq(&ctx->ctx_lock); | |
264 | ||
265 | memset(res, 0, sizeof(*res)); | |
266 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
267 | res->data = kiocb->ki_user_data; | |
268 | ret = cancel(kiocb, res); | |
269 | ||
270 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c KO |
271 | |
272 | return ret; | |
273 | } | |
274 | ||
36f55889 KO |
275 | static void free_ioctx_rcu(struct rcu_head *head) |
276 | { | |
277 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
278 | kmem_cache_free(kioctx_cachep, ctx); | |
279 | } | |
280 | ||
281 | /* | |
282 | * When this function runs, the kioctx has been removed from the "hash table" | |
283 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
284 | * now it's safe to cancel any that need to be. | |
285 | */ | |
286 | static void free_ioctx(struct kioctx *ctx) | |
287 | { | |
3e845ce0 | 288 | struct aio_ring *ring; |
36f55889 KO |
289 | struct io_event res; |
290 | struct kiocb *req; | |
3e845ce0 | 291 | unsigned head, avail; |
36f55889 KO |
292 | |
293 | spin_lock_irq(&ctx->ctx_lock); | |
294 | ||
295 | while (!list_empty(&ctx->active_reqs)) { | |
296 | req = list_first_entry(&ctx->active_reqs, | |
297 | struct kiocb, ki_list); | |
298 | ||
299 | list_del_init(&req->ki_list); | |
300 | kiocb_cancel(ctx, req, &res); | |
301 | } | |
302 | ||
303 | spin_unlock_irq(&ctx->ctx_lock); | |
304 | ||
58c85dc2 | 305 | ring = kmap_atomic(ctx->ring_pages[0]); |
3e845ce0 KO |
306 | head = ring->head; |
307 | kunmap_atomic(ring); | |
308 | ||
309 | while (atomic_read(&ctx->reqs_active) > 0) { | |
58c85dc2 | 310 | wait_event(ctx->wait, head != ctx->tail); |
3e845ce0 | 311 | |
58c85dc2 | 312 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
3e845ce0 KO |
313 | |
314 | atomic_sub(avail, &ctx->reqs_active); | |
315 | head += avail; | |
58c85dc2 | 316 | head %= ctx->nr_events; |
3e845ce0 KO |
317 | } |
318 | ||
319 | WARN_ON(atomic_read(&ctx->reqs_active) < 0); | |
36f55889 KO |
320 | |
321 | aio_free_ring(ctx); | |
322 | ||
323 | spin_lock(&aio_nr_lock); | |
324 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); | |
325 | aio_nr -= ctx->max_reqs; | |
326 | spin_unlock(&aio_nr_lock); | |
327 | ||
328 | pr_debug("freeing %p\n", ctx); | |
329 | ||
330 | /* | |
331 | * Here the call_rcu() is between the wait_event() for reqs_active to | |
332 | * hit 0, and freeing the ioctx. | |
333 | * | |
334 | * aio_complete() decrements reqs_active, but it has to touch the ioctx | |
335 | * after to issue a wakeup so we use rcu. | |
336 | */ | |
337 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); | |
338 | } | |
339 | ||
340 | static void put_ioctx(struct kioctx *ctx) | |
341 | { | |
342 | if (unlikely(atomic_dec_and_test(&ctx->users))) | |
343 | free_ioctx(ctx); | |
344 | } | |
345 | ||
1da177e4 LT |
346 | /* ioctx_alloc |
347 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
348 | */ | |
349 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
350 | { | |
41003a7b | 351 | struct mm_struct *mm = current->mm; |
1da177e4 | 352 | struct kioctx *ctx; |
e23754f8 | 353 | int err = -ENOMEM; |
1da177e4 LT |
354 | |
355 | /* Prevent overflows */ | |
356 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
357 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
358 | pr_debug("ENOMEM: nr_events too high\n"); | |
359 | return ERR_PTR(-EINVAL); | |
360 | } | |
361 | ||
2dd542b7 | 362 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
1da177e4 LT |
363 | return ERR_PTR(-EAGAIN); |
364 | ||
c3762229 | 365 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
366 | if (!ctx) |
367 | return ERR_PTR(-ENOMEM); | |
368 | ||
1da177e4 | 369 | ctx->max_reqs = nr_events; |
1da177e4 | 370 | |
86b62a2c | 371 | atomic_set(&ctx->users, 2); |
36f55889 | 372 | atomic_set(&ctx->dead, 0); |
1da177e4 | 373 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 374 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 375 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
376 | init_waitqueue_head(&ctx->wait); |
377 | ||
378 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 LT |
379 | |
380 | if (aio_setup_ring(ctx) < 0) | |
381 | goto out_freectx; | |
382 | ||
383 | /* limit the number of system wide aios */ | |
9fa1cb39 | 384 | spin_lock(&aio_nr_lock); |
2dd542b7 AV |
385 | if (aio_nr + nr_events > aio_max_nr || |
386 | aio_nr + nr_events < aio_nr) { | |
9fa1cb39 | 387 | spin_unlock(&aio_nr_lock); |
1da177e4 | 388 | goto out_cleanup; |
2dd542b7 AV |
389 | } |
390 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 391 | spin_unlock(&aio_nr_lock); |
1da177e4 | 392 | |
39fa0031 | 393 | /* now link into global list. */ |
abf137dd JA |
394 | spin_lock(&mm->ioctx_lock); |
395 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
396 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 | 397 | |
caf4167a | 398 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 399 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
400 | return ctx; |
401 | ||
402 | out_cleanup: | |
e23754f8 AV |
403 | err = -EAGAIN; |
404 | aio_free_ring(ctx); | |
1da177e4 | 405 | out_freectx: |
1da177e4 | 406 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 407 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 408 | return ERR_PTR(err); |
1da177e4 LT |
409 | } |
410 | ||
36f55889 | 411 | static void kill_ioctx_work(struct work_struct *work) |
1da177e4 | 412 | { |
36f55889 | 413 | struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); |
06af121e | 414 | |
36f55889 KO |
415 | wake_up_all(&ctx->wait); |
416 | put_ioctx(ctx); | |
417 | } | |
906b973c | 418 | |
36f55889 KO |
419 | static void kill_ioctx_rcu(struct rcu_head *head) |
420 | { | |
421 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
1da177e4 | 422 | |
36f55889 KO |
423 | INIT_WORK(&ctx->rcu_work, kill_ioctx_work); |
424 | schedule_work(&ctx->rcu_work); | |
425 | } | |
1da177e4 | 426 | |
36f55889 KO |
427 | /* kill_ioctx |
428 | * Cancels all outstanding aio requests on an aio context. Used | |
429 | * when the processes owning a context have all exited to encourage | |
430 | * the rapid destruction of the kioctx. | |
431 | */ | |
432 | static void kill_ioctx(struct kioctx *ctx) | |
433 | { | |
434 | if (!atomic_xchg(&ctx->dead, 1)) { | |
435 | hlist_del_rcu(&ctx->list); | |
436 | /* Between hlist_del_rcu() and dropping the initial ref */ | |
437 | synchronize_rcu(); | |
dee11c23 | 438 | |
36f55889 KO |
439 | /* |
440 | * We can't punt to workqueue here because put_ioctx() -> | |
441 | * free_ioctx() will unmap the ringbuffer, and that has to be | |
442 | * done in the original process's context. kill_ioctx_rcu/work() | |
443 | * exist for exit_aio(), as in that path free_ioctx() won't do | |
444 | * the unmap. | |
445 | */ | |
446 | kill_ioctx_work(&ctx->rcu_work); | |
447 | } | |
1da177e4 LT |
448 | } |
449 | ||
450 | /* wait_on_sync_kiocb: | |
451 | * Waits on the given sync kiocb to complete. | |
452 | */ | |
fc9b52cd | 453 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
1da177e4 | 454 | { |
11599eba | 455 | while (atomic_read(&iocb->ki_users)) { |
1da177e4 | 456 | set_current_state(TASK_UNINTERRUPTIBLE); |
11599eba | 457 | if (!atomic_read(&iocb->ki_users)) |
1da177e4 | 458 | break; |
41d10da3 | 459 | io_schedule(); |
1da177e4 LT |
460 | } |
461 | __set_current_state(TASK_RUNNING); | |
462 | return iocb->ki_user_data; | |
463 | } | |
385773e0 | 464 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 465 | |
36f55889 KO |
466 | /* |
467 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
468 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
469 | * called on the context. | |
470 | * | |
471 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
472 | * them. | |
1da177e4 | 473 | */ |
fc9b52cd | 474 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 475 | { |
abf137dd | 476 | struct kioctx *ctx; |
36f55889 | 477 | struct hlist_node *n; |
abf137dd | 478 | |
36f55889 | 479 | hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { |
1da177e4 LT |
480 | if (1 != atomic_read(&ctx->users)) |
481 | printk(KERN_DEBUG | |
482 | "exit_aio:ioctx still alive: %d %d %d\n", | |
36f55889 KO |
483 | atomic_read(&ctx->users), |
484 | atomic_read(&ctx->dead), | |
11599eba | 485 | atomic_read(&ctx->reqs_active)); |
936af157 AV |
486 | /* |
487 | * We don't need to bother with munmap() here - | |
488 | * exit_mmap(mm) is coming and it'll unmap everything. | |
489 | * Since aio_free_ring() uses non-zero ->mmap_size | |
490 | * as indicator that it needs to unmap the area, | |
491 | * just set it to 0; aio_free_ring() is the only | |
492 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 493 | */ |
58c85dc2 | 494 | ctx->mmap_size = 0; |
36f55889 KO |
495 | |
496 | if (!atomic_xchg(&ctx->dead, 1)) { | |
497 | hlist_del_rcu(&ctx->list); | |
498 | call_rcu(&ctx->rcu_head, kill_ioctx_rcu); | |
499 | } | |
1da177e4 LT |
500 | } |
501 | } | |
502 | ||
1da177e4 | 503 | /* aio_get_req |
11599eba | 504 | * Allocate a slot for an aio request. Increments the ki_users count |
1da177e4 LT |
505 | * of the kioctx so that the kioctx stays around until all requests are |
506 | * complete. Returns NULL if no requests are free. | |
507 | * | |
11599eba | 508 | * Returns with kiocb->ki_users set to 2. The io submit code path holds |
1da177e4 LT |
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 | */ | |
a1c8eae7 | 513 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 514 | { |
a1c8eae7 KO |
515 | struct kiocb *req; |
516 | ||
58c85dc2 | 517 | if (atomic_read(&ctx->reqs_active) >= ctx->nr_events) |
a1c8eae7 KO |
518 | return NULL; |
519 | ||
58c85dc2 | 520 | if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1) |
a1c8eae7 | 521 | goto out_put; |
1da177e4 | 522 | |
0460fef2 | 523 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 524 | if (unlikely(!req)) |
a1c8eae7 | 525 | goto out_put; |
1da177e4 | 526 | |
11599eba | 527 | atomic_set(&req->ki_users, 2); |
1da177e4 | 528 | req->ki_ctx = ctx; |
1da177e4 | 529 | |
080d676d | 530 | return req; |
a1c8eae7 KO |
531 | out_put: |
532 | atomic_dec(&ctx->reqs_active); | |
533 | return NULL; | |
1da177e4 LT |
534 | } |
535 | ||
11599eba | 536 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 537 | { |
1d98ebfc KO |
538 | if (req->ki_filp) |
539 | fput(req->ki_filp); | |
13389010 DL |
540 | if (req->ki_eventfd != NULL) |
541 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 LT |
542 | if (req->ki_dtor) |
543 | req->ki_dtor(req); | |
eed4e51f BP |
544 | if (req->ki_iovec != &req->ki_inline_vec) |
545 | kfree(req->ki_iovec); | |
1da177e4 | 546 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
547 | } |
548 | ||
2d68449e | 549 | void aio_put_req(struct kiocb *req) |
1da177e4 | 550 | { |
11599eba KO |
551 | if (atomic_dec_and_test(&req->ki_users)) |
552 | kiocb_free(req); | |
1da177e4 | 553 | } |
385773e0 | 554 | EXPORT_SYMBOL(aio_put_req); |
1da177e4 | 555 | |
d5470b59 | 556 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 557 | { |
abf137dd | 558 | struct mm_struct *mm = current->mm; |
65c24491 | 559 | struct kioctx *ctx, *ret = NULL; |
1da177e4 | 560 | |
abf137dd JA |
561 | rcu_read_lock(); |
562 | ||
b67bfe0d | 563 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
36f55889 KO |
564 | if (ctx->user_id == ctx_id) { |
565 | atomic_inc(&ctx->users); | |
65c24491 | 566 | ret = ctx; |
1da177e4 LT |
567 | break; |
568 | } | |
abf137dd | 569 | } |
1da177e4 | 570 | |
abf137dd | 571 | rcu_read_unlock(); |
65c24491 | 572 | return ret; |
1da177e4 LT |
573 | } |
574 | ||
1da177e4 LT |
575 | /* aio_complete |
576 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 577 | */ |
2d68449e | 578 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
579 | { |
580 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 581 | struct aio_ring *ring; |
21b40200 | 582 | struct io_event *ev_page, *event; |
1da177e4 | 583 | unsigned long flags; |
21b40200 | 584 | unsigned tail, pos; |
1da177e4 | 585 | |
20dcae32 ZB |
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 | |
1da177e4 LT |
592 | */ |
593 | if (is_sync_kiocb(iocb)) { | |
11599eba | 594 | BUG_ON(atomic_read(&iocb->ki_users) != 1); |
1da177e4 | 595 | iocb->ki_user_data = res; |
11599eba | 596 | atomic_set(&iocb->ki_users, 0); |
1da177e4 | 597 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 598 | return; |
1da177e4 LT |
599 | } |
600 | ||
36f55889 | 601 | /* |
36f55889 KO |
602 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we |
603 | * need to issue a wakeup after decrementing reqs_active. | |
1da177e4 | 604 | */ |
36f55889 | 605 | rcu_read_lock(); |
1da177e4 | 606 | |
0460fef2 KO |
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 | } | |
11599eba | 614 | |
1da177e4 LT |
615 | /* |
616 | * cancelled requests don't get events, userland was given one | |
617 | * when the event got cancelled. | |
618 | */ | |
0460fef2 | 619 | if (unlikely(xchg(&iocb->ki_cancel, |
3e845ce0 KO |
620 | KIOCB_CANCELLED) == KIOCB_CANCELLED)) { |
621 | atomic_dec(&ctx->reqs_active); | |
622 | /* Still need the wake_up in case free_ioctx is waiting */ | |
1da177e4 | 623 | goto put_rq; |
3e845ce0 | 624 | } |
1da177e4 | 625 | |
0460fef2 KO |
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 | ||
58c85dc2 | 633 | tail = ctx->tail; |
21b40200 KO |
634 | pos = tail + AIO_EVENTS_OFFSET; |
635 | ||
58c85dc2 | 636 | if (++tail >= ctx->nr_events) |
4bf69b2a | 637 | tail = 0; |
1da177e4 | 638 | |
58c85dc2 | 639 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
640 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
641 | ||
1da177e4 LT |
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 | ||
21b40200 | 647 | kunmap_atomic(ev_page); |
58c85dc2 | 648 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
649 | |
650 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
651 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
652 | res, res2); | |
1da177e4 LT |
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 | ||
58c85dc2 | 659 | ctx->tail = tail; |
1da177e4 | 660 | |
58c85dc2 | 661 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 662 | ring->tail = tail; |
e8e3c3d6 | 663 | kunmap_atomic(ring); |
58c85dc2 | 664 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 665 | |
0460fef2 KO |
666 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
667 | ||
21b40200 | 668 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
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 | */ | |
87c3a86e | 675 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
676 | eventfd_signal(iocb->ki_eventfd, 1); |
677 | ||
1da177e4 LT |
678 | put_rq: |
679 | /* everything turned out well, dispose of the aiocb. */ | |
11599eba | 680 | aio_put_req(iocb); |
1da177e4 | 681 | |
6cb2a210 QB |
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 | ||
1da177e4 LT |
690 | if (waitqueue_active(&ctx->wait)) |
691 | wake_up(&ctx->wait); | |
692 | ||
36f55889 | 693 | rcu_read_unlock(); |
1da177e4 | 694 | } |
385773e0 | 695 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 696 | |
a31ad380 KO |
697 | /* aio_read_events |
698 | * Pull an event off of the ioctx's event ring. Returns the number of | |
699 | * events fetched | |
1da177e4 | 700 | */ |
a31ad380 KO |
701 | static long aio_read_events_ring(struct kioctx *ctx, |
702 | struct io_event __user *event, long nr) | |
1da177e4 | 703 | { |
1da177e4 | 704 | struct aio_ring *ring; |
a31ad380 KO |
705 | unsigned head, pos; |
706 | long ret = 0; | |
707 | int copy_ret; | |
708 | ||
58c85dc2 | 709 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 710 | |
58c85dc2 | 711 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 KO |
712 | head = ring->head; |
713 | kunmap_atomic(ring); | |
714 | ||
58c85dc2 | 715 | pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events); |
1da177e4 | 716 | |
58c85dc2 | 717 | if (head == ctx->tail) |
1da177e4 LT |
718 | goto out; |
719 | ||
a31ad380 KO |
720 | while (ret < nr) { |
721 | long avail; | |
722 | struct io_event *ev; | |
723 | struct page *page; | |
724 | ||
58c85dc2 KO |
725 | avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head; |
726 | if (head == ctx->tail) | |
a31ad380 KO |
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; | |
58c85dc2 | 734 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
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; | |
58c85dc2 | 749 | head %= ctx->nr_events; |
1da177e4 | 750 | } |
1da177e4 | 751 | |
58c85dc2 | 752 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 753 | ring->head = head; |
91d80a84 | 754 | kunmap_atomic(ring); |
58c85dc2 | 755 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 756 | |
58c85dc2 | 757 | pr_debug("%li h%u t%u\n", ret, head, ctx->tail); |
3e845ce0 KO |
758 | |
759 | atomic_sub(ret, &ctx->reqs_active); | |
a31ad380 | 760 | out: |
58c85dc2 | 761 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 762 | |
1da177e4 LT |
763 | return ret; |
764 | } | |
765 | ||
a31ad380 KO |
766 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
767 | struct io_event __user *event, long *i) | |
1da177e4 | 768 | { |
a31ad380 | 769 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 770 | |
a31ad380 KO |
771 | if (ret > 0) |
772 | *i += ret; | |
1da177e4 | 773 | |
a31ad380 KO |
774 | if (unlikely(atomic_read(&ctx->dead))) |
775 | ret = -EINVAL; | |
1da177e4 | 776 | |
a31ad380 KO |
777 | if (!*i) |
778 | *i = ret; | |
1da177e4 | 779 | |
a31ad380 | 780 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
781 | } |
782 | ||
a31ad380 | 783 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
784 | struct io_event __user *event, |
785 | struct timespec __user *timeout) | |
786 | { | |
a31ad380 KO |
787 | ktime_t until = { .tv64 = KTIME_MAX }; |
788 | long ret = 0; | |
1da177e4 | 789 | |
1da177e4 LT |
790 | if (timeout) { |
791 | struct timespec ts; | |
a31ad380 | 792 | |
1da177e4 | 793 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 794 | return -EFAULT; |
1da177e4 | 795 | |
a31ad380 | 796 | until = timespec_to_ktime(ts); |
1da177e4 LT |
797 | } |
798 | ||
a31ad380 KO |
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); | |
1da177e4 | 815 | |
a31ad380 KO |
816 | if (!ret && signal_pending(current)) |
817 | ret = -EINTR; | |
1da177e4 | 818 | |
a31ad380 | 819 | return ret; |
1da177e4 LT |
820 | } |
821 | ||
1da177e4 LT |
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 | */ | |
002c8976 | 835 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
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; | |
d55b5fda ZB |
846 | if (unlikely(ctx || nr_events == 0)) { |
847 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
848 | ctx, nr_events); | |
1da177e4 LT |
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); | |
a2e1859a | 856 | if (ret) |
36f55889 | 857 | kill_ioctx(ioctx); |
a2e1859a | 858 | put_ioctx(ioctx); |
1da177e4 LT |
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 | |
642b5123 | 868 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
869 | * is invalid. |
870 | */ | |
002c8976 | 871 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
872 | { |
873 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
874 | if (likely(NULL != ioctx)) { | |
36f55889 | 875 | kill_ioctx(ioctx); |
a2e1859a | 876 | put_ioctx(ioctx); |
1da177e4 LT |
877 | return 0; |
878 | } | |
879 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
880 | return -EINVAL; | |
881 | } | |
882 | ||
eed4e51f | 883 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1da177e4 | 884 | { |
eed4e51f BP |
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++; | |
897f15fb | 898 | } |
eed4e51f | 899 | } |
1da177e4 | 900 | |
eed4e51f BP |
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); | |
1da177e4 LT |
904 | } |
905 | ||
eed4e51f | 906 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
1da177e4 LT |
907 | { |
908 | struct file *file = iocb->ki_filp; | |
eed4e51f BP |
909 | struct address_space *mapping = file->f_mapping; |
910 | struct inode *inode = mapping->host; | |
911 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, | |
912 | unsigned long, loff_t); | |
1da177e4 | 913 | ssize_t ret = 0; |
eed4e51f BP |
914 | unsigned short opcode; |
915 | ||
916 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || | |
917 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { | |
918 | rw_op = file->f_op->aio_read; | |
919 | opcode = IOCB_CMD_PREADV; | |
920 | } else { | |
921 | rw_op = file->f_op->aio_write; | |
922 | opcode = IOCB_CMD_PWRITEV; | |
923 | } | |
1da177e4 | 924 | |
c2ec6682 RR |
925 | /* This matches the pread()/pwrite() logic */ |
926 | if (iocb->ki_pos < 0) | |
927 | return -EINVAL; | |
928 | ||
8d71db4f AV |
929 | if (opcode == IOCB_CMD_PWRITEV) |
930 | file_start_write(file); | |
897f15fb | 931 | do { |
eed4e51f BP |
932 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
933 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
934 | iocb->ki_pos); | |
935 | if (ret > 0) | |
936 | aio_advance_iovec(iocb, ret); | |
937 | ||
938 | /* retry all partial writes. retry partial reads as long as its a | |
939 | * regular file. */ | |
940 | } while (ret > 0 && iocb->ki_left > 0 && | |
941 | (opcode == IOCB_CMD_PWRITEV || | |
942 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); | |
8d71db4f AV |
943 | if (opcode == IOCB_CMD_PWRITEV) |
944 | file_end_write(file); | |
1da177e4 | 945 | |
eed4e51f BP |
946 | /* This means we must have transferred all that we could */ |
947 | /* No need to retry anymore */ | |
1da177e4 LT |
948 | if ((ret == 0) || (iocb->ki_left == 0)) |
949 | ret = iocb->ki_nbytes - iocb->ki_left; | |
950 | ||
7adfa2ff RR |
951 | /* If we managed to write some out we return that, rather than |
952 | * the eventual error. */ | |
953 | if (opcode == IOCB_CMD_PWRITEV | |
41003a7b | 954 | && ret < 0 && ret != -EIOCBQUEUED |
7adfa2ff RR |
955 | && iocb->ki_nbytes - iocb->ki_left) |
956 | ret = iocb->ki_nbytes - iocb->ki_left; | |
957 | ||
1da177e4 LT |
958 | return ret; |
959 | } | |
960 | ||
961 | static ssize_t aio_fdsync(struct kiocb *iocb) | |
962 | { | |
963 | struct file *file = iocb->ki_filp; | |
964 | ssize_t ret = -EINVAL; | |
965 | ||
966 | if (file->f_op->aio_fsync) | |
967 | ret = file->f_op->aio_fsync(iocb, 1); | |
968 | return ret; | |
969 | } | |
970 | ||
971 | static ssize_t aio_fsync(struct kiocb *iocb) | |
972 | { | |
973 | struct file *file = iocb->ki_filp; | |
974 | ssize_t ret = -EINVAL; | |
975 | ||
976 | if (file->f_op->aio_fsync) | |
977 | ret = file->f_op->aio_fsync(iocb, 0); | |
978 | return ret; | |
979 | } | |
980 | ||
9d85cba7 | 981 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
eed4e51f BP |
982 | { |
983 | ssize_t ret; | |
984 | ||
9d85cba7 JM |
985 | #ifdef CONFIG_COMPAT |
986 | if (compat) | |
987 | ret = compat_rw_copy_check_uvector(type, | |
988 | (struct compat_iovec __user *)kiocb->ki_buf, | |
989 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 990 | &kiocb->ki_iovec); |
9d85cba7 JM |
991 | else |
992 | #endif | |
993 | ret = rw_copy_check_uvector(type, | |
994 | (struct iovec __user *)kiocb->ki_buf, | |
995 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 996 | &kiocb->ki_iovec); |
eed4e51f BP |
997 | if (ret < 0) |
998 | goto out; | |
999 | ||
a70b52ec LT |
1000 | ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret); |
1001 | if (ret < 0) | |
1002 | goto out; | |
1003 | ||
eed4e51f BP |
1004 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1005 | kiocb->ki_cur_seg = 0; | |
1006 | /* ki_nbytes/left now reflect bytes instead of segs */ | |
1007 | kiocb->ki_nbytes = ret; | |
1008 | kiocb->ki_left = ret; | |
1009 | ||
1010 | ret = 0; | |
1011 | out: | |
1012 | return ret; | |
1013 | } | |
1014 | ||
a70b52ec | 1015 | static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb) |
eed4e51f | 1016 | { |
a70b52ec LT |
1017 | int bytes; |
1018 | ||
1019 | bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left); | |
1020 | if (bytes < 0) | |
1021 | return bytes; | |
1022 | ||
eed4e51f BP |
1023 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1024 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
a70b52ec | 1025 | kiocb->ki_iovec->iov_len = bytes; |
eed4e51f BP |
1026 | kiocb->ki_nr_segs = 1; |
1027 | kiocb->ki_cur_seg = 0; | |
eed4e51f BP |
1028 | return 0; |
1029 | } | |
1030 | ||
1da177e4 LT |
1031 | /* |
1032 | * aio_setup_iocb: | |
1033 | * Performs the initial checks and aio retry method | |
1034 | * setup for the kiocb at the time of io submission. | |
1035 | */ | |
9d85cba7 | 1036 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
1da177e4 LT |
1037 | { |
1038 | struct file *file = kiocb->ki_filp; | |
1039 | ssize_t ret = 0; | |
1040 | ||
1041 | switch (kiocb->ki_opcode) { | |
1042 | case IOCB_CMD_PREAD: | |
1043 | ret = -EBADF; | |
1044 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1045 | break; | |
1046 | ret = -EFAULT; | |
1047 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, | |
1048 | kiocb->ki_left))) | |
1049 | break; | |
a70b52ec | 1050 | ret = aio_setup_single_vector(READ, file, kiocb); |
eed4e51f BP |
1051 | if (ret) |
1052 | break; | |
1da177e4 LT |
1053 | ret = -EINVAL; |
1054 | if (file->f_op->aio_read) | |
eed4e51f | 1055 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1056 | break; |
1057 | case IOCB_CMD_PWRITE: | |
1058 | ret = -EBADF; | |
1059 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1060 | break; | |
1061 | ret = -EFAULT; | |
1062 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, | |
1063 | kiocb->ki_left))) | |
1064 | break; | |
a70b52ec | 1065 | ret = aio_setup_single_vector(WRITE, file, kiocb); |
eed4e51f BP |
1066 | if (ret) |
1067 | break; | |
1068 | ret = -EINVAL; | |
1069 | if (file->f_op->aio_write) | |
1070 | kiocb->ki_retry = aio_rw_vect_retry; | |
1071 | break; | |
1072 | case IOCB_CMD_PREADV: | |
1073 | ret = -EBADF; | |
1074 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1075 | break; | |
9d85cba7 | 1076 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
eed4e51f BP |
1077 | if (ret) |
1078 | break; | |
1079 | ret = -EINVAL; | |
1080 | if (file->f_op->aio_read) | |
1081 | kiocb->ki_retry = aio_rw_vect_retry; | |
1082 | break; | |
1083 | case IOCB_CMD_PWRITEV: | |
1084 | ret = -EBADF; | |
1085 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1086 | break; | |
9d85cba7 | 1087 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
eed4e51f BP |
1088 | if (ret) |
1089 | break; | |
1da177e4 LT |
1090 | ret = -EINVAL; |
1091 | if (file->f_op->aio_write) | |
eed4e51f | 1092 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1093 | break; |
1094 | case IOCB_CMD_FDSYNC: | |
1095 | ret = -EINVAL; | |
1096 | if (file->f_op->aio_fsync) | |
1097 | kiocb->ki_retry = aio_fdsync; | |
1098 | break; | |
1099 | case IOCB_CMD_FSYNC: | |
1100 | ret = -EINVAL; | |
1101 | if (file->f_op->aio_fsync) | |
1102 | kiocb->ki_retry = aio_fsync; | |
1103 | break; | |
1104 | default: | |
caf4167a | 1105 | pr_debug("EINVAL: no operation provided\n"); |
1da177e4 LT |
1106 | ret = -EINVAL; |
1107 | } | |
1108 | ||
1109 | if (!kiocb->ki_retry) | |
1110 | return ret; | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
d5470b59 | 1115 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
a1c8eae7 | 1116 | struct iocb *iocb, bool compat) |
1da177e4 LT |
1117 | { |
1118 | struct kiocb *req; | |
1da177e4 LT |
1119 | ssize_t ret; |
1120 | ||
1121 | /* enforce forwards compatibility on users */ | |
9c3060be | 1122 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1123 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1124 | return -EINVAL; |
1125 | } | |
1126 | ||
1127 | /* prevent overflows */ | |
1128 | if (unlikely( | |
1129 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1130 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1131 | ((ssize_t)iocb->aio_nbytes < 0) | |
1132 | )) { | |
1133 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1134 | return -EINVAL; | |
1135 | } | |
1136 | ||
a1c8eae7 | 1137 | req = aio_get_req(ctx); /* returns with 2 references to req */ |
1d98ebfc | 1138 | if (unlikely(!req)) |
1da177e4 | 1139 | return -EAGAIN; |
1d98ebfc KO |
1140 | |
1141 | req->ki_filp = fget(iocb->aio_fildes); | |
1142 | if (unlikely(!req->ki_filp)) { | |
1143 | ret = -EBADF; | |
1144 | goto out_put_req; | |
1da177e4 | 1145 | } |
1d98ebfc | 1146 | |
9c3060be DL |
1147 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1148 | /* | |
1149 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1150 | * instance of the file* now. The file descriptor must be | |
1151 | * an eventfd() fd, and will be signaled for each completed | |
1152 | * event using the eventfd_signal() function. | |
1153 | */ | |
13389010 | 1154 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1155 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1156 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1157 | req->ki_eventfd = NULL; |
9c3060be DL |
1158 | goto out_put_req; |
1159 | } | |
1160 | } | |
1da177e4 | 1161 | |
212079cf | 1162 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
1da177e4 | 1163 | if (unlikely(ret)) { |
caf4167a | 1164 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1165 | goto out_put_req; |
1166 | } | |
1167 | ||
1168 | req->ki_obj.user = user_iocb; | |
1169 | req->ki_user_data = iocb->aio_data; | |
1170 | req->ki_pos = iocb->aio_offset; | |
1171 | ||
1172 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1173 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1174 | req->ki_opcode = iocb->aio_lio_opcode; | |
1da177e4 | 1175 | |
9d85cba7 | 1176 | ret = aio_setup_iocb(req, compat); |
41003a7b | 1177 | if (ret) |
7137c6bd | 1178 | goto out_put_req; |
41003a7b | 1179 | |
0460fef2 | 1180 | ret = req->ki_retry(req); |
41003a7b ZB |
1181 | if (ret != -EIOCBQUEUED) { |
1182 | /* | |
1183 | * There's no easy way to restart the syscall since other AIO's | |
1184 | * may be already running. Just fail this IO with EINTR. | |
1185 | */ | |
1186 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1187 | ret == -ERESTARTNOHAND || | |
1188 | ret == -ERESTART_RESTARTBLOCK)) | |
1189 | ret = -EINTR; | |
1190 | aio_complete(req, ret, 0); | |
7137c6bd | 1191 | } |
cfb1e33e | 1192 | |
1da177e4 LT |
1193 | aio_put_req(req); /* drop extra ref to req */ |
1194 | return 0; | |
1195 | ||
1196 | out_put_req: | |
11599eba | 1197 | atomic_dec(&ctx->reqs_active); |
1da177e4 LT |
1198 | aio_put_req(req); /* drop extra ref to req */ |
1199 | aio_put_req(req); /* drop i/o ref to req */ | |
1200 | return ret; | |
1201 | } | |
1202 | ||
9d85cba7 JM |
1203 | long do_io_submit(aio_context_t ctx_id, long nr, |
1204 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1205 | { |
1206 | struct kioctx *ctx; | |
1207 | long ret = 0; | |
080d676d | 1208 | int i = 0; |
9f5b9425 | 1209 | struct blk_plug plug; |
1da177e4 LT |
1210 | |
1211 | if (unlikely(nr < 0)) | |
1212 | return -EINVAL; | |
1213 | ||
75e1c70f JM |
1214 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1215 | nr = LONG_MAX/sizeof(*iocbpp); | |
1216 | ||
1da177e4 LT |
1217 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1218 | return -EFAULT; | |
1219 | ||
1220 | ctx = lookup_ioctx(ctx_id); | |
1221 | if (unlikely(!ctx)) { | |
caf4167a | 1222 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1223 | return -EINVAL; |
1224 | } | |
1225 | ||
9f5b9425 SL |
1226 | blk_start_plug(&plug); |
1227 | ||
1da177e4 LT |
1228 | /* |
1229 | * AKPM: should this return a partial result if some of the IOs were | |
1230 | * successfully submitted? | |
1231 | */ | |
1232 | for (i=0; i<nr; i++) { | |
1233 | struct iocb __user *user_iocb; | |
1234 | struct iocb tmp; | |
1235 | ||
1236 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1237 | ret = -EFAULT; | |
1238 | break; | |
1239 | } | |
1240 | ||
1241 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1242 | ret = -EFAULT; | |
1243 | break; | |
1244 | } | |
1245 | ||
a1c8eae7 | 1246 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1247 | if (ret) |
1248 | break; | |
1249 | } | |
9f5b9425 | 1250 | blk_finish_plug(&plug); |
1da177e4 LT |
1251 | |
1252 | put_ioctx(ctx); | |
1253 | return i ? i : ret; | |
1254 | } | |
1255 | ||
9d85cba7 JM |
1256 | /* sys_io_submit: |
1257 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1258 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1259 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1260 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1261 | * is invalid for the file descriptor in the iocb. May fail with | |
1262 | * -EFAULT if any of the data structures point to invalid data. May | |
1263 | * fail with -EBADF if the file descriptor specified in the first | |
1264 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1265 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1266 | * fail with -ENOSYS if not implemented. | |
1267 | */ | |
1268 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1269 | struct iocb __user * __user *, iocbpp) | |
1270 | { | |
1271 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1272 | } | |
1273 | ||
1da177e4 LT |
1274 | /* lookup_kiocb |
1275 | * Finds a given iocb for cancellation. | |
1da177e4 | 1276 | */ |
25ee7e38 AB |
1277 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1278 | u32 key) | |
1da177e4 LT |
1279 | { |
1280 | struct list_head *pos; | |
d00689af ZB |
1281 | |
1282 | assert_spin_locked(&ctx->ctx_lock); | |
1283 | ||
1da177e4 LT |
1284 | /* TODO: use a hash or array, this sucks. */ |
1285 | list_for_each(pos, &ctx->active_reqs) { | |
1286 | struct kiocb *kiocb = list_kiocb(pos); | |
1287 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) | |
1288 | return kiocb; | |
1289 | } | |
1290 | return NULL; | |
1291 | } | |
1292 | ||
1293 | /* sys_io_cancel: | |
1294 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1295 | * the operation is successfully cancelled, the resulting event is | |
1296 | * copied into the memory pointed to by result without being placed | |
1297 | * into the completion queue and 0 is returned. May fail with | |
1298 | * -EFAULT if any of the data structures pointed to are invalid. | |
1299 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1300 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1301 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1302 | */ | |
002c8976 HC |
1303 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1304 | struct io_event __user *, result) | |
1da177e4 | 1305 | { |
906b973c | 1306 | struct io_event res; |
1da177e4 LT |
1307 | struct kioctx *ctx; |
1308 | struct kiocb *kiocb; | |
1309 | u32 key; | |
1310 | int ret; | |
1311 | ||
1312 | ret = get_user(key, &iocb->aio_key); | |
1313 | if (unlikely(ret)) | |
1314 | return -EFAULT; | |
1315 | ||
1316 | ctx = lookup_ioctx(ctx_id); | |
1317 | if (unlikely(!ctx)) | |
1318 | return -EINVAL; | |
1319 | ||
1320 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1321 | |
1da177e4 | 1322 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c KO |
1323 | if (kiocb) |
1324 | ret = kiocb_cancel(ctx, kiocb, &res); | |
1325 | else | |
1326 | ret = -EINVAL; | |
1327 | ||
1da177e4 LT |
1328 | spin_unlock_irq(&ctx->ctx_lock); |
1329 | ||
906b973c KO |
1330 | if (!ret) { |
1331 | /* Cancellation succeeded -- copy the result | |
1332 | * into the user's buffer. | |
1333 | */ | |
1334 | if (copy_to_user(result, &res, sizeof(res))) | |
1335 | ret = -EFAULT; | |
1336 | } | |
1da177e4 LT |
1337 | |
1338 | put_ioctx(ctx); | |
1339 | ||
1340 | return ret; | |
1341 | } | |
1342 | ||
1343 | /* io_getevents: | |
1344 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1345 | * the completion queue for the aio_context specified by ctx_id. If |
1346 | * it succeeds, the number of read events is returned. May fail with | |
1347 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1348 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1349 | * if any of the memory specified is invalid. May return 0 or | |
1350 | * < min_nr if the timeout specified by timeout has elapsed | |
1351 | * before sufficient events are available, where timeout == NULL | |
1352 | * specifies an infinite timeout. Note that the timeout pointed to by | |
1353 | * timeout is relative and will be updated if not NULL and the | |
1354 | * operation blocks. Will fail with -ENOSYS if not implemented. | |
1da177e4 | 1355 | */ |
002c8976 HC |
1356 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1357 | long, min_nr, | |
1358 | long, nr, | |
1359 | struct io_event __user *, events, | |
1360 | struct timespec __user *, timeout) | |
1da177e4 LT |
1361 | { |
1362 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1363 | long ret = -EINVAL; | |
1364 | ||
1365 | if (likely(ioctx)) { | |
2e410255 | 1366 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 LT |
1367 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1368 | put_ioctx(ioctx); | |
1369 | } | |
1da177e4 LT |
1370 | return ret; |
1371 | } |