]>
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 | */ | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/errno.h> | |
14 | #include <linux/time.h> | |
15 | #include <linux/aio_abi.h> | |
630d9c47 | 16 | #include <linux/export.h> |
1da177e4 | 17 | #include <linux/syscalls.h> |
b9d128f1 | 18 | #include <linux/backing-dev.h> |
027445c3 | 19 | #include <linux/uio.h> |
1da177e4 LT |
20 | |
21 | #define DEBUG 0 | |
22 | ||
23 | #include <linux/sched.h> | |
24 | #include <linux/fs.h> | |
25 | #include <linux/file.h> | |
26 | #include <linux/mm.h> | |
27 | #include <linux/mman.h> | |
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 LT |
41 | |
42 | #if DEBUG > 1 | |
43 | #define dprintk printk | |
44 | #else | |
45 | #define dprintk(x...) do { ; } while (0) | |
46 | #endif | |
47 | ||
1da177e4 | 48 | /*------ sysctl variables----*/ |
d55b5fda ZB |
49 | static DEFINE_SPINLOCK(aio_nr_lock); |
50 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
51 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
52 | /*----end sysctl variables---*/ |
53 | ||
e18b890b CL |
54 | static struct kmem_cache *kiocb_cachep; |
55 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 56 | |
1da177e4 LT |
57 | /* aio_setup |
58 | * Creates the slab caches used by the aio routines, panic on | |
59 | * failure as this is done early during the boot sequence. | |
60 | */ | |
61 | static int __init aio_setup(void) | |
62 | { | |
0a31bd5f CL |
63 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
64 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 65 | |
1da177e4 LT |
66 | pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page)); |
67 | ||
68 | return 0; | |
69 | } | |
385773e0 | 70 | __initcall(aio_setup); |
1da177e4 LT |
71 | |
72 | static void aio_free_ring(struct kioctx *ctx) | |
73 | { | |
74 | struct aio_ring_info *info = &ctx->ring_info; | |
75 | long i; | |
76 | ||
77 | for (i=0; i<info->nr_pages; i++) | |
78 | put_page(info->ring_pages[i]); | |
79 | ||
936af157 | 80 | if (info->mmap_size) { |
bfce281c | 81 | vm_munmap(info->mmap_base, info->mmap_size); |
936af157 | 82 | } |
1da177e4 LT |
83 | |
84 | if (info->ring_pages && info->ring_pages != info->internal_pages) | |
85 | kfree(info->ring_pages); | |
86 | info->ring_pages = NULL; | |
87 | info->nr = 0; | |
88 | } | |
89 | ||
90 | static int aio_setup_ring(struct kioctx *ctx) | |
91 | { | |
92 | struct aio_ring *ring; | |
93 | struct aio_ring_info *info = &ctx->ring_info; | |
94 | unsigned nr_events = ctx->max_reqs; | |
41003a7b | 95 | struct mm_struct *mm = current->mm; |
41badc15 | 96 | unsigned long size, populate; |
1da177e4 LT |
97 | int nr_pages; |
98 | ||
99 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
100 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
101 | ||
102 | size = sizeof(struct aio_ring); | |
103 | size += sizeof(struct io_event) * nr_events; | |
104 | nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT; | |
105 | ||
106 | if (nr_pages < 0) | |
107 | return -EINVAL; | |
108 | ||
109 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event); | |
110 | ||
111 | info->nr = 0; | |
112 | info->ring_pages = info->internal_pages; | |
113 | if (nr_pages > AIO_RING_PAGES) { | |
11b0b5ab | 114 | info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
115 | if (!info->ring_pages) |
116 | return -ENOMEM; | |
1da177e4 LT |
117 | } |
118 | ||
119 | info->mmap_size = nr_pages * PAGE_SIZE; | |
120 | dprintk("attempting mmap of %lu bytes\n", info->mmap_size); | |
41003a7b | 121 | down_write(&mm->mmap_sem); |
e3fc629d AV |
122 | info->mmap_base = do_mmap_pgoff(NULL, 0, info->mmap_size, |
123 | PROT_READ|PROT_WRITE, | |
bebeb3d6 ML |
124 | MAP_ANONYMOUS|MAP_PRIVATE, 0, |
125 | &populate); | |
1da177e4 | 126 | if (IS_ERR((void *)info->mmap_base)) { |
41003a7b | 127 | up_write(&mm->mmap_sem); |
1da177e4 LT |
128 | info->mmap_size = 0; |
129 | aio_free_ring(ctx); | |
130 | return -EAGAIN; | |
131 | } | |
132 | ||
133 | dprintk("mmap address: 0x%08lx\n", info->mmap_base); | |
41003a7b | 134 | info->nr_pages = get_user_pages(current, mm, info->mmap_base, nr_pages, |
1da177e4 | 135 | 1, 0, info->ring_pages, NULL); |
41003a7b | 136 | up_write(&mm->mmap_sem); |
1da177e4 LT |
137 | |
138 | if (unlikely(info->nr_pages != nr_pages)) { | |
139 | aio_free_ring(ctx); | |
140 | return -EAGAIN; | |
141 | } | |
bebeb3d6 | 142 | if (populate) |
41badc15 | 143 | mm_populate(info->mmap_base, populate); |
1da177e4 LT |
144 | |
145 | ctx->user_id = info->mmap_base; | |
146 | ||
147 | info->nr = nr_events; /* trusted copy */ | |
148 | ||
e8e3c3d6 | 149 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
150 | ring->nr = nr_events; /* user copy */ |
151 | ring->id = ctx->user_id; | |
152 | ring->head = ring->tail = 0; | |
153 | ring->magic = AIO_RING_MAGIC; | |
154 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
155 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
156 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 157 | kunmap_atomic(ring); |
1da177e4 LT |
158 | |
159 | return 0; | |
160 | } | |
161 | ||
162 | ||
163 | /* aio_ring_event: returns a pointer to the event at the given index from | |
e8e3c3d6 | 164 | * kmap_atomic(). Release the pointer with put_aio_ring_event(); |
1da177e4 LT |
165 | */ |
166 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) | |
167 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
168 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
169 | ||
e8e3c3d6 | 170 | #define aio_ring_event(info, nr) ({ \ |
1da177e4 LT |
171 | unsigned pos = (nr) + AIO_EVENTS_OFFSET; \ |
172 | struct io_event *__event; \ | |
173 | __event = kmap_atomic( \ | |
e8e3c3d6 | 174 | (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE]); \ |
1da177e4 LT |
175 | __event += pos % AIO_EVENTS_PER_PAGE; \ |
176 | __event; \ | |
177 | }) | |
178 | ||
e8e3c3d6 | 179 | #define put_aio_ring_event(event) do { \ |
1da177e4 LT |
180 | struct io_event *__event = (event); \ |
181 | (void)__event; \ | |
e8e3c3d6 | 182 | kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK)); \ |
1da177e4 LT |
183 | } while(0) |
184 | ||
abf137dd JA |
185 | static void ctx_rcu_free(struct rcu_head *head) |
186 | { | |
187 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
abf137dd | 188 | kmem_cache_free(kioctx_cachep, ctx); |
abf137dd | 189 | } |
d5470b59 AB |
190 | |
191 | /* __put_ioctx | |
192 | * Called when the last user of an aio context has gone away, | |
193 | * and the struct needs to be freed. | |
194 | */ | |
195 | static void __put_ioctx(struct kioctx *ctx) | |
196 | { | |
2dd542b7 | 197 | unsigned nr_events = ctx->max_reqs; |
d5470b59 AB |
198 | BUG_ON(ctx->reqs_active); |
199 | ||
d5470b59 | 200 | aio_free_ring(ctx); |
2dd542b7 AV |
201 | if (nr_events) { |
202 | spin_lock(&aio_nr_lock); | |
203 | BUG_ON(aio_nr - nr_events > aio_nr); | |
204 | aio_nr -= nr_events; | |
205 | spin_unlock(&aio_nr_lock); | |
206 | } | |
d5470b59 | 207 | pr_debug("__put_ioctx: freeing %p\n", ctx); |
abf137dd | 208 | call_rcu(&ctx->rcu_head, ctx_rcu_free); |
d5470b59 AB |
209 | } |
210 | ||
3bd9a5d7 NP |
211 | static inline int try_get_ioctx(struct kioctx *kioctx) |
212 | { | |
213 | return atomic_inc_not_zero(&kioctx->users); | |
214 | } | |
215 | ||
216 | static inline void put_ioctx(struct kioctx *kioctx) | |
217 | { | |
218 | BUG_ON(atomic_read(&kioctx->users) <= 0); | |
219 | if (unlikely(atomic_dec_and_test(&kioctx->users))) | |
220 | __put_ioctx(kioctx); | |
221 | } | |
d5470b59 | 222 | |
1da177e4 LT |
223 | /* ioctx_alloc |
224 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
225 | */ | |
226 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
227 | { | |
41003a7b | 228 | struct mm_struct *mm = current->mm; |
1da177e4 | 229 | struct kioctx *ctx; |
e23754f8 | 230 | int err = -ENOMEM; |
1da177e4 LT |
231 | |
232 | /* Prevent overflows */ | |
233 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
234 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
235 | pr_debug("ENOMEM: nr_events too high\n"); | |
236 | return ERR_PTR(-EINVAL); | |
237 | } | |
238 | ||
2dd542b7 | 239 | if (!nr_events || (unsigned long)nr_events > aio_max_nr) |
1da177e4 LT |
240 | return ERR_PTR(-EAGAIN); |
241 | ||
c3762229 | 242 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
243 | if (!ctx) |
244 | return ERR_PTR(-ENOMEM); | |
245 | ||
1da177e4 | 246 | ctx->max_reqs = nr_events; |
1da177e4 | 247 | |
86b62a2c | 248 | atomic_set(&ctx->users, 2); |
1da177e4 LT |
249 | spin_lock_init(&ctx->ctx_lock); |
250 | spin_lock_init(&ctx->ring_info.ring_lock); | |
251 | init_waitqueue_head(&ctx->wait); | |
252 | ||
253 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 LT |
254 | |
255 | if (aio_setup_ring(ctx) < 0) | |
256 | goto out_freectx; | |
257 | ||
258 | /* limit the number of system wide aios */ | |
9fa1cb39 | 259 | spin_lock(&aio_nr_lock); |
2dd542b7 AV |
260 | if (aio_nr + nr_events > aio_max_nr || |
261 | aio_nr + nr_events < aio_nr) { | |
9fa1cb39 | 262 | spin_unlock(&aio_nr_lock); |
1da177e4 | 263 | goto out_cleanup; |
2dd542b7 AV |
264 | } |
265 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 266 | spin_unlock(&aio_nr_lock); |
1da177e4 | 267 | |
39fa0031 | 268 | /* now link into global list. */ |
abf137dd JA |
269 | spin_lock(&mm->ioctx_lock); |
270 | hlist_add_head_rcu(&ctx->list, &mm->ioctx_list); | |
271 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 LT |
272 | |
273 | dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", | |
41003a7b | 274 | ctx, ctx->user_id, mm, ctx->ring_info.nr); |
1da177e4 LT |
275 | return ctx; |
276 | ||
277 | out_cleanup: | |
e23754f8 AV |
278 | err = -EAGAIN; |
279 | aio_free_ring(ctx); | |
1da177e4 | 280 | out_freectx: |
1da177e4 | 281 | kmem_cache_free(kioctx_cachep, ctx); |
e23754f8 AV |
282 | dprintk("aio: error allocating ioctx %d\n", err); |
283 | return ERR_PTR(err); | |
1da177e4 LT |
284 | } |
285 | ||
06af121e | 286 | /* kill_ctx |
1da177e4 LT |
287 | * Cancels all outstanding aio requests on an aio context. Used |
288 | * when the processes owning a context have all exited to encourage | |
289 | * the rapid destruction of the kioctx. | |
290 | */ | |
06af121e | 291 | static void kill_ctx(struct kioctx *ctx) |
1da177e4 LT |
292 | { |
293 | int (*cancel)(struct kiocb *, struct io_event *); | |
06af121e AV |
294 | struct task_struct *tsk = current; |
295 | DECLARE_WAITQUEUE(wait, tsk); | |
1da177e4 | 296 | struct io_event res; |
06af121e | 297 | |
1da177e4 LT |
298 | spin_lock_irq(&ctx->ctx_lock); |
299 | ctx->dead = 1; | |
300 | while (!list_empty(&ctx->active_reqs)) { | |
301 | struct list_head *pos = ctx->active_reqs.next; | |
302 | struct kiocb *iocb = list_kiocb(pos); | |
303 | list_del_init(&iocb->ki_list); | |
304 | cancel = iocb->ki_cancel; | |
305 | kiocbSetCancelled(iocb); | |
306 | if (cancel) { | |
307 | iocb->ki_users++; | |
308 | spin_unlock_irq(&ctx->ctx_lock); | |
309 | cancel(iocb, &res); | |
310 | spin_lock_irq(&ctx->ctx_lock); | |
311 | } | |
312 | } | |
1da177e4 | 313 | |
1da177e4 | 314 | if (!ctx->reqs_active) |
dee11c23 | 315 | goto out; |
1da177e4 LT |
316 | |
317 | add_wait_queue(&ctx->wait, &wait); | |
318 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); | |
319 | while (ctx->reqs_active) { | |
dee11c23 | 320 | spin_unlock_irq(&ctx->ctx_lock); |
41d10da3 | 321 | io_schedule(); |
1da177e4 | 322 | set_task_state(tsk, TASK_UNINTERRUPTIBLE); |
dee11c23 | 323 | spin_lock_irq(&ctx->ctx_lock); |
1da177e4 LT |
324 | } |
325 | __set_task_state(tsk, TASK_RUNNING); | |
326 | remove_wait_queue(&ctx->wait, &wait); | |
dee11c23 KC |
327 | |
328 | out: | |
329 | spin_unlock_irq(&ctx->ctx_lock); | |
1da177e4 LT |
330 | } |
331 | ||
332 | /* wait_on_sync_kiocb: | |
333 | * Waits on the given sync kiocb to complete. | |
334 | */ | |
fc9b52cd | 335 | ssize_t wait_on_sync_kiocb(struct kiocb *iocb) |
1da177e4 LT |
336 | { |
337 | while (iocb->ki_users) { | |
338 | set_current_state(TASK_UNINTERRUPTIBLE); | |
339 | if (!iocb->ki_users) | |
340 | break; | |
41d10da3 | 341 | io_schedule(); |
1da177e4 LT |
342 | } |
343 | __set_current_state(TASK_RUNNING); | |
344 | return iocb->ki_user_data; | |
345 | } | |
385773e0 | 346 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 LT |
347 | |
348 | /* exit_aio: called when the last user of mm goes away. At this point, | |
349 | * there is no way for any new requests to be submited or any of the | |
350 | * io_* syscalls to be called on the context. However, there may be | |
351 | * outstanding requests which hold references to the context; as they | |
352 | * go away, they will call put_ioctx and release any pinned memory | |
353 | * associated with the request (held via struct page * references). | |
354 | */ | |
fc9b52cd | 355 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 356 | { |
abf137dd JA |
357 | struct kioctx *ctx; |
358 | ||
359 | while (!hlist_empty(&mm->ioctx_list)) { | |
360 | ctx = hlist_entry(mm->ioctx_list.first, struct kioctx, list); | |
361 | hlist_del_rcu(&ctx->list); | |
362 | ||
06af121e | 363 | kill_ctx(ctx); |
1da177e4 LT |
364 | |
365 | if (1 != atomic_read(&ctx->users)) | |
366 | printk(KERN_DEBUG | |
367 | "exit_aio:ioctx still alive: %d %d %d\n", | |
368 | atomic_read(&ctx->users), ctx->dead, | |
369 | ctx->reqs_active); | |
936af157 AV |
370 | /* |
371 | * We don't need to bother with munmap() here - | |
372 | * exit_mmap(mm) is coming and it'll unmap everything. | |
373 | * Since aio_free_ring() uses non-zero ->mmap_size | |
374 | * as indicator that it needs to unmap the area, | |
375 | * just set it to 0; aio_free_ring() is the only | |
376 | * place that uses ->mmap_size, so it's safe. | |
936af157 AV |
377 | */ |
378 | ctx->ring_info.mmap_size = 0; | |
1da177e4 | 379 | put_ioctx(ctx); |
1da177e4 LT |
380 | } |
381 | } | |
382 | ||
1da177e4 LT |
383 | /* aio_get_req |
384 | * Allocate a slot for an aio request. Increments the users count | |
385 | * of the kioctx so that the kioctx stays around until all requests are | |
386 | * complete. Returns NULL if no requests are free. | |
387 | * | |
388 | * Returns with kiocb->users set to 2. The io submit code path holds | |
389 | * an extra reference while submitting the i/o. | |
390 | * This prevents races between the aio code path referencing the | |
391 | * req (after submitting it) and aio_complete() freeing the req. | |
392 | */ | |
fc9b52cd | 393 | static struct kiocb *__aio_get_req(struct kioctx *ctx) |
1da177e4 LT |
394 | { |
395 | struct kiocb *req = NULL; | |
1da177e4 LT |
396 | |
397 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL); | |
398 | if (unlikely(!req)) | |
399 | return NULL; | |
400 | ||
4faa5285 | 401 | req->ki_flags = 0; |
1da177e4 LT |
402 | req->ki_users = 2; |
403 | req->ki_key = 0; | |
404 | req->ki_ctx = ctx; | |
405 | req->ki_cancel = NULL; | |
406 | req->ki_retry = NULL; | |
1da177e4 LT |
407 | req->ki_dtor = NULL; |
408 | req->private = NULL; | |
eed4e51f | 409 | req->ki_iovec = NULL; |
87c3a86e | 410 | req->ki_eventfd = NULL; |
1da177e4 | 411 | |
080d676d JM |
412 | return req; |
413 | } | |
414 | ||
415 | /* | |
416 | * struct kiocb's are allocated in batches to reduce the number of | |
417 | * times the ctx lock is acquired and released. | |
418 | */ | |
419 | #define KIOCB_BATCH_SIZE 32L | |
420 | struct kiocb_batch { | |
421 | struct list_head head; | |
422 | long count; /* number of requests left to allocate */ | |
423 | }; | |
424 | ||
425 | static void kiocb_batch_init(struct kiocb_batch *batch, long total) | |
426 | { | |
427 | INIT_LIST_HEAD(&batch->head); | |
428 | batch->count = total; | |
429 | } | |
430 | ||
69e4747e | 431 | static void kiocb_batch_free(struct kioctx *ctx, struct kiocb_batch *batch) |
080d676d JM |
432 | { |
433 | struct kiocb *req, *n; | |
434 | ||
69e4747e GN |
435 | if (list_empty(&batch->head)) |
436 | return; | |
437 | ||
438 | spin_lock_irq(&ctx->ctx_lock); | |
080d676d JM |
439 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { |
440 | list_del(&req->ki_batch); | |
69e4747e | 441 | list_del(&req->ki_list); |
080d676d | 442 | kmem_cache_free(kiocb_cachep, req); |
69e4747e | 443 | ctx->reqs_active--; |
080d676d | 444 | } |
880641bb JM |
445 | if (unlikely(!ctx->reqs_active && ctx->dead)) |
446 | wake_up_all(&ctx->wait); | |
69e4747e | 447 | spin_unlock_irq(&ctx->ctx_lock); |
080d676d JM |
448 | } |
449 | ||
450 | /* | |
451 | * Allocate a batch of kiocbs. This avoids taking and dropping the | |
452 | * context lock a lot during setup. | |
453 | */ | |
454 | static int kiocb_batch_refill(struct kioctx *ctx, struct kiocb_batch *batch) | |
455 | { | |
456 | unsigned short allocated, to_alloc; | |
457 | long avail; | |
080d676d JM |
458 | struct kiocb *req, *n; |
459 | struct aio_ring *ring; | |
460 | ||
461 | to_alloc = min(batch->count, KIOCB_BATCH_SIZE); | |
462 | for (allocated = 0; allocated < to_alloc; allocated++) { | |
463 | req = __aio_get_req(ctx); | |
464 | if (!req) | |
465 | /* allocation failed, go with what we've got */ | |
466 | break; | |
467 | list_add(&req->ki_batch, &batch->head); | |
468 | } | |
469 | ||
470 | if (allocated == 0) | |
471 | goto out; | |
472 | ||
1da177e4 | 473 | spin_lock_irq(&ctx->ctx_lock); |
080d676d JM |
474 | ring = kmap_atomic(ctx->ring_info.ring_pages[0]); |
475 | ||
476 | avail = aio_ring_avail(&ctx->ring_info, ring) - ctx->reqs_active; | |
477 | BUG_ON(avail < 0); | |
080d676d JM |
478 | if (avail < allocated) { |
479 | /* Trim back the number of requests. */ | |
480 | list_for_each_entry_safe(req, n, &batch->head, ki_batch) { | |
481 | list_del(&req->ki_batch); | |
482 | kmem_cache_free(kiocb_cachep, req); | |
483 | if (--allocated <= avail) | |
484 | break; | |
485 | } | |
486 | } | |
487 | ||
488 | batch->count -= allocated; | |
489 | list_for_each_entry(req, &batch->head, ki_batch) { | |
1da177e4 | 490 | list_add(&req->ki_list, &ctx->active_reqs); |
1da177e4 | 491 | ctx->reqs_active++; |
1da177e4 | 492 | } |
1da177e4 | 493 | |
080d676d JM |
494 | kunmap_atomic(ring); |
495 | spin_unlock_irq(&ctx->ctx_lock); | |
1da177e4 | 496 | |
080d676d JM |
497 | out: |
498 | return allocated; | |
1da177e4 LT |
499 | } |
500 | ||
080d676d JM |
501 | static inline struct kiocb *aio_get_req(struct kioctx *ctx, |
502 | struct kiocb_batch *batch) | |
1da177e4 LT |
503 | { |
504 | struct kiocb *req; | |
080d676d JM |
505 | |
506 | if (list_empty(&batch->head)) | |
507 | if (kiocb_batch_refill(ctx, batch) == 0) | |
508 | return NULL; | |
509 | req = list_first_entry(&batch->head, struct kiocb, ki_batch); | |
510 | list_del(&req->ki_batch); | |
1da177e4 LT |
511 | return req; |
512 | } | |
513 | ||
514 | static inline void really_put_req(struct kioctx *ctx, struct kiocb *req) | |
515 | { | |
d00689af ZB |
516 | assert_spin_locked(&ctx->ctx_lock); |
517 | ||
13389010 DL |
518 | if (req->ki_eventfd != NULL) |
519 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 LT |
520 | if (req->ki_dtor) |
521 | req->ki_dtor(req); | |
eed4e51f BP |
522 | if (req->ki_iovec != &req->ki_inline_vec) |
523 | kfree(req->ki_iovec); | |
1da177e4 LT |
524 | kmem_cache_free(kiocb_cachep, req); |
525 | ctx->reqs_active--; | |
526 | ||
527 | if (unlikely(!ctx->reqs_active && ctx->dead)) | |
e91f90bb | 528 | wake_up_all(&ctx->wait); |
1da177e4 LT |
529 | } |
530 | ||
1da177e4 LT |
531 | /* __aio_put_req |
532 | * Returns true if this put was the last user of the request. | |
533 | */ | |
2d68449e | 534 | static void __aio_put_req(struct kioctx *ctx, struct kiocb *req) |
1da177e4 | 535 | { |
516e0cc5 AV |
536 | dprintk(KERN_DEBUG "aio_put(%p): f_count=%ld\n", |
537 | req, atomic_long_read(&req->ki_filp->f_count)); | |
1da177e4 | 538 | |
d00689af ZB |
539 | assert_spin_locked(&ctx->ctx_lock); |
540 | ||
87c3a86e | 541 | req->ki_users--; |
93e06b41 | 542 | BUG_ON(req->ki_users < 0); |
1da177e4 | 543 | if (likely(req->ki_users)) |
2d68449e | 544 | return; |
1da177e4 LT |
545 | list_del(&req->ki_list); /* remove from active_reqs */ |
546 | req->ki_cancel = NULL; | |
547 | req->ki_retry = NULL; | |
548 | ||
3ffa3c0e AV |
549 | fput(req->ki_filp); |
550 | req->ki_filp = NULL; | |
551 | really_put_req(ctx, req); | |
1da177e4 LT |
552 | } |
553 | ||
554 | /* aio_put_req | |
555 | * Returns true if this put was the last user of the kiocb, | |
556 | * false if the request is still in use. | |
557 | */ | |
2d68449e | 558 | void aio_put_req(struct kiocb *req) |
1da177e4 LT |
559 | { |
560 | struct kioctx *ctx = req->ki_ctx; | |
1da177e4 | 561 | spin_lock_irq(&ctx->ctx_lock); |
2d68449e | 562 | __aio_put_req(ctx, req); |
1da177e4 | 563 | spin_unlock_irq(&ctx->ctx_lock); |
1da177e4 | 564 | } |
385773e0 | 565 | EXPORT_SYMBOL(aio_put_req); |
1da177e4 | 566 | |
d5470b59 | 567 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 568 | { |
abf137dd | 569 | struct mm_struct *mm = current->mm; |
65c24491 | 570 | struct kioctx *ctx, *ret = NULL; |
1da177e4 | 571 | |
abf137dd JA |
572 | rcu_read_lock(); |
573 | ||
b67bfe0d | 574 | hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) { |
3bd9a5d7 NP |
575 | /* |
576 | * RCU protects us against accessing freed memory but | |
577 | * we have to be careful not to get a reference when the | |
578 | * reference count already dropped to 0 (ctx->dead test | |
579 | * is unreliable because of races). | |
580 | */ | |
581 | if (ctx->user_id == ctx_id && !ctx->dead && try_get_ioctx(ctx)){ | |
65c24491 | 582 | ret = ctx; |
1da177e4 LT |
583 | break; |
584 | } | |
abf137dd | 585 | } |
1da177e4 | 586 | |
abf137dd | 587 | rcu_read_unlock(); |
65c24491 | 588 | return ret; |
1da177e4 LT |
589 | } |
590 | ||
1da177e4 LT |
591 | /* aio_complete |
592 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 593 | */ |
2d68449e | 594 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
595 | { |
596 | struct kioctx *ctx = iocb->ki_ctx; | |
597 | struct aio_ring_info *info; | |
598 | struct aio_ring *ring; | |
599 | struct io_event *event; | |
600 | unsigned long flags; | |
601 | unsigned long tail; | |
1da177e4 | 602 | |
20dcae32 ZB |
603 | /* |
604 | * Special case handling for sync iocbs: | |
605 | * - events go directly into the iocb for fast handling | |
606 | * - the sync task with the iocb in its stack holds the single iocb | |
607 | * ref, no other paths have a way to get another ref | |
608 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
609 | */ |
610 | if (is_sync_kiocb(iocb)) { | |
20dcae32 | 611 | BUG_ON(iocb->ki_users != 1); |
1da177e4 | 612 | iocb->ki_user_data = res; |
20dcae32 | 613 | iocb->ki_users = 0; |
1da177e4 | 614 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 615 | return; |
1da177e4 LT |
616 | } |
617 | ||
618 | info = &ctx->ring_info; | |
619 | ||
620 | /* add a completion event to the ring buffer. | |
621 | * must be done holding ctx->ctx_lock to prevent | |
622 | * other code from messing with the tail | |
623 | * pointer since we might be called from irq | |
624 | * context. | |
625 | */ | |
626 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
627 | ||
1da177e4 LT |
628 | /* |
629 | * cancelled requests don't get events, userland was given one | |
630 | * when the event got cancelled. | |
631 | */ | |
632 | if (kiocbIsCancelled(iocb)) | |
633 | goto put_rq; | |
634 | ||
e8e3c3d6 | 635 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
636 | |
637 | tail = info->tail; | |
e8e3c3d6 | 638 | event = aio_ring_event(info, tail); |
4bf69b2a KC |
639 | if (++tail >= info->nr) |
640 | tail = 0; | |
1da177e4 LT |
641 | |
642 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; | |
643 | event->data = iocb->ki_user_data; | |
644 | event->res = res; | |
645 | event->res2 = res2; | |
646 | ||
647 | dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n", | |
648 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, | |
649 | res, res2); | |
650 | ||
651 | /* after flagging the request as done, we | |
652 | * must never even look at it again | |
653 | */ | |
654 | smp_wmb(); /* make event visible before updating tail */ | |
655 | ||
656 | info->tail = tail; | |
657 | ring->tail = tail; | |
658 | ||
e8e3c3d6 CW |
659 | put_aio_ring_event(event); |
660 | kunmap_atomic(ring); | |
1da177e4 LT |
661 | |
662 | pr_debug("added to ring %p at [%lu]\n", iocb, tail); | |
8d1c98b0 DL |
663 | |
664 | /* | |
665 | * Check if the user asked us to deliver the result through an | |
666 | * eventfd. The eventfd_signal() function is safe to be called | |
667 | * from IRQ context. | |
668 | */ | |
87c3a86e | 669 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
670 | eventfd_signal(iocb->ki_eventfd, 1); |
671 | ||
1da177e4 LT |
672 | put_rq: |
673 | /* everything turned out well, dispose of the aiocb. */ | |
2d68449e | 674 | __aio_put_req(ctx, iocb); |
1da177e4 | 675 | |
6cb2a210 QB |
676 | /* |
677 | * We have to order our ring_info tail store above and test | |
678 | * of the wait list below outside the wait lock. This is | |
679 | * like in wake_up_bit() where clearing a bit has to be | |
680 | * ordered with the unlocked test. | |
681 | */ | |
682 | smp_mb(); | |
683 | ||
1da177e4 LT |
684 | if (waitqueue_active(&ctx->wait)) |
685 | wake_up(&ctx->wait); | |
686 | ||
dee11c23 | 687 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); |
1da177e4 | 688 | } |
385773e0 | 689 | EXPORT_SYMBOL(aio_complete); |
1da177e4 LT |
690 | |
691 | /* aio_read_evt | |
692 | * Pull an event off of the ioctx's event ring. Returns the number of | |
693 | * events fetched (0 or 1 ;-) | |
694 | * FIXME: make this use cmpxchg. | |
695 | * TODO: make the ringbuffer user mmap()able (requires FIXME). | |
696 | */ | |
697 | static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent) | |
698 | { | |
699 | struct aio_ring_info *info = &ioctx->ring_info; | |
700 | struct aio_ring *ring; | |
701 | unsigned long head; | |
702 | int ret = 0; | |
703 | ||
e8e3c3d6 | 704 | ring = kmap_atomic(info->ring_pages[0]); |
1da177e4 LT |
705 | dprintk("in aio_read_evt h%lu t%lu m%lu\n", |
706 | (unsigned long)ring->head, (unsigned long)ring->tail, | |
707 | (unsigned long)ring->nr); | |
708 | ||
709 | if (ring->head == ring->tail) | |
710 | goto out; | |
711 | ||
712 | spin_lock(&info->ring_lock); | |
713 | ||
714 | head = ring->head % info->nr; | |
715 | if (head != ring->tail) { | |
e8e3c3d6 | 716 | struct io_event *evp = aio_ring_event(info, head); |
1da177e4 LT |
717 | *ent = *evp; |
718 | head = (head + 1) % info->nr; | |
719 | smp_mb(); /* finish reading the event before updatng the head */ | |
720 | ring->head = head; | |
721 | ret = 1; | |
e8e3c3d6 | 722 | put_aio_ring_event(evp); |
1da177e4 LT |
723 | } |
724 | spin_unlock(&info->ring_lock); | |
725 | ||
726 | out: | |
1da177e4 LT |
727 | dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret, |
728 | (unsigned long)ring->head, (unsigned long)ring->tail); | |
91d80a84 | 729 | kunmap_atomic(ring); |
1da177e4 LT |
730 | return ret; |
731 | } | |
732 | ||
733 | struct aio_timeout { | |
734 | struct timer_list timer; | |
735 | int timed_out; | |
736 | struct task_struct *p; | |
737 | }; | |
738 | ||
739 | static void timeout_func(unsigned long data) | |
740 | { | |
741 | struct aio_timeout *to = (struct aio_timeout *)data; | |
742 | ||
743 | to->timed_out = 1; | |
744 | wake_up_process(to->p); | |
745 | } | |
746 | ||
747 | static inline void init_timeout(struct aio_timeout *to) | |
748 | { | |
c6f3a97f | 749 | setup_timer_on_stack(&to->timer, timeout_func, (unsigned long) to); |
1da177e4 LT |
750 | to->timed_out = 0; |
751 | to->p = current; | |
752 | } | |
753 | ||
754 | static inline void set_timeout(long start_jiffies, struct aio_timeout *to, | |
755 | const struct timespec *ts) | |
756 | { | |
757 | to->timer.expires = start_jiffies + timespec_to_jiffies(ts); | |
758 | if (time_after(to->timer.expires, jiffies)) | |
759 | add_timer(&to->timer); | |
760 | else | |
761 | to->timed_out = 1; | |
762 | } | |
763 | ||
764 | static inline void clear_timeout(struct aio_timeout *to) | |
765 | { | |
766 | del_singleshot_timer_sync(&to->timer); | |
767 | } | |
768 | ||
769 | static int read_events(struct kioctx *ctx, | |
770 | long min_nr, long nr, | |
771 | struct io_event __user *event, | |
772 | struct timespec __user *timeout) | |
773 | { | |
774 | long start_jiffies = jiffies; | |
775 | struct task_struct *tsk = current; | |
776 | DECLARE_WAITQUEUE(wait, tsk); | |
777 | int ret; | |
778 | int i = 0; | |
779 | struct io_event ent; | |
780 | struct aio_timeout to; | |
1da177e4 LT |
781 | |
782 | /* needed to zero any padding within an entry (there shouldn't be | |
783 | * any, but C is fun! | |
784 | */ | |
785 | memset(&ent, 0, sizeof(ent)); | |
1da177e4 LT |
786 | ret = 0; |
787 | while (likely(i < nr)) { | |
788 | ret = aio_read_evt(ctx, &ent); | |
789 | if (unlikely(ret <= 0)) | |
790 | break; | |
791 | ||
792 | dprintk("read event: %Lx %Lx %Lx %Lx\n", | |
793 | ent.data, ent.obj, ent.res, ent.res2); | |
794 | ||
795 | /* Could we split the check in two? */ | |
796 | ret = -EFAULT; | |
797 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { | |
798 | dprintk("aio: lost an event due to EFAULT.\n"); | |
799 | break; | |
800 | } | |
801 | ret = 0; | |
802 | ||
803 | /* Good, event copied to userland, update counts. */ | |
804 | event ++; | |
805 | i ++; | |
806 | } | |
807 | ||
808 | if (min_nr <= i) | |
809 | return i; | |
810 | if (ret) | |
811 | return ret; | |
812 | ||
813 | /* End fast path */ | |
814 | ||
1da177e4 LT |
815 | init_timeout(&to); |
816 | if (timeout) { | |
817 | struct timespec ts; | |
818 | ret = -EFAULT; | |
819 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) | |
820 | goto out; | |
821 | ||
822 | set_timeout(start_jiffies, &to, &ts); | |
823 | } | |
824 | ||
825 | while (likely(i < nr)) { | |
826 | add_wait_queue_exclusive(&ctx->wait, &wait); | |
827 | do { | |
828 | set_task_state(tsk, TASK_INTERRUPTIBLE); | |
829 | ret = aio_read_evt(ctx, &ent); | |
830 | if (ret) | |
831 | break; | |
832 | if (min_nr <= i) | |
833 | break; | |
e92adcba JM |
834 | if (unlikely(ctx->dead)) { |
835 | ret = -EINVAL; | |
836 | break; | |
837 | } | |
1da177e4 LT |
838 | if (to.timed_out) /* Only check after read evt */ |
839 | break; | |
e00ba3da JM |
840 | /* Try to only show up in io wait if there are ops |
841 | * in flight */ | |
842 | if (ctx->reqs_active) | |
843 | io_schedule(); | |
844 | else | |
845 | schedule(); | |
1da177e4 LT |
846 | if (signal_pending(tsk)) { |
847 | ret = -EINTR; | |
848 | break; | |
849 | } | |
850 | /*ret = aio_read_evt(ctx, &ent);*/ | |
851 | } while (1) ; | |
852 | ||
853 | set_task_state(tsk, TASK_RUNNING); | |
854 | remove_wait_queue(&ctx->wait, &wait); | |
855 | ||
856 | if (unlikely(ret <= 0)) | |
857 | break; | |
858 | ||
859 | ret = -EFAULT; | |
860 | if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) { | |
861 | dprintk("aio: lost an event due to EFAULT.\n"); | |
862 | break; | |
863 | } | |
864 | ||
865 | /* Good, event copied to userland, update counts. */ | |
866 | event ++; | |
867 | i ++; | |
868 | } | |
869 | ||
870 | if (timeout) | |
871 | clear_timeout(&to); | |
872 | out: | |
c6f3a97f | 873 | destroy_timer_on_stack(&to.timer); |
1da177e4 LT |
874 | return i ? i : ret; |
875 | } | |
876 | ||
877 | /* Take an ioctx and remove it from the list of ioctx's. Protects | |
878 | * against races with itself via ->dead. | |
879 | */ | |
880 | static void io_destroy(struct kioctx *ioctx) | |
881 | { | |
882 | struct mm_struct *mm = current->mm; | |
1da177e4 LT |
883 | int was_dead; |
884 | ||
885 | /* delete the entry from the list is someone else hasn't already */ | |
abf137dd | 886 | spin_lock(&mm->ioctx_lock); |
1da177e4 LT |
887 | was_dead = ioctx->dead; |
888 | ioctx->dead = 1; | |
abf137dd JA |
889 | hlist_del_rcu(&ioctx->list); |
890 | spin_unlock(&mm->ioctx_lock); | |
1da177e4 LT |
891 | |
892 | dprintk("aio_release(%p)\n", ioctx); | |
893 | if (likely(!was_dead)) | |
894 | put_ioctx(ioctx); /* twice for the list */ | |
895 | ||
06af121e | 896 | kill_ctx(ioctx); |
e92adcba JM |
897 | |
898 | /* | |
899 | * Wake up any waiters. The setting of ctx->dead must be seen | |
900 | * by other CPUs at this point. Right now, we rely on the | |
901 | * locking done by the above calls to ensure this consistency. | |
902 | */ | |
e91f90bb | 903 | wake_up_all(&ioctx->wait); |
1da177e4 LT |
904 | } |
905 | ||
906 | /* sys_io_setup: | |
907 | * Create an aio_context capable of receiving at least nr_events. | |
908 | * ctxp must not point to an aio_context that already exists, and | |
909 | * must be initialized to 0 prior to the call. On successful | |
910 | * creation of the aio_context, *ctxp is filled in with the resulting | |
911 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
912 | * if the specified nr_events exceeds internal limits. May fail | |
913 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
914 | * of available events. May fail with -ENOMEM if insufficient kernel | |
915 | * resources are available. May fail with -EFAULT if an invalid | |
916 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
917 | * implemented. | |
918 | */ | |
002c8976 | 919 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
920 | { |
921 | struct kioctx *ioctx = NULL; | |
922 | unsigned long ctx; | |
923 | long ret; | |
924 | ||
925 | ret = get_user(ctx, ctxp); | |
926 | if (unlikely(ret)) | |
927 | goto out; | |
928 | ||
929 | ret = -EINVAL; | |
d55b5fda ZB |
930 | if (unlikely(ctx || nr_events == 0)) { |
931 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
932 | ctx, nr_events); | |
1da177e4 LT |
933 | goto out; |
934 | } | |
935 | ||
936 | ioctx = ioctx_alloc(nr_events); | |
937 | ret = PTR_ERR(ioctx); | |
938 | if (!IS_ERR(ioctx)) { | |
939 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a AV |
940 | if (ret) |
941 | io_destroy(ioctx); | |
942 | put_ioctx(ioctx); | |
1da177e4 LT |
943 | } |
944 | ||
945 | out: | |
946 | return ret; | |
947 | } | |
948 | ||
949 | /* sys_io_destroy: | |
950 | * Destroy the aio_context specified. May cancel any outstanding | |
951 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 952 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
953 | * is invalid. |
954 | */ | |
002c8976 | 955 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
956 | { |
957 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
958 | if (likely(NULL != ioctx)) { | |
959 | io_destroy(ioctx); | |
a2e1859a | 960 | put_ioctx(ioctx); |
1da177e4 LT |
961 | return 0; |
962 | } | |
963 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
964 | return -EINVAL; | |
965 | } | |
966 | ||
eed4e51f | 967 | static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret) |
1da177e4 | 968 | { |
eed4e51f BP |
969 | struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg]; |
970 | ||
971 | BUG_ON(ret <= 0); | |
972 | ||
973 | while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) { | |
974 | ssize_t this = min((ssize_t)iov->iov_len, ret); | |
975 | iov->iov_base += this; | |
976 | iov->iov_len -= this; | |
977 | iocb->ki_left -= this; | |
978 | ret -= this; | |
979 | if (iov->iov_len == 0) { | |
980 | iocb->ki_cur_seg++; | |
981 | iov++; | |
897f15fb | 982 | } |
eed4e51f | 983 | } |
1da177e4 | 984 | |
eed4e51f BP |
985 | /* the caller should not have done more io than what fit in |
986 | * the remaining iovecs */ | |
987 | BUG_ON(ret > 0 && iocb->ki_left == 0); | |
1da177e4 LT |
988 | } |
989 | ||
eed4e51f | 990 | static ssize_t aio_rw_vect_retry(struct kiocb *iocb) |
1da177e4 LT |
991 | { |
992 | struct file *file = iocb->ki_filp; | |
eed4e51f BP |
993 | struct address_space *mapping = file->f_mapping; |
994 | struct inode *inode = mapping->host; | |
995 | ssize_t (*rw_op)(struct kiocb *, const struct iovec *, | |
996 | unsigned long, loff_t); | |
1da177e4 | 997 | ssize_t ret = 0; |
eed4e51f BP |
998 | unsigned short opcode; |
999 | ||
1000 | if ((iocb->ki_opcode == IOCB_CMD_PREADV) || | |
1001 | (iocb->ki_opcode == IOCB_CMD_PREAD)) { | |
1002 | rw_op = file->f_op->aio_read; | |
1003 | opcode = IOCB_CMD_PREADV; | |
1004 | } else { | |
1005 | rw_op = file->f_op->aio_write; | |
1006 | opcode = IOCB_CMD_PWRITEV; | |
1007 | } | |
1da177e4 | 1008 | |
c2ec6682 RR |
1009 | /* This matches the pread()/pwrite() logic */ |
1010 | if (iocb->ki_pos < 0) | |
1011 | return -EINVAL; | |
1012 | ||
8d71db4f AV |
1013 | if (opcode == IOCB_CMD_PWRITEV) |
1014 | file_start_write(file); | |
897f15fb | 1015 | do { |
eed4e51f BP |
1016 | ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg], |
1017 | iocb->ki_nr_segs - iocb->ki_cur_seg, | |
1018 | iocb->ki_pos); | |
1019 | if (ret > 0) | |
1020 | aio_advance_iovec(iocb, ret); | |
1021 | ||
1022 | /* retry all partial writes. retry partial reads as long as its a | |
1023 | * regular file. */ | |
1024 | } while (ret > 0 && iocb->ki_left > 0 && | |
1025 | (opcode == IOCB_CMD_PWRITEV || | |
1026 | (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode)))); | |
8d71db4f AV |
1027 | if (opcode == IOCB_CMD_PWRITEV) |
1028 | file_end_write(file); | |
1da177e4 | 1029 | |
eed4e51f BP |
1030 | /* This means we must have transferred all that we could */ |
1031 | /* No need to retry anymore */ | |
1da177e4 LT |
1032 | if ((ret == 0) || (iocb->ki_left == 0)) |
1033 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1034 | ||
7adfa2ff RR |
1035 | /* If we managed to write some out we return that, rather than |
1036 | * the eventual error. */ | |
1037 | if (opcode == IOCB_CMD_PWRITEV | |
41003a7b | 1038 | && ret < 0 && ret != -EIOCBQUEUED |
7adfa2ff RR |
1039 | && iocb->ki_nbytes - iocb->ki_left) |
1040 | ret = iocb->ki_nbytes - iocb->ki_left; | |
1041 | ||
1da177e4 LT |
1042 | return ret; |
1043 | } | |
1044 | ||
1045 | static ssize_t aio_fdsync(struct kiocb *iocb) | |
1046 | { | |
1047 | struct file *file = iocb->ki_filp; | |
1048 | ssize_t ret = -EINVAL; | |
1049 | ||
1050 | if (file->f_op->aio_fsync) | |
1051 | ret = file->f_op->aio_fsync(iocb, 1); | |
1052 | return ret; | |
1053 | } | |
1054 | ||
1055 | static ssize_t aio_fsync(struct kiocb *iocb) | |
1056 | { | |
1057 | struct file *file = iocb->ki_filp; | |
1058 | ssize_t ret = -EINVAL; | |
1059 | ||
1060 | if (file->f_op->aio_fsync) | |
1061 | ret = file->f_op->aio_fsync(iocb, 0); | |
1062 | return ret; | |
1063 | } | |
1064 | ||
9d85cba7 | 1065 | static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat) |
eed4e51f BP |
1066 | { |
1067 | ssize_t ret; | |
1068 | ||
9d85cba7 JM |
1069 | #ifdef CONFIG_COMPAT |
1070 | if (compat) | |
1071 | ret = compat_rw_copy_check_uvector(type, | |
1072 | (struct compat_iovec __user *)kiocb->ki_buf, | |
1073 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1074 | &kiocb->ki_iovec); |
9d85cba7 JM |
1075 | else |
1076 | #endif | |
1077 | ret = rw_copy_check_uvector(type, | |
1078 | (struct iovec __user *)kiocb->ki_buf, | |
1079 | kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec, | |
ac34ebb3 | 1080 | &kiocb->ki_iovec); |
eed4e51f BP |
1081 | if (ret < 0) |
1082 | goto out; | |
1083 | ||
a70b52ec LT |
1084 | ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret); |
1085 | if (ret < 0) | |
1086 | goto out; | |
1087 | ||
eed4e51f BP |
1088 | kiocb->ki_nr_segs = kiocb->ki_nbytes; |
1089 | kiocb->ki_cur_seg = 0; | |
1090 | /* ki_nbytes/left now reflect bytes instead of segs */ | |
1091 | kiocb->ki_nbytes = ret; | |
1092 | kiocb->ki_left = ret; | |
1093 | ||
1094 | ret = 0; | |
1095 | out: | |
1096 | return ret; | |
1097 | } | |
1098 | ||
a70b52ec | 1099 | static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb) |
eed4e51f | 1100 | { |
a70b52ec LT |
1101 | int bytes; |
1102 | ||
1103 | bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left); | |
1104 | if (bytes < 0) | |
1105 | return bytes; | |
1106 | ||
eed4e51f BP |
1107 | kiocb->ki_iovec = &kiocb->ki_inline_vec; |
1108 | kiocb->ki_iovec->iov_base = kiocb->ki_buf; | |
a70b52ec | 1109 | kiocb->ki_iovec->iov_len = bytes; |
eed4e51f BP |
1110 | kiocb->ki_nr_segs = 1; |
1111 | kiocb->ki_cur_seg = 0; | |
eed4e51f BP |
1112 | return 0; |
1113 | } | |
1114 | ||
1da177e4 LT |
1115 | /* |
1116 | * aio_setup_iocb: | |
1117 | * Performs the initial checks and aio retry method | |
1118 | * setup for the kiocb at the time of io submission. | |
1119 | */ | |
9d85cba7 | 1120 | static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat) |
1da177e4 LT |
1121 | { |
1122 | struct file *file = kiocb->ki_filp; | |
1123 | ssize_t ret = 0; | |
1124 | ||
1125 | switch (kiocb->ki_opcode) { | |
1126 | case IOCB_CMD_PREAD: | |
1127 | ret = -EBADF; | |
1128 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1129 | break; | |
1130 | ret = -EFAULT; | |
1131 | if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf, | |
1132 | kiocb->ki_left))) | |
1133 | break; | |
a70b52ec | 1134 | ret = aio_setup_single_vector(READ, file, kiocb); |
eed4e51f BP |
1135 | if (ret) |
1136 | break; | |
1da177e4 LT |
1137 | ret = -EINVAL; |
1138 | if (file->f_op->aio_read) | |
eed4e51f | 1139 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1140 | break; |
1141 | case IOCB_CMD_PWRITE: | |
1142 | ret = -EBADF; | |
1143 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1144 | break; | |
1145 | ret = -EFAULT; | |
1146 | if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf, | |
1147 | kiocb->ki_left))) | |
1148 | break; | |
a70b52ec | 1149 | ret = aio_setup_single_vector(WRITE, file, kiocb); |
eed4e51f BP |
1150 | if (ret) |
1151 | break; | |
1152 | ret = -EINVAL; | |
1153 | if (file->f_op->aio_write) | |
1154 | kiocb->ki_retry = aio_rw_vect_retry; | |
1155 | break; | |
1156 | case IOCB_CMD_PREADV: | |
1157 | ret = -EBADF; | |
1158 | if (unlikely(!(file->f_mode & FMODE_READ))) | |
1159 | break; | |
9d85cba7 | 1160 | ret = aio_setup_vectored_rw(READ, kiocb, compat); |
eed4e51f BP |
1161 | if (ret) |
1162 | break; | |
1163 | ret = -EINVAL; | |
1164 | if (file->f_op->aio_read) | |
1165 | kiocb->ki_retry = aio_rw_vect_retry; | |
1166 | break; | |
1167 | case IOCB_CMD_PWRITEV: | |
1168 | ret = -EBADF; | |
1169 | if (unlikely(!(file->f_mode & FMODE_WRITE))) | |
1170 | break; | |
9d85cba7 | 1171 | ret = aio_setup_vectored_rw(WRITE, kiocb, compat); |
eed4e51f BP |
1172 | if (ret) |
1173 | break; | |
1da177e4 LT |
1174 | ret = -EINVAL; |
1175 | if (file->f_op->aio_write) | |
eed4e51f | 1176 | kiocb->ki_retry = aio_rw_vect_retry; |
1da177e4 LT |
1177 | break; |
1178 | case IOCB_CMD_FDSYNC: | |
1179 | ret = -EINVAL; | |
1180 | if (file->f_op->aio_fsync) | |
1181 | kiocb->ki_retry = aio_fdsync; | |
1182 | break; | |
1183 | case IOCB_CMD_FSYNC: | |
1184 | ret = -EINVAL; | |
1185 | if (file->f_op->aio_fsync) | |
1186 | kiocb->ki_retry = aio_fsync; | |
1187 | break; | |
1188 | default: | |
1189 | dprintk("EINVAL: io_submit: no operation provided\n"); | |
1190 | ret = -EINVAL; | |
1191 | } | |
1192 | ||
1193 | if (!kiocb->ki_retry) | |
1194 | return ret; | |
1195 | ||
1196 | return 0; | |
1197 | } | |
1198 | ||
d5470b59 | 1199 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
080d676d JM |
1200 | struct iocb *iocb, struct kiocb_batch *batch, |
1201 | bool compat) | |
1da177e4 LT |
1202 | { |
1203 | struct kiocb *req; | |
1204 | struct file *file; | |
1205 | ssize_t ret; | |
1206 | ||
1207 | /* enforce forwards compatibility on users */ | |
9c3060be | 1208 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
1da177e4 LT |
1209 | pr_debug("EINVAL: io_submit: reserve field set\n"); |
1210 | return -EINVAL; | |
1211 | } | |
1212 | ||
1213 | /* prevent overflows */ | |
1214 | if (unlikely( | |
1215 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1216 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1217 | ((ssize_t)iocb->aio_nbytes < 0) | |
1218 | )) { | |
1219 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1220 | return -EINVAL; | |
1221 | } | |
1222 | ||
1223 | file = fget(iocb->aio_fildes); | |
1224 | if (unlikely(!file)) | |
1225 | return -EBADF; | |
1226 | ||
080d676d | 1227 | req = aio_get_req(ctx, batch); /* returns with 2 references to req */ |
1da177e4 LT |
1228 | if (unlikely(!req)) { |
1229 | fput(file); | |
1230 | return -EAGAIN; | |
1231 | } | |
87e2831c | 1232 | req->ki_filp = file; |
9c3060be DL |
1233 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1234 | /* | |
1235 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1236 | * instance of the file* now. The file descriptor must be | |
1237 | * an eventfd() fd, and will be signaled for each completed | |
1238 | * event using the eventfd_signal() function. | |
1239 | */ | |
13389010 | 1240 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1241 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1242 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1243 | req->ki_eventfd = NULL; |
9c3060be DL |
1244 | goto out_put_req; |
1245 | } | |
1246 | } | |
1da177e4 | 1247 | |
212079cf | 1248 | ret = put_user(req->ki_key, &user_iocb->aio_key); |
1da177e4 LT |
1249 | if (unlikely(ret)) { |
1250 | dprintk("EFAULT: aio_key\n"); | |
1251 | goto out_put_req; | |
1252 | } | |
1253 | ||
1254 | req->ki_obj.user = user_iocb; | |
1255 | req->ki_user_data = iocb->aio_data; | |
1256 | req->ki_pos = iocb->aio_offset; | |
1257 | ||
1258 | req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf; | |
1259 | req->ki_left = req->ki_nbytes = iocb->aio_nbytes; | |
1260 | req->ki_opcode = iocb->aio_lio_opcode; | |
1da177e4 | 1261 | |
9d85cba7 | 1262 | ret = aio_setup_iocb(req, compat); |
1da177e4 LT |
1263 | |
1264 | if (ret) | |
1265 | goto out_put_req; | |
1266 | ||
1267 | spin_lock_irq(&ctx->ctx_lock); | |
7137c6bd JK |
1268 | /* |
1269 | * We could have raced with io_destroy() and are currently holding a | |
1270 | * reference to ctx which should be destroyed. We cannot submit IO | |
1271 | * since ctx gets freed as soon as io_submit() puts its reference. The | |
1272 | * check here is reliable: io_destroy() sets ctx->dead before waiting | |
1273 | * for outstanding IO and the barrier between these two is realized by | |
1274 | * unlock of mm->ioctx_lock and lock of ctx->ctx_lock. Analogously we | |
1275 | * increment ctx->reqs_active before checking for ctx->dead and the | |
1276 | * barrier is realized by unlock and lock of ctx->ctx_lock. Thus if we | |
1277 | * don't see ctx->dead set here, io_destroy() waits for our IO to | |
1278 | * finish. | |
1279 | */ | |
41003a7b | 1280 | if (ctx->dead) |
7137c6bd | 1281 | ret = -EINVAL; |
41003a7b ZB |
1282 | spin_unlock_irq(&ctx->ctx_lock); |
1283 | if (ret) | |
7137c6bd | 1284 | goto out_put_req; |
41003a7b ZB |
1285 | |
1286 | if (unlikely(kiocbIsCancelled(req))) | |
1287 | ret = -EINTR; | |
1288 | else | |
1289 | ret = req->ki_retry(req); | |
1290 | ||
1291 | if (ret != -EIOCBQUEUED) { | |
1292 | /* | |
1293 | * There's no easy way to restart the syscall since other AIO's | |
1294 | * may be already running. Just fail this IO with EINTR. | |
1295 | */ | |
1296 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1297 | ret == -ERESTARTNOHAND || | |
1298 | ret == -ERESTART_RESTARTBLOCK)) | |
1299 | ret = -EINTR; | |
1300 | aio_complete(req, ret, 0); | |
7137c6bd | 1301 | } |
cfb1e33e | 1302 | |
1da177e4 LT |
1303 | aio_put_req(req); /* drop extra ref to req */ |
1304 | return 0; | |
1305 | ||
1306 | out_put_req: | |
1307 | aio_put_req(req); /* drop extra ref to req */ | |
1308 | aio_put_req(req); /* drop i/o ref to req */ | |
1309 | return ret; | |
1310 | } | |
1311 | ||
9d85cba7 JM |
1312 | long do_io_submit(aio_context_t ctx_id, long nr, |
1313 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1314 | { |
1315 | struct kioctx *ctx; | |
1316 | long ret = 0; | |
080d676d | 1317 | int i = 0; |
9f5b9425 | 1318 | struct blk_plug plug; |
080d676d | 1319 | struct kiocb_batch batch; |
1da177e4 LT |
1320 | |
1321 | if (unlikely(nr < 0)) | |
1322 | return -EINVAL; | |
1323 | ||
75e1c70f JM |
1324 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1325 | nr = LONG_MAX/sizeof(*iocbpp); | |
1326 | ||
1da177e4 LT |
1327 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1328 | return -EFAULT; | |
1329 | ||
1330 | ctx = lookup_ioctx(ctx_id); | |
1331 | if (unlikely(!ctx)) { | |
1332 | pr_debug("EINVAL: io_submit: invalid context id\n"); | |
1333 | return -EINVAL; | |
1334 | } | |
1335 | ||
080d676d JM |
1336 | kiocb_batch_init(&batch, nr); |
1337 | ||
9f5b9425 SL |
1338 | blk_start_plug(&plug); |
1339 | ||
1da177e4 LT |
1340 | /* |
1341 | * AKPM: should this return a partial result if some of the IOs were | |
1342 | * successfully submitted? | |
1343 | */ | |
1344 | for (i=0; i<nr; i++) { | |
1345 | struct iocb __user *user_iocb; | |
1346 | struct iocb tmp; | |
1347 | ||
1348 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1349 | ret = -EFAULT; | |
1350 | break; | |
1351 | } | |
1352 | ||
1353 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1354 | ret = -EFAULT; | |
1355 | break; | |
1356 | } | |
1357 | ||
080d676d | 1358 | ret = io_submit_one(ctx, user_iocb, &tmp, &batch, compat); |
1da177e4 LT |
1359 | if (ret) |
1360 | break; | |
1361 | } | |
9f5b9425 | 1362 | blk_finish_plug(&plug); |
1da177e4 | 1363 | |
69e4747e | 1364 | kiocb_batch_free(ctx, &batch); |
1da177e4 LT |
1365 | put_ioctx(ctx); |
1366 | return i ? i : ret; | |
1367 | } | |
1368 | ||
9d85cba7 JM |
1369 | /* sys_io_submit: |
1370 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1371 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1372 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1373 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1374 | * is invalid for the file descriptor in the iocb. May fail with | |
1375 | * -EFAULT if any of the data structures point to invalid data. May | |
1376 | * fail with -EBADF if the file descriptor specified in the first | |
1377 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1378 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1379 | * fail with -ENOSYS if not implemented. | |
1380 | */ | |
1381 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1382 | struct iocb __user * __user *, iocbpp) | |
1383 | { | |
1384 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1385 | } | |
1386 | ||
1da177e4 LT |
1387 | /* lookup_kiocb |
1388 | * Finds a given iocb for cancellation. | |
1da177e4 | 1389 | */ |
25ee7e38 AB |
1390 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1391 | u32 key) | |
1da177e4 LT |
1392 | { |
1393 | struct list_head *pos; | |
d00689af ZB |
1394 | |
1395 | assert_spin_locked(&ctx->ctx_lock); | |
1396 | ||
1da177e4 LT |
1397 | /* TODO: use a hash or array, this sucks. */ |
1398 | list_for_each(pos, &ctx->active_reqs) { | |
1399 | struct kiocb *kiocb = list_kiocb(pos); | |
1400 | if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key) | |
1401 | return kiocb; | |
1402 | } | |
1403 | return NULL; | |
1404 | } | |
1405 | ||
1406 | /* sys_io_cancel: | |
1407 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1408 | * the operation is successfully cancelled, the resulting event is | |
1409 | * copied into the memory pointed to by result without being placed | |
1410 | * into the completion queue and 0 is returned. May fail with | |
1411 | * -EFAULT if any of the data structures pointed to are invalid. | |
1412 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1413 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1414 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1415 | */ | |
002c8976 HC |
1416 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1417 | struct io_event __user *, result) | |
1da177e4 LT |
1418 | { |
1419 | int (*cancel)(struct kiocb *iocb, struct io_event *res); | |
1420 | struct kioctx *ctx; | |
1421 | struct kiocb *kiocb; | |
1422 | u32 key; | |
1423 | int ret; | |
1424 | ||
1425 | ret = get_user(key, &iocb->aio_key); | |
1426 | if (unlikely(ret)) | |
1427 | return -EFAULT; | |
1428 | ||
1429 | ctx = lookup_ioctx(ctx_id); | |
1430 | if (unlikely(!ctx)) | |
1431 | return -EINVAL; | |
1432 | ||
1433 | spin_lock_irq(&ctx->ctx_lock); | |
1434 | ret = -EAGAIN; | |
1435 | kiocb = lookup_kiocb(ctx, iocb, key); | |
1436 | if (kiocb && kiocb->ki_cancel) { | |
1437 | cancel = kiocb->ki_cancel; | |
1438 | kiocb->ki_users ++; | |
1439 | kiocbSetCancelled(kiocb); | |
1440 | } else | |
1441 | cancel = NULL; | |
1442 | spin_unlock_irq(&ctx->ctx_lock); | |
1443 | ||
1444 | if (NULL != cancel) { | |
1445 | struct io_event tmp; | |
1446 | pr_debug("calling cancel\n"); | |
1447 | memset(&tmp, 0, sizeof(tmp)); | |
1448 | tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user; | |
1449 | tmp.data = kiocb->ki_user_data; | |
1450 | ret = cancel(kiocb, &tmp); | |
1451 | if (!ret) { | |
1452 | /* Cancellation succeeded -- copy the result | |
1453 | * into the user's buffer. | |
1454 | */ | |
1455 | if (copy_to_user(result, &tmp, sizeof(tmp))) | |
1456 | ret = -EFAULT; | |
1457 | } | |
1458 | } else | |
8f58202b | 1459 | ret = -EINVAL; |
1da177e4 LT |
1460 | |
1461 | put_ioctx(ctx); | |
1462 | ||
1463 | return ret; | |
1464 | } | |
1465 | ||
1466 | /* io_getevents: | |
1467 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1468 | * the completion queue for the aio_context specified by ctx_id. If |
1469 | * it succeeds, the number of read events is returned. May fail with | |
1470 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1471 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1472 | * if any of the memory specified is invalid. May return 0 or | |
1473 | * < min_nr if the timeout specified by timeout has elapsed | |
1474 | * before sufficient events are available, where timeout == NULL | |
1475 | * specifies an infinite timeout. Note that the timeout pointed to by | |
1476 | * timeout is relative and will be updated if not NULL and the | |
1477 | * operation blocks. Will fail with -ENOSYS if not implemented. | |
1da177e4 | 1478 | */ |
002c8976 HC |
1479 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1480 | long, min_nr, | |
1481 | long, nr, | |
1482 | struct io_event __user *, events, | |
1483 | struct timespec __user *, timeout) | |
1da177e4 LT |
1484 | { |
1485 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1486 | long ret = -EINVAL; | |
1487 | ||
1488 | if (likely(ioctx)) { | |
2e410255 | 1489 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 LT |
1490 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
1491 | put_ioctx(ioctx); | |
1492 | } | |
1da177e4 LT |
1493 | return ret; |
1494 | } |