blob: 228368610dfa64e7af06ec9dd864074cf550851e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
16#include <linux/module.h>
17#include <linux/syscalls.h>
Badari Pulavarty027445c2006-09-30 23:28:46 -070018#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20#define DEBUG 0
21
22#include <linux/sched.h>
23#include <linux/fs.h>
24#include <linux/file.h>
25#include <linux/mm.h>
26#include <linux/mman.h>
27#include <linux/slab.h>
28#include <linux/timer.h>
29#include <linux/aio.h>
30#include <linux/highmem.h>
31#include <linux/workqueue.h>
32#include <linux/security.h>
Davide Libenzi9c3060b2007-05-10 22:23:21 -070033#include <linux/eventfd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35#include <asm/kmap_types.h>
36#include <asm/uaccess.h>
37#include <asm/mmu_context.h>
38
39#if DEBUG > 1
40#define dprintk printk
41#else
42#define dprintk(x...) do { ; } while (0)
43#endif
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*------ sysctl variables----*/
Zach Brownd55b5fd2005-11-07 00:59:31 -080046static DEFINE_SPINLOCK(aio_nr_lock);
47unsigned long aio_nr; /* current system wide number of aio requests */
48unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
Linus Torvalds1da177e2005-04-16 15:20:36 -070049/*----end sysctl variables---*/
50
Christoph Lametere18b8902006-12-06 20:33:20 -080051static struct kmem_cache *kiocb_cachep;
52static struct kmem_cache *kioctx_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54static struct workqueue_struct *aio_wq;
55
56/* Used for rare fput completion. */
David Howells65f27f32006-11-22 14:55:48 +000057static void aio_fput_routine(struct work_struct *);
58static DECLARE_WORK(fput_work, aio_fput_routine);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
60static DEFINE_SPINLOCK(fput_lock);
Adrian Bunk25ee7e32005-04-25 08:18:14 -070061static LIST_HEAD(fput_head);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
David Howells65f27f32006-11-22 14:55:48 +000063static void aio_kick_handler(struct work_struct *);
Sébastien Duguc016e222005-06-28 20:44:59 -070064static void aio_queue_work(struct kioctx *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66/* aio_setup
67 * Creates the slab caches used by the aio routines, panic on
68 * failure as this is done early during the boot sequence.
69 */
70static int __init aio_setup(void)
71{
Christoph Lameter0a31bd52007-05-06 14:49:57 -070072 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
73 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75 aio_wq = create_workqueue("aio");
76
77 pr_debug("aio_setup: sizeof(struct page) = %d\n", (int)sizeof(struct page));
78
79 return 0;
80}
81
82static void aio_free_ring(struct kioctx *ctx)
83{
84 struct aio_ring_info *info = &ctx->ring_info;
85 long i;
86
87 for (i=0; i<info->nr_pages; i++)
88 put_page(info->ring_pages[i]);
89
90 if (info->mmap_size) {
91 down_write(&ctx->mm->mmap_sem);
92 do_munmap(ctx->mm, info->mmap_base, info->mmap_size);
93 up_write(&ctx->mm->mmap_sem);
94 }
95
96 if (info->ring_pages && info->ring_pages != info->internal_pages)
97 kfree(info->ring_pages);
98 info->ring_pages = NULL;
99 info->nr = 0;
100}
101
102static int aio_setup_ring(struct kioctx *ctx)
103{
104 struct aio_ring *ring;
105 struct aio_ring_info *info = &ctx->ring_info;
106 unsigned nr_events = ctx->max_reqs;
107 unsigned long size;
108 int nr_pages;
109
110 /* Compensate for the ring buffer's head/tail overlap entry */
111 nr_events += 2; /* 1 is required, 2 for good luck */
112
113 size = sizeof(struct aio_ring);
114 size += sizeof(struct io_event) * nr_events;
115 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
116
117 if (nr_pages < 0)
118 return -EINVAL;
119
120 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
121
122 info->nr = 0;
123 info->ring_pages = info->internal_pages;
124 if (nr_pages > AIO_RING_PAGES) {
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800125 info->ring_pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (!info->ring_pages)
127 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 }
129
130 info->mmap_size = nr_pages * PAGE_SIZE;
131 dprintk("attempting mmap of %lu bytes\n", info->mmap_size);
132 down_write(&ctx->mm->mmap_sem);
133 info->mmap_base = do_mmap(NULL, 0, info->mmap_size,
Robert P. J. Daye10a4432007-02-10 01:42:56 -0800134 PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 0);
136 if (IS_ERR((void *)info->mmap_base)) {
137 up_write(&ctx->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 info->mmap_size = 0;
139 aio_free_ring(ctx);
140 return -EAGAIN;
141 }
142
143 dprintk("mmap address: 0x%08lx\n", info->mmap_base);
144 info->nr_pages = get_user_pages(current, ctx->mm,
145 info->mmap_base, nr_pages,
146 1, 0, info->ring_pages, NULL);
147 up_write(&ctx->mm->mmap_sem);
148
149 if (unlikely(info->nr_pages != nr_pages)) {
150 aio_free_ring(ctx);
151 return -EAGAIN;
152 }
153
154 ctx->user_id = info->mmap_base;
155
156 info->nr = nr_events; /* trusted copy */
157
158 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
159 ring->nr = nr_events; /* user copy */
160 ring->id = ctx->user_id;
161 ring->head = ring->tail = 0;
162 ring->magic = AIO_RING_MAGIC;
163 ring->compat_features = AIO_RING_COMPAT_FEATURES;
164 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
165 ring->header_length = sizeof(struct aio_ring);
166 kunmap_atomic(ring, KM_USER0);
167
168 return 0;
169}
170
171
172/* aio_ring_event: returns a pointer to the event at the given index from
173 * kmap_atomic(, km). Release the pointer with put_aio_ring_event();
174 */
175#define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
176#define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
177#define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
178
179#define aio_ring_event(info, nr, km) ({ \
180 unsigned pos = (nr) + AIO_EVENTS_OFFSET; \
181 struct io_event *__event; \
182 __event = kmap_atomic( \
183 (info)->ring_pages[pos / AIO_EVENTS_PER_PAGE], km); \
184 __event += pos % AIO_EVENTS_PER_PAGE; \
185 __event; \
186})
187
188#define put_aio_ring_event(event, km) do { \
189 struct io_event *__event = (event); \
190 (void)__event; \
191 kunmap_atomic((void *)((unsigned long)__event & PAGE_MASK), km); \
192} while(0)
193
194/* ioctx_alloc
195 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
196 */
197static struct kioctx *ioctx_alloc(unsigned nr_events)
198{
199 struct mm_struct *mm;
200 struct kioctx *ctx;
201
202 /* Prevent overflows */
203 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
204 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
205 pr_debug("ENOMEM: nr_events too high\n");
206 return ERR_PTR(-EINVAL);
207 }
208
Zach Brownd55b5fd2005-11-07 00:59:31 -0800209 if ((unsigned long)nr_events > aio_max_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 return ERR_PTR(-EAGAIN);
211
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800212 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 if (!ctx)
214 return ERR_PTR(-ENOMEM);
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 ctx->max_reqs = nr_events;
217 mm = ctx->mm = current->mm;
218 atomic_inc(&mm->mm_count);
219
220 atomic_set(&ctx->users, 1);
221 spin_lock_init(&ctx->ctx_lock);
222 spin_lock_init(&ctx->ring_info.ring_lock);
223 init_waitqueue_head(&ctx->wait);
224
225 INIT_LIST_HEAD(&ctx->active_reqs);
226 INIT_LIST_HEAD(&ctx->run_list);
David Howells65f27f32006-11-22 14:55:48 +0000227 INIT_DELAYED_WORK(&ctx->wq, aio_kick_handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229 if (aio_setup_ring(ctx) < 0)
230 goto out_freectx;
231
232 /* limit the number of system wide aios */
Zach Brownd55b5fd2005-11-07 00:59:31 -0800233 spin_lock(&aio_nr_lock);
234 if (aio_nr + ctx->max_reqs > aio_max_nr ||
235 aio_nr + ctx->max_reqs < aio_nr)
236 ctx->max_reqs = 0;
237 else
238 aio_nr += ctx->max_reqs;
239 spin_unlock(&aio_nr_lock);
240 if (ctx->max_reqs == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 goto out_cleanup;
242
243 /* now link into global list. kludge. FIXME */
244 write_lock(&mm->ioctx_list_lock);
245 ctx->next = mm->ioctx_list;
246 mm->ioctx_list = ctx;
247 write_unlock(&mm->ioctx_list_lock);
248
249 dprintk("aio: allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
250 ctx, ctx->user_id, current->mm, ctx->ring_info.nr);
251 return ctx;
252
253out_cleanup:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 __put_ioctx(ctx);
255 return ERR_PTR(-EAGAIN);
256
257out_freectx:
258 mmdrop(mm);
259 kmem_cache_free(kioctx_cachep, ctx);
260 ctx = ERR_PTR(-ENOMEM);
261
262 dprintk("aio: error allocating ioctx %p\n", ctx);
263 return ctx;
264}
265
266/* aio_cancel_all
267 * Cancels all outstanding aio requests on an aio context. Used
268 * when the processes owning a context have all exited to encourage
269 * the rapid destruction of the kioctx.
270 */
271static void aio_cancel_all(struct kioctx *ctx)
272{
273 int (*cancel)(struct kiocb *, struct io_event *);
274 struct io_event res;
275 spin_lock_irq(&ctx->ctx_lock);
276 ctx->dead = 1;
277 while (!list_empty(&ctx->active_reqs)) {
278 struct list_head *pos = ctx->active_reqs.next;
279 struct kiocb *iocb = list_kiocb(pos);
280 list_del_init(&iocb->ki_list);
281 cancel = iocb->ki_cancel;
282 kiocbSetCancelled(iocb);
283 if (cancel) {
284 iocb->ki_users++;
285 spin_unlock_irq(&ctx->ctx_lock);
286 cancel(iocb, &res);
287 spin_lock_irq(&ctx->ctx_lock);
288 }
289 }
290 spin_unlock_irq(&ctx->ctx_lock);
291}
292
Adrian Bunk25ee7e32005-04-25 08:18:14 -0700293static void wait_for_all_aios(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294{
295 struct task_struct *tsk = current;
296 DECLARE_WAITQUEUE(wait, tsk);
297
Ken Chendee11c22007-02-03 01:13:45 -0800298 spin_lock_irq(&ctx->ctx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 if (!ctx->reqs_active)
Ken Chendee11c22007-02-03 01:13:45 -0800300 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
302 add_wait_queue(&ctx->wait, &wait);
303 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
304 while (ctx->reqs_active) {
Ken Chendee11c22007-02-03 01:13:45 -0800305 spin_unlock_irq(&ctx->ctx_lock);
Jeff Moyer41d10da2007-10-16 23:27:20 -0700306 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
Ken Chendee11c22007-02-03 01:13:45 -0800308 spin_lock_irq(&ctx->ctx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 }
310 __set_task_state(tsk, TASK_RUNNING);
311 remove_wait_queue(&ctx->wait, &wait);
Ken Chendee11c22007-02-03 01:13:45 -0800312
313out:
314 spin_unlock_irq(&ctx->ctx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
317/* wait_on_sync_kiocb:
318 * Waits on the given sync kiocb to complete.
319 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800320ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 while (iocb->ki_users) {
323 set_current_state(TASK_UNINTERRUPTIBLE);
324 if (!iocb->ki_users)
325 break;
Jeff Moyer41d10da2007-10-16 23:27:20 -0700326 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328 __set_current_state(TASK_RUNNING);
329 return iocb->ki_user_data;
330}
331
332/* exit_aio: called when the last user of mm goes away. At this point,
333 * there is no way for any new requests to be submited or any of the
334 * io_* syscalls to be called on the context. However, there may be
335 * outstanding requests which hold references to the context; as they
336 * go away, they will call put_ioctx and release any pinned memory
337 * associated with the request (held via struct page * references).
338 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800339void exit_aio(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 struct kioctx *ctx = mm->ioctx_list;
342 mm->ioctx_list = NULL;
343 while (ctx) {
344 struct kioctx *next = ctx->next;
345 ctx->next = NULL;
346 aio_cancel_all(ctx);
347
348 wait_for_all_aios(ctx);
349 /*
Andrew Mortona9df62c2007-05-09 02:33:54 -0700350 * Ensure we don't leave the ctx on the aio_wq
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 */
Oleg Nesterov28e53bd2007-05-09 02:34:22 -0700352 cancel_work_sync(&ctx->wq.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 if (1 != atomic_read(&ctx->users))
355 printk(KERN_DEBUG
356 "exit_aio:ioctx still alive: %d %d %d\n",
357 atomic_read(&ctx->users), ctx->dead,
358 ctx->reqs_active);
359 put_ioctx(ctx);
360 ctx = next;
361 }
362}
363
364/* __put_ioctx
365 * Called when the last user of an aio context has gone away,
366 * and the struct needs to be freed.
367 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800368void __put_ioctx(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 unsigned nr_events = ctx->max_reqs;
371
Eric Sesterhenn93e06b42006-11-30 05:29:23 +0100372 BUG_ON(ctx->reqs_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374 cancel_delayed_work(&ctx->wq);
Oleg Nesterov28e53bd2007-05-09 02:34:22 -0700375 cancel_work_sync(&ctx->wq.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 aio_free_ring(ctx);
377 mmdrop(ctx->mm);
378 ctx->mm = NULL;
379 pr_debug("__put_ioctx: freeing %p\n", ctx);
380 kmem_cache_free(kioctx_cachep, ctx);
381
Zach Brownd55b5fd2005-11-07 00:59:31 -0800382 if (nr_events) {
383 spin_lock(&aio_nr_lock);
384 BUG_ON(aio_nr - nr_events > aio_nr);
385 aio_nr -= nr_events;
386 spin_unlock(&aio_nr_lock);
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
390/* aio_get_req
391 * Allocate a slot for an aio request. Increments the users count
392 * of the kioctx so that the kioctx stays around until all requests are
393 * complete. Returns NULL if no requests are free.
394 *
395 * Returns with kiocb->users set to 2. The io submit code path holds
396 * an extra reference while submitting the i/o.
397 * This prevents races between the aio code path referencing the
398 * req (after submitting it) and aio_complete() freeing the req.
399 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800400static struct kiocb *__aio_get_req(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
402 struct kiocb *req = NULL;
403 struct aio_ring *ring;
404 int okay = 0;
405
406 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL);
407 if (unlikely(!req))
408 return NULL;
409
Zach Brown4faa5282005-10-17 16:43:33 -0700410 req->ki_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 req->ki_users = 2;
412 req->ki_key = 0;
413 req->ki_ctx = ctx;
414 req->ki_cancel = NULL;
415 req->ki_retry = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 req->ki_dtor = NULL;
417 req->private = NULL;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700418 req->ki_iovec = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 INIT_LIST_HEAD(&req->ki_run_list);
Davide Libenzi9c3060b2007-05-10 22:23:21 -0700420 req->ki_eventfd = ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422 /* Check if the completion queue has enough free space to
423 * accept an event from this io.
424 */
425 spin_lock_irq(&ctx->ctx_lock);
426 ring = kmap_atomic(ctx->ring_info.ring_pages[0], KM_USER0);
427 if (ctx->reqs_active < aio_ring_avail(&ctx->ring_info, ring)) {
428 list_add(&req->ki_list, &ctx->active_reqs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 ctx->reqs_active++;
430 okay = 1;
431 }
432 kunmap_atomic(ring, KM_USER0);
433 spin_unlock_irq(&ctx->ctx_lock);
434
435 if (!okay) {
436 kmem_cache_free(kiocb_cachep, req);
437 req = NULL;
438 }
439
440 return req;
441}
442
443static inline struct kiocb *aio_get_req(struct kioctx *ctx)
444{
445 struct kiocb *req;
446 /* Handle a potential starvation case -- should be exceedingly rare as
447 * requests will be stuck on fput_head only if the aio_fput_routine is
448 * delayed and the requests were the last user of the struct file.
449 */
450 req = __aio_get_req(ctx);
451 if (unlikely(NULL == req)) {
452 aio_fput_routine(NULL);
453 req = __aio_get_req(ctx);
454 }
455 return req;
456}
457
458static inline void really_put_req(struct kioctx *ctx, struct kiocb *req)
459{
Zach Brownd00689a2005-11-13 16:07:34 -0800460 assert_spin_locked(&ctx->ctx_lock);
461
Davide Libenzi9c3060b2007-05-10 22:23:21 -0700462 if (!IS_ERR(req->ki_eventfd))
463 fput(req->ki_eventfd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 if (req->ki_dtor)
465 req->ki_dtor(req);
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700466 if (req->ki_iovec != &req->ki_inline_vec)
467 kfree(req->ki_iovec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 kmem_cache_free(kiocb_cachep, req);
469 ctx->reqs_active--;
470
471 if (unlikely(!ctx->reqs_active && ctx->dead))
472 wake_up(&ctx->wait);
473}
474
David Howells65f27f32006-11-22 14:55:48 +0000475static void aio_fput_routine(struct work_struct *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 spin_lock_irq(&fput_lock);
478 while (likely(!list_empty(&fput_head))) {
479 struct kiocb *req = list_kiocb(fput_head.next);
480 struct kioctx *ctx = req->ki_ctx;
481
482 list_del(&req->ki_list);
483 spin_unlock_irq(&fput_lock);
484
485 /* Complete the fput */
486 __fput(req->ki_filp);
487
488 /* Link the iocb into the context's free list */
489 spin_lock_irq(&ctx->ctx_lock);
490 really_put_req(ctx, req);
491 spin_unlock_irq(&ctx->ctx_lock);
492
493 put_ioctx(ctx);
494 spin_lock_irq(&fput_lock);
495 }
496 spin_unlock_irq(&fput_lock);
497}
498
499/* __aio_put_req
500 * Returns true if this put was the last user of the request.
501 */
502static int __aio_put_req(struct kioctx *ctx, struct kiocb *req)
503{
504 dprintk(KERN_DEBUG "aio_put(%p): f_count=%d\n",
505 req, atomic_read(&req->ki_filp->f_count));
506
Zach Brownd00689a2005-11-13 16:07:34 -0800507 assert_spin_locked(&ctx->ctx_lock);
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 req->ki_users --;
Eric Sesterhenn93e06b42006-11-30 05:29:23 +0100510 BUG_ON(req->ki_users < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 if (likely(req->ki_users))
512 return 0;
513 list_del(&req->ki_list); /* remove from active_reqs */
514 req->ki_cancel = NULL;
515 req->ki_retry = NULL;
516
517 /* Must be done under the lock to serialise against cancellation.
518 * Call this aio_fput as it duplicates fput via the fput_work.
519 */
Nick Piggin095975d2006-01-08 01:02:19 -0800520 if (unlikely(atomic_dec_and_test(&req->ki_filp->f_count))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 get_ioctx(ctx);
522 spin_lock(&fput_lock);
523 list_add(&req->ki_list, &fput_head);
524 spin_unlock(&fput_lock);
525 queue_work(aio_wq, &fput_work);
526 } else
527 really_put_req(ctx, req);
528 return 1;
529}
530
531/* aio_put_req
532 * Returns true if this put was the last user of the kiocb,
533 * false if the request is still in use.
534 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800535int aio_put_req(struct kiocb *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
537 struct kioctx *ctx = req->ki_ctx;
538 int ret;
539 spin_lock_irq(&ctx->ctx_lock);
540 ret = __aio_put_req(ctx, req);
541 spin_unlock_irq(&ctx->ctx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 return ret;
543}
544
545/* Lookup an ioctx id. ioctx_list is lockless for reads.
546 * FIXME: this is O(n) and is only suitable for development.
547 */
548struct kioctx *lookup_ioctx(unsigned long ctx_id)
549{
550 struct kioctx *ioctx;
551 struct mm_struct *mm;
552
553 mm = current->mm;
554 read_lock(&mm->ioctx_list_lock);
555 for (ioctx = mm->ioctx_list; ioctx; ioctx = ioctx->next)
556 if (likely(ioctx->user_id == ctx_id && !ioctx->dead)) {
557 get_ioctx(ioctx);
558 break;
559 }
560 read_unlock(&mm->ioctx_list_lock);
561
562 return ioctx;
563}
564
565/*
566 * use_mm
567 * Makes the calling kernel thread take on the specified
568 * mm context.
569 * Called by the retry thread execute retries within the
570 * iocb issuer's mm context, so that copy_from/to_user
571 * operations work seamlessly for aio.
572 * (Note: this routine is intended to be called only
573 * from a kernel thread context)
574 */
575static void use_mm(struct mm_struct *mm)
576{
577 struct mm_struct *active_mm;
578 struct task_struct *tsk = current;
579
580 task_lock(tsk);
581 tsk->flags |= PF_BORROWED_MM;
582 active_mm = tsk->active_mm;
583 atomic_inc(&mm->mm_count);
584 tsk->mm = mm;
585 tsk->active_mm = mm;
Paolo 'Blaisorblade' Giarrusso1e40cd32005-09-03 15:57:25 -0700586 /*
587 * Note that on UML this *requires* PF_BORROWED_MM to be set, otherwise
588 * it won't work. Update it accordingly if you change it here
589 */
Jeremy Fitzhardinge90aef122006-12-13 00:34:49 -0800590 switch_mm(active_mm, mm, tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 task_unlock(tsk);
592
593 mmdrop(active_mm);
594}
595
596/*
597 * unuse_mm
598 * Reverses the effect of use_mm, i.e. releases the
599 * specified mm context which was earlier taken on
600 * by the calling kernel thread
601 * (Note: this routine is intended to be called only
602 * from a kernel thread context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -0700604static void unuse_mm(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
606 struct task_struct *tsk = current;
607
608 task_lock(tsk);
609 tsk->flags &= ~PF_BORROWED_MM;
610 tsk->mm = NULL;
611 /* active_mm is still 'mm' */
612 enter_lazy_tlb(mm, tsk);
613 task_unlock(tsk);
614}
615
616/*
617 * Queue up a kiocb to be retried. Assumes that the kiocb
618 * has already been marked as kicked, and places it on
619 * the retry run list for the corresponding ioctx, if it
620 * isn't already queued. Returns 1 if it actually queued
621 * the kiocb (to tell the caller to activate the work
622 * queue to process it), or 0, if it found that it was
623 * already queued.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
625static inline int __queue_kicked_iocb(struct kiocb *iocb)
626{
627 struct kioctx *ctx = iocb->ki_ctx;
628
Zach Brownd00689a2005-11-13 16:07:34 -0800629 assert_spin_locked(&ctx->ctx_lock);
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (list_empty(&iocb->ki_run_list)) {
632 list_add_tail(&iocb->ki_run_list,
633 &ctx->run_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 return 1;
635 }
636 return 0;
637}
638
639/* aio_run_iocb
640 * This is the core aio execution routine. It is
641 * invoked both for initial i/o submission and
642 * subsequent retries via the aio_kick_handler.
643 * Expects to be invoked with iocb->ki_ctx->lock
Andreas Mohrd6e05ed2006-06-26 18:35:02 +0200644 * already held. The lock is released and reacquired
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 * as needed during processing.
646 *
647 * Calls the iocb retry method (already setup for the
648 * iocb on initial submission) for operation specific
649 * handling, but takes care of most of common retry
650 * execution details for a given iocb. The retry method
651 * needs to be non-blocking as far as possible, to avoid
652 * holding up other iocbs waiting to be serviced by the
653 * retry kernel thread.
654 *
655 * The trickier parts in this code have to do with
656 * ensuring that only one retry instance is in progress
657 * for a given iocb at any time. Providing that guarantee
658 * simplifies the coding of individual aio operations as
659 * it avoids various potential races.
660 */
661static ssize_t aio_run_iocb(struct kiocb *iocb)
662{
663 struct kioctx *ctx = iocb->ki_ctx;
664 ssize_t (*retry)(struct kiocb *);
665 ssize_t ret;
666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 if (!(retry = iocb->ki_retry)) {
668 printk("aio_run_iocb: iocb->ki_retry = NULL\n");
669 return 0;
670 }
671
672 /*
673 * We don't want the next retry iteration for this
674 * operation to start until this one has returned and
675 * updated the iocb state. However, wait_queue functions
676 * can trigger a kick_iocb from interrupt context in the
677 * meantime, indicating that data is available for the next
678 * iteration. We want to remember that and enable the
679 * next retry iteration _after_ we are through with
680 * this one.
681 *
682 * So, in order to be able to register a "kick", but
683 * prevent it from being queued now, we clear the kick
684 * flag, but make the kick code *think* that the iocb is
685 * still on the run list until we are actually done.
686 * When we are done with this iteration, we check if
687 * the iocb was kicked in the meantime and if so, queue
688 * it up afresh.
689 */
690
691 kiocbClearKicked(iocb);
692
693 /*
694 * This is so that aio_complete knows it doesn't need to
695 * pull the iocb off the run list (We can't just call
696 * INIT_LIST_HEAD because we don't want a kick_iocb to
697 * queue this on the run list yet)
698 */
699 iocb->ki_run_list.next = iocb->ki_run_list.prev = NULL;
700 spin_unlock_irq(&ctx->ctx_lock);
701
702 /* Quit retrying if the i/o has been cancelled */
703 if (kiocbIsCancelled(iocb)) {
704 ret = -EINTR;
705 aio_complete(iocb, ret, 0);
706 /* must not access the iocb after this */
707 goto out;
708 }
709
710 /*
711 * Now we are all set to call the retry method in async
Alexey Dobriyan6212e3a2007-10-18 03:04:56 -0700712 * context.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 ret = retry(iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Zach Brown897f15f2005-09-30 11:58:55 -0700716 if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
717 BUG_ON(!list_empty(&iocb->ki_wait.task_list));
718 aio_complete(iocb, ret, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720out:
721 spin_lock_irq(&ctx->ctx_lock);
722
723 if (-EIOCBRETRY == ret) {
724 /*
725 * OK, now that we are done with this iteration
726 * and know that there is more left to go,
727 * this is where we let go so that a subsequent
728 * "kick" can start the next iteration
729 */
730
731 /* will make __queue_kicked_iocb succeed from here on */
732 INIT_LIST_HEAD(&iocb->ki_run_list);
733 /* we must queue the next iteration ourselves, if it
734 * has already been kicked */
735 if (kiocbIsKicked(iocb)) {
736 __queue_kicked_iocb(iocb);
Sébastien Duguc016e222005-06-28 20:44:59 -0700737
738 /*
739 * __queue_kicked_iocb will always return 1 here, because
740 * iocb->ki_run_list is empty at this point so it should
741 * be safe to unconditionally queue the context into the
742 * work queue.
743 */
744 aio_queue_work(ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
746 }
747 return ret;
748}
749
750/*
751 * __aio_run_iocbs:
752 * Process all pending retries queued on the ioctx
753 * run list.
754 * Assumes it is operating within the aio issuer's mm
Zach Brownd00689a2005-11-13 16:07:34 -0800755 * context.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
757static int __aio_run_iocbs(struct kioctx *ctx)
758{
759 struct kiocb *iocb;
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700760 struct list_head run_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Zach Brownd00689a2005-11-13 16:07:34 -0800762 assert_spin_locked(&ctx->ctx_lock);
763
Oleg Nesterov626ab0e2006-06-23 02:05:55 -0700764 list_replace_init(&ctx->run_list, &run_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 while (!list_empty(&run_list)) {
766 iocb = list_entry(run_list.next, struct kiocb,
767 ki_run_list);
768 list_del(&iocb->ki_run_list);
769 /*
770 * Hold an extra reference while retrying i/o.
771 */
772 iocb->ki_users++; /* grab extra reference */
773 aio_run_iocb(iocb);
Ken Chendee11c22007-02-03 01:13:45 -0800774 __aio_put_req(ctx, iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 if (!list_empty(&ctx->run_list))
777 return 1;
778 return 0;
779}
780
781static void aio_queue_work(struct kioctx * ctx)
782{
783 unsigned long timeout;
784 /*
785 * if someone is waiting, get the work started right
786 * away, otherwise, use a longer delay
787 */
788 smp_mb();
789 if (waitqueue_active(&ctx->wait))
790 timeout = 1;
791 else
792 timeout = HZ/10;
793 queue_delayed_work(aio_wq, &ctx->wq, timeout);
794}
795
796
797/*
798 * aio_run_iocbs:
799 * Process all pending retries queued on the ioctx
800 * run list.
801 * Assumes it is operating within the aio issuer's mm
802 * context.
803 */
804static inline void aio_run_iocbs(struct kioctx *ctx)
805{
806 int requeue;
807
808 spin_lock_irq(&ctx->ctx_lock);
809
810 requeue = __aio_run_iocbs(ctx);
811 spin_unlock_irq(&ctx->ctx_lock);
812 if (requeue)
813 aio_queue_work(ctx);
814}
815
816/*
817 * just like aio_run_iocbs, but keeps running them until
818 * the list stays empty
819 */
820static inline void aio_run_all_iocbs(struct kioctx *ctx)
821{
822 spin_lock_irq(&ctx->ctx_lock);
823 while (__aio_run_iocbs(ctx))
824 ;
825 spin_unlock_irq(&ctx->ctx_lock);
826}
827
828/*
829 * aio_kick_handler:
830 * Work queue handler triggered to process pending
831 * retries on an ioctx. Takes on the aio issuer's
832 * mm context before running the iocbs, so that
833 * copy_xxx_user operates on the issuer's address
834 * space.
835 * Run on aiod's context.
836 */
David Howells65f27f32006-11-22 14:55:48 +0000837static void aio_kick_handler(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
David Howells65f27f32006-11-22 14:55:48 +0000839 struct kioctx *ctx = container_of(work, struct kioctx, wq.work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 mm_segment_t oldfs = get_fs();
Zach Brown1ebb1102006-12-29 16:47:02 -0800841 struct mm_struct *mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 int requeue;
843
844 set_fs(USER_DS);
845 use_mm(ctx->mm);
846 spin_lock_irq(&ctx->ctx_lock);
847 requeue =__aio_run_iocbs(ctx);
Zach Brown1ebb1102006-12-29 16:47:02 -0800848 mm = ctx->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 spin_unlock_irq(&ctx->ctx_lock);
Zach Brown1ebb1102006-12-29 16:47:02 -0800850 unuse_mm(mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 set_fs(oldfs);
852 /*
853 * we're in a worker thread already, don't use queue_delayed_work,
854 */
855 if (requeue)
David Howells52bad642006-11-22 14:54:01 +0000856 queue_delayed_work(aio_wq, &ctx->wq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857}
858
859
860/*
861 * Called by kick_iocb to queue the kiocb for retry
862 * and if required activate the aio work queue to process
863 * it
864 */
Zach Brown998765e2005-09-30 11:58:54 -0700865static void try_queue_kicked_iocb(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
867 struct kioctx *ctx = iocb->ki_ctx;
868 unsigned long flags;
869 int run = 0;
870
Zach Brown998765e2005-09-30 11:58:54 -0700871 /* We're supposed to be the only path putting the iocb back on the run
872 * list. If we find that the iocb is *back* on a wait queue already
873 * than retry has happened before we could queue the iocb. This also
874 * means that the retry could have completed and freed our iocb, no
875 * good. */
876 BUG_ON((!list_empty(&iocb->ki_wait.task_list)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 spin_lock_irqsave(&ctx->ctx_lock, flags);
Zach Brown998765e2005-09-30 11:58:54 -0700879 /* set this inside the lock so that we can't race with aio_run_iocb()
880 * testing it and putting the iocb on the run list under the lock */
881 if (!kiocbTryKick(iocb))
882 run = __queue_kicked_iocb(iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
Ken Chen644d3a02005-05-01 08:59:15 -0700884 if (run)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 aio_queue_work(ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
887
888/*
889 * kick_iocb:
890 * Called typically from a wait queue callback context
891 * (aio_wake_function) to trigger a retry of the iocb.
892 * The retry is usually executed by aio workqueue
893 * threads (See aio_kick_handler).
894 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800895void kick_iocb(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896{
897 /* sync iocbs are easy: they can only ever be executing from a
898 * single context. */
899 if (is_sync_kiocb(iocb)) {
900 kiocbSetKicked(iocb);
901 wake_up_process(iocb->ki_obj.tsk);
902 return;
903 }
904
Zach Brown998765e2005-09-30 11:58:54 -0700905 try_queue_kicked_iocb(iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906}
907EXPORT_SYMBOL(kick_iocb);
908
909/* aio_complete
910 * Called when the io request on the given iocb is complete.
911 * Returns true if this is the last user of the request. The
912 * only other user of the request can be the cancellation code.
913 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800914int aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915{
916 struct kioctx *ctx = iocb->ki_ctx;
917 struct aio_ring_info *info;
918 struct aio_ring *ring;
919 struct io_event *event;
920 unsigned long flags;
921 unsigned long tail;
922 int ret;
923
Zach Brown20dcae32005-11-13 16:07:33 -0800924 /*
925 * Special case handling for sync iocbs:
926 * - events go directly into the iocb for fast handling
927 * - the sync task with the iocb in its stack holds the single iocb
928 * ref, no other paths have a way to get another ref
929 * - the sync task helpfully left a reference to itself in the iocb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 */
931 if (is_sync_kiocb(iocb)) {
Zach Brown20dcae32005-11-13 16:07:33 -0800932 BUG_ON(iocb->ki_users != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 iocb->ki_user_data = res;
Zach Brown20dcae32005-11-13 16:07:33 -0800934 iocb->ki_users = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 wake_up_process(iocb->ki_obj.tsk);
Zach Brown20dcae32005-11-13 16:07:33 -0800936 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
938
939 info = &ctx->ring_info;
940
941 /* add a completion event to the ring buffer.
942 * must be done holding ctx->ctx_lock to prevent
943 * other code from messing with the tail
944 * pointer since we might be called from irq
945 * context.
946 */
947 spin_lock_irqsave(&ctx->ctx_lock, flags);
948
949 if (iocb->ki_run_list.prev && !list_empty(&iocb->ki_run_list))
950 list_del_init(&iocb->ki_run_list);
951
952 /*
953 * cancelled requests don't get events, userland was given one
954 * when the event got cancelled.
955 */
956 if (kiocbIsCancelled(iocb))
957 goto put_rq;
958
959 ring = kmap_atomic(info->ring_pages[0], KM_IRQ1);
960
961 tail = info->tail;
962 event = aio_ring_event(info, tail, KM_IRQ0);
Ken Chen4bf69b22005-05-01 08:59:15 -0700963 if (++tail >= info->nr)
964 tail = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
966 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
967 event->data = iocb->ki_user_data;
968 event->res = res;
969 event->res2 = res2;
970
971 dprintk("aio_complete: %p[%lu]: %p: %p %Lx %lx %lx\n",
972 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
973 res, res2);
974
975 /* after flagging the request as done, we
976 * must never even look at it again
977 */
978 smp_wmb(); /* make event visible before updating tail */
979
980 info->tail = tail;
981 ring->tail = tail;
982
983 put_aio_ring_event(event, KM_IRQ0);
984 kunmap_atomic(ring, KM_IRQ1);
985
986 pr_debug("added to ring %p at [%lu]\n", iocb, tail);
Davide Libenzi8d1c98b2008-04-10 21:29:19 -0700987
988 /*
989 * Check if the user asked us to deliver the result through an
990 * eventfd. The eventfd_signal() function is safe to be called
991 * from IRQ context.
992 */
993 if (!IS_ERR(iocb->ki_eventfd))
994 eventfd_signal(iocb->ki_eventfd, 1);
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996put_rq:
997 /* everything turned out well, dispose of the aiocb. */
998 ret = __aio_put_req(ctx, iocb);
999
Quentin Barnes6cb2a212008-03-19 17:00:39 -07001000 /*
1001 * We have to order our ring_info tail store above and test
1002 * of the wait list below outside the wait lock. This is
1003 * like in wake_up_bit() where clearing a bit has to be
1004 * ordered with the unlocked test.
1005 */
1006 smp_mb();
1007
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (waitqueue_active(&ctx->wait))
1009 wake_up(&ctx->wait);
1010
Ken Chendee11c22007-02-03 01:13:45 -08001011 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 return ret;
1013}
1014
1015/* aio_read_evt
1016 * Pull an event off of the ioctx's event ring. Returns the number of
1017 * events fetched (0 or 1 ;-)
1018 * FIXME: make this use cmpxchg.
1019 * TODO: make the ringbuffer user mmap()able (requires FIXME).
1020 */
1021static int aio_read_evt(struct kioctx *ioctx, struct io_event *ent)
1022{
1023 struct aio_ring_info *info = &ioctx->ring_info;
1024 struct aio_ring *ring;
1025 unsigned long head;
1026 int ret = 0;
1027
1028 ring = kmap_atomic(info->ring_pages[0], KM_USER0);
1029 dprintk("in aio_read_evt h%lu t%lu m%lu\n",
1030 (unsigned long)ring->head, (unsigned long)ring->tail,
1031 (unsigned long)ring->nr);
1032
1033 if (ring->head == ring->tail)
1034 goto out;
1035
1036 spin_lock(&info->ring_lock);
1037
1038 head = ring->head % info->nr;
1039 if (head != ring->tail) {
1040 struct io_event *evp = aio_ring_event(info, head, KM_USER1);
1041 *ent = *evp;
1042 head = (head + 1) % info->nr;
1043 smp_mb(); /* finish reading the event before updatng the head */
1044 ring->head = head;
1045 ret = 1;
1046 put_aio_ring_event(evp, KM_USER1);
1047 }
1048 spin_unlock(&info->ring_lock);
1049
1050out:
1051 kunmap_atomic(ring, KM_USER0);
1052 dprintk("leaving aio_read_evt: %d h%lu t%lu\n", ret,
1053 (unsigned long)ring->head, (unsigned long)ring->tail);
1054 return ret;
1055}
1056
1057struct aio_timeout {
1058 struct timer_list timer;
1059 int timed_out;
1060 struct task_struct *p;
1061};
1062
1063static void timeout_func(unsigned long data)
1064{
1065 struct aio_timeout *to = (struct aio_timeout *)data;
1066
1067 to->timed_out = 1;
1068 wake_up_process(to->p);
1069}
1070
1071static inline void init_timeout(struct aio_timeout *to)
1072{
1073 init_timer(&to->timer);
1074 to->timer.data = (unsigned long)to;
1075 to->timer.function = timeout_func;
1076 to->timed_out = 0;
1077 to->p = current;
1078}
1079
1080static inline void set_timeout(long start_jiffies, struct aio_timeout *to,
1081 const struct timespec *ts)
1082{
1083 to->timer.expires = start_jiffies + timespec_to_jiffies(ts);
1084 if (time_after(to->timer.expires, jiffies))
1085 add_timer(&to->timer);
1086 else
1087 to->timed_out = 1;
1088}
1089
1090static inline void clear_timeout(struct aio_timeout *to)
1091{
1092 del_singleshot_timer_sync(&to->timer);
1093}
1094
1095static int read_events(struct kioctx *ctx,
1096 long min_nr, long nr,
1097 struct io_event __user *event,
1098 struct timespec __user *timeout)
1099{
1100 long start_jiffies = jiffies;
1101 struct task_struct *tsk = current;
1102 DECLARE_WAITQUEUE(wait, tsk);
1103 int ret;
1104 int i = 0;
1105 struct io_event ent;
1106 struct aio_timeout to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 int retry = 0;
1108
1109 /* needed to zero any padding within an entry (there shouldn't be
1110 * any, but C is fun!
1111 */
1112 memset(&ent, 0, sizeof(ent));
1113retry:
1114 ret = 0;
1115 while (likely(i < nr)) {
1116 ret = aio_read_evt(ctx, &ent);
1117 if (unlikely(ret <= 0))
1118 break;
1119
1120 dprintk("read event: %Lx %Lx %Lx %Lx\n",
1121 ent.data, ent.obj, ent.res, ent.res2);
1122
1123 /* Could we split the check in two? */
1124 ret = -EFAULT;
1125 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1126 dprintk("aio: lost an event due to EFAULT.\n");
1127 break;
1128 }
1129 ret = 0;
1130
1131 /* Good, event copied to userland, update counts. */
1132 event ++;
1133 i ++;
1134 }
1135
1136 if (min_nr <= i)
1137 return i;
1138 if (ret)
1139 return ret;
1140
1141 /* End fast path */
1142
1143 /* racey check, but it gets redone */
1144 if (!retry && unlikely(!list_empty(&ctx->run_list))) {
1145 retry = 1;
1146 aio_run_all_iocbs(ctx);
1147 goto retry;
1148 }
1149
1150 init_timeout(&to);
1151 if (timeout) {
1152 struct timespec ts;
1153 ret = -EFAULT;
1154 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
1155 goto out;
1156
1157 set_timeout(start_jiffies, &to, &ts);
1158 }
1159
1160 while (likely(i < nr)) {
1161 add_wait_queue_exclusive(&ctx->wait, &wait);
1162 do {
1163 set_task_state(tsk, TASK_INTERRUPTIBLE);
1164 ret = aio_read_evt(ctx, &ent);
1165 if (ret)
1166 break;
1167 if (min_nr <= i)
1168 break;
1169 ret = 0;
1170 if (to.timed_out) /* Only check after read evt */
1171 break;
Jeff Moyere00ba3d2007-12-04 23:45:02 -08001172 /* Try to only show up in io wait if there are ops
1173 * in flight */
1174 if (ctx->reqs_active)
1175 io_schedule();
1176 else
1177 schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 if (signal_pending(tsk)) {
1179 ret = -EINTR;
1180 break;
1181 }
1182 /*ret = aio_read_evt(ctx, &ent);*/
1183 } while (1) ;
1184
1185 set_task_state(tsk, TASK_RUNNING);
1186 remove_wait_queue(&ctx->wait, &wait);
1187
1188 if (unlikely(ret <= 0))
1189 break;
1190
1191 ret = -EFAULT;
1192 if (unlikely(copy_to_user(event, &ent, sizeof(ent)))) {
1193 dprintk("aio: lost an event due to EFAULT.\n");
1194 break;
1195 }
1196
1197 /* Good, event copied to userland, update counts. */
1198 event ++;
1199 i ++;
1200 }
1201
1202 if (timeout)
1203 clear_timeout(&to);
1204out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 return i ? i : ret;
1206}
1207
1208/* Take an ioctx and remove it from the list of ioctx's. Protects
1209 * against races with itself via ->dead.
1210 */
1211static void io_destroy(struct kioctx *ioctx)
1212{
1213 struct mm_struct *mm = current->mm;
1214 struct kioctx **tmp;
1215 int was_dead;
1216
1217 /* delete the entry from the list is someone else hasn't already */
1218 write_lock(&mm->ioctx_list_lock);
1219 was_dead = ioctx->dead;
1220 ioctx->dead = 1;
1221 for (tmp = &mm->ioctx_list; *tmp && *tmp != ioctx;
1222 tmp = &(*tmp)->next)
1223 ;
1224 if (*tmp)
1225 *tmp = ioctx->next;
1226 write_unlock(&mm->ioctx_list_lock);
1227
1228 dprintk("aio_release(%p)\n", ioctx);
1229 if (likely(!was_dead))
1230 put_ioctx(ioctx); /* twice for the list */
1231
1232 aio_cancel_all(ioctx);
1233 wait_for_all_aios(ioctx);
1234 put_ioctx(ioctx); /* once for the lookup */
1235}
1236
1237/* sys_io_setup:
1238 * Create an aio_context capable of receiving at least nr_events.
1239 * ctxp must not point to an aio_context that already exists, and
1240 * must be initialized to 0 prior to the call. On successful
1241 * creation of the aio_context, *ctxp is filled in with the resulting
1242 * handle. May fail with -EINVAL if *ctxp is not initialized,
1243 * if the specified nr_events exceeds internal limits. May fail
1244 * with -EAGAIN if the specified nr_events exceeds the user's limit
1245 * of available events. May fail with -ENOMEM if insufficient kernel
1246 * resources are available. May fail with -EFAULT if an invalid
1247 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1248 * implemented.
1249 */
1250asmlinkage long sys_io_setup(unsigned nr_events, aio_context_t __user *ctxp)
1251{
1252 struct kioctx *ioctx = NULL;
1253 unsigned long ctx;
1254 long ret;
1255
1256 ret = get_user(ctx, ctxp);
1257 if (unlikely(ret))
1258 goto out;
1259
1260 ret = -EINVAL;
Zach Brownd55b5fd2005-11-07 00:59:31 -08001261 if (unlikely(ctx || nr_events == 0)) {
1262 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1263 ctx, nr_events);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 goto out;
1265 }
1266
1267 ioctx = ioctx_alloc(nr_events);
1268 ret = PTR_ERR(ioctx);
1269 if (!IS_ERR(ioctx)) {
1270 ret = put_user(ioctx->user_id, ctxp);
1271 if (!ret)
1272 return 0;
1273
1274 get_ioctx(ioctx); /* io_destroy() expects us to hold a ref */
1275 io_destroy(ioctx);
1276 }
1277
1278out:
1279 return ret;
1280}
1281
1282/* sys_io_destroy:
1283 * Destroy the aio_context specified. May cancel any outstanding
1284 * AIOs and block on completion. Will fail with -ENOSYS if not
1285 * implemented. May fail with -EFAULT if the context pointed to
1286 * is invalid.
1287 */
1288asmlinkage long sys_io_destroy(aio_context_t ctx)
1289{
1290 struct kioctx *ioctx = lookup_ioctx(ctx);
1291 if (likely(NULL != ioctx)) {
1292 io_destroy(ioctx);
1293 return 0;
1294 }
1295 pr_debug("EINVAL: io_destroy: invalid context id\n");
1296 return -EINVAL;
1297}
1298
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001299static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
1300{
1301 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
1302
1303 BUG_ON(ret <= 0);
1304
1305 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
1306 ssize_t this = min((ssize_t)iov->iov_len, ret);
1307 iov->iov_base += this;
1308 iov->iov_len -= this;
1309 iocb->ki_left -= this;
1310 ret -= this;
1311 if (iov->iov_len == 0) {
1312 iocb->ki_cur_seg++;
1313 iov++;
1314 }
1315 }
1316
1317 /* the caller should not have done more io than what fit in
1318 * the remaining iovecs */
1319 BUG_ON(ret > 0 && iocb->ki_left == 0);
1320}
1321
1322static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
1324 struct file *file = iocb->ki_filp;
1325 struct address_space *mapping = file->f_mapping;
1326 struct inode *inode = mapping->host;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001327 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
1328 unsigned long, loff_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 ssize_t ret = 0;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001330 unsigned short opcode;
1331
1332 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
1333 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
1334 rw_op = file->f_op->aio_read;
1335 opcode = IOCB_CMD_PREADV;
1336 } else {
1337 rw_op = file->f_op->aio_write;
1338 opcode = IOCB_CMD_PWRITEV;
1339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
Rusty Russellc2ec6682008-02-08 04:20:15 -08001341 /* This matches the pread()/pwrite() logic */
1342 if (iocb->ki_pos < 0)
1343 return -EINVAL;
1344
Zach Brown897f15f2005-09-30 11:58:55 -07001345 do {
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001346 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
1347 iocb->ki_nr_segs - iocb->ki_cur_seg,
1348 iocb->ki_pos);
1349 if (ret > 0)
1350 aio_advance_iovec(iocb, ret);
Badari Pulavarty027445c2006-09-30 23:28:46 -07001351
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001352 /* retry all partial writes. retry partial reads as long as its a
1353 * regular file. */
Zach Brown353fb072005-09-30 11:58:56 -07001354 } while (ret > 0 && iocb->ki_left > 0 &&
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001355 (opcode == IOCB_CMD_PWRITEV ||
1356 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 /* This means we must have transferred all that we could */
1359 /* No need to retry anymore */
1360 if ((ret == 0) || (iocb->ki_left == 0))
1361 ret = iocb->ki_nbytes - iocb->ki_left;
1362
Rusty Russell7adfa2f2008-02-08 04:20:14 -08001363 /* If we managed to write some out we return that, rather than
1364 * the eventual error. */
1365 if (opcode == IOCB_CMD_PWRITEV
1366 && ret < 0 && ret != -EIOCBQUEUED && ret != -EIOCBRETRY
1367 && iocb->ki_nbytes - iocb->ki_left)
1368 ret = iocb->ki_nbytes - iocb->ki_left;
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 return ret;
1371}
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373static ssize_t aio_fdsync(struct kiocb *iocb)
1374{
1375 struct file *file = iocb->ki_filp;
1376 ssize_t ret = -EINVAL;
1377
1378 if (file->f_op->aio_fsync)
1379 ret = file->f_op->aio_fsync(iocb, 1);
1380 return ret;
1381}
1382
1383static ssize_t aio_fsync(struct kiocb *iocb)
1384{
1385 struct file *file = iocb->ki_filp;
1386 ssize_t ret = -EINVAL;
1387
1388 if (file->f_op->aio_fsync)
1389 ret = file->f_op->aio_fsync(iocb, 0);
1390 return ret;
1391}
1392
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001393static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb)
1394{
1395 ssize_t ret;
1396
1397 ret = rw_copy_check_uvector(type, (struct iovec __user *)kiocb->ki_buf,
1398 kiocb->ki_nbytes, 1,
1399 &kiocb->ki_inline_vec, &kiocb->ki_iovec);
1400 if (ret < 0)
1401 goto out;
1402
1403 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1404 kiocb->ki_cur_seg = 0;
1405 /* ki_nbytes/left now reflect bytes instead of segs */
1406 kiocb->ki_nbytes = ret;
1407 kiocb->ki_left = ret;
1408
1409 ret = 0;
1410out:
1411 return ret;
1412}
1413
1414static ssize_t aio_setup_single_vector(struct kiocb *kiocb)
1415{
1416 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1417 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
1418 kiocb->ki_iovec->iov_len = kiocb->ki_left;
1419 kiocb->ki_nr_segs = 1;
1420 kiocb->ki_cur_seg = 0;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001421 return 0;
1422}
1423
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424/*
1425 * aio_setup_iocb:
1426 * Performs the initial checks and aio retry method
1427 * setup for the kiocb at the time of io submission.
1428 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -07001429static ssize_t aio_setup_iocb(struct kiocb *kiocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
1431 struct file *file = kiocb->ki_filp;
1432 ssize_t ret = 0;
1433
1434 switch (kiocb->ki_opcode) {
1435 case IOCB_CMD_PREAD:
1436 ret = -EBADF;
1437 if (unlikely(!(file->f_mode & FMODE_READ)))
1438 break;
1439 ret = -EFAULT;
1440 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1441 kiocb->ki_left)))
1442 break;
Kostik Belousov8766ce42005-10-23 12:57:13 -07001443 ret = security_file_permission(file, MAY_READ);
1444 if (unlikely(ret))
1445 break;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001446 ret = aio_setup_single_vector(kiocb);
1447 if (ret)
1448 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 ret = -EINVAL;
1450 if (file->f_op->aio_read)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001451 kiocb->ki_retry = aio_rw_vect_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 break;
1453 case IOCB_CMD_PWRITE:
1454 ret = -EBADF;
1455 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1456 break;
1457 ret = -EFAULT;
1458 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1459 kiocb->ki_left)))
1460 break;
Kostik Belousov8766ce42005-10-23 12:57:13 -07001461 ret = security_file_permission(file, MAY_WRITE);
1462 if (unlikely(ret))
1463 break;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001464 ret = aio_setup_single_vector(kiocb);
1465 if (ret)
1466 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 ret = -EINVAL;
1468 if (file->f_op->aio_write)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001469 kiocb->ki_retry = aio_rw_vect_retry;
1470 break;
1471 case IOCB_CMD_PREADV:
1472 ret = -EBADF;
1473 if (unlikely(!(file->f_mode & FMODE_READ)))
1474 break;
1475 ret = security_file_permission(file, MAY_READ);
1476 if (unlikely(ret))
1477 break;
1478 ret = aio_setup_vectored_rw(READ, kiocb);
1479 if (ret)
1480 break;
1481 ret = -EINVAL;
1482 if (file->f_op->aio_read)
1483 kiocb->ki_retry = aio_rw_vect_retry;
1484 break;
1485 case IOCB_CMD_PWRITEV:
1486 ret = -EBADF;
1487 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1488 break;
1489 ret = security_file_permission(file, MAY_WRITE);
1490 if (unlikely(ret))
1491 break;
1492 ret = aio_setup_vectored_rw(WRITE, kiocb);
1493 if (ret)
1494 break;
1495 ret = -EINVAL;
1496 if (file->f_op->aio_write)
1497 kiocb->ki_retry = aio_rw_vect_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 break;
1499 case IOCB_CMD_FDSYNC:
1500 ret = -EINVAL;
1501 if (file->f_op->aio_fsync)
1502 kiocb->ki_retry = aio_fdsync;
1503 break;
1504 case IOCB_CMD_FSYNC:
1505 ret = -EINVAL;
1506 if (file->f_op->aio_fsync)
1507 kiocb->ki_retry = aio_fsync;
1508 break;
1509 default:
1510 dprintk("EINVAL: io_submit: no operation provided\n");
1511 ret = -EINVAL;
1512 }
1513
1514 if (!kiocb->ki_retry)
1515 return ret;
1516
1517 return 0;
1518}
1519
1520/*
1521 * aio_wake_function:
1522 * wait queue callback function for aio notification,
1523 * Simply triggers a retry of the operation via kick_iocb.
1524 *
1525 * This callback is specified in the wait queue entry in
Alexey Dobriyan6212e3a2007-10-18 03:04:56 -07001526 * a kiocb.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 *
1528 * Note:
1529 * This routine is executed with the wait queue lock held.
1530 * Since kick_iocb acquires iocb->ctx->ctx_lock, it nests
1531 * the ioctx lock inside the wait queue lock. This is safe
1532 * because this callback isn't used for wait queues which
1533 * are nested inside ioctx lock (i.e. ctx->wait)
1534 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -07001535static int aio_wake_function(wait_queue_t *wait, unsigned mode,
1536 int sync, void *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537{
1538 struct kiocb *iocb = container_of(wait, struct kiocb, ki_wait);
1539
1540 list_del_init(&wait->task_list);
1541 kick_iocb(iocb);
1542 return 1;
1543}
1544
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001545int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 struct iocb *iocb)
1547{
1548 struct kiocb *req;
1549 struct file *file;
1550 ssize_t ret;
1551
1552 /* enforce forwards compatibility on users */
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001553 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 pr_debug("EINVAL: io_submit: reserve field set\n");
1555 return -EINVAL;
1556 }
1557
1558 /* prevent overflows */
1559 if (unlikely(
1560 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1561 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1562 ((ssize_t)iocb->aio_nbytes < 0)
1563 )) {
1564 pr_debug("EINVAL: io_submit: overflow check\n");
1565 return -EINVAL;
1566 }
1567
1568 file = fget(iocb->aio_fildes);
1569 if (unlikely(!file))
1570 return -EBADF;
1571
1572 req = aio_get_req(ctx); /* returns with 2 references to req */
1573 if (unlikely(!req)) {
1574 fput(file);
1575 return -EAGAIN;
1576 }
Yan Zheng87e28312007-10-08 12:16:20 -07001577 req->ki_filp = file;
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001578 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1579 /*
1580 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1581 * instance of the file* now. The file descriptor must be
1582 * an eventfd() fd, and will be signaled for each completed
1583 * event using the eventfd_signal() function.
1584 */
1585 req->ki_eventfd = eventfd_fget((int) iocb->aio_resfd);
1586 if (unlikely(IS_ERR(req->ki_eventfd))) {
1587 ret = PTR_ERR(req->ki_eventfd);
1588 goto out_put_req;
1589 }
1590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
Ken Chen212079c2005-05-01 08:59:15 -07001592 ret = put_user(req->ki_key, &user_iocb->aio_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 if (unlikely(ret)) {
1594 dprintk("EFAULT: aio_key\n");
1595 goto out_put_req;
1596 }
1597
1598 req->ki_obj.user = user_iocb;
1599 req->ki_user_data = iocb->aio_data;
1600 req->ki_pos = iocb->aio_offset;
1601
1602 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1603 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1604 req->ki_opcode = iocb->aio_lio_opcode;
1605 init_waitqueue_func_entry(&req->ki_wait, aio_wake_function);
1606 INIT_LIST_HEAD(&req->ki_wait.task_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 ret = aio_setup_iocb(req);
1609
1610 if (ret)
1611 goto out_put_req;
1612
1613 spin_lock_irq(&ctx->ctx_lock);
Benjamin LaHaiseac0b1bc2005-09-09 13:02:09 -07001614 aio_run_iocb(req);
Benjamin LaHaiseac0b1bc2005-09-09 13:02:09 -07001615 if (!list_empty(&ctx->run_list)) {
Ken Chen954d3e92005-05-01 08:59:16 -07001616 /* drain the run list */
1617 while (__aio_run_iocbs(ctx))
1618 ;
1619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 spin_unlock_irq(&ctx->ctx_lock);
1621 aio_put_req(req); /* drop extra ref to req */
1622 return 0;
1623
1624out_put_req:
1625 aio_put_req(req); /* drop extra ref to req */
1626 aio_put_req(req); /* drop i/o ref to req */
1627 return ret;
1628}
1629
1630/* sys_io_submit:
1631 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1632 * the number of iocbs queued. May return -EINVAL if the aio_context
1633 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1634 * *iocbpp[0] is not properly initialized, if the operation specified
1635 * is invalid for the file descriptor in the iocb. May fail with
1636 * -EFAULT if any of the data structures point to invalid data. May
1637 * fail with -EBADF if the file descriptor specified in the first
1638 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1639 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1640 * fail with -ENOSYS if not implemented.
1641 */
1642asmlinkage long sys_io_submit(aio_context_t ctx_id, long nr,
1643 struct iocb __user * __user *iocbpp)
1644{
1645 struct kioctx *ctx;
1646 long ret = 0;
1647 int i;
1648
1649 if (unlikely(nr < 0))
1650 return -EINVAL;
1651
1652 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1653 return -EFAULT;
1654
1655 ctx = lookup_ioctx(ctx_id);
1656 if (unlikely(!ctx)) {
1657 pr_debug("EINVAL: io_submit: invalid context id\n");
1658 return -EINVAL;
1659 }
1660
1661 /*
1662 * AKPM: should this return a partial result if some of the IOs were
1663 * successfully submitted?
1664 */
1665 for (i=0; i<nr; i++) {
1666 struct iocb __user *user_iocb;
1667 struct iocb tmp;
1668
1669 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1670 ret = -EFAULT;
1671 break;
1672 }
1673
1674 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1675 ret = -EFAULT;
1676 break;
1677 }
1678
1679 ret = io_submit_one(ctx, user_iocb, &tmp);
1680 if (ret)
1681 break;
1682 }
1683
1684 put_ioctx(ctx);
1685 return i ? i : ret;
1686}
1687
1688/* lookup_kiocb
1689 * Finds a given iocb for cancellation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -07001691static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1692 u32 key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
1694 struct list_head *pos;
Zach Brownd00689a2005-11-13 16:07:34 -08001695
1696 assert_spin_locked(&ctx->ctx_lock);
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 /* TODO: use a hash or array, this sucks. */
1699 list_for_each(pos, &ctx->active_reqs) {
1700 struct kiocb *kiocb = list_kiocb(pos);
1701 if (kiocb->ki_obj.user == iocb && kiocb->ki_key == key)
1702 return kiocb;
1703 }
1704 return NULL;
1705}
1706
1707/* sys_io_cancel:
1708 * Attempts to cancel an iocb previously passed to io_submit. If
1709 * the operation is successfully cancelled, the resulting event is
1710 * copied into the memory pointed to by result without being placed
1711 * into the completion queue and 0 is returned. May fail with
1712 * -EFAULT if any of the data structures pointed to are invalid.
1713 * May fail with -EINVAL if aio_context specified by ctx_id is
1714 * invalid. May fail with -EAGAIN if the iocb specified was not
1715 * cancelled. Will fail with -ENOSYS if not implemented.
1716 */
1717asmlinkage long sys_io_cancel(aio_context_t ctx_id, struct iocb __user *iocb,
1718 struct io_event __user *result)
1719{
1720 int (*cancel)(struct kiocb *iocb, struct io_event *res);
1721 struct kioctx *ctx;
1722 struct kiocb *kiocb;
1723 u32 key;
1724 int ret;
1725
1726 ret = get_user(key, &iocb->aio_key);
1727 if (unlikely(ret))
1728 return -EFAULT;
1729
1730 ctx = lookup_ioctx(ctx_id);
1731 if (unlikely(!ctx))
1732 return -EINVAL;
1733
1734 spin_lock_irq(&ctx->ctx_lock);
1735 ret = -EAGAIN;
1736 kiocb = lookup_kiocb(ctx, iocb, key);
1737 if (kiocb && kiocb->ki_cancel) {
1738 cancel = kiocb->ki_cancel;
1739 kiocb->ki_users ++;
1740 kiocbSetCancelled(kiocb);
1741 } else
1742 cancel = NULL;
1743 spin_unlock_irq(&ctx->ctx_lock);
1744
1745 if (NULL != cancel) {
1746 struct io_event tmp;
1747 pr_debug("calling cancel\n");
1748 memset(&tmp, 0, sizeof(tmp));
1749 tmp.obj = (u64)(unsigned long)kiocb->ki_obj.user;
1750 tmp.data = kiocb->ki_user_data;
1751 ret = cancel(kiocb, &tmp);
1752 if (!ret) {
1753 /* Cancellation succeeded -- copy the result
1754 * into the user's buffer.
1755 */
1756 if (copy_to_user(result, &tmp, sizeof(tmp)))
1757 ret = -EFAULT;
1758 }
1759 } else
Wendy Cheng8f582022005-09-09 13:02:08 -07001760 ret = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761
1762 put_ioctx(ctx);
1763
1764 return ret;
1765}
1766
1767/* io_getevents:
1768 * Attempts to read at least min_nr events and up to nr events from
1769 * the completion queue for the aio_context specified by ctx_id. May
1770 * fail with -EINVAL if ctx_id is invalid, if min_nr is out of range,
1771 * if nr is out of range, if when is out of range. May fail with
1772 * -EFAULT if any of the memory specified to is invalid. May return
1773 * 0 or < min_nr if no events are available and the timeout specified
1774 * by when has elapsed, where when == NULL specifies an infinite
1775 * timeout. Note that the timeout pointed to by when is relative and
1776 * will be updated if not NULL and the operation blocks. Will fail
1777 * with -ENOSYS if not implemented.
1778 */
1779asmlinkage long sys_io_getevents(aio_context_t ctx_id,
1780 long min_nr,
1781 long nr,
1782 struct io_event __user *events,
1783 struct timespec __user *timeout)
1784{
1785 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1786 long ret = -EINVAL;
1787
1788 if (likely(ioctx)) {
1789 if (likely(min_nr <= nr && min_nr >= 0 && nr >= 0))
1790 ret = read_events(ioctx, min_nr, nr, events, timeout);
1791 put_ioctx(ioctx);
1792 }
1793
Roland McGrath598af052008-04-10 15:38:45 -07001794 asmlinkage_protect(5, ret, ctx_id, min_nr, nr, events, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 return ret;
1796}
1797
1798__initcall(aio_setup);
1799
1800EXPORT_SYMBOL(aio_complete);
1801EXPORT_SYMBOL(aio_put_req);
1802EXPORT_SYMBOL(wait_on_sync_kiocb);