]>
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; | |
a31ad380 | 66 | struct mutex ring_lock; |
4e179bca KO |
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; | |
36f55889 | 82 | atomic_t dead; |
4e179bca KO |
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 | ||
11599eba | 92 | atomic_t reqs_active; |
4e179bca KO |
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; | |
36f55889 | 101 | struct work_struct rcu_work; |
4e179bca KO |
102 | }; |
103 | ||
1da177e4 | 104 | /*------ sysctl variables----*/ |
d55b5fda ZB |
105 | static DEFINE_SPINLOCK(aio_nr_lock); |
106 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
107 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
108 | /*----end sysctl variables---*/ |
109 | ||
e18b890b CL |
110 | static struct kmem_cache *kiocb_cachep; |
111 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 112 | |
1da177e4 LT |
113 | /* aio_setup |
114 | * Creates the slab caches used by the aio routines, panic on | |
115 | * failure as this is done early during the boot sequence. | |
116 | */ | |
117 | static int __init aio_setup(void) | |
118 | { | |
0a31bd5f CL |
119 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
120 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 121 | |
caf4167a | 122 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
123 | |
124 | return 0; | |
125 | } | |
385773e0 | 126 | __initcall(aio_setup); |
1da177e4 LT |
127 | |
128 | static void aio_free_ring(struct kioctx *ctx) | |
129 | { | |
130 | struct aio_ring_info *info = &ctx->ring_info; | |
131 | long i; | |
132 | ||
133 | for (i=0; i<info->nr_pages; i++) | |
134 | put_page(info->ring_pages[i]); | |
135 | ||
936af157 | 136 | if (info->mmap_size) { |
bfce281c | 137 | vm_munmap(info->mmap_base, info->mmap_size); |
936af157 | 138 | } |
1da177e4 LT |
139 | |
140 | if (info->ring_pages && info->ring_pages != info->internal_pages) | |
141 | kfree(info->ring_pages); | |
142 | info->ring_pages = NULL; | |
143 | info->nr = 0; | |
144 | } | |
145 | ||
146 | static int aio_setup_ring(struct kioctx *ctx) | |
147 | { | |
148 | struct aio_ring *ring; | |
149 | struct aio_ring_info *info = &ctx->ring_info; | |
150 | unsigned nr_events = ctx->max_reqs; | |
41003a7b | 151 | struct mm_struct *mm = current->mm; |
41badc15 | 152 | unsigned long size, populate; |
1da177e4 LT |
153 | int nr_pages; |
154 | ||
155 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
156 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
157 | ||
158 | size = sizeof(struct aio_ring); | |
159 | size += sizeof(struct io_event) * nr_events; | |
160 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; | |
161 | ||
162 | if (nr_pages < 0) | |
163 | return -EINVAL; | |
164 | ||
165 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); | |
166 | ||
167 | info->nr = 0; | |
168 | info->ring_pages = info->internal_pages; | |
169 | if (nr_pages > AIO_RING_PAGES) { | |
11b0b5ab | 170 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
171 | if (!info->ring_pages) |
172 | return -ENOMEM; | |
1da177e4 LT |
173 | } |
174 | ||
175 | info->mmap_size = nr_pages * PAGE_SIZE; | |
caf4167a | 176 | pr_debug("attempting mmap of %lu bytes\n", info->mmap_size); |
41003a7b | 177 | down_write(&mm->mmap_sem); |
e3fc629d AV |
178 | info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size, |
179 | PROT_READ|PROT_WRITE, | |
bebeb3d6 ML |
180 | MAP_ANONYMOUS|MAP_PRIVATE, 0, |
181 | &populate); | |
1da177e4 | 182 | if (IS_ERR((void *)info->mmap_base)) { |
41003a7b | 183 | up_write(&mm->mmap_sem); |
1da177e4 LT |
184 | info->mmap_size = 0; |
185 | aio_free_ring(ctx); | |
186 | return -EAGAIN; | |
187 | } | |
188 | ||
caf4167a | 189 | pr_debug("mmap address: 0x%08lx\n", info->mmap_base); |
41003a7b | 190 | info->nr_pages = get_user_pages(current, mm, info->mmap_base, nr_pages, |
1da177e4 | 191 | 1, 0, info->ring_pages, NULL); |
41003a7b | 192 | up_write(&mm->mmap_sem); |
1da177e4 LT |
193 | |
194 | if (unlikely(info->nr_pages != nr_pages)) { | |
195 | aio_free_ring(ctx); | |
196 | return -EAGAIN; | |
197 | } | |
bebeb3d6 | 198 | if (populate) |
41badc15 | 199 | mm_populate(info->mmap_base, populate); |
1da177e4 LT |
200 | |
201 | ctx->user_id = info->mmap_base; | |
202 | ||
203 | info->nr = nr_events; /* trusted copy */ | |
204 | ||
e8e3c3d6 | 205 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
206 | ring->nr = nr_events; /* user copy */ |
207 | ring->id = ctx->user_id; | |
208 | ring->head = ring->tail = 0; | |
209 | ring->magic = AIO_RING_MAGIC; | |
210 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
211 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
212 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 213 | kunmap_atomic(ring); |
1da177e4 LT |
214 | |
215 | return 0; | |
216 | } | |
217 | ||
218 | ||
219 | /* aio_ring_event: returns a pointer to the event at the given index from | |
e8e3c3d6 | 220 | * kmap_atomic(). Release the pointer with put_aio_ring_event(); |
1da177e4 LT |
221 | */ |
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 | ||
e8e3c3d6 | 226 | #define aio_ring_event(info, nr) ({ \ |
1da177e4 LT |
227 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
228 | struct io_event *__event; \ | |
229 | __event = kmap_atomic( \ | |
e8e3c3d6 | 230 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \ |
1da177e4 LT |
231 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
232 | __event; \ | |
233 | }) | |
234 | ||
e8e3c3d6 | 235 | #define put_aio_ring_event(event) do { \ |
1da177e4 LT |
236 | struct io_event *__event = (event); \ |
237 | (void)__event; \ | |
e8e3c3d6 | 238 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \ |
1da177e4 LT |
239 | } while(0) |
240 | ||
906b973c KO |
241 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb, |
242 | struct io_event *res) | |
243 | { | |
244 | int (*cancel)(struct kiocb *, struct io_event *); | |
245 | int ret = -EINVAL; | |
246 | ||
247 | cancel = kiocb->ki_cancel; | |
248 | kiocbSetCancelled(kiocb); | |
249 | if (cancel) { | |
11599eba | 250 | atomic_inc(&kiocb->ki_users); |
906b973c KO |
251 | spin_unlock_irq(&ctx->ctx_lock); |
252 | ||
253 | memset(res, 0, sizeof(*res)); | |
254 | res->obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
255 | res->data = kiocb->ki_user_data; | |
256 | ret = cancel(kiocb, res); | |
257 | ||
258 | spin_lock_irq(&ctx->ctx_lock); | |
259 | } | |
260 | ||
261 | return ret; | |
262 | } | |
263 | ||
36f55889 KO |
264 | static void free_ioctx_rcu(struct rcu_head *head) |
265 | { | |
266 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
267 | kmem_cache_free(kioctx_cachep, ctx); | |
268 | } | |
269 | ||
270 | /* | |
271 | * When this function runs, the kioctx has been removed from the "hash table" | |
272 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
273 | * now it's safe to cancel any that need to be. | |
274 | */ | |
275 | static void free_ioctx(struct kioctx *ctx) | |
276 | { | |
277 | struct io_event res; | |
278 | struct kiocb *req; | |
279 | ||
280 | spin_lock_irq(&ctx->ctx_lock); | |
281 | ||
282 | while (!list_empty(&ctx->active_reqs)) { | |
283 | req = list_first_entry(&ctx->active_reqs, | |
284 | struct kiocb, ki_list); | |
285 | ||
286 | list_del_init(&req->ki_list); | |
287 | kiocb_cancel(ctx, req, &res); | |
288 | } | |
289 | ||
290 | spin_unlock_irq(&ctx->ctx_lock); | |
291 | ||
292 | wait_event(ctx->wait, !atomic_read(&ctx->reqs_active)); | |
293 | ||
294 | aio_free_ring(ctx); | |
295 | ||
296 | spin_lock(&aio_nr_lock); | |
297 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); | |
298 | aio_nr -= ctx->max_reqs; | |
299 | spin_unlock(&aio_nr_lock); | |
300 | ||
301 | pr_debug("freeing %p\n", ctx); | |
302 | ||
303 | /* | |
304 | * Here the call_rcu() is between the wait_event() for reqs_active to | |
305 | * hit 0, and freeing the ioctx. | |
306 | * | |
307 | * aio_complete() decrements reqs_active, but it has to touch the ioctx | |
308 | * after to issue a wakeup so we use rcu. | |
309 | */ | |
310 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); | |
311 | } | |
312 | ||
313 | static void put_ioctx(struct kioctx *ctx) | |
314 | { | |
315 | if (unlikely(atomic_dec_and_test(&ctx->users))) | |
316 | free_ioctx(ctx); | |
317 | } | |
318 | ||
1da177e4 LT |
319 | /* ioctx_alloc |
320 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
321 | */ | |
322 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
323 | { | |
41003a7b | 324 | struct mm_struct *mm = current->mm; |
1da177e4 | 325 | struct kioctx *ctx; |
e23754f8 | 326 | int err = -ENOMEM; |
1da177e4 LT |
327 | |
328 | /* Prevent overflows */ | |
329 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
330 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
331 | pr_debug("ENOMEM: nr_events too high\n"); | |
332 | return ERR_PTR(-EINVAL); | |
333 | } | |
334 | ||
2dd542b7 | 335 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
1da177e4 LT |
336 | return ERR_PTR(-EAGAIN); |
337 | ||
c3762229 | 338 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
339 | if (!ctx) |
340 | return ERR_PTR(-ENOMEM); | |
341 | ||
1da177e4 | 342 | ctx->max_reqs = nr_events; |
1da177e4 | 343 | |
86b62a2c | 344 | atomic_set(&ctx->users, 2); |
36f55889 | 345 | atomic_set(&ctx->dead, 0); |
1da177e4 | 346 | spin_lock_init(&ctx->ctx_lock); |
a31ad380 | 347 | mutex_init(&ctx->ring_info.ring_lock); |
1da177e4 LT |
348 | init_waitqueue_head(&ctx->wait); |
349 | ||
350 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 LT |
351 | |
352 | if (aio_setup_ring(ctx) < 0) | |
353 | goto out_freectx; | |
354 | ||
355 | /* limit the number of system wide aios */ | |
9fa1cb39 | 356 | spin_lock(&aio_nr_lock); |
2dd542b7 AV |
357 | if (aio_nr + nr_events > aio_max_nr || |
358 | aio_nr + nr_events < aio_nr) { | |
9fa1cb39 | 359 | spin_unlock(&aio_nr_lock); |
1da177e4 | 360 | goto out_cleanup; |
2dd542b7 AV |
361 | } |
362 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 363 | spin_unlock(&aio_nr_lock); |
1da177e4 | 364 | |
39fa0031 | 365 | /* now link into global list. */ |
abf137dd JA |
366 | spin_lock(&mm->ioctx_lock); |
367 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
368 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 | 369 | |
caf4167a | 370 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
41003a7b | 371 | ctx, ctx->user_id, mm, ctx->ring_info.nr); |
1da177e4 LT |
372 | return ctx; |
373 | ||
374 | out_cleanup: | |
e23754f8 AV |
375 | err = -EAGAIN; |
376 | aio_free_ring(ctx); | |
1da177e4 | 377 | out_freectx: |
1da177e4 | 378 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 379 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 380 | return ERR_PTR(err); |
1da177e4 LT |
381 | } |
382 | ||
36f55889 | 383 | static void kill_ioctx_work(struct work_struct *work) |
1da177e4 | 384 | { |
36f55889 | 385 | struct kioctx *ctx = container_of(work, struct kioctx, rcu_work); |
06af121e | 386 | |
36f55889 KO |
387 | wake_up_all(&ctx->wait); |
388 | put_ioctx(ctx); | |
389 | } | |
906b973c | 390 | |
36f55889 KO |
391 | static void kill_ioctx_rcu(struct rcu_head *head) |
392 | { | |
393 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
1da177e4 | 394 | |
36f55889 KO |
395 | INIT_WORK(&ctx->rcu_work, kill_ioctx_work); |
396 | schedule_work(&ctx->rcu_work); | |
397 | } | |
1da177e4 | 398 | |
36f55889 KO |
399 | /* kill_ioctx |
400 | * Cancels all outstanding aio requests on an aio context. Used | |
401 | * when the processes owning a context have all exited to encourage | |
402 | * the rapid destruction of the kioctx. | |
403 | */ | |
404 | static void kill_ioctx(struct kioctx *ctx) | |
405 | { | |
406 | if (!atomic_xchg(&ctx->dead, 1)) { | |
407 | hlist_del_rcu(&ctx->list); | |
408 | /* Between hlist_del_rcu() and dropping the initial ref */ | |
409 | synchronize_rcu(); | |
dee11c23 | 410 | |
36f55889 KO |
411 | /* |
412 | * We can't punt to workqueue here because put_ioctx() -> | |
413 | * free_ioctx() will unmap the ringbuffer, and that has to be | |
414 | * done in the original process's context. kill_ioctx_rcu/work() | |
415 | * exist for exit_aio(), as in that path free_ioctx() won't do | |
416 | * the unmap. | |
417 | */ | |
418 | kill_ioctx_work(&ctx->rcu_work); | |
419 | } | |
1da177e4 LT |
420 | } |
421 | ||
422 | /* wait_on_sync_kiocb: | |
423 | * Waits on the given sync kiocb to complete. | |
424 | */ | |
fc9b52cd | 425 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
1da177e4 | 426 | { |
11599eba | 427 | while (atomic_read(&iocb->ki_users)) { |
1da177e4 | 428 | set_current_state(TASK_UNINTERRUPTIBLE); |
11599eba | 429 | if (!atomic_read(&iocb->ki_users)) |
1da177e4 | 430 | break; |
41d10da3 | 431 | io_schedule(); |
1da177e4 LT |
432 | } |
433 | __set_current_state(TASK_RUNNING); | |
434 | return iocb->ki_user_data; | |
435 | } | |
385773e0 | 436 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 437 | |
36f55889 KO |
438 | /* |
439 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
440 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
441 | * called on the context. | |
442 | * | |
443 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
444 | * them. | |
1da177e4 | 445 | */ |
fc9b52cd | 446 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 447 | { |
abf137dd | 448 | struct kioctx *ctx; |
36f55889 | 449 | struct hlist_node *n; |
abf137dd | 450 | |
36f55889 | 451 | hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) { |
1da177e4 LT |
452 | if (1 != atomic_read(&ctx->users)) |
453 | printk(KERN_DEBUG | |
454 | "exit_aio:ioctx still alive: %d %d %d\n", | |
36f55889 KO |
455 | atomic_read(&ctx->users), |
456 | atomic_read(&ctx->dead), | |
11599eba | 457 | atomic_read(&ctx->reqs_active)); |
936af157 AV |
458 | /* |
459 | * We don't need to bother with munmap() here - | |
460 | * exit_mmap(mm) is coming and it'll unmap everything. | |
461 | * Since aio_free_ring() uses non-zero ->mmap_size | |
462 | * as indicator that it needs to unmap the area, | |
463 | * just set it to 0; aio_free_ring() is the only | |
464 | * place that uses ->mmap_size, so it's safe. | |
936af157 AV |
465 | */ |
466 | ctx->ring_info.mmap_size = 0; | |
36f55889 KO |
467 | |
468 | if (!atomic_xchg(&ctx->dead, 1)) { | |
469 | hlist_del_rcu(&ctx->list); | |
470 | call_rcu(&ctx->rcu_head, kill_ioctx_rcu); | |
471 | } | |
1da177e4 LT |
472 | } |
473 | } | |
474 | ||
1da177e4 | 475 | /* aio_get_req |
11599eba | 476 | * Allocate a slot for an aio request. Increments the ki_users count |
1da177e4 LT |
477 | * of the kioctx so that the kioctx stays around until all requests are |
478 | * complete. Returns NULL if no requests are free. | |
479 | * | |
11599eba | 480 | * Returns with kiocb->ki_users set to 2. The io submit code path holds |
1da177e4 LT |
481 | * an extra reference while submitting the i/o. |
482 | * This prevents races between the aio code path referencing the | |
483 | * req (after submitting it) and aio_complete() freeing the req. | |
484 | */ | |
fc9b52cd | 485 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
1da177e4 LT |
486 | { |
487 | struct kiocb *req = NULL; | |
1da177e4 LT |
488 | |
489 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); | |
490 | if (unlikely(!req)) | |
491 | return NULL; | |
492 | ||
4faa5285 | 493 | req->ki_flags = 0; |
11599eba | 494 | atomic_set(&req->ki_users, 2); |
1da177e4 LT |
495 | req->ki_key = 0; |
496 | req->ki_ctx = ctx; | |
497 | req->ki_cancel = NULL; | |
498 | req->ki_retry = NULL; | |
1da177e4 LT |
499 | req->ki_dtor = NULL; |
500 | req->private = NULL; | |
eed4e51f | 501 | req->ki_iovec = NULL; |
87c3a86e | 502 | req->ki_eventfd = NULL; |
1da177e4 | 503 | |
080d676d JM |
504 | return req; |
505 | } | |
506 | ||
507 | /* | |
508 | * struct kiocb's are allocated in batches to reduce the number of | |
509 | * times the ctx lock is acquired and released. | |
510 | */ | |
511 | #define KIOCB_BATCH_SIZE 32L | |
512 | struct kiocb_batch { | |
513 | struct list_head head; | |
514 | long count; /* number of requests left to allocate */ | |
515 | }; | |
516 | ||
517 | static void kiocb_batch_init(struct kiocb_batch *batch, long total) | |
518 | { | |
519 | INIT_LIST_HEAD(&batch->head); | |
520 | batch->count = total; | |
521 | } | |
522 | ||
69e4747e | 523 | static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) |
080d676d JM |
524 | { |
525 | struct kiocb *req, *n; | |
526 | ||
69e4747e GN |
527 | if (list_empty(&batch->head)) |
528 | return; | |
529 | ||
530 | spin_lock_irq(&ctx->ctx_lock); | |
080d676d JM |
531 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
532 | list_del(&req->ki_batch); | |
69e4747e | 533 | list_del(&req->ki_list); |
080d676d | 534 | kmem_cache_free(kiocb_cachep, req); |
11599eba | 535 | atomic_dec(&ctx->reqs_active); |
080d676d | 536 | } |
69e4747e | 537 | spin_unlock_irq(&ctx->ctx_lock); |
080d676d JM |
538 | } |
539 | ||
540 | /* | |
541 | * Allocate a batch of kiocbs. This avoids taking and dropping the | |
542 | * context lock a lot during setup. | |
543 | */ | |
544 | static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) | |
545 | { | |
546 | unsigned short allocated, to_alloc; | |
547 | long avail; | |
080d676d JM |
548 | struct kiocb *req, *n; |
549 | struct aio_ring *ring; | |
550 | ||
551 | to_alloc = min(batch->count, KIOCB_BATCH_SIZE); | |
552 | for (allocated = 0; allocated < to_alloc; allocated++) { | |
553 | req = __aio_get_req(ctx); | |
554 | if (!req) | |
555 | /* allocation failed, go with what we've got */ | |
556 | break; | |
557 | list_add(&req->ki_batch, &batch->head); | |
558 | } | |
559 | ||
560 | if (allocated == 0) | |
561 | goto out; | |
562 | ||
1da177e4 | 563 | spin_lock_irq(&ctx->ctx_lock); |
080d676d JM |
564 | ring = kmap_atomic(ctx->ring_info.ring_pages[0]); |
565 | ||
11599eba KO |
566 | avail = aio_ring_avail(&ctx->ring_info, ring) - |
567 | atomic_read(&ctx->reqs_active); | |
080d676d | 568 | BUG_ON(avail < 0); |
080d676d JM |
569 | if (avail < allocated) { |
570 | /* Trim back the number of requests. */ | |
571 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { | |
572 | list_del(&req->ki_batch); | |
573 | kmem_cache_free(kiocb_cachep, req); | |
574 | if (--allocated <= avail) | |
575 | break; | |
576 | } | |
577 | } | |
578 | ||
579 | batch->count -= allocated; | |
580 | list_for_each_entry(req, &batch->head, ki_batch) { | |
1da177e4 | 581 | list_add(&req->ki_list, &ctx->active_reqs); |
11599eba | 582 | atomic_inc(&ctx->reqs_active); |
1da177e4 | 583 | } |
1da177e4 | 584 | |
080d676d JM |
585 | kunmap_atomic(ring); |
586 | spin_unlock_irq(&ctx->ctx_lock); | |
1da177e4 | 587 | |
080d676d JM |
588 | out: |
589 | return allocated; | |
1da177e4 LT |
590 | } |
591 | ||
080d676d JM |
592 | static inline struct kiocb *aio_get_req(struct kioctx *ctx, |
593 | struct kiocb_batch *batch) | |
1da177e4 LT |
594 | { |
595 | struct kiocb *req; | |
080d676d JM |
596 | |
597 | if (list_empty(&batch->head)) | |
598 | if (kiocb_batch_refill(ctx, batch) == 0) | |
599 | return NULL; | |
600 | req = list_first_entry(&batch->head, struct kiocb, ki_batch); | |
601 | list_del(&req->ki_batch); | |
1da177e4 LT |
602 | return req; |
603 | } | |
604 | ||
11599eba | 605 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 606 | { |
1d98ebfc KO |
607 | if (req->ki_filp) |
608 | fput(req->ki_filp); | |
13389010 DL |
609 | if (req->ki_eventfd != NULL) |
610 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 LT |
611 | if (req->ki_dtor) |
612 | req->ki_dtor(req); | |
eed4e51f BP |
613 | if (req->ki_iovec != &req->ki_inline_vec) |
614 | kfree(req->ki_iovec); | |
1da177e4 | 615 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
616 | } |
617 | ||
2d68449e | 618 | void aio_put_req(struct kiocb *req) |
1da177e4 | 619 | { |
11599eba KO |
620 | if (atomic_dec_and_test(&req->ki_users)) |
621 | kiocb_free(req); | |
1da177e4 | 622 | } |
385773e0 | 623 | EXPORT_SYMBOL(aio_put_req); |
1da177e4 | 624 | |
d5470b59 | 625 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 626 | { |
abf137dd | 627 | struct mm_struct *mm = current->mm; |
65c24491 | 628 | struct kioctx *ctx, *ret = NULL; |
1da177e4 | 629 | |
abf137dd JA |
630 | rcu_read_lock(); |
631 | ||
b67bfe0d | 632 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
36f55889 KO |
633 | if (ctx->user_id == ctx_id) { |
634 | atomic_inc(&ctx->users); | |
65c24491 | 635 | ret = ctx; |
1da177e4 LT |
636 | break; |
637 | } | |
abf137dd | 638 | } |
1da177e4 | 639 | |
abf137dd | 640 | rcu_read_unlock(); |
65c24491 | 641 | return ret; |
1da177e4 LT |
642 | } |
643 | ||
1da177e4 LT |
644 | /* aio_complete |
645 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 646 | */ |
2d68449e | 647 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
648 | { |
649 | struct kioctx *ctx = iocb->ki_ctx; | |
650 | struct aio_ring_info *info; | |
651 | struct aio_ring *ring; | |
652 | struct io_event *event; | |
653 | unsigned long flags; | |
654 | unsigned long tail; | |
1da177e4 | 655 | |
20dcae32 ZB |
656 | /* |
657 | * Special case handling for sync iocbs: | |
658 | * - events go directly into the iocb for fast handling | |
659 | * - the sync task with the iocb in its stack holds the single iocb | |
660 | * ref, no other paths have a way to get another ref | |
661 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
662 | */ |
663 | if (is_sync_kiocb(iocb)) { | |
11599eba | 664 | BUG_ON(atomic_read(&iocb->ki_users) != 1); |
1da177e4 | 665 | iocb->ki_user_data = res; |
11599eba | 666 | atomic_set(&iocb->ki_users, 0); |
1da177e4 | 667 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 668 | return; |
1da177e4 LT |
669 | } |
670 | ||
671 | info = &ctx->ring_info; | |
672 | ||
36f55889 KO |
673 | /* |
674 | * Add a completion event to the ring buffer. Must be done holding | |
675 | * ctx->ctx_lock to prevent other code from messing with the tail | |
676 | * pointer since we might be called from irq context. | |
677 | * | |
678 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we | |
679 | * need to issue a wakeup after decrementing reqs_active. | |
1da177e4 | 680 | */ |
36f55889 | 681 | rcu_read_lock(); |
1da177e4 LT |
682 | spin_lock_irqsave(&ctx->ctx_lock, flags); |
683 | ||
11599eba KO |
684 | list_del(&iocb->ki_list); /* remove from active_reqs */ |
685 | ||
1da177e4 LT |
686 | /* |
687 | * cancelled requests don't get events, userland was given one | |
688 | * when the event got cancelled. | |
689 | */ | |
690 | if (kiocbIsCancelled(iocb)) | |
691 | goto put_rq; | |
692 | ||
e8e3c3d6 | 693 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
694 | |
695 | tail = info->tail; | |
e8e3c3d6 | 696 | event = aio_ring_event(info, tail); |
4bf69b2a KC |
697 | if (++tail >= info->nr) |
698 | tail = 0; | |
1da177e4 LT |
699 | |
700 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; | |
701 | event->data = iocb->ki_user_data; | |
702 | event->res = res; | |
703 | event->res2 = res2; | |
704 | ||
caf4167a KO |
705 | pr_debug("%p[%lu]: %p: %p %Lx %lx %lx\n", |
706 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, | |
707 | res, res2); | |
1da177e4 LT |
708 | |
709 | /* after flagging the request as done, we | |
710 | * must never even look at it again | |
711 | */ | |
712 | smp_wmb(); /* make event visible before updating tail */ | |
713 | ||
714 | info->tail = tail; | |
715 | ring->tail = tail; | |
716 | ||
e8e3c3d6 CW |
717 | put_aio_ring_event(event); |
718 | kunmap_atomic(ring); | |
1da177e4 LT |
719 | |
720 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); | |
8d1c98b0 DL |
721 | |
722 | /* | |
723 | * Check if the user asked us to deliver the result through an | |
724 | * eventfd. The eventfd_signal() function is safe to be called | |
725 | * from IRQ context. | |
726 | */ | |
87c3a86e | 727 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
728 | eventfd_signal(iocb->ki_eventfd, 1); |
729 | ||
1da177e4 LT |
730 | put_rq: |
731 | /* everything turned out well, dispose of the aiocb. */ | |
11599eba KO |
732 | aio_put_req(iocb); |
733 | atomic_dec(&ctx->reqs_active); | |
1da177e4 | 734 | |
6cb2a210 QB |
735 | /* |
736 | * We have to order our ring_info tail store above and test | |
737 | * of the wait list below outside the wait lock. This is | |
738 | * like in wake_up_bit() where clearing a bit has to be | |
739 | * ordered with the unlocked test. | |
740 | */ | |
741 | smp_mb(); | |
742 | ||
1da177e4 LT |
743 | if (waitqueue_active(&ctx->wait)) |
744 | wake_up(&ctx->wait); | |
745 | ||
dee11c23 | 746 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
36f55889 | 747 | rcu_read_unlock(); |
1da177e4 | 748 | } |
385773e0 | 749 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 750 | |
a31ad380 KO |
751 | /* aio_read_events |
752 | * Pull an event off of the ioctx's event ring. Returns the number of | |
753 | * events fetched | |
1da177e4 | 754 | */ |
a31ad380 KO |
755 | static long aio_read_events_ring(struct kioctx *ctx, |
756 | struct io_event __user *event, long nr) | |
1da177e4 | 757 | { |
a31ad380 | 758 | struct aio_ring_info *info = &ctx->ring_info; |
1da177e4 | 759 | struct aio_ring *ring; |
a31ad380 KO |
760 | unsigned head, pos; |
761 | long ret = 0; | |
762 | int copy_ret; | |
763 | ||
764 | mutex_lock(&info->ring_lock); | |
1da177e4 | 765 | |
e8e3c3d6 | 766 | ring = kmap_atomic(info->ring_pages[0]); |
a31ad380 KO |
767 | head = ring->head; |
768 | kunmap_atomic(ring); | |
769 | ||
770 | pr_debug("h%u t%u m%u\n", head, info->tail, info->nr); | |
1da177e4 | 771 | |
a31ad380 | 772 | if (head == info->tail) |
1da177e4 LT |
773 | goto out; |
774 | ||
a31ad380 KO |
775 | while (ret < nr) { |
776 | long avail; | |
777 | struct io_event *ev; | |
778 | struct page *page; | |
779 | ||
780 | avail = (head <= info->tail ? info->tail : info->nr) - head; | |
781 | if (head == info->tail) | |
782 | break; | |
783 | ||
784 | avail = min(avail, nr - ret); | |
785 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
786 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
787 | ||
788 | pos = head + AIO_EVENTS_OFFSET; | |
789 | page = info->ring_pages[pos / AIO_EVENTS_PER_PAGE]; | |
790 | pos %= AIO_EVENTS_PER_PAGE; | |
791 | ||
792 | ev = kmap(page); | |
793 | copy_ret = copy_to_user(event + ret, ev + pos, | |
794 | sizeof(*ev) * avail); | |
795 | kunmap(page); | |
796 | ||
797 | if (unlikely(copy_ret)) { | |
798 | ret = -EFAULT; | |
799 | goto out; | |
800 | } | |
801 | ||
802 | ret += avail; | |
803 | head += avail; | |
804 | head %= info->nr; | |
1da177e4 | 805 | } |
1da177e4 | 806 | |
a31ad380 KO |
807 | ring = kmap_atomic(info->ring_pages[0]); |
808 | ring->head = head; | |
91d80a84 | 809 | kunmap_atomic(ring); |
a31ad380 KO |
810 | |
811 | pr_debug("%li h%u t%u\n", ret, head, info->tail); | |
812 | out: | |
813 | mutex_unlock(&info->ring_lock); | |
814 | ||
1da177e4 LT |
815 | return ret; |
816 | } | |
817 | ||
a31ad380 KO |
818 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
819 | struct io_event __user *event, long *i) | |
1da177e4 | 820 | { |
a31ad380 | 821 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 822 | |
a31ad380 KO |
823 | if (ret > 0) |
824 | *i += ret; | |
1da177e4 | 825 | |
a31ad380 KO |
826 | if (unlikely(atomic_read(&ctx->dead))) |
827 | ret = -EINVAL; | |
1da177e4 | 828 | |
a31ad380 KO |
829 | if (!*i) |
830 | *i = ret; | |
1da177e4 | 831 | |
a31ad380 | 832 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
833 | } |
834 | ||
a31ad380 | 835 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
836 | struct io_event __user *event, |
837 | struct timespec __user *timeout) | |
838 | { | |
a31ad380 KO |
839 | ktime_t until = { .tv64 = KTIME_MAX }; |
840 | long ret = 0; | |
1da177e4 | 841 | |
1da177e4 LT |
842 | if (timeout) { |
843 | struct timespec ts; | |
a31ad380 | 844 | |
1da177e4 | 845 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 846 | return -EFAULT; |
1da177e4 | 847 | |
a31ad380 | 848 | until = timespec_to_ktime(ts); |
1da177e4 LT |
849 | } |
850 | ||
a31ad380 KO |
851 | /* |
852 | * Note that aio_read_events() is being called as the conditional - i.e. | |
853 | * we're calling it after prepare_to_wait() has set task state to | |
854 | * TASK_INTERRUPTIBLE. | |
855 | * | |
856 | * But aio_read_events() can block, and if it blocks it's going to flip | |
857 | * the task state back to TASK_RUNNING. | |
858 | * | |
859 | * This should be ok, provided it doesn't flip the state back to | |
860 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
861 | * will only happen if the mutex_lock() call blocks, and we then find | |
862 | * the ringbuffer empty. So in practice we should be ok, but it's | |
863 | * something to be aware of when touching this code. | |
864 | */ | |
865 | wait_event_interruptible_hrtimeout(ctx->wait, | |
866 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 867 | |
a31ad380 KO |
868 | if (!ret && signal_pending(current)) |
869 | ret = -EINTR; | |
1da177e4 | 870 | |
a31ad380 | 871 | return ret; |
1da177e4 LT |
872 | } |
873 | ||
1da177e4 LT |
874 | /* sys_io_setup: |
875 | * Create an aio_context capable of receiving at least nr_events. | |
876 | * ctxp must not point to an aio_context that already exists, and | |
877 | * must be initialized to 0 prior to the call. On successful | |
878 | * creation of the aio_context, *ctxp is filled in with the resulting | |
879 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
880 | * if the specified nr_events exceeds internal limits. May fail | |
881 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
882 | * of available events. May fail with -ENOMEM if insufficient kernel | |
883 | * resources are available. May fail with -EFAULT if an invalid | |
884 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
885 | * implemented. | |
886 | */ | |
002c8976 | 887 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
888 | { |
889 | struct kioctx *ioctx = NULL; | |
890 | unsigned long ctx; | |
891 | long ret; | |
892 | ||
893 | ret = get_user(ctx, ctxp); | |
894 | if (unlikely(ret)) | |
895 | goto out; | |
896 | ||
897 | ret = -EINVAL; | |
d55b5fda ZB |
898 | if (unlikely(ctx || nr_events == 0)) { |
899 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
900 | ctx, nr_events); | |
1da177e4 LT |
901 | goto out; |
902 | } | |
903 | ||
904 | ioctx = ioctx_alloc(nr_events); | |
905 | ret = PTR_ERR(ioctx); | |
906 | if (!IS_ERR(ioctx)) { | |
907 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 908 | if (ret) |
36f55889 | 909 | kill_ioctx(ioctx); |
a2e1859a | 910 | put_ioctx(ioctx); |
1da177e4 LT |
911 | } |
912 | ||
913 | out: | |
914 | return ret; | |
915 | } | |
916 | ||
917 | /* sys_io_destroy: | |
918 | * Destroy the aio_context specified. May cancel any outstanding | |
919 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 920 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
921 | * is invalid. |
922 | */ | |
002c8976 | 923 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
924 | { |
925 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
926 | if (likely(NULL != ioctx)) { | |
36f55889 | 927 | kill_ioctx(ioctx); |
a2e1859a | 928 | put_ioctx(ioctx); |
1da177e4 LT |
929 | return 0; |
930 | } | |
931 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
932 | return -EINVAL; | |
933 | } | |
934 | ||
eed4e51f | 935 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1da177e4 | 936 | { |
eed4e51f BP |
937 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
938 | ||
939 | BUG_ON(ret <= 0); | |
940 | ||
941 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { | |
942 | ssize_t this = min((ssize_t)iov->iov_len, ret); | |
943 | iov->iov_base += this; | |
944 | iov->iov_len -= this; | |
945 | iocb->ki_left -= this; | |
946 | ret -= this; | |
947 | if (iov->iov_len == 0) { | |
948 | iocb->ki_cur_seg++; | |
949 | iov++; | |
897f15fb | 950 | } |
eed4e51f | 951 | } |
1da177e4 | 952 | |
eed4e51f BP |
953 | /* the caller should not have done more io than what fit in |
954 | * the remaining iovecs */ | |
955 | BUG_ON(ret > 0 && iocb->ki_left == 0); | |
1da177e4 LT |
956 | } |
957 | ||
eed4e51f | 958 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
1da177e4 LT |
959 | { |
960 | struct file *file = iocb->ki_filp; | |
eed4e51f BP |
961 | struct address_space *mapping = file->f_mapping; |
962 | struct inode *inode = mapping->host; | |
963 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, | |
964 | unsigned long, loff_t); | |
1da177e4 | 965 | ssize_t ret = 0; |
eed4e51f BP |
966 | unsigned short opcode; |
967 | ||
968 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || | |
969 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { | |
970 | rw_op = file->f_op->aio_read; | |
971 | opcode = IOCB_CMD_PREADV; | |
972 | } else { | |
973 | rw_op = file->f_op->aio_write; | |
974 | opcode = IOCB_CMD_PWRITEV; | |
975 | } | |
1da177e4 | 976 | |
c2ec6682 RR |
977 | /* This matches the pread()/pwrite() logic */ |
978 | if (iocb->ki_pos < 0) | |
979 | return -EINVAL; | |
980 | ||
8d71db4f AV |
981 | if (opcode == IOCB_CMD_PWRITEV) |
982 | file_start_write(file); | |
897f15fb | 983 | do { |
eed4e51f BP |
984 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
985 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
986 | iocb->ki_pos); | |
987 | if (ret > 0) | |
988 | aio_advance_iovec(iocb, ret); | |
989 | ||
990 | /* retry all partial writes. retry partial reads as long as its a | |
991 | * regular file. */ | |
992 | } while (ret > 0 && iocb->ki_left > 0 && | |
993 | (opcode == IOCB_CMD_PWRITEV || | |
994 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); | |
8d71db4f AV |
995 | if (opcode == IOCB_CMD_PWRITEV) |
996 | file_end_write(file); | |
1da177e4 | 997 | |
eed4e51f BP |
998 | /* This means we must have transferred all that we could */ |
999 | /* No need to retry anymore */ | |
1da177e4 LT |
1000 | if ((ret == 0) || (iocb->ki_left == 0)) |
1001 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1002 | ||
7adfa2ff RR |
1003 | /* If we managed to write some out we return that, rather than |
1004 | * the eventual error. */ | |
1005 | if (opcode == IOCB_CMD_PWRITEV | |
41003a7b | 1006 | && ret < 0 && ret != -EIOCBQUEUED |
7adfa2ff RR |
1007 | && iocb->ki_nbytes - iocb->ki_left) |
1008 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1009 | ||
1da177e4 LT |
1010 | return ret; |
1011 | } | |
1012 | ||
1013 | static ssize_t aio_fdsync(struct kiocb *iocb) | |
1014 | { | |
1015 | struct file *file = iocb->ki_filp; | |
1016 | ssize_t ret = -EINVAL; | |
1017 | ||
1018 | if (file->f_op->aio_fsync) | |
1019 | ret = file->f_op->aio_fsync(iocb, 1); | |
1020 | return ret; | |
1021 | } | |
1022 | ||
1023 | static ssize_t aio_fsync(struct kiocb *iocb) | |
1024 | { | |
1025 | struct file *file = iocb->ki_filp; | |
1026 | ssize_t ret = -EINVAL; | |
1027 | ||
1028 | if (file->f_op->aio_fsync) | |
1029 | ret = file->f_op->aio_fsync(iocb, 0); | |
1030 | return ret; | |
1031 | } | |
1032 | ||
9d85cba7 | 1033 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
eed4e51f BP |
1034 | { |
1035 | ssize_t ret; | |
1036 | ||
9d85cba7 JM |
1037 | #ifdef CONFIG_COMPAT |
1038 | if (compat) | |
1039 | ret = compat_rw_copy_check_uvector(type, | |
1040 | (struct compat_iovec __user *)kiocb->ki_buf, | |
1041 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1042 | &kiocb->ki_iovec); |
9d85cba7 JM |
1043 | else |
1044 | #endif | |
1045 | ret = rw_copy_check_uvector(type, | |
1046 | (struct iovec __user *)kiocb->ki_buf, | |
1047 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1048 | &kiocb->ki_iovec); |
eed4e51f BP |
1049 | if (ret < 0) |
1050 | goto out; | |
1051 | ||
a70b52ec LT |
1052 | ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret); |
1053 | if (ret < 0) | |
1054 | goto out; | |
1055 | ||
eed4e51f BP |
1056 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1057 | kiocb->ki_cur_seg = 0; | |
1058 | /* ki_nbytes/left now reflect bytes instead of segs */ | |
1059 | kiocb->ki_nbytes = ret; | |
1060 | kiocb->ki_left = ret; | |
1061 | ||
1062 | ret = 0; | |
1063 | out: | |
1064 | return ret; | |
1065 | } | |
1066 | ||
a70b52ec | 1067 | static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb) |
eed4e51f | 1068 | { |
a70b52ec LT |
1069 | int bytes; |
1070 | ||
1071 | bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left); | |
1072 | if (bytes < 0) | |
1073 | return bytes; | |
1074 | ||
eed4e51f BP |
1075 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1076 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
a70b52ec | 1077 | kiocb->ki_iovec->iov_len = bytes; |
eed4e51f BP |
1078 | kiocb->ki_nr_segs = 1; |
1079 | kiocb->ki_cur_seg = 0; | |
eed4e51f BP |
1080 | return 0; |
1081 | } | |
1082 | ||
1da177e4 LT |
1083 | /* |
1084 | * aio_setup_iocb: | |
1085 | * Performs the initial checks and aio retry method | |
1086 | * setup for the kiocb at the time of io submission. | |
1087 | */ | |
9d85cba7 | 1088 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
1da177e4 LT |
1089 | { |
1090 | struct file *file = kiocb->ki_filp; | |
1091 | ssize_t ret = 0; | |
1092 | ||
1093 | switch (kiocb->ki_opcode) { | |
1094 | case IOCB_CMD_PREAD: | |
1095 | ret = -EBADF; | |
1096 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1097 | break; | |
1098 | ret = -EFAULT; | |
1099 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, | |
1100 | kiocb->ki_left))) | |
1101 | break; | |
a70b52ec | 1102 | ret = aio_setup_single_vector(READ, file, kiocb); |
eed4e51f BP |
1103 | if (ret) |
1104 | break; | |
1da177e4 LT |
1105 | ret = -EINVAL; |
1106 | if (file->f_op->aio_read) | |
eed4e51f | 1107 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1108 | break; |
1109 | case IOCB_CMD_PWRITE: | |
1110 | ret = -EBADF; | |
1111 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1112 | break; | |
1113 | ret = -EFAULT; | |
1114 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, | |
1115 | kiocb->ki_left))) | |
1116 | break; | |
a70b52ec | 1117 | ret = aio_setup_single_vector(WRITE, file, kiocb); |
eed4e51f BP |
1118 | if (ret) |
1119 | break; | |
1120 | ret = -EINVAL; | |
1121 | if (file->f_op->aio_write) | |
1122 | kiocb->ki_retry = aio_rw_vect_retry; | |
1123 | break; | |
1124 | case IOCB_CMD_PREADV: | |
1125 | ret = -EBADF; | |
1126 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1127 | break; | |
9d85cba7 | 1128 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
eed4e51f BP |
1129 | if (ret) |
1130 | break; | |
1131 | ret = -EINVAL; | |
1132 | if (file->f_op->aio_read) | |
1133 | kiocb->ki_retry = aio_rw_vect_retry; | |
1134 | break; | |
1135 | case IOCB_CMD_PWRITEV: | |
1136 | ret = -EBADF; | |
1137 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1138 | break; | |
9d85cba7 | 1139 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
eed4e51f BP |
1140 | if (ret) |
1141 | break; | |
1da177e4 LT |
1142 | ret = -EINVAL; |
1143 | if (file->f_op->aio_write) | |
eed4e51f | 1144 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1145 | break; |
1146 | case IOCB_CMD_FDSYNC: | |
1147 | ret = -EINVAL; | |
1148 | if (file->f_op->aio_fsync) | |
1149 | kiocb->ki_retry = aio_fdsync; | |
1150 | break; | |
1151 | case IOCB_CMD_FSYNC: | |
1152 | ret = -EINVAL; | |
1153 | if (file->f_op->aio_fsync) | |
1154 | kiocb->ki_retry = aio_fsync; | |
1155 | break; | |
1156 | default: | |
caf4167a | 1157 | pr_debug("EINVAL: no operation provided\n"); |
1da177e4 LT |
1158 | ret = -EINVAL; |
1159 | } | |
1160 | ||
1161 | if (!kiocb->ki_retry) | |
1162 | return ret; | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
d5470b59 | 1167 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
080d676d JM |
1168 | struct iocb *iocb, struct kiocb_batch *batch, |
1169 | bool compat) | |
1da177e4 LT |
1170 | { |
1171 | struct kiocb *req; | |
1da177e4 LT |
1172 | ssize_t ret; |
1173 | ||
1174 | /* enforce forwards compatibility on users */ | |
9c3060be | 1175 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1176 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1177 | return -EINVAL; |
1178 | } | |
1179 | ||
1180 | /* prevent overflows */ | |
1181 | if (unlikely( | |
1182 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1183 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1184 | ((ssize_t)iocb->aio_nbytes < 0) | |
1185 | )) { | |
1186 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1187 | return -EINVAL; | |
1188 | } | |
1189 | ||
080d676d | 1190 | req = aio_get_req(ctx, batch); /* returns with 2 references to req */ |
1d98ebfc | 1191 | if (unlikely(!req)) |
1da177e4 | 1192 | return -EAGAIN; |
1d98ebfc KO |
1193 | |
1194 | req->ki_filp = fget(iocb->aio_fildes); | |
1195 | if (unlikely(!req->ki_filp)) { | |
1196 | ret = -EBADF; | |
1197 | goto out_put_req; | |
1da177e4 | 1198 | } |
1d98ebfc | 1199 | |
9c3060be DL |
1200 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1201 | /* | |
1202 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1203 | * instance of the file* now. The file descriptor must be | |
1204 | * an eventfd() fd, and will be signaled for each completed | |
1205 | * event using the eventfd_signal() function. | |
1206 | */ | |
13389010 | 1207 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1208 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1209 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1210 | req->ki_eventfd = NULL; |
9c3060be DL |
1211 | goto out_put_req; |
1212 | } | |
1213 | } | |
1da177e4 | 1214 | |
212079cf | 1215 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
1da177e4 | 1216 | if (unlikely(ret)) { |
caf4167a | 1217 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1218 | goto out_put_req; |
1219 | } | |
1220 | ||
1221 | req->ki_obj.user = user_iocb; | |
1222 | req->ki_user_data = iocb->aio_data; | |
1223 | req->ki_pos = iocb->aio_offset; | |
1224 | ||
1225 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1226 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1227 | req->ki_opcode = iocb->aio_lio_opcode; | |
1da177e4 | 1228 | |
9d85cba7 | 1229 | ret = aio_setup_iocb(req, compat); |
1da177e4 | 1230 | |
41003a7b | 1231 | if (ret) |
7137c6bd | 1232 | goto out_put_req; |
41003a7b ZB |
1233 | |
1234 | if (unlikely(kiocbIsCancelled(req))) | |
1235 | ret = -EINTR; | |
1236 | else | |
1237 | ret = req->ki_retry(req); | |
1238 | ||
1239 | if (ret != -EIOCBQUEUED) { | |
1240 | /* | |
1241 | * There's no easy way to restart the syscall since other AIO's | |
1242 | * may be already running. Just fail this IO with EINTR. | |
1243 | */ | |
1244 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1245 | ret == -ERESTARTNOHAND || | |
1246 | ret == -ERESTART_RESTARTBLOCK)) | |
1247 | ret = -EINTR; | |
1248 | aio_complete(req, ret, 0); | |
7137c6bd | 1249 | } |
cfb1e33e | 1250 | |
1da177e4 LT |
1251 | aio_put_req(req); /* drop extra ref to req */ |
1252 | return 0; | |
1253 | ||
1254 | out_put_req: | |
11599eba KO |
1255 | spin_lock_irq(&ctx->ctx_lock); |
1256 | list_del(&req->ki_list); | |
1257 | spin_unlock_irq(&ctx->ctx_lock); | |
1258 | ||
1259 | atomic_dec(&ctx->reqs_active); | |
1da177e4 LT |
1260 | aio_put_req(req); /* drop extra ref to req */ |
1261 | aio_put_req(req); /* drop i/o ref to req */ | |
1262 | return ret; | |
1263 | } | |
1264 | ||
9d85cba7 JM |
1265 | long do_io_submit(aio_context_t ctx_id, long nr, |
1266 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1267 | { |
1268 | struct kioctx *ctx; | |
1269 | long ret = 0; | |
080d676d | 1270 | int i = 0; |
9f5b9425 | 1271 | struct blk_plug plug; |
080d676d | 1272 | struct kiocb_batch batch; |
1da177e4 LT |
1273 | |
1274 | if (unlikely(nr < 0)) | |
1275 | return -EINVAL; | |
1276 | ||
75e1c70f JM |
1277 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1278 | nr = LONG_MAX/sizeof(*iocbpp); | |
1279 | ||
1da177e4 LT |
1280 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1281 | return -EFAULT; | |
1282 | ||
1283 | ctx = lookup_ioctx(ctx_id); | |
1284 | if (unlikely(!ctx)) { | |
caf4167a | 1285 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1286 | return -EINVAL; |
1287 | } | |
1288 | ||
080d676d JM |
1289 | kiocb_batch_init(&batch, nr); |
1290 | ||
9f5b9425 SL |
1291 | blk_start_plug(&plug); |
1292 | ||
1da177e4 LT |
1293 | /* |
1294 | * AKPM: should this return a partial result if some of the IOs were | |
1295 | * successfully submitted? | |
1296 | */ | |
1297 | for (i=0; i<nr; i++) { | |
1298 | struct iocb __user *user_iocb; | |
1299 | struct iocb tmp; | |
1300 | ||
1301 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1302 | ret = -EFAULT; | |
1303 | break; | |
1304 | } | |
1305 | ||
1306 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1307 | ret = -EFAULT; | |
1308 | break; | |
1309 | } | |
1310 | ||
080d676d | 1311 | ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat); |
1da177e4 LT |
1312 | if (ret) |
1313 | break; | |
1314 | } | |
9f5b9425 | 1315 | blk_finish_plug(&plug); |
1da177e4 | 1316 | |
69e4747e | 1317 | kiocb_batch_free(ctx, &batch); |
1da177e4 LT |
1318 | put_ioctx(ctx); |
1319 | return i ? i : ret; | |
1320 | } | |
1321 | ||
9d85cba7 JM |
1322 | /* sys_io_submit: |
1323 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1324 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1325 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1326 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1327 | * is invalid for the file descriptor in the iocb. May fail with | |
1328 | * -EFAULT if any of the data structures point to invalid data. May | |
1329 | * fail with -EBADF if the file descriptor specified in the first | |
1330 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1331 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1332 | * fail with -ENOSYS if not implemented. | |
1333 | */ | |
1334 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1335 | struct iocb __user * __user *, iocbpp) | |
1336 | { | |
1337 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1338 | } | |
1339 | ||
1da177e4 LT |
1340 | /* lookup_kiocb |
1341 | * Finds a given iocb for cancellation. | |
1da177e4 | 1342 | */ |
25ee7e38 AB |
1343 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1344 | u32 key) | |
1da177e4 LT |
1345 | { |
1346 | struct list_head *pos; | |
d00689af ZB |
1347 | |
1348 | assert_spin_locked(&ctx->ctx_lock); | |
1349 | ||
1da177e4 LT |
1350 | /* TODO: use a hash or array, this sucks. */ |
1351 | list_for_each(pos, &ctx->active_reqs) { | |
1352 | struct kiocb *kiocb = list_kiocb(pos); | |
1353 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) | |
1354 | return kiocb; | |
1355 | } | |
1356 | return NULL; | |
1357 | } | |
1358 | ||
1359 | /* sys_io_cancel: | |
1360 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1361 | * the operation is successfully cancelled, the resulting event is | |
1362 | * copied into the memory pointed to by result without being placed | |
1363 | * into the completion queue and 0 is returned. May fail with | |
1364 | * -EFAULT if any of the data structures pointed to are invalid. | |
1365 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1366 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1367 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1368 | */ | |
002c8976 HC |
1369 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1370 | struct io_event __user *, result) | |
1da177e4 | 1371 | { |
906b973c | 1372 | struct io_event res; |
1da177e4 LT |
1373 | struct kioctx *ctx; |
1374 | struct kiocb *kiocb; | |
1375 | u32 key; | |
1376 | int ret; | |
1377 | ||
1378 | ret = get_user(key, &iocb->aio_key); | |
1379 | if (unlikely(ret)) | |
1380 | return -EFAULT; | |
1381 | ||
1382 | ctx = lookup_ioctx(ctx_id); | |
1383 | if (unlikely(!ctx)) | |
1384 | return -EINVAL; | |
1385 | ||
1386 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1387 | |
1da177e4 | 1388 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c KO |
1389 | if (kiocb) |
1390 | ret = kiocb_cancel(ctx, kiocb, &res); | |
1391 | else | |
1392 | ret = -EINVAL; | |
1393 | ||
1da177e4 LT |
1394 | spin_unlock_irq(&ctx->ctx_lock); |
1395 | ||
906b973c KO |
1396 | if (!ret) { |
1397 | /* Cancellation succeeded -- copy the result | |
1398 | * into the user's buffer. | |
1399 | */ | |
1400 | if (copy_to_user(result, &res, sizeof(res))) | |
1401 | ret = -EFAULT; | |
1402 | } | |
1da177e4 LT |
1403 | |
1404 | put_ioctx(ctx); | |
1405 | ||
1406 | return ret; | |
1407 | } | |
1408 | ||
1409 | /* io_getevents: | |
1410 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1411 | * the completion queue for the aio_context specified by ctx_id. If |
1412 | * it succeeds, the number of read events is returned. May fail with | |
1413 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1414 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1415 | * if any of the memory specified is invalid. May return 0 or | |
1416 | * < min_nr if the timeout specified by timeout has elapsed | |
1417 | * before sufficient events are available, where timeout == NULL | |
1418 | * specifies an infinite timeout. Note that the timeout pointed to by | |
1419 | * timeout is relative and will be updated if not NULL and the | |
1420 | * operation blocks. Will fail with -ENOSYS if not implemented. | |
1da177e4 | 1421 | */ |
002c8976 HC |
1422 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1423 | long, min_nr, | |
1424 | long, nr, | |
1425 | struct io_event __user *, events, | |
1426 | struct timespec __user *, timeout) | |
1da177e4 LT |
1427 | { |
1428 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1429 | long ret = -EINVAL; | |
1430 | ||
1431 | if (likely(ioctx)) { | |
2e410255 | 1432 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 LT |
1433 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1434 | put_ioctx(ioctx); | |
1435 | } | |
1da177e4 LT |
1436 | return ret; |
1437 | } |