blob: 5d7dad365f5fe25f0f394132f2481d4c3456e914 [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 */
Kent Overstreetcaf41672013-05-07 16:18:35 -070011#define pr_fmt(fmt) "%s: " fmt, __func__
12
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/kernel.h>
14#include <linux/init.h>
15#include <linux/errno.h>
16#include <linux/time.h>
17#include <linux/aio_abi.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/syscalls.h>
Jens Axboeb9d128f2009-10-29 13:59:26 +010020#include <linux/backing-dev.h>
Badari Pulavarty027445c2006-09-30 23:28:46 -070021#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/sched.h>
24#include <linux/fs.h>
25#include <linux/file.h>
26#include <linux/mm.h>
27#include <linux/mman.h>
Michael S. Tsirkin3d2d8272009-09-21 17:03:51 -070028#include <linux/mmu_context.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#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>
Davide Libenzi9c3060b2007-05-10 22:23:21 -070035#include <linux/eventfd.h>
Jeff Moyercfb1e332009-10-02 18:57:36 -040036#include <linux/blkdev.h>
Jeff Moyer9d85cba2010-05-26 14:44:26 -070037#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <asm/kmap_types.h>
40#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Kent Overstreet4e179bc2013-05-07 16:18:33 -070042#define AIO_RING_MAGIC 0xa10a10a1
43#define AIO_RING_COMPAT_FEATURES 1
44#define AIO_RING_INCOMPAT_FEATURES 0
45struct aio_ring {
46 unsigned id; /* kernel internal index number */
47 unsigned nr; /* number of io_events */
48 unsigned head;
49 unsigned tail;
50
51 unsigned magic;
52 unsigned compat_features;
53 unsigned incompat_features;
54 unsigned header_length; /* size of aio_ring */
55
56
57 struct io_event io_events[0];
58}; /* 128 bytes + ring size */
59
60#define AIO_RING_PAGES 8
Kent Overstreet4e179bc2013-05-07 16:18:33 -070061
Kent Overstreet4e179bc2013-05-07 16:18:33 -070062struct kioctx {
63 atomic_t users;
Kent Overstreet36f55882013-05-07 16:18:41 -070064 atomic_t dead;
Kent Overstreet4e179bc2013-05-07 16:18:33 -070065
66 /* This needs improving */
67 unsigned long user_id;
68 struct hlist_node list;
69
Kent Overstreet3e845ce2013-05-07 16:18:51 -070070 /*
71 * This is what userspace passed to io_setup(), it's not used for
72 * anything but counting against the global max_reqs quota.
73 *
Kent Overstreet58c85dc2013-05-07 16:18:55 -070074 * The real limit is nr_events - 1, which will be larger (see
Kent Overstreet3e845ce2013-05-07 16:18:51 -070075 * aio_setup_ring())
76 */
Kent Overstreet4e179bc2013-05-07 16:18:33 -070077 unsigned max_reqs;
78
Kent Overstreet58c85dc2013-05-07 16:18:55 -070079 /* Size of ringbuffer, in units of struct io_event */
80 unsigned nr_events;
Kent Overstreet4e179bc2013-05-07 16:18:33 -070081
Kent Overstreet58c85dc2013-05-07 16:18:55 -070082 unsigned long mmap_base;
83 unsigned long mmap_size;
84
85 struct page **ring_pages;
86 long nr_pages;
87
Kent Overstreet4e23bca2013-05-07 16:18:56 -070088 struct rcu_head rcu_head;
89 struct work_struct rcu_work;
90
91 struct {
92 atomic_t reqs_active;
93 } ____cacheline_aligned_in_smp;
94
95 struct {
96 spinlock_t ctx_lock;
97 struct list_head active_reqs; /* used for cancellation */
98 } ____cacheline_aligned_in_smp;
99
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700100 struct {
101 struct mutex ring_lock;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700102 wait_queue_head_t wait;
103 } ____cacheline_aligned_in_smp;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700104
105 struct {
106 unsigned tail;
107 spinlock_t completion_lock;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700108 } ____cacheline_aligned_in_smp;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700109
110 struct page *internal_pages[AIO_RING_PAGES];
Kent Overstreet4e179bc2013-05-07 16:18:33 -0700111};
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/*------ sysctl variables----*/
Zach Brownd55b5fd2005-11-07 00:59:31 -0800114static DEFINE_SPINLOCK(aio_nr_lock);
115unsigned long aio_nr; /* current system wide number of aio requests */
116unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117/*----end sysctl variables---*/
118
Christoph Lametere18b8902006-12-06 20:33:20 -0800119static struct kmem_cache *kiocb_cachep;
120static struct kmem_cache *kioctx_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122/* aio_setup
123 * Creates the slab caches used by the aio routines, panic on
124 * failure as this is done early during the boot sequence.
125 */
126static int __init aio_setup(void)
127{
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700128 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
129 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Kent Overstreetcaf41672013-05-07 16:18:35 -0700131 pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
133 return 0;
134}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700135__initcall(aio_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
137static void aio_free_ring(struct kioctx *ctx)
138{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 long i;
140
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700141 for (i = 0; i < ctx->nr_pages; i++)
142 put_page(ctx->ring_pages[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700144 if (ctx->mmap_size)
145 vm_munmap(ctx->mmap_base, ctx->mmap_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700147 if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages)
148 kfree(ctx->ring_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151static int aio_setup_ring(struct kioctx *ctx)
152{
153 struct aio_ring *ring;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 unsigned nr_events = ctx->max_reqs;
Zach Brown41003a72013-05-07 16:18:25 -0700155 struct mm_struct *mm = current->mm;
Michel Lespinasse41badc12013-02-22 16:32:47 -0800156 unsigned long size, populate;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 int nr_pages;
158
159 /* Compensate for the ring buffer's head/tail overlap entry */
160 nr_events += 2; /* 1 is required, 2 for good luck */
161
162 size = sizeof(struct aio_ring);
163 size += sizeof(struct io_event) * nr_events;
164 nr_pages = (size + PAGE_SIZE-1) >> PAGE_SHIFT;
165
166 if (nr_pages < 0)
167 return -EINVAL;
168
169 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) / sizeof(struct io_event);
170
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700171 ctx->nr_events = 0;
172 ctx->ring_pages = ctx->internal_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 if (nr_pages > AIO_RING_PAGES) {
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700174 ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *),
175 GFP_KERNEL);
176 if (!ctx->ring_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 }
179
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700180 ctx->mmap_size = nr_pages * PAGE_SIZE;
181 pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
Zach Brown41003a72013-05-07 16:18:25 -0700182 down_write(&mm->mmap_sem);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700183 ctx->mmap_base = do_mmap_pgoff(NULL, 0, ctx->mmap_size,
184 PROT_READ|PROT_WRITE,
185 MAP_ANONYMOUS|MAP_PRIVATE, 0, &populate);
186 if (IS_ERR((void *)ctx->mmap_base)) {
Zach Brown41003a72013-05-07 16:18:25 -0700187 up_write(&mm->mmap_sem);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700188 ctx->mmap_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 aio_free_ring(ctx);
190 return -EAGAIN;
191 }
192
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700193 pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
194 ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages,
195 1, 0, ctx->ring_pages, NULL);
Zach Brown41003a72013-05-07 16:18:25 -0700196 up_write(&mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700198 if (unlikely(ctx->nr_pages != nr_pages)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 aio_free_ring(ctx);
200 return -EAGAIN;
201 }
Michel Lespinassebebeb3d2013-02-22 16:32:37 -0800202 if (populate)
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700203 mm_populate(ctx->mmap_base, populate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700205 ctx->user_id = ctx->mmap_base;
206 ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700208 ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 ring->nr = nr_events; /* user copy */
210 ring->id = ctx->user_id;
211 ring->head = ring->tail = 0;
212 ring->magic = AIO_RING_MAGIC;
213 ring->compat_features = AIO_RING_COMPAT_FEATURES;
214 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
215 ring->header_length = sizeof(struct aio_ring);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800216 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700217 flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 return 0;
220}
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222#define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
223#define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
224#define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
225
Kent Overstreet0460fef2013-05-07 16:18:49 -0700226void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel)
227{
228 struct kioctx *ctx = req->ki_ctx;
229 unsigned long flags;
230
231 spin_lock_irqsave(&ctx->ctx_lock, flags);
232
233 if (!req->ki_list.next)
234 list_add(&req->ki_list, &ctx->active_reqs);
235
236 req->ki_cancel = cancel;
237
238 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
239}
240EXPORT_SYMBOL(kiocb_set_cancel_fn);
241
Kent Overstreet906b9732013-05-07 16:18:31 -0700242static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb,
243 struct io_event *res)
244{
Kent Overstreet0460fef2013-05-07 16:18:49 -0700245 kiocb_cancel_fn *old, *cancel;
Kent Overstreet906b9732013-05-07 16:18:31 -0700246 int ret = -EINVAL;
247
Kent Overstreet0460fef2013-05-07 16:18:49 -0700248 /*
249 * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it
250 * actually has a cancel function, hence the cmpxchg()
251 */
Kent Overstreet906b9732013-05-07 16:18:31 -0700252
Kent Overstreet0460fef2013-05-07 16:18:49 -0700253 cancel = ACCESS_ONCE(kiocb->ki_cancel);
254 do {
255 if (!cancel || cancel == KIOCB_CANCELLED)
256 return ret;
Kent Overstreet906b9732013-05-07 16:18:31 -0700257
Kent Overstreet0460fef2013-05-07 16:18:49 -0700258 old = cancel;
259 cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
260 } while (cancel != old);
261
262 atomic_inc(&kiocb->ki_users);
263 spin_unlock_irq(&ctx->ctx_lock);
264
265 memset(res, 0, sizeof(*res));
266 res->obj = (u64)(unsigned long)kiocb->ki_obj.user;
267 res->data = kiocb->ki_user_data;
268 ret = cancel(kiocb, res);
269
270 spin_lock_irq(&ctx->ctx_lock);
Kent Overstreet906b9732013-05-07 16:18:31 -0700271
272 return ret;
273}
274
Kent Overstreet36f55882013-05-07 16:18:41 -0700275static void free_ioctx_rcu(struct rcu_head *head)
276{
277 struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
278 kmem_cache_free(kioctx_cachep, ctx);
279}
280
281/*
282 * When this function runs, the kioctx has been removed from the "hash table"
283 * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted -
284 * now it's safe to cancel any that need to be.
285 */
286static void free_ioctx(struct kioctx *ctx)
287{
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700288 struct aio_ring *ring;
Kent Overstreet36f55882013-05-07 16:18:41 -0700289 struct io_event res;
290 struct kiocb *req;
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700291 unsigned head, avail;
Kent Overstreet36f55882013-05-07 16:18:41 -0700292
293 spin_lock_irq(&ctx->ctx_lock);
294
295 while (!list_empty(&ctx->active_reqs)) {
296 req = list_first_entry(&ctx->active_reqs,
297 struct kiocb, ki_list);
298
299 list_del_init(&req->ki_list);
300 kiocb_cancel(ctx, req, &res);
301 }
302
303 spin_unlock_irq(&ctx->ctx_lock);
304
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700305 ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700306 head = ring->head;
307 kunmap_atomic(ring);
308
309 while (atomic_read(&ctx->reqs_active) > 0) {
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700310 wait_event(ctx->wait, head != ctx->tail);
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700311
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700312 avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head;
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700313
314 atomic_sub(avail, &ctx->reqs_active);
315 head += avail;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700316 head %= ctx->nr_events;
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700317 }
318
319 WARN_ON(atomic_read(&ctx->reqs_active) < 0);
Kent Overstreet36f55882013-05-07 16:18:41 -0700320
321 aio_free_ring(ctx);
322
323 spin_lock(&aio_nr_lock);
324 BUG_ON(aio_nr - ctx->max_reqs > aio_nr);
325 aio_nr -= ctx->max_reqs;
326 spin_unlock(&aio_nr_lock);
327
328 pr_debug("freeing %p\n", ctx);
329
330 /*
331 * Here the call_rcu() is between the wait_event() for reqs_active to
332 * hit 0, and freeing the ioctx.
333 *
334 * aio_complete() decrements reqs_active, but it has to touch the ioctx
335 * after to issue a wakeup so we use rcu.
336 */
337 call_rcu(&ctx->rcu_head, free_ioctx_rcu);
338}
339
340static void put_ioctx(struct kioctx *ctx)
341{
342 if (unlikely(atomic_dec_and_test(&ctx->users)))
343 free_ioctx(ctx);
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/* ioctx_alloc
347 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
348 */
349static struct kioctx *ioctx_alloc(unsigned nr_events)
350{
Zach Brown41003a72013-05-07 16:18:25 -0700351 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 struct kioctx *ctx;
Al Viroe23754f2012-03-06 14:33:22 -0500353 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 /* Prevent overflows */
356 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
357 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
358 pr_debug("ENOMEM: nr_events too high\n");
359 return ERR_PTR(-EINVAL);
360 }
361
Al Viro2dd542b2012-03-10 23:10:35 -0500362 if (!nr_events || (unsigned long)nr_events > aio_max_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 return ERR_PTR(-EAGAIN);
364
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800365 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 if (!ctx)
367 return ERR_PTR(-ENOMEM);
368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 ctx->max_reqs = nr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Al Viro86b62a22012-03-07 05:16:35 +0000371 atomic_set(&ctx->users, 2);
Kent Overstreet36f55882013-05-07 16:18:41 -0700372 atomic_set(&ctx->dead, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 spin_lock_init(&ctx->ctx_lock);
Kent Overstreet0460fef2013-05-07 16:18:49 -0700374 spin_lock_init(&ctx->completion_lock);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700375 mutex_init(&ctx->ring_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 init_waitqueue_head(&ctx->wait);
377
378 INIT_LIST_HEAD(&ctx->active_reqs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
380 if (aio_setup_ring(ctx) < 0)
381 goto out_freectx;
382
383 /* limit the number of system wide aios */
Al Viro9fa1cb32012-03-10 23:14:05 -0500384 spin_lock(&aio_nr_lock);
Al Viro2dd542b2012-03-10 23:10:35 -0500385 if (aio_nr + nr_events > aio_max_nr ||
386 aio_nr + nr_events < aio_nr) {
Al Viro9fa1cb32012-03-10 23:14:05 -0500387 spin_unlock(&aio_nr_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 goto out_cleanup;
Al Viro2dd542b2012-03-10 23:10:35 -0500389 }
390 aio_nr += ctx->max_reqs;
Al Viro9fa1cb32012-03-10 23:14:05 -0500391 spin_unlock(&aio_nr_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Jeff Moyer39fa0032008-04-29 01:03:48 -0700393 /* now link into global list. */
Jens Axboeabf137d2008-12-09 08:11:22 +0100394 spin_lock(&mm->ioctx_lock);
395 hlist_add_head_rcu(&ctx->list, &mm->ioctx_list);
396 spin_unlock(&mm->ioctx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
Kent Overstreetcaf41672013-05-07 16:18:35 -0700398 pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700399 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return ctx;
401
402out_cleanup:
Al Viroe23754f2012-03-06 14:33:22 -0500403 err = -EAGAIN;
404 aio_free_ring(ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405out_freectx:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreetcaf41672013-05-07 16:18:35 -0700407 pr_debug("error allocating ioctx %d\n", err);
Al Viroe23754f2012-03-06 14:33:22 -0500408 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409}
410
Kent Overstreet36f55882013-05-07 16:18:41 -0700411static void kill_ioctx_work(struct work_struct *work)
412{
413 struct kioctx *ctx = container_of(work, struct kioctx, rcu_work);
414
415 wake_up_all(&ctx->wait);
416 put_ioctx(ctx);
417}
418
419static void kill_ioctx_rcu(struct rcu_head *head)
420{
421 struct kioctx *ctx = container_of(head, struct kioctx, rcu_head);
422
423 INIT_WORK(&ctx->rcu_work, kill_ioctx_work);
424 schedule_work(&ctx->rcu_work);
425}
426
427/* kill_ioctx
428 * Cancels all outstanding aio requests on an aio context. Used
429 * when the processes owning a context have all exited to encourage
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 * the rapid destruction of the kioctx.
431 */
Kent Overstreet36f55882013-05-07 16:18:41 -0700432static void kill_ioctx(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
Kent Overstreet36f55882013-05-07 16:18:41 -0700434 if (!atomic_xchg(&ctx->dead, 1)) {
435 hlist_del_rcu(&ctx->list);
436 /* Between hlist_del_rcu() and dropping the initial ref */
437 synchronize_rcu();
Al Viro06af1212012-03-20 16:26:24 -0400438
Kent Overstreet36f55882013-05-07 16:18:41 -0700439 /*
440 * We can't punt to workqueue here because put_ioctx() ->
441 * free_ioctx() will unmap the ringbuffer, and that has to be
442 * done in the original process's context. kill_ioctx_rcu/work()
443 * exist for exit_aio(), as in that path free_ioctx() won't do
444 * the unmap.
445 */
446 kill_ioctx_work(&ctx->rcu_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450/* wait_on_sync_kiocb:
451 * Waits on the given sync kiocb to complete.
452 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800453ssize_t wait_on_sync_kiocb(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
Kent Overstreet11599eb2013-05-07 16:18:39 -0700455 while (atomic_read(&iocb->ki_users)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet11599eb2013-05-07 16:18:39 -0700457 if (!atomic_read(&iocb->ki_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 break;
Jeff Moyer41d10da2007-10-16 23:27:20 -0700459 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
461 __set_current_state(TASK_RUNNING);
462 return iocb->ki_user_data;
463}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700464EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Kent Overstreet36f55882013-05-07 16:18:41 -0700466/*
467 * exit_aio: called when the last user of mm goes away. At this point, there is
468 * no way for any new requests to be submited or any of the io_* syscalls to be
469 * called on the context.
470 *
471 * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on
472 * them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800474void exit_aio(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Jens Axboeabf137d2008-12-09 08:11:22 +0100476 struct kioctx *ctx;
Kent Overstreet36f55882013-05-07 16:18:41 -0700477 struct hlist_node *n;
Jens Axboeabf137d2008-12-09 08:11:22 +0100478
Kent Overstreet36f55882013-05-07 16:18:41 -0700479 hlist_for_each_entry_safe(ctx, n, &mm->ioctx_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (1 != atomic_read(&ctx->users))
481 printk(KERN_DEBUG
482 "exit_aio:ioctx still alive: %d %d %d\n",
Kent Overstreet36f55882013-05-07 16:18:41 -0700483 atomic_read(&ctx->users),
484 atomic_read(&ctx->dead),
Kent Overstreet11599eb2013-05-07 16:18:39 -0700485 atomic_read(&ctx->reqs_active));
Al Viro936af152012-04-20 21:49:41 -0400486 /*
487 * We don't need to bother with munmap() here -
488 * exit_mmap(mm) is coming and it'll unmap everything.
489 * Since aio_free_ring() uses non-zero ->mmap_size
490 * as indicator that it needs to unmap the area,
491 * just set it to 0; aio_free_ring() is the only
492 * place that uses ->mmap_size, so it's safe.
Al Viro936af152012-04-20 21:49:41 -0400493 */
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700494 ctx->mmap_size = 0;
Kent Overstreet36f55882013-05-07 16:18:41 -0700495
496 if (!atomic_xchg(&ctx->dead, 1)) {
497 hlist_del_rcu(&ctx->list);
498 call_rcu(&ctx->rcu_head, kill_ioctx_rcu);
499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 }
501}
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/* aio_get_req
Kent Overstreet11599eb2013-05-07 16:18:39 -0700504 * Allocate a slot for an aio request. Increments the ki_users count
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 * of the kioctx so that the kioctx stays around until all requests are
506 * complete. Returns NULL if no requests are free.
507 *
Kent Overstreet11599eb2013-05-07 16:18:39 -0700508 * Returns with kiocb->ki_users set to 2. The io submit code path holds
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 * an extra reference while submitting the i/o.
510 * This prevents races between the aio code path referencing the
511 * req (after submitting it) and aio_complete() freeing the req.
512 */
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700513static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700515 struct kiocb *req;
516
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700517 if (atomic_read(&ctx->reqs_active) >= ctx->nr_events)
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700518 return NULL;
519
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700520 if (atomic_inc_return(&ctx->reqs_active) > ctx->nr_events - 1)
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700521 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Kent Overstreet0460fef2013-05-07 16:18:49 -0700523 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (unlikely(!req))
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700525 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Kent Overstreet11599eb2013-05-07 16:18:39 -0700527 atomic_set(&req->ki_users, 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 req->ki_ctx = ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return req;
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700531out_put:
532 atomic_dec(&ctx->reqs_active);
533 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535
Kent Overstreet11599eb2013-05-07 16:18:39 -0700536static void kiocb_free(struct kiocb *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
Kent Overstreet1d98ebf2013-05-07 16:18:37 -0700538 if (req->ki_filp)
539 fput(req->ki_filp);
Davide Libenzi13389012009-06-30 11:41:11 -0700540 if (req->ki_eventfd != NULL)
541 eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (req->ki_dtor)
543 req->ki_dtor(req);
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700544 if (req->ki_iovec != &req->ki_inline_vec)
545 kfree(req->ki_iovec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 kmem_cache_free(kiocb_cachep, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547}
548
Kent Overstreet2d684492013-05-07 16:18:29 -0700549void aio_put_req(struct kiocb *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Kent Overstreet11599eb2013-05-07 16:18:39 -0700551 if (atomic_dec_and_test(&req->ki_users))
552 kiocb_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700554EXPORT_SYMBOL(aio_put_req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Adrian Bunkd5470b52008-04-29 00:58:57 -0700556static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
Jens Axboeabf137d2008-12-09 08:11:22 +0100558 struct mm_struct *mm = current->mm;
Jeff Moyer65c24492009-03-18 17:04:21 -0700559 struct kioctx *ctx, *ret = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Jens Axboeabf137d2008-12-09 08:11:22 +0100561 rcu_read_lock();
562
Sasha Levinb67bfe02013-02-27 17:06:00 -0800563 hlist_for_each_entry_rcu(ctx, &mm->ioctx_list, list) {
Kent Overstreet36f55882013-05-07 16:18:41 -0700564 if (ctx->user_id == ctx_id) {
565 atomic_inc(&ctx->users);
Jeff Moyer65c24492009-03-18 17:04:21 -0700566 ret = ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 break;
568 }
Jens Axboeabf137d2008-12-09 08:11:22 +0100569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Jens Axboeabf137d2008-12-09 08:11:22 +0100571 rcu_read_unlock();
Jeff Moyer65c24492009-03-18 17:04:21 -0700572 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573}
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575/* aio_complete
576 * Called when the io request on the given iocb is complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 */
Kent Overstreet2d684492013-05-07 16:18:29 -0700578void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 struct kioctx *ctx = iocb->ki_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 struct aio_ring *ring;
Kent Overstreet21b40202013-05-07 16:18:47 -0700582 struct io_event *ev_page, *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 unsigned long flags;
Kent Overstreet21b40202013-05-07 16:18:47 -0700584 unsigned tail, pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Zach Brown20dcae32005-11-13 16:07:33 -0800586 /*
587 * Special case handling for sync iocbs:
588 * - events go directly into the iocb for fast handling
589 * - the sync task with the iocb in its stack holds the single iocb
590 * ref, no other paths have a way to get another ref
591 * - the sync task helpfully left a reference to itself in the iocb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 */
593 if (is_sync_kiocb(iocb)) {
Kent Overstreet11599eb2013-05-07 16:18:39 -0700594 BUG_ON(atomic_read(&iocb->ki_users) != 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 iocb->ki_user_data = res;
Kent Overstreet11599eb2013-05-07 16:18:39 -0700596 atomic_set(&iocb->ki_users, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 wake_up_process(iocb->ki_obj.tsk);
Kent Overstreet2d684492013-05-07 16:18:29 -0700598 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 }
600
Kent Overstreet36f55882013-05-07 16:18:41 -0700601 /*
Kent Overstreet36f55882013-05-07 16:18:41 -0700602 * Take rcu_read_lock() in case the kioctx is being destroyed, as we
603 * need to issue a wakeup after decrementing reqs_active.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 */
Kent Overstreet36f55882013-05-07 16:18:41 -0700605 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
Kent Overstreet0460fef2013-05-07 16:18:49 -0700607 if (iocb->ki_list.next) {
608 unsigned long flags;
609
610 spin_lock_irqsave(&ctx->ctx_lock, flags);
611 list_del(&iocb->ki_list);
612 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
613 }
Kent Overstreet11599eb2013-05-07 16:18:39 -0700614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 /*
616 * cancelled requests don't get events, userland was given one
617 * when the event got cancelled.
618 */
Kent Overstreet0460fef2013-05-07 16:18:49 -0700619 if (unlikely(xchg(&iocb->ki_cancel,
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700620 KIOCB_CANCELLED) == KIOCB_CANCELLED)) {
621 atomic_dec(&ctx->reqs_active);
622 /* Still need the wake_up in case free_ioctx is waiting */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 goto put_rq;
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Kent Overstreet0460fef2013-05-07 16:18:49 -0700626 /*
627 * Add a completion event to the ring buffer. Must be done holding
628 * ctx->ctx_lock to prevent other code from messing with the tail
629 * pointer since we might be called from irq context.
630 */
631 spin_lock_irqsave(&ctx->completion_lock, flags);
632
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700633 tail = ctx->tail;
Kent Overstreet21b40202013-05-07 16:18:47 -0700634 pos = tail + AIO_EVENTS_OFFSET;
635
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700636 if (++tail >= ctx->nr_events)
Ken Chen4bf69b22005-05-01 08:59:15 -0700637 tail = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700639 ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet21b40202013-05-07 16:18:47 -0700640 event = ev_page + pos % AIO_EVENTS_PER_PAGE;
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 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
Kent Overstreet21b40202013-05-07 16:18:47 -0700647 kunmap_atomic(ev_page);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700648 flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet21b40202013-05-07 16:18:47 -0700649
650 pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreetcaf41672013-05-07 16:18:35 -0700651 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
652 res, res2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 /* after flagging the request as done, we
655 * must never even look at it again
656 */
657 smp_wmb(); /* make event visible before updating tail */
658
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700659 ctx->tail = tail;
Kent Overstreet21b40202013-05-07 16:18:47 -0700660
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700661 ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 ring->tail = tail;
Cong Wange8e3c3d2011-11-25 23:14:27 +0800663 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700664 flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Kent Overstreet0460fef2013-05-07 16:18:49 -0700666 spin_unlock_irqrestore(&ctx->completion_lock, flags);
667
Kent Overstreet21b40202013-05-07 16:18:47 -0700668 pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi8d1c98b2008-04-10 21:29:19 -0700669
670 /*
671 * Check if the user asked us to deliver the result through an
672 * eventfd. The eventfd_signal() function is safe to be called
673 * from IRQ context.
674 */
Davide Libenzi87c3a862009-03-18 17:04:19 -0700675 if (iocb->ki_eventfd != NULL)
Davide Libenzi8d1c98b2008-04-10 21:29:19 -0700676 eventfd_signal(iocb->ki_eventfd, 1);
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678put_rq:
679 /* everything turned out well, dispose of the aiocb. */
Kent Overstreet11599eb2013-05-07 16:18:39 -0700680 aio_put_req(iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Quentin Barnes6cb2a212008-03-19 17:00:39 -0700682 /*
683 * We have to order our ring_info tail store above and test
684 * of the wait list below outside the wait lock. This is
685 * like in wake_up_bit() where clearing a bit has to be
686 * ordered with the unlocked test.
687 */
688 smp_mb();
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (waitqueue_active(&ctx->wait))
691 wake_up(&ctx->wait);
692
Kent Overstreet36f55882013-05-07 16:18:41 -0700693 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700695EXPORT_SYMBOL(aio_complete);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Kent Overstreeta31ad382013-05-07 16:18:45 -0700697/* aio_read_events
698 * Pull an event off of the ioctx's event ring. Returns the number of
699 * events fetched
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 */
Kent Overstreeta31ad382013-05-07 16:18:45 -0700701static long aio_read_events_ring(struct kioctx *ctx,
702 struct io_event __user *event, long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct aio_ring *ring;
Kent Overstreeta31ad382013-05-07 16:18:45 -0700705 unsigned head, pos;
706 long ret = 0;
707 int copy_ret;
708
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700709 mutex_lock(&ctx->ring_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700711 ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700712 head = ring->head;
713 kunmap_atomic(ring);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700715 pr_debug("h%u t%u m%u\n", head, ctx->tail, ctx->nr_events);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700716
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700717 if (head == ctx->tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 goto out;
719
Kent Overstreeta31ad382013-05-07 16:18:45 -0700720 while (ret < nr) {
721 long avail;
722 struct io_event *ev;
723 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700725 avail = (head <= ctx->tail ? ctx->tail : ctx->nr_events) - head;
726 if (head == ctx->tail)
Kent Overstreeta31ad382013-05-07 16:18:45 -0700727 break;
728
729 avail = min(avail, nr - ret);
730 avail = min_t(long, avail, AIO_EVENTS_PER_PAGE -
731 ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE));
732
733 pos = head + AIO_EVENTS_OFFSET;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700734 page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
Kent Overstreeta31ad382013-05-07 16:18:45 -0700735 pos %= AIO_EVENTS_PER_PAGE;
736
737 ev = kmap(page);
738 copy_ret = copy_to_user(event + ret, ev + pos,
739 sizeof(*ev) * avail);
740 kunmap(page);
741
742 if (unlikely(copy_ret)) {
743 ret = -EFAULT;
744 goto out;
745 }
746
747 ret += avail;
748 head += avail;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700749 head %= ctx->nr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700752 ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700753 ring->head = head;
Zhao Hongjiang91d80a82013-04-26 11:03:53 +0800754 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700755 flush_dcache_page(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700756
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700757 pr_debug("%li h%u t%u\n", ret, head, ctx->tail);
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700758
759 atomic_sub(ret, &ctx->reqs_active);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700760out:
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700761 mutex_unlock(&ctx->ring_lock);
Kent Overstreeta31ad382013-05-07 16:18:45 -0700762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return ret;
764}
765
Kent Overstreeta31ad382013-05-07 16:18:45 -0700766static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
767 struct io_event __user *event, long *i)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Kent Overstreeta31ad382013-05-07 16:18:45 -0700769 long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Kent Overstreeta31ad382013-05-07 16:18:45 -0700771 if (ret > 0)
772 *i += ret;
773
774 if (unlikely(atomic_read(&ctx->dead)))
775 ret = -EINVAL;
776
777 if (!*i)
778 *i = ret;
779
780 return ret < 0 || *i >= min_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
Kent Overstreeta31ad382013-05-07 16:18:45 -0700783static long read_events(struct kioctx *ctx, long min_nr, long nr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 struct io_event __user *event,
785 struct timespec __user *timeout)
786{
Kent Overstreeta31ad382013-05-07 16:18:45 -0700787 ktime_t until = { .tv64 = KTIME_MAX };
788 long ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 if (timeout) {
791 struct timespec ts;
Kent Overstreeta31ad382013-05-07 16:18:45 -0700792
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
Kent Overstreeta31ad382013-05-07 16:18:45 -0700794 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Kent Overstreeta31ad382013-05-07 16:18:45 -0700796 until = timespec_to_ktime(ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 }
798
Kent Overstreeta31ad382013-05-07 16:18:45 -0700799 /*
800 * Note that aio_read_events() is being called as the conditional - i.e.
801 * we're calling it after prepare_to_wait() has set task state to
802 * TASK_INTERRUPTIBLE.
803 *
804 * But aio_read_events() can block, and if it blocks it's going to flip
805 * the task state back to TASK_RUNNING.
806 *
807 * This should be ok, provided it doesn't flip the state back to
808 * TASK_RUNNING and return 0 too much - that causes us to spin. That
809 * will only happen if the mutex_lock() call blocks, and we then find
810 * the ringbuffer empty. So in practice we should be ok, but it's
811 * something to be aware of when touching this code.
812 */
813 wait_event_interruptible_hrtimeout(ctx->wait,
814 aio_read_events(ctx, min_nr, nr, event, &ret), until);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Kent Overstreeta31ad382013-05-07 16:18:45 -0700816 if (!ret && signal_pending(current))
817 ret = -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Kent Overstreeta31ad382013-05-07 16:18:45 -0700819 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
821
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822/* sys_io_setup:
823 * Create an aio_context capable of receiving at least nr_events.
824 * ctxp must not point to an aio_context that already exists, and
825 * must be initialized to 0 prior to the call. On successful
826 * creation of the aio_context, *ctxp is filled in with the resulting
827 * handle. May fail with -EINVAL if *ctxp is not initialized,
828 * if the specified nr_events exceeds internal limits. May fail
829 * with -EAGAIN if the specified nr_events exceeds the user's limit
830 * of available events. May fail with -ENOMEM if insufficient kernel
831 * resources are available. May fail with -EFAULT if an invalid
832 * pointer is passed for ctxp. Will fail with -ENOSYS if not
833 * implemented.
834 */
Heiko Carstens002c8972009-01-14 14:14:18 +0100835SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836{
837 struct kioctx *ioctx = NULL;
838 unsigned long ctx;
839 long ret;
840
841 ret = get_user(ctx, ctxp);
842 if (unlikely(ret))
843 goto out;
844
845 ret = -EINVAL;
Zach Brownd55b5fd2005-11-07 00:59:31 -0800846 if (unlikely(ctx || nr_events == 0)) {
847 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
848 ctx, nr_events);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 goto out;
850 }
851
852 ioctx = ioctx_alloc(nr_events);
853 ret = PTR_ERR(ioctx);
854 if (!IS_ERR(ioctx)) {
855 ret = put_user(ioctx->user_id, ctxp);
Al Viroa2e18592012-03-20 16:27:57 -0400856 if (ret)
Kent Overstreet36f55882013-05-07 16:18:41 -0700857 kill_ioctx(ioctx);
Al Viroa2e18592012-03-20 16:27:57 -0400858 put_ioctx(ioctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 }
860
861out:
862 return ret;
863}
864
865/* sys_io_destroy:
866 * Destroy the aio_context specified. May cancel any outstanding
867 * AIOs and block on completion. Will fail with -ENOSYS if not
Satoru Takeuchi642b5122010-08-05 11:23:11 -0700868 * implemented. May fail with -EINVAL if the context pointed to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 * is invalid.
870 */
Heiko Carstens002c8972009-01-14 14:14:18 +0100871SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872{
873 struct kioctx *ioctx = lookup_ioctx(ctx);
874 if (likely(NULL != ioctx)) {
Kent Overstreet36f55882013-05-07 16:18:41 -0700875 kill_ioctx(ioctx);
Al Viroa2e18592012-03-20 16:27:57 -0400876 put_ioctx(ioctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return 0;
878 }
879 pr_debug("EINVAL: io_destroy: invalid context id\n");
880 return -EINVAL;
881}
882
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700883static void aio_advance_iovec(struct kiocb *iocb, ssize_t ret)
884{
885 struct iovec *iov = &iocb->ki_iovec[iocb->ki_cur_seg];
886
887 BUG_ON(ret <= 0);
888
889 while (iocb->ki_cur_seg < iocb->ki_nr_segs && ret > 0) {
890 ssize_t this = min((ssize_t)iov->iov_len, ret);
891 iov->iov_base += this;
892 iov->iov_len -= this;
893 iocb->ki_left -= this;
894 ret -= this;
895 if (iov->iov_len == 0) {
896 iocb->ki_cur_seg++;
897 iov++;
898 }
899 }
900
901 /* the caller should not have done more io than what fit in
902 * the remaining iovecs */
903 BUG_ON(ret > 0 && iocb->ki_left == 0);
904}
905
906static ssize_t aio_rw_vect_retry(struct kiocb *iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907{
908 struct file *file = iocb->ki_filp;
909 struct address_space *mapping = file->f_mapping;
910 struct inode *inode = mapping->host;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700911 ssize_t (*rw_op)(struct kiocb *, const struct iovec *,
912 unsigned long, loff_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 ssize_t ret = 0;
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700914 unsigned short opcode;
915
916 if ((iocb->ki_opcode == IOCB_CMD_PREADV) ||
917 (iocb->ki_opcode == IOCB_CMD_PREAD)) {
918 rw_op = file->f_op->aio_read;
919 opcode = IOCB_CMD_PREADV;
920 } else {
921 rw_op = file->f_op->aio_write;
922 opcode = IOCB_CMD_PWRITEV;
923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
Rusty Russellc2ec6682008-02-08 04:20:15 -0800925 /* This matches the pread()/pwrite() logic */
926 if (iocb->ki_pos < 0)
927 return -EINVAL;
928
Al Viro8d71db42013-03-19 21:01:03 -0400929 if (opcode == IOCB_CMD_PWRITEV)
930 file_start_write(file);
Zach Brown897f15f2005-09-30 11:58:55 -0700931 do {
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700932 ret = rw_op(iocb, &iocb->ki_iovec[iocb->ki_cur_seg],
933 iocb->ki_nr_segs - iocb->ki_cur_seg,
934 iocb->ki_pos);
935 if (ret > 0)
936 aio_advance_iovec(iocb, ret);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700937
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700938 /* retry all partial writes. retry partial reads as long as its a
939 * regular file. */
Zach Brown353fb072005-09-30 11:58:56 -0700940 } while (ret > 0 && iocb->ki_left > 0 &&
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700941 (opcode == IOCB_CMD_PWRITEV ||
942 (!S_ISFIFO(inode->i_mode) && !S_ISSOCK(inode->i_mode))));
Al Viro8d71db42013-03-19 21:01:03 -0400943 if (opcode == IOCB_CMD_PWRITEV)
944 file_end_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 /* This means we must have transferred all that we could */
947 /* No need to retry anymore */
948 if ((ret == 0) || (iocb->ki_left == 0))
949 ret = iocb->ki_nbytes - iocb->ki_left;
950
Rusty Russell7adfa2f2008-02-08 04:20:14 -0800951 /* If we managed to write some out we return that, rather than
952 * the eventual error. */
953 if (opcode == IOCB_CMD_PWRITEV
Zach Brown41003a72013-05-07 16:18:25 -0700954 && ret < 0 && ret != -EIOCBQUEUED
Rusty Russell7adfa2f2008-02-08 04:20:14 -0800955 && iocb->ki_nbytes - iocb->ki_left)
956 ret = iocb->ki_nbytes - iocb->ki_left;
957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 return ret;
959}
960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961static ssize_t aio_fdsync(struct kiocb *iocb)
962{
963 struct file *file = iocb->ki_filp;
964 ssize_t ret = -EINVAL;
965
966 if (file->f_op->aio_fsync)
967 ret = file->f_op->aio_fsync(iocb, 1);
968 return ret;
969}
970
971static ssize_t aio_fsync(struct kiocb *iocb)
972{
973 struct file *file = iocb->ki_filp;
974 ssize_t ret = -EINVAL;
975
976 if (file->f_op->aio_fsync)
977 ret = file->f_op->aio_fsync(iocb, 0);
978 return ret;
979}
980
Jeff Moyer9d85cba2010-05-26 14:44:26 -0700981static ssize_t aio_setup_vectored_rw(int type, struct kiocb *kiocb, bool compat)
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700982{
983 ssize_t ret;
984
Jeff Moyer9d85cba2010-05-26 14:44:26 -0700985#ifdef CONFIG_COMPAT
986 if (compat)
987 ret = compat_rw_copy_check_uvector(type,
988 (struct compat_iovec __user *)kiocb->ki_buf,
989 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700990 &kiocb->ki_iovec);
Jeff Moyer9d85cba2010-05-26 14:44:26 -0700991 else
992#endif
993 ret = rw_copy_check_uvector(type,
994 (struct iovec __user *)kiocb->ki_buf,
995 kiocb->ki_nbytes, 1, &kiocb->ki_inline_vec,
Christopher Yeohac34ebb2012-05-31 16:26:42 -0700996 &kiocb->ki_iovec);
Badari Pulavartyeed4e512006-09-30 23:28:49 -0700997 if (ret < 0)
998 goto out;
999
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001000 ret = rw_verify_area(type, kiocb->ki_filp, &kiocb->ki_pos, ret);
1001 if (ret < 0)
1002 goto out;
1003
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001004 kiocb->ki_nr_segs = kiocb->ki_nbytes;
1005 kiocb->ki_cur_seg = 0;
1006 /* ki_nbytes/left now reflect bytes instead of segs */
1007 kiocb->ki_nbytes = ret;
1008 kiocb->ki_left = ret;
1009
1010 ret = 0;
1011out:
1012 return ret;
1013}
1014
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001015static ssize_t aio_setup_single_vector(int type, struct file * file, struct kiocb *kiocb)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001016{
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001017 int bytes;
1018
1019 bytes = rw_verify_area(type, file, &kiocb->ki_pos, kiocb->ki_left);
1020 if (bytes < 0)
1021 return bytes;
1022
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001023 kiocb->ki_iovec = &kiocb->ki_inline_vec;
1024 kiocb->ki_iovec->iov_base = kiocb->ki_buf;
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001025 kiocb->ki_iovec->iov_len = bytes;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001026 kiocb->ki_nr_segs = 1;
1027 kiocb->ki_cur_seg = 0;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001028 return 0;
1029}
1030
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031/*
1032 * aio_setup_iocb:
1033 * Performs the initial checks and aio retry method
1034 * setup for the kiocb at the time of io submission.
1035 */
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001036static ssize_t aio_setup_iocb(struct kiocb *kiocb, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 struct file *file = kiocb->ki_filp;
1039 ssize_t ret = 0;
1040
1041 switch (kiocb->ki_opcode) {
1042 case IOCB_CMD_PREAD:
1043 ret = -EBADF;
1044 if (unlikely(!(file->f_mode & FMODE_READ)))
1045 break;
1046 ret = -EFAULT;
1047 if (unlikely(!access_ok(VERIFY_WRITE, kiocb->ki_buf,
1048 kiocb->ki_left)))
1049 break;
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001050 ret = aio_setup_single_vector(READ, file, kiocb);
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001051 if (ret)
1052 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 ret = -EINVAL;
1054 if (file->f_op->aio_read)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001055 kiocb->ki_retry = aio_rw_vect_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 break;
1057 case IOCB_CMD_PWRITE:
1058 ret = -EBADF;
1059 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1060 break;
1061 ret = -EFAULT;
1062 if (unlikely(!access_ok(VERIFY_READ, kiocb->ki_buf,
1063 kiocb->ki_left)))
1064 break;
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001065 ret = aio_setup_single_vector(WRITE, file, kiocb);
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001066 if (ret)
1067 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 ret = -EINVAL;
1069 if (file->f_op->aio_write)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001070 kiocb->ki_retry = aio_rw_vect_retry;
1071 break;
1072 case IOCB_CMD_PREADV:
1073 ret = -EBADF;
1074 if (unlikely(!(file->f_mode & FMODE_READ)))
1075 break;
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001076 ret = aio_setup_vectored_rw(READ, kiocb, compat);
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001077 if (ret)
1078 break;
1079 ret = -EINVAL;
1080 if (file->f_op->aio_read)
1081 kiocb->ki_retry = aio_rw_vect_retry;
1082 break;
1083 case IOCB_CMD_PWRITEV:
1084 ret = -EBADF;
1085 if (unlikely(!(file->f_mode & FMODE_WRITE)))
1086 break;
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001087 ret = aio_setup_vectored_rw(WRITE, kiocb, compat);
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001088 if (ret)
1089 break;
1090 ret = -EINVAL;
1091 if (file->f_op->aio_write)
1092 kiocb->ki_retry = aio_rw_vect_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 break;
1094 case IOCB_CMD_FDSYNC:
1095 ret = -EINVAL;
1096 if (file->f_op->aio_fsync)
1097 kiocb->ki_retry = aio_fdsync;
1098 break;
1099 case IOCB_CMD_FSYNC:
1100 ret = -EINVAL;
1101 if (file->f_op->aio_fsync)
1102 kiocb->ki_retry = aio_fsync;
1103 break;
1104 default:
Kent Overstreetcaf41672013-05-07 16:18:35 -07001105 pr_debug("EINVAL: no operation provided\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 ret = -EINVAL;
1107 }
1108
1109 if (!kiocb->ki_retry)
1110 return ret;
1111
1112 return 0;
1113}
1114
Adrian Bunkd5470b52008-04-29 00:58:57 -07001115static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
Kent Overstreeta1c8eae2013-05-07 16:18:53 -07001116 struct iocb *iocb, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117{
1118 struct kiocb *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 ssize_t ret;
1120
1121 /* enforce forwards compatibility on users */
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001122 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001123 pr_debug("EINVAL: reserve field set\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return -EINVAL;
1125 }
1126
1127 /* prevent overflows */
1128 if (unlikely(
1129 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1130 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1131 ((ssize_t)iocb->aio_nbytes < 0)
1132 )) {
1133 pr_debug("EINVAL: io_submit: overflow check\n");
1134 return -EINVAL;
1135 }
1136
Kent Overstreeta1c8eae2013-05-07 16:18:53 -07001137 req = aio_get_req(ctx); /* returns with 2 references to req */
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001138 if (unlikely(!req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 return -EAGAIN;
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001140
1141 req->ki_filp = fget(iocb->aio_fildes);
1142 if (unlikely(!req->ki_filp)) {
1143 ret = -EBADF;
1144 goto out_put_req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 }
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001146
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001147 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1148 /*
1149 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1150 * instance of the file* now. The file descriptor must be
1151 * an eventfd() fd, and will be signaled for each completed
1152 * event using the eventfd_signal() function.
1153 */
Davide Libenzi13389012009-06-30 11:41:11 -07001154 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -07001155 if (IS_ERR(req->ki_eventfd)) {
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001156 ret = PTR_ERR(req->ki_eventfd);
Davide Libenzi87c3a862009-03-18 17:04:19 -07001157 req->ki_eventfd = NULL;
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001158 goto out_put_req;
1159 }
1160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Kent Overstreet8a660892013-05-07 16:19:10 -07001162 ret = put_user(KIOCB_KEY, &user_iocb->aio_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 if (unlikely(ret)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001164 pr_debug("EFAULT: aio_key\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 goto out_put_req;
1166 }
1167
1168 req->ki_obj.user = user_iocb;
1169 req->ki_user_data = iocb->aio_data;
1170 req->ki_pos = iocb->aio_offset;
1171
1172 req->ki_buf = (char __user *)(unsigned long)iocb->aio_buf;
1173 req->ki_left = req->ki_nbytes = iocb->aio_nbytes;
1174 req->ki_opcode = iocb->aio_lio_opcode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001176 ret = aio_setup_iocb(req, compat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (ret)
1178 goto out_put_req;
1179
Kent Overstreet0460fef2013-05-07 16:18:49 -07001180 ret = req->ki_retry(req);
Zach Brown41003a72013-05-07 16:18:25 -07001181 if (ret != -EIOCBQUEUED) {
1182 /*
1183 * There's no easy way to restart the syscall since other AIO's
1184 * may be already running. Just fail this IO with EINTR.
1185 */
1186 if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
1187 ret == -ERESTARTNOHAND ||
1188 ret == -ERESTART_RESTARTBLOCK))
1189 ret = -EINTR;
1190 aio_complete(req, ret, 0);
1191 }
Jeff Moyercfb1e332009-10-02 18:57:36 -04001192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 aio_put_req(req); /* drop extra ref to req */
1194 return 0;
1195
1196out_put_req:
Kent Overstreet11599eb2013-05-07 16:18:39 -07001197 atomic_dec(&ctx->reqs_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 aio_put_req(req); /* drop extra ref to req */
1199 aio_put_req(req); /* drop i/o ref to req */
1200 return ret;
1201}
1202
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001203long do_io_submit(aio_context_t ctx_id, long nr,
1204 struct iocb __user *__user *iocbpp, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
1206 struct kioctx *ctx;
1207 long ret = 0;
Jeff Moyer080d6762011-11-02 13:40:10 -07001208 int i = 0;
Shaohua Li9f5b9422010-07-01 07:55:01 +02001209 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 if (unlikely(nr < 0))
1212 return -EINVAL;
1213
Jeff Moyer75e1c702010-09-10 14:16:00 -07001214 if (unlikely(nr > LONG_MAX/sizeof(*iocbpp)))
1215 nr = LONG_MAX/sizeof(*iocbpp);
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1218 return -EFAULT;
1219
1220 ctx = lookup_ioctx(ctx_id);
1221 if (unlikely(!ctx)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001222 pr_debug("EINVAL: invalid context id\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 return -EINVAL;
1224 }
1225
Shaohua Li9f5b9422010-07-01 07:55:01 +02001226 blk_start_plug(&plug);
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 /*
1229 * AKPM: should this return a partial result if some of the IOs were
1230 * successfully submitted?
1231 */
1232 for (i=0; i<nr; i++) {
1233 struct iocb __user *user_iocb;
1234 struct iocb tmp;
1235
1236 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1237 ret = -EFAULT;
1238 break;
1239 }
1240
1241 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1242 ret = -EFAULT;
1243 break;
1244 }
1245
Kent Overstreeta1c8eae2013-05-07 16:18:53 -07001246 ret = io_submit_one(ctx, user_iocb, &tmp, compat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 if (ret)
1248 break;
1249 }
Shaohua Li9f5b9422010-07-01 07:55:01 +02001250 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 put_ioctx(ctx);
1253 return i ? i : ret;
1254}
1255
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001256/* sys_io_submit:
1257 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1258 * the number of iocbs queued. May return -EINVAL if the aio_context
1259 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1260 * *iocbpp[0] is not properly initialized, if the operation specified
1261 * is invalid for the file descriptor in the iocb. May fail with
1262 * -EFAULT if any of the data structures point to invalid data. May
1263 * fail with -EBADF if the file descriptor specified in the first
1264 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1265 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1266 * fail with -ENOSYS if not implemented.
1267 */
1268SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
1269 struct iocb __user * __user *, iocbpp)
1270{
1271 return do_io_submit(ctx_id, nr, iocbpp, 0);
1272}
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274/* lookup_kiocb
1275 * Finds a given iocb for cancellation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -07001277static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1278 u32 key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279{
1280 struct list_head *pos;
Zach Brownd00689a2005-11-13 16:07:34 -08001281
1282 assert_spin_locked(&ctx->ctx_lock);
1283
Kent Overstreet8a660892013-05-07 16:19:10 -07001284 if (key != KIOCB_KEY)
1285 return NULL;
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 /* TODO: use a hash or array, this sucks. */
1288 list_for_each(pos, &ctx->active_reqs) {
1289 struct kiocb *kiocb = list_kiocb(pos);
Kent Overstreet8a660892013-05-07 16:19:10 -07001290 if (kiocb->ki_obj.user == iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 return kiocb;
1292 }
1293 return NULL;
1294}
1295
1296/* sys_io_cancel:
1297 * Attempts to cancel an iocb previously passed to io_submit. If
1298 * the operation is successfully cancelled, the resulting event is
1299 * copied into the memory pointed to by result without being placed
1300 * into the completion queue and 0 is returned. May fail with
1301 * -EFAULT if any of the data structures pointed to are invalid.
1302 * May fail with -EINVAL if aio_context specified by ctx_id is
1303 * invalid. May fail with -EAGAIN if the iocb specified was not
1304 * cancelled. Will fail with -ENOSYS if not implemented.
1305 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001306SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb,
1307 struct io_event __user *, result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
Kent Overstreet906b9732013-05-07 16:18:31 -07001309 struct io_event res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 struct kioctx *ctx;
1311 struct kiocb *kiocb;
1312 u32 key;
1313 int ret;
1314
1315 ret = get_user(key, &iocb->aio_key);
1316 if (unlikely(ret))
1317 return -EFAULT;
1318
1319 ctx = lookup_ioctx(ctx_id);
1320 if (unlikely(!ctx))
1321 return -EINVAL;
1322
1323 spin_lock_irq(&ctx->ctx_lock);
Kent Overstreet906b9732013-05-07 16:18:31 -07001324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 kiocb = lookup_kiocb(ctx, iocb, key);
Kent Overstreet906b9732013-05-07 16:18:31 -07001326 if (kiocb)
1327 ret = kiocb_cancel(ctx, kiocb, &res);
1328 else
1329 ret = -EINVAL;
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 spin_unlock_irq(&ctx->ctx_lock);
1332
Kent Overstreet906b9732013-05-07 16:18:31 -07001333 if (!ret) {
1334 /* Cancellation succeeded -- copy the result
1335 * into the user's buffer.
1336 */
1337 if (copy_to_user(result, &res, sizeof(res)))
1338 ret = -EFAULT;
1339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
1341 put_ioctx(ctx);
1342
1343 return ret;
1344}
1345
1346/* io_getevents:
1347 * Attempts to read at least min_nr events and up to nr events from
Satoru Takeuchi642b5122010-08-05 11:23:11 -07001348 * the completion queue for the aio_context specified by ctx_id. If
1349 * it succeeds, the number of read events is returned. May fail with
1350 * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is
1351 * out of range, if timeout is out of range. May fail with -EFAULT
1352 * if any of the memory specified is invalid. May return 0 or
1353 * < min_nr if the timeout specified by timeout has elapsed
1354 * before sufficient events are available, where timeout == NULL
1355 * specifies an infinite timeout. Note that the timeout pointed to by
1356 * timeout is relative and will be updated if not NULL and the
1357 * operation blocks. Will fail with -ENOSYS if not implemented.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001359SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
1360 long, min_nr,
1361 long, nr,
1362 struct io_event __user *, events,
1363 struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
1365 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1366 long ret = -EINVAL;
1367
1368 if (likely(ioctx)) {
Namhyung Kim2e410252011-01-12 17:01:08 -08001369 if (likely(min_nr <= nr && min_nr >= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 ret = read_events(ioctx, min_nr, nr, events, timeout);
1371 put_ioctx(ioctx);
1372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 return ret;
1374}