blob: a0ed6c7d2cd2a3e91a5d12e48af705d75afe315a [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>
Kent Overstreete1bdd5f2013-04-26 10:58:39 +100029#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/slab.h>
31#include <linux/timer.h>
32#include <linux/aio.h>
33#include <linux/highmem.h>
34#include <linux/workqueue.h>
35#include <linux/security.h>
Davide Libenzi9c3060b2007-05-10 22:23:21 -070036#include <linux/eventfd.h>
Jeff Moyercfb1e332009-10-02 18:57:36 -040037#include <linux/blkdev.h>
Jeff Moyer9d85cba2010-05-26 14:44:26 -070038#include <linux/compat.h>
Gu Zheng36bc08c2013-07-16 17:56:16 +080039#include <linux/migrate.h>
40#include <linux/ramfs.h>
Kent Overstreet723be6e2013-05-28 15:14:48 -070041#include <linux/percpu-refcount.h>
Benjamin LaHaise71ad7492013-09-17 10:18:25 -040042#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44#include <asm/kmap_types.h>
45#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Al Viro68d70d02013-06-19 15:26:04 +040047#include "internal.h"
48
Kent Overstreet4e179bc2013-05-07 16:18:33 -070049#define AIO_RING_MAGIC 0xa10a10a1
50#define AIO_RING_COMPAT_FEATURES 1
51#define AIO_RING_INCOMPAT_FEATURES 0
52struct aio_ring {
53 unsigned id; /* kernel internal index number */
54 unsigned nr; /* number of io_events */
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -040055 unsigned head; /* Written to by userland or under ring_lock
56 * mutex by aio_read_events_ring(). */
Kent Overstreet4e179bc2013-05-07 16:18:33 -070057 unsigned tail;
58
59 unsigned magic;
60 unsigned compat_features;
61 unsigned incompat_features;
62 unsigned header_length; /* size of aio_ring */
63
64
65 struct io_event io_events[0];
66}; /* 128 bytes + ring size */
67
68#define AIO_RING_PAGES 8
Kent Overstreet4e179bc2013-05-07 16:18:33 -070069
Benjamin LaHaisedb446a02013-07-30 12:54:40 -040070struct kioctx_table {
71 struct rcu_head rcu;
72 unsigned nr;
73 struct kioctx *table[];
74};
75
Kent Overstreete1bdd5f2013-04-26 10:58:39 +100076struct kioctx_cpu {
77 unsigned reqs_available;
78};
79
Kent Overstreet4e179bc2013-05-07 16:18:33 -070080struct kioctx {
Kent Overstreet723be6e2013-05-28 15:14:48 -070081 struct percpu_ref users;
Kent Overstreet36f55882013-05-07 16:18:41 -070082 atomic_t dead;
Kent Overstreet4e179bc2013-05-07 16:18:33 -070083
Kent Overstreete34ecee2013-10-10 19:31:47 -070084 struct percpu_ref reqs;
85
Kent Overstreet4e179bc2013-05-07 16:18:33 -070086 unsigned long user_id;
Kent Overstreet4e179bc2013-05-07 16:18:33 -070087
Kent Overstreete1bdd5f2013-04-26 10:58:39 +100088 struct __percpu kioctx_cpu *cpu;
89
90 /*
91 * For percpu reqs_available, number of slots we move to/from global
92 * counter at a time:
93 */
94 unsigned req_batch;
Kent Overstreet3e845ce2013-05-07 16:18:51 -070095 /*
96 * This is what userspace passed to io_setup(), it's not used for
97 * anything but counting against the global max_reqs quota.
98 *
Kent Overstreet58c85dc2013-05-07 16:18:55 -070099 * The real limit is nr_events - 1, which will be larger (see
Kent Overstreet3e845ce2013-05-07 16:18:51 -0700100 * aio_setup_ring())
101 */
Kent Overstreet4e179bc2013-05-07 16:18:33 -0700102 unsigned max_reqs;
103
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700104 /* Size of ringbuffer, in units of struct io_event */
105 unsigned nr_events;
Kent Overstreet4e179bc2013-05-07 16:18:33 -0700106
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700107 unsigned long mmap_base;
108 unsigned long mmap_size;
109
110 struct page **ring_pages;
111 long nr_pages;
112
Kent Overstreet723be6e2013-05-28 15:14:48 -0700113 struct work_struct free_work;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700114
Anatol Pomozove02ba722014-04-15 11:31:33 -0700115 /*
116 * signals when all in-flight requests are done
117 */
118 struct completion *requests_done;
119
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700120 struct {
Kent Overstreet34e83fc2013-04-26 10:58:39 +1000121 /*
122 * This counts the number of available slots in the ringbuffer,
123 * so we avoid overflowing it: it's decremented (if positive)
124 * when allocating a kiocb and incremented when the resulting
125 * io_event is pulled off the ringbuffer.
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000126 *
127 * We batch accesses to it with a percpu version.
Kent Overstreet34e83fc2013-04-26 10:58:39 +1000128 */
129 atomic_t reqs_available;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700130 } ____cacheline_aligned_in_smp;
131
132 struct {
133 spinlock_t ctx_lock;
134 struct list_head active_reqs; /* used for cancellation */
135 } ____cacheline_aligned_in_smp;
136
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700137 struct {
138 struct mutex ring_lock;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700139 wait_queue_head_t wait;
140 } ____cacheline_aligned_in_smp;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700141
142 struct {
143 unsigned tail;
144 spinlock_t completion_lock;
Kent Overstreet4e23bca2013-05-07 16:18:56 -0700145 } ____cacheline_aligned_in_smp;
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700146
147 struct page *internal_pages[AIO_RING_PAGES];
Gu Zheng36bc08c2013-07-16 17:56:16 +0800148 struct file *aio_ring_file;
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400149
150 unsigned id;
Kent Overstreet4e179bc2013-05-07 16:18:33 -0700151};
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/*------ sysctl variables----*/
Zach Brownd55b5fd2005-11-07 00:59:31 -0800154static DEFINE_SPINLOCK(aio_nr_lock);
155unsigned long aio_nr; /* current system wide number of aio requests */
156unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157/*----end sysctl variables---*/
158
Christoph Lametere18b8902006-12-06 20:33:20 -0800159static struct kmem_cache *kiocb_cachep;
160static struct kmem_cache *kioctx_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Benjamin LaHaise71ad7492013-09-17 10:18:25 -0400162static struct vfsmount *aio_mnt;
163
164static const struct file_operations aio_ring_fops;
165static const struct address_space_operations aio_ctx_aops;
166
167static struct file *aio_private_file(struct kioctx *ctx, loff_t nr_pages)
168{
169 struct qstr this = QSTR_INIT("[aio]", 5);
170 struct file *file;
171 struct path path;
172 struct inode *inode = alloc_anon_inode(aio_mnt->mnt_sb);
Dan Carpenter7f626562013-11-13 10:49:40 +0300173 if (IS_ERR(inode))
174 return ERR_CAST(inode);
Benjamin LaHaise71ad7492013-09-17 10:18:25 -0400175
176 inode->i_mapping->a_ops = &aio_ctx_aops;
177 inode->i_mapping->private_data = ctx;
178 inode->i_size = PAGE_SIZE * nr_pages;
179
180 path.dentry = d_alloc_pseudo(aio_mnt->mnt_sb, &this);
181 if (!path.dentry) {
182 iput(inode);
183 return ERR_PTR(-ENOMEM);
184 }
185 path.mnt = mntget(aio_mnt);
186
187 d_instantiate(path.dentry, inode);
188 file = alloc_file(&path, FMODE_READ | FMODE_WRITE, &aio_ring_fops);
189 if (IS_ERR(file)) {
190 path_put(&path);
191 return file;
192 }
193
194 file->f_flags = O_RDWR;
195 file->private_data = ctx;
196 return file;
197}
198
199static struct dentry *aio_mount(struct file_system_type *fs_type,
200 int flags, const char *dev_name, void *data)
201{
202 static const struct dentry_operations ops = {
203 .d_dname = simple_dname,
204 };
205 return mount_pseudo(fs_type, "aio:", NULL, &ops, 0xa10a10a1);
206}
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208/* aio_setup
209 * Creates the slab caches used by the aio routines, panic on
210 * failure as this is done early during the boot sequence.
211 */
212static int __init aio_setup(void)
213{
Benjamin LaHaise71ad7492013-09-17 10:18:25 -0400214 static struct file_system_type aio_fs = {
215 .name = "aio",
216 .mount = aio_mount,
217 .kill_sb = kill_anon_super,
218 };
219 aio_mnt = kern_mount(&aio_fs);
220 if (IS_ERR(aio_mnt))
221 panic("Failed to create aio fs mount.");
222
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700223 kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
224 kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Kent Overstreetcaf41672013-05-07 16:18:35 -0700226 pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 return 0;
229}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700230__initcall(aio_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Benjamin LaHaise5e9ae2e2013-09-26 20:34:51 -0400232static void put_aio_ring_file(struct kioctx *ctx)
233{
234 struct file *aio_ring_file = ctx->aio_ring_file;
235 if (aio_ring_file) {
236 truncate_setsize(aio_ring_file->f_inode, 0);
237
238 /* Prevent further access to the kioctx from migratepages */
239 spin_lock(&aio_ring_file->f_inode->i_mapping->private_lock);
240 aio_ring_file->f_inode->i_mapping->private_data = NULL;
241 ctx->aio_ring_file = NULL;
242 spin_unlock(&aio_ring_file->f_inode->i_mapping->private_lock);
243
244 fput(aio_ring_file);
245 }
246}
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248static void aio_free_ring(struct kioctx *ctx)
249{
Gu Zheng36bc08c2013-07-16 17:56:16 +0800250 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400252 /* Disconnect the kiotx from the ring file. This prevents future
253 * accesses to the kioctx from page migration.
254 */
255 put_aio_ring_file(ctx);
256
Gu Zheng36bc08c2013-07-16 17:56:16 +0800257 for (i = 0; i < ctx->nr_pages; i++) {
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500258 struct page *page;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800259 pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i,
260 page_count(ctx->ring_pages[i]));
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500261 page = ctx->ring_pages[i];
262 if (!page)
263 continue;
264 ctx->ring_pages[i] = NULL;
265 put_page(page);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Sasha Levinddb8c452013-11-19 17:33:03 -0500268 if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) {
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700269 kfree(ctx->ring_pages);
Sasha Levinddb8c452013-11-19 17:33:03 -0500270 ctx->ring_pages = NULL;
271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Gu Zheng36bc08c2013-07-16 17:56:16 +0800274static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma)
275{
276 vma->vm_ops = &generic_file_vm_ops;
277 return 0;
278}
279
280static const struct file_operations aio_ring_fops = {
281 .mmap = aio_ring_mmap,
282};
283
284static int aio_set_page_dirty(struct page *page)
285{
286 return 0;
287}
288
Benjamin LaHaise0c453552013-07-17 09:34:24 -0400289#if IS_ENABLED(CONFIG_MIGRATION)
Gu Zheng36bc08c2013-07-16 17:56:16 +0800290static int aio_migratepage(struct address_space *mapping, struct page *new,
291 struct page *old, enum migrate_mode mode)
292{
Benjamin LaHaise5e9ae2e2013-09-26 20:34:51 -0400293 struct kioctx *ctx;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800294 unsigned long flags;
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400295 pgoff_t idx;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800296 int rc;
297
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500298 rc = 0;
299
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400300 /* mapping->private_lock here protects against the kioctx teardown. */
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500301 spin_lock(&mapping->private_lock);
302 ctx = mapping->private_data;
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400303 if (!ctx) {
304 rc = -EINVAL;
305 goto out;
306 }
307
308 /* The ring_lock mutex. The prevents aio_read_events() from writing
309 * to the ring's head, and prevents page migration from mucking in
310 * a partially initialized kiotx.
311 */
312 if (!mutex_trylock(&ctx->ring_lock)) {
313 rc = -EAGAIN;
314 goto out;
315 }
316
317 idx = old->index;
318 if (idx < (pgoff_t)ctx->nr_pages) {
319 /* Make sure the old page hasn't already been changed */
320 if (ctx->ring_pages[idx] != old)
321 rc = -EAGAIN;
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500322 } else
323 rc = -EINVAL;
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500324
325 if (rc != 0)
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400326 goto out_unlock;
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500327
Gu Zheng36bc08c2013-07-16 17:56:16 +0800328 /* Writeback must be complete */
329 BUG_ON(PageWriteback(old));
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500330 get_page(new);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800331
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500332 rc = migrate_page_move_mapping(mapping, new, old, NULL, mode, 1);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800333 if (rc != MIGRATEPAGE_SUCCESS) {
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500334 put_page(new);
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400335 goto out_unlock;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800336 }
337
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400338 /* Take completion_lock to prevent other writes to the ring buffer
339 * while the old page is copied to the new. This prevents new
340 * events from being lost.
Benjamin LaHaise5e9ae2e2013-09-26 20:34:51 -0400341 */
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400342 spin_lock_irqsave(&ctx->completion_lock, flags);
343 migrate_page_copy(new, old);
344 BUG_ON(ctx->ring_pages[idx] != old);
345 ctx->ring_pages[idx] = new;
346 spin_unlock_irqrestore(&ctx->completion_lock, flags);
347
348 /* The old page is no longer accessible. */
349 put_page(old);
350
351out_unlock:
352 mutex_unlock(&ctx->ring_lock);
353out:
Benjamin LaHaise5e9ae2e2013-09-26 20:34:51 -0400354 spin_unlock(&mapping->private_lock);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800355 return rc;
356}
Benjamin LaHaise0c453552013-07-17 09:34:24 -0400357#endif
Gu Zheng36bc08c2013-07-16 17:56:16 +0800358
359static const struct address_space_operations aio_ctx_aops = {
360 .set_page_dirty = aio_set_page_dirty,
Benjamin LaHaise0c453552013-07-17 09:34:24 -0400361#if IS_ENABLED(CONFIG_MIGRATION)
Gu Zheng36bc08c2013-07-16 17:56:16 +0800362 .migratepage = aio_migratepage,
Benjamin LaHaise0c453552013-07-17 09:34:24 -0400363#endif
Gu Zheng36bc08c2013-07-16 17:56:16 +0800364};
365
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366static int aio_setup_ring(struct kioctx *ctx)
367{
368 struct aio_ring *ring;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 unsigned nr_events = ctx->max_reqs;
Zach Brown41003a72013-05-07 16:18:25 -0700370 struct mm_struct *mm = current->mm;
Linus Torvalds3dc9acb2013-12-20 05:11:12 +0900371 unsigned long size, unused;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 int nr_pages;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800373 int i;
374 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376 /* Compensate for the ring buffer's head/tail overlap entry */
377 nr_events += 2; /* 1 is required, 2 for good luck */
378
379 size = sizeof(struct aio_ring);
380 size += sizeof(struct io_event) * nr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Gu Zheng36bc08c2013-07-16 17:56:16 +0800382 nr_pages = PFN_UP(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 if (nr_pages < 0)
384 return -EINVAL;
385
Benjamin LaHaise71ad7492013-09-17 10:18:25 -0400386 file = aio_private_file(ctx, nr_pages);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800387 if (IS_ERR(file)) {
388 ctx->aio_ring_file = NULL;
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400389 return -ENOMEM;
Gu Zheng36bc08c2013-07-16 17:56:16 +0800390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Gu Zheng36bc08c2013-07-16 17:56:16 +0800392 ctx->aio_ring_file = file;
393 nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring))
394 / sizeof(struct io_event);
395
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700396 ctx->ring_pages = ctx->internal_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 if (nr_pages > AIO_RING_PAGES) {
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700398 ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *),
399 GFP_KERNEL);
Gu Zhengd1b94322013-12-04 18:19:06 +0800400 if (!ctx->ring_pages) {
401 put_aio_ring_file(ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 return -ENOMEM;
Gu Zhengd1b94322013-12-04 18:19:06 +0800403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 }
405
Linus Torvalds3dc9acb2013-12-20 05:11:12 +0900406 for (i = 0; i < nr_pages; i++) {
407 struct page *page;
408 page = find_or_create_page(file->f_inode->i_mapping,
409 i, GFP_HIGHUSER | __GFP_ZERO);
410 if (!page)
411 break;
412 pr_debug("pid(%d) page[%d]->count=%d\n",
413 current->pid, i, page_count(page));
414 SetPageUptodate(page);
415 SetPageDirty(page);
416 unlock_page(page);
417
418 ctx->ring_pages[i] = page;
419 }
420 ctx->nr_pages = i;
421
422 if (unlikely(i != nr_pages)) {
423 aio_free_ring(ctx);
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400424 return -ENOMEM;
Linus Torvalds3dc9acb2013-12-20 05:11:12 +0900425 }
426
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700427 ctx->mmap_size = nr_pages * PAGE_SIZE;
428 pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800429
Zach Brown41003a72013-05-07 16:18:25 -0700430 down_write(&mm->mmap_sem);
Gu Zheng36bc08c2013-07-16 17:56:16 +0800431 ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size,
432 PROT_READ | PROT_WRITE,
Linus Torvalds3dc9acb2013-12-20 05:11:12 +0900433 MAP_SHARED, 0, &unused);
434 up_write(&mm->mmap_sem);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700435 if (IS_ERR((void *)ctx->mmap_base)) {
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700436 ctx->mmap_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 aio_free_ring(ctx);
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400438 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 }
440
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700441 pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base);
Benjamin LaHaised6c355c2013-09-09 11:57:59 -0400442
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700443 ctx->user_id = ctx->mmap_base;
444 ctx->nr_events = nr_events; /* trusted copy */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700446 ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 ring->nr = nr_events; /* user copy */
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400448 ring->id = ~0U;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 ring->head = ring->tail = 0;
450 ring->magic = AIO_RING_MAGIC;
451 ring->compat_features = AIO_RING_COMPAT_FEATURES;
452 ring->incompat_features = AIO_RING_INCOMPAT_FEATURES;
453 ring->header_length = sizeof(struct aio_ring);
Cong Wange8e3c3d2011-11-25 23:14:27 +0800454 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700455 flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 return 0;
458}
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460#define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event))
461#define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event))
462#define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE)
463
Kent Overstreet0460fef2013-05-07 16:18:49 -0700464void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel)
465{
466 struct kioctx *ctx = req->ki_ctx;
467 unsigned long flags;
468
469 spin_lock_irqsave(&ctx->ctx_lock, flags);
470
471 if (!req->ki_list.next)
472 list_add(&req->ki_list, &ctx->active_reqs);
473
474 req->ki_cancel = cancel;
475
476 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
477}
478EXPORT_SYMBOL(kiocb_set_cancel_fn);
479
Kent Overstreetbec68faa2013-05-13 14:45:08 -0700480static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb)
Kent Overstreet906b9732013-05-07 16:18:31 -0700481{
Kent Overstreet0460fef2013-05-07 16:18:49 -0700482 kiocb_cancel_fn *old, *cancel;
Kent Overstreet906b9732013-05-07 16:18:31 -0700483
Kent Overstreet0460fef2013-05-07 16:18:49 -0700484 /*
485 * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it
486 * actually has a cancel function, hence the cmpxchg()
487 */
Kent Overstreet906b9732013-05-07 16:18:31 -0700488
Kent Overstreet0460fef2013-05-07 16:18:49 -0700489 cancel = ACCESS_ONCE(kiocb->ki_cancel);
490 do {
491 if (!cancel || cancel == KIOCB_CANCELLED)
Kent Overstreet57282d82013-05-13 13:42:52 -0700492 return -EINVAL;
Kent Overstreet906b9732013-05-07 16:18:31 -0700493
Kent Overstreet0460fef2013-05-07 16:18:49 -0700494 old = cancel;
495 cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED);
496 } while (cancel != old);
497
Kent Overstreet57282d82013-05-13 13:42:52 -0700498 return cancel(kiocb);
Kent Overstreet906b9732013-05-07 16:18:31 -0700499}
500
Kent Overstreete34ecee2013-10-10 19:31:47 -0700501static void free_ioctx(struct work_struct *work)
Kent Overstreet36f55882013-05-07 16:18:41 -0700502{
Kent Overstreete34ecee2013-10-10 19:31:47 -0700503 struct kioctx *ctx = container_of(work, struct kioctx, free_work);
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000504
Kent Overstreete34ecee2013-10-10 19:31:47 -0700505 pr_debug("freeing %p\n", ctx);
506
507 aio_free_ring(ctx);
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000508 free_percpu(ctx->cpu);
Kent Overstreet36f55882013-05-07 16:18:41 -0700509 kmem_cache_free(kioctx_cachep, ctx);
510}
511
Kent Overstreete34ecee2013-10-10 19:31:47 -0700512static void free_ioctx_reqs(struct percpu_ref *ref)
513{
514 struct kioctx *ctx = container_of(ref, struct kioctx, reqs);
515
Anatol Pomozove02ba722014-04-15 11:31:33 -0700516 /* At this point we know that there are no any in-flight requests */
517 if (ctx->requests_done)
518 complete(ctx->requests_done);
519
Kent Overstreete34ecee2013-10-10 19:31:47 -0700520 INIT_WORK(&ctx->free_work, free_ioctx);
521 schedule_work(&ctx->free_work);
522}
523
Kent Overstreet36f55882013-05-07 16:18:41 -0700524/*
525 * When this function runs, the kioctx has been removed from the "hash table"
526 * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted -
527 * now it's safe to cancel any that need to be.
528 */
Kent Overstreete34ecee2013-10-10 19:31:47 -0700529static void free_ioctx_users(struct percpu_ref *ref)
Kent Overstreet36f55882013-05-07 16:18:41 -0700530{
Kent Overstreete34ecee2013-10-10 19:31:47 -0700531 struct kioctx *ctx = container_of(ref, struct kioctx, users);
Kent Overstreet36f55882013-05-07 16:18:41 -0700532 struct kiocb *req;
533
534 spin_lock_irq(&ctx->ctx_lock);
535
536 while (!list_empty(&ctx->active_reqs)) {
537 req = list_first_entry(&ctx->active_reqs,
538 struct kiocb, ki_list);
539
540 list_del_init(&req->ki_list);
Kent Overstreetbec68faa2013-05-13 14:45:08 -0700541 kiocb_cancel(ctx, req);
Kent Overstreet36f55882013-05-07 16:18:41 -0700542 }
543
544 spin_unlock_irq(&ctx->ctx_lock);
545
Kent Overstreete34ecee2013-10-10 19:31:47 -0700546 percpu_ref_kill(&ctx->reqs);
547 percpu_ref_put(&ctx->reqs);
Kent Overstreet36f55882013-05-07 16:18:41 -0700548}
549
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400550static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm)
551{
552 unsigned i, new_nr;
553 struct kioctx_table *table, *old;
554 struct aio_ring *ring;
555
556 spin_lock(&mm->ioctx_lock);
Artem Savkovd9b2c872013-09-09 12:29:35 -0400557 rcu_read_lock();
Benjamin LaHaise77d30b12013-08-30 11:12:50 -0400558 table = rcu_dereference(mm->ioctx_table);
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400559
560 while (1) {
561 if (table)
562 for (i = 0; i < table->nr; i++)
563 if (!table->table[i]) {
564 ctx->id = i;
565 table->table[i] = ctx;
Artem Savkovd9b2c872013-09-09 12:29:35 -0400566 rcu_read_unlock();
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400567 spin_unlock(&mm->ioctx_lock);
568
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400569 /* While kioctx setup is in progress,
570 * we are protected from page migration
571 * changes ring_pages by ->ring_lock.
572 */
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400573 ring = kmap_atomic(ctx->ring_pages[0]);
574 ring->id = ctx->id;
575 kunmap_atomic(ring);
576 return 0;
577 }
578
579 new_nr = (table ? table->nr : 1) * 4;
580
Artem Savkovd9b2c872013-09-09 12:29:35 -0400581 rcu_read_unlock();
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400582 spin_unlock(&mm->ioctx_lock);
583
584 table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) *
585 new_nr, GFP_KERNEL);
586 if (!table)
587 return -ENOMEM;
588
589 table->nr = new_nr;
590
591 spin_lock(&mm->ioctx_lock);
Artem Savkovd9b2c872013-09-09 12:29:35 -0400592 rcu_read_lock();
Benjamin LaHaise77d30b12013-08-30 11:12:50 -0400593 old = rcu_dereference(mm->ioctx_table);
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400594
595 if (!old) {
596 rcu_assign_pointer(mm->ioctx_table, table);
597 } else if (table->nr > old->nr) {
598 memcpy(table->table, old->table,
599 old->nr * sizeof(struct kioctx *));
600
601 rcu_assign_pointer(mm->ioctx_table, table);
602 kfree_rcu(old, rcu);
603 } else {
604 kfree(table);
605 table = old;
606 }
607 }
608}
609
Kent Overstreete34ecee2013-10-10 19:31:47 -0700610static void aio_nr_sub(unsigned nr)
611{
612 spin_lock(&aio_nr_lock);
613 if (WARN_ON(aio_nr - nr > aio_nr))
614 aio_nr = 0;
615 else
616 aio_nr -= nr;
617 spin_unlock(&aio_nr_lock);
618}
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620/* ioctx_alloc
621 * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed.
622 */
623static struct kioctx *ioctx_alloc(unsigned nr_events)
624{
Zach Brown41003a72013-05-07 16:18:25 -0700625 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 struct kioctx *ctx;
Al Viroe23754f2012-03-06 14:33:22 -0500627 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000629 /*
630 * We keep track of the number of available ringbuffer slots, to prevent
631 * overflow (reqs_available), and we also use percpu counters for this.
632 *
633 * So since up to half the slots might be on other cpu's percpu counters
634 * and unavailable, double nr_events so userspace sees what they
635 * expected: additionally, we move req_batch slots to/from percpu
636 * counters at a time, so make sure that isn't 0:
637 */
638 nr_events = max(nr_events, num_possible_cpus() * 4);
639 nr_events *= 2;
640
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 /* Prevent overflows */
642 if ((nr_events > (0x10000000U / sizeof(struct io_event))) ||
643 (nr_events > (0x10000000U / sizeof(struct kiocb)))) {
644 pr_debug("ENOMEM: nr_events too high\n");
645 return ERR_PTR(-EINVAL);
646 }
647
Benjamin LaHaise4cd81c32013-07-30 12:06:37 -0400648 if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 return ERR_PTR(-EAGAIN);
650
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800651 ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 if (!ctx)
653 return ERR_PTR(-ENOMEM);
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 ctx->max_reqs = nr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400657 spin_lock_init(&ctx->ctx_lock);
658 spin_lock_init(&ctx->completion_lock);
659 mutex_init(&ctx->ring_lock);
660 /* Protect against page migration throughout kiotx setup by keeping
661 * the ring_lock mutex held until setup is complete. */
662 mutex_lock(&ctx->ring_lock);
663 init_waitqueue_head(&ctx->wait);
664
665 INIT_LIST_HEAD(&ctx->active_reqs);
666
Kent Overstreete34ecee2013-10-10 19:31:47 -0700667 if (percpu_ref_init(&ctx->users, free_ioctx_users))
668 goto err;
669
670 if (percpu_ref_init(&ctx->reqs, free_ioctx_reqs))
671 goto err;
Kent Overstreet723be6e2013-05-28 15:14:48 -0700672
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000673 ctx->cpu = alloc_percpu(struct kioctx_cpu);
674 if (!ctx->cpu)
Kent Overstreete34ecee2013-10-10 19:31:47 -0700675 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400677 err = aio_setup_ring(ctx);
678 if (err < 0)
Kent Overstreete34ecee2013-10-10 19:31:47 -0700679 goto err;
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000680
Kent Overstreet34e83fc2013-04-26 10:58:39 +1000681 atomic_set(&ctx->reqs_available, ctx->nr_events - 1);
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000682 ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4);
Benjamin LaHaise6878ea72013-07-31 10:34:18 -0400683 if (ctx->req_batch < 1)
684 ctx->req_batch = 1;
Kent Overstreet34e83fc2013-04-26 10:58:39 +1000685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 /* limit the number of system wide aios */
Al Viro9fa1cb32012-03-10 23:14:05 -0500687 spin_lock(&aio_nr_lock);
Benjamin LaHaise4cd81c32013-07-30 12:06:37 -0400688 if (aio_nr + nr_events > (aio_max_nr * 2UL) ||
Al Viro2dd542b2012-03-10 23:10:35 -0500689 aio_nr + nr_events < aio_nr) {
Al Viro9fa1cb32012-03-10 23:14:05 -0500690 spin_unlock(&aio_nr_lock);
Kent Overstreete34ecee2013-10-10 19:31:47 -0700691 err = -EAGAIN;
Gu Zhengd1b94322013-12-04 18:19:06 +0800692 goto err_ctx;
Al Viro2dd542b2012-03-10 23:10:35 -0500693 }
694 aio_nr += ctx->max_reqs;
Al Viro9fa1cb32012-03-10 23:14:05 -0500695 spin_unlock(&aio_nr_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Benjamin LaHaise18816862013-12-21 15:49:28 -0500697 percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */
698 percpu_ref_get(&ctx->reqs); /* free_ioctx_users() will drop this */
Kent Overstreet723be6e2013-05-28 15:14:48 -0700699
Benjamin LaHaiseda903822013-08-05 13:21:43 -0400700 err = ioctx_add_table(ctx, mm);
701 if (err)
Kent Overstreete34ecee2013-10-10 19:31:47 -0700702 goto err_cleanup;
Benjamin LaHaiseda903822013-08-05 13:21:43 -0400703
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400704 /* Release the ring_lock mutex now that all setup is complete. */
705 mutex_unlock(&ctx->ring_lock);
706
Kent Overstreetcaf41672013-05-07 16:18:35 -0700707 pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n",
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700708 ctx, ctx->user_id, mm, ctx->nr_events);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 return ctx;
710
Kent Overstreete34ecee2013-10-10 19:31:47 -0700711err_cleanup:
712 aio_nr_sub(ctx->max_reqs);
Gu Zhengd1b94322013-12-04 18:19:06 +0800713err_ctx:
714 aio_free_ring(ctx);
Kent Overstreete34ecee2013-10-10 19:31:47 -0700715err:
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -0400716 mutex_unlock(&ctx->ring_lock);
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000717 free_percpu(ctx->cpu);
Kent Overstreete34ecee2013-10-10 19:31:47 -0700718 free_percpu(ctx->reqs.pcpu_count);
Kent Overstreet723be6e2013-05-28 15:14:48 -0700719 free_percpu(ctx->users.pcpu_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 kmem_cache_free(kioctx_cachep, ctx);
Kent Overstreetcaf41672013-05-07 16:18:35 -0700721 pr_debug("error allocating ioctx %d\n", err);
Al Viroe23754f2012-03-06 14:33:22 -0500722 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
724
Kent Overstreet36f55882013-05-07 16:18:41 -0700725/* kill_ioctx
726 * Cancels all outstanding aio requests on an aio context. Used
727 * when the processes owning a context have all exited to encourage
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 * the rapid destruction of the kioctx.
729 */
Anatol Pomozove02ba722014-04-15 11:31:33 -0700730static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx,
731 struct completion *requests_done)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
Kent Overstreet36f55882013-05-07 16:18:41 -0700733 if (!atomic_xchg(&ctx->dead, 1)) {
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400734 struct kioctx_table *table;
735
736 spin_lock(&mm->ioctx_lock);
Artem Savkovd9b2c872013-09-09 12:29:35 -0400737 rcu_read_lock();
Benjamin LaHaise77d30b12013-08-30 11:12:50 -0400738 table = rcu_dereference(mm->ioctx_table);
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400739
740 WARN_ON(ctx != table->table[ctx->id]);
741 table->table[ctx->id] = NULL;
Artem Savkovd9b2c872013-09-09 12:29:35 -0400742 rcu_read_unlock();
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400743 spin_unlock(&mm->ioctx_lock);
744
Kent Overstreet723be6e2013-05-28 15:14:48 -0700745 /* percpu_ref_kill() will do the necessary call_rcu() */
746 wake_up_all(&ctx->wait);
Al Viro06af1212012-03-20 16:26:24 -0400747
Kent Overstreet36f55882013-05-07 16:18:41 -0700748 /*
Kent Overstreet4fcc7122013-06-12 14:04:59 -0700749 * It'd be more correct to do this in free_ioctx(), after all
750 * the outstanding kiocbs have finished - but by then io_destroy
751 * has already returned, so io_setup() could potentially return
752 * -EAGAIN with no ioctxs actually in use (as far as userspace
753 * could tell).
Kent Overstreet36f55882013-05-07 16:18:41 -0700754 */
Kent Overstreete34ecee2013-10-10 19:31:47 -0700755 aio_nr_sub(ctx->max_reqs);
Kent Overstreet4fcc7122013-06-12 14:04:59 -0700756
757 if (ctx->mmap_size)
758 vm_munmap(ctx->mmap_base, ctx->mmap_size);
759
Anatol Pomozove02ba722014-04-15 11:31:33 -0700760 ctx->requests_done = requests_done;
Kent Overstreet723be6e2013-05-28 15:14:48 -0700761 percpu_ref_kill(&ctx->users);
Anatol Pomozove02ba722014-04-15 11:31:33 -0700762 } else {
763 if (requests_done)
764 complete(requests_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
768/* wait_on_sync_kiocb:
769 * Waits on the given sync kiocb to complete.
770 */
Kent Overstreet57282d82013-05-13 13:42:52 -0700771ssize_t wait_on_sync_kiocb(struct kiocb *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Kent Overstreet57282d82013-05-13 13:42:52 -0700773 while (!req->ki_ctx) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 set_current_state(TASK_UNINTERRUPTIBLE);
Kent Overstreet57282d82013-05-13 13:42:52 -0700775 if (req->ki_ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 break;
Jeff Moyer41d10da2007-10-16 23:27:20 -0700777 io_schedule();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
779 __set_current_state(TASK_RUNNING);
Kent Overstreet57282d82013-05-13 13:42:52 -0700780 return req->ki_user_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
H Hartley Sweeten385773e2009-09-22 16:43:53 -0700782EXPORT_SYMBOL(wait_on_sync_kiocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Kent Overstreet36f55882013-05-07 16:18:41 -0700784/*
785 * exit_aio: called when the last user of mm goes away. At this point, there is
786 * no way for any new requests to be submited or any of the io_* syscalls to be
787 * called on the context.
788 *
789 * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on
790 * them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -0800792void exit_aio(struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400794 struct kioctx_table *table;
Jens Axboeabf137d2008-12-09 08:11:22 +0100795 struct kioctx *ctx;
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400796 unsigned i = 0;
Jens Axboeabf137d2008-12-09 08:11:22 +0100797
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400798 while (1) {
799 rcu_read_lock();
800 table = rcu_dereference(mm->ioctx_table);
801
802 do {
803 if (!table || i >= table->nr) {
804 rcu_read_unlock();
805 rcu_assign_pointer(mm->ioctx_table, NULL);
806 if (table)
807 kfree(table);
808 return;
809 }
810
811 ctx = table->table[i++];
812 } while (!ctx);
813
814 rcu_read_unlock();
815
Al Viro936af152012-04-20 21:49:41 -0400816 /*
817 * We don't need to bother with munmap() here -
818 * exit_mmap(mm) is coming and it'll unmap everything.
819 * Since aio_free_ring() uses non-zero ->mmap_size
820 * as indicator that it needs to unmap the area,
821 * just set it to 0; aio_free_ring() is the only
822 * place that uses ->mmap_size, so it's safe.
Al Viro936af152012-04-20 21:49:41 -0400823 */
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700824 ctx->mmap_size = 0;
Kent Overstreet36f55882013-05-07 16:18:41 -0700825
Anatol Pomozove02ba722014-04-15 11:31:33 -0700826 kill_ioctx(mm, ctx, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 }
828}
829
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000830static void put_reqs_available(struct kioctx *ctx, unsigned nr)
831{
832 struct kioctx_cpu *kcpu;
833
834 preempt_disable();
835 kcpu = this_cpu_ptr(ctx->cpu);
836
837 kcpu->reqs_available += nr;
838 while (kcpu->reqs_available >= ctx->req_batch * 2) {
839 kcpu->reqs_available -= ctx->req_batch;
840 atomic_add(ctx->req_batch, &ctx->reqs_available);
841 }
842
843 preempt_enable();
844}
845
846static bool get_reqs_available(struct kioctx *ctx)
847{
848 struct kioctx_cpu *kcpu;
849 bool ret = false;
850
851 preempt_disable();
852 kcpu = this_cpu_ptr(ctx->cpu);
853
854 if (!kcpu->reqs_available) {
855 int old, avail = atomic_read(&ctx->reqs_available);
856
857 do {
858 if (avail < ctx->req_batch)
859 goto out;
860
861 old = avail;
862 avail = atomic_cmpxchg(&ctx->reqs_available,
863 avail, avail - ctx->req_batch);
864 } while (avail != old);
865
866 kcpu->reqs_available += ctx->req_batch;
867 }
868
869 ret = true;
870 kcpu->reqs_available--;
871out:
872 preempt_enable();
873 return ret;
874}
875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876/* aio_get_req
Kent Overstreet57282d82013-05-13 13:42:52 -0700877 * Allocate a slot for an aio request.
878 * Returns NULL if no requests are free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 */
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700880static inline struct kiocb *aio_get_req(struct kioctx *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700882 struct kiocb *req;
883
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000884 if (!get_reqs_available(ctx))
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700885 return NULL;
886
Kent Overstreet0460fef2013-05-07 16:18:49 -0700887 req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (unlikely(!req))
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700889 goto out_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Kent Overstreete34ecee2013-10-10 19:31:47 -0700891 percpu_ref_get(&ctx->reqs);
892
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 req->ki_ctx = ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 return req;
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700895out_put:
Kent Overstreete1bdd5f2013-04-26 10:58:39 +1000896 put_reqs_available(ctx, 1);
Kent Overstreeta1c8eae2013-05-07 16:18:53 -0700897 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
899
Kent Overstreet11599eb2013-05-07 16:18:39 -0700900static void kiocb_free(struct kiocb *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901{
Kent Overstreet1d98ebf2013-05-07 16:18:37 -0700902 if (req->ki_filp)
903 fput(req->ki_filp);
Davide Libenzi13389012009-06-30 11:41:11 -0700904 if (req->ki_eventfd != NULL)
905 eventfd_ctx_put(req->ki_eventfd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 kmem_cache_free(kiocb_cachep, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Adrian Bunkd5470b52008-04-29 00:58:57 -0700909static struct kioctx *lookup_ioctx(unsigned long ctx_id)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400911 struct aio_ring __user *ring = (void __user *)ctx_id;
Jens Axboeabf137d2008-12-09 08:11:22 +0100912 struct mm_struct *mm = current->mm;
Jeff Moyer65c24492009-03-18 17:04:21 -0700913 struct kioctx *ctx, *ret = NULL;
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400914 struct kioctx_table *table;
915 unsigned id;
916
917 if (get_user(id, &ring->id))
918 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Jens Axboeabf137d2008-12-09 08:11:22 +0100920 rcu_read_lock();
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400921 table = rcu_dereference(mm->ioctx_table);
Jens Axboeabf137d2008-12-09 08:11:22 +0100922
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400923 if (!table || id >= table->nr)
924 goto out;
925
926 ctx = table->table[id];
Benjamin LaHaisef30d7042013-08-07 18:23:48 -0400927 if (ctx && ctx->user_id == ctx_id) {
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400928 percpu_ref_get(&ctx->users);
929 ret = ctx;
Jens Axboeabf137d2008-12-09 08:11:22 +0100930 }
Benjamin LaHaisedb446a02013-07-30 12:54:40 -0400931out:
Jens Axboeabf137d2008-12-09 08:11:22 +0100932 rcu_read_unlock();
Jeff Moyer65c24492009-03-18 17:04:21 -0700933 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936/* aio_complete
937 * Called when the io request on the given iocb is complete.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 */
Kent Overstreet2d684492013-05-07 16:18:29 -0700939void aio_complete(struct kiocb *iocb, long res, long res2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 struct kioctx *ctx = iocb->ki_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 struct aio_ring *ring;
Kent Overstreet21b40202013-05-07 16:18:47 -0700943 struct io_event *ev_page, *event;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 unsigned long flags;
Kent Overstreet21b40202013-05-07 16:18:47 -0700945 unsigned tail, pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Zach Brown20dcae32005-11-13 16:07:33 -0800947 /*
948 * Special case handling for sync iocbs:
949 * - events go directly into the iocb for fast handling
950 * - the sync task with the iocb in its stack holds the single iocb
951 * ref, no other paths have a way to get another ref
952 * - the sync task helpfully left a reference to itself in the iocb
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 */
954 if (is_sync_kiocb(iocb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 iocb->ki_user_data = res;
Kent Overstreet57282d82013-05-13 13:42:52 -0700956 smp_wmb();
957 iocb->ki_ctx = ERR_PTR(-EXDEV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 wake_up_process(iocb->ki_obj.tsk);
Kent Overstreet2d684492013-05-07 16:18:29 -0700959 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
Kent Overstreet0460fef2013-05-07 16:18:49 -0700962 if (iocb->ki_list.next) {
963 unsigned long flags;
964
965 spin_lock_irqsave(&ctx->ctx_lock, flags);
966 list_del(&iocb->ki_list);
967 spin_unlock_irqrestore(&ctx->ctx_lock, flags);
968 }
Kent Overstreet11599eb2013-05-07 16:18:39 -0700969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 /*
Kent Overstreet0460fef2013-05-07 16:18:49 -0700971 * Add a completion event to the ring buffer. Must be done holding
Tang Chen4b30f072013-07-03 15:09:16 -0700972 * ctx->completion_lock to prevent other code from messing with the tail
Kent Overstreet0460fef2013-05-07 16:18:49 -0700973 * pointer since we might be called from irq context.
974 */
975 spin_lock_irqsave(&ctx->completion_lock, flags);
976
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700977 tail = ctx->tail;
Kent Overstreet21b40202013-05-07 16:18:47 -0700978 pos = tail + AIO_EVENTS_OFFSET;
979
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700980 if (++tail >= ctx->nr_events)
Ken Chen4bf69b22005-05-01 08:59:15 -0700981 tail = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700983 ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet21b40202013-05-07 16:18:47 -0700984 event = ev_page + pos % AIO_EVENTS_PER_PAGE;
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 event->obj = (u64)(unsigned long)iocb->ki_obj.user;
987 event->data = iocb->ki_user_data;
988 event->res = res;
989 event->res2 = res2;
990
Kent Overstreet21b40202013-05-07 16:18:47 -0700991 kunmap_atomic(ev_page);
Kent Overstreet58c85dc2013-05-07 16:18:55 -0700992 flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]);
Kent Overstreet21b40202013-05-07 16:18:47 -0700993
994 pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n",
Kent Overstreetcaf41672013-05-07 16:18:35 -0700995 ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data,
996 res, res2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /* after flagging the request as done, we
999 * must never even look at it again
1000 */
1001 smp_wmb(); /* make event visible before updating tail */
1002
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001003 ctx->tail = tail;
Kent Overstreet21b40202013-05-07 16:18:47 -07001004
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001005 ring = kmap_atomic(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 ring->tail = tail;
Cong Wange8e3c3d2011-11-25 23:14:27 +08001007 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001008 flush_dcache_page(ctx->ring_pages[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Kent Overstreet0460fef2013-05-07 16:18:49 -07001010 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1011
Kent Overstreet21b40202013-05-07 16:18:47 -07001012 pr_debug("added to ring %p at [%u]\n", iocb, tail);
Davide Libenzi8d1c98b2008-04-10 21:29:19 -07001013
1014 /*
1015 * Check if the user asked us to deliver the result through an
1016 * eventfd. The eventfd_signal() function is safe to be called
1017 * from IRQ context.
1018 */
Davide Libenzi87c3a862009-03-18 17:04:19 -07001019 if (iocb->ki_eventfd != NULL)
Davide Libenzi8d1c98b2008-04-10 21:29:19 -07001020 eventfd_signal(iocb->ki_eventfd, 1);
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /* everything turned out well, dispose of the aiocb. */
Kent Overstreet57282d82013-05-13 13:42:52 -07001023 kiocb_free(iocb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
Quentin Barnes6cb2a212008-03-19 17:00:39 -07001025 /*
1026 * We have to order our ring_info tail store above and test
1027 * of the wait list below outside the wait lock. This is
1028 * like in wake_up_bit() where clearing a bit has to be
1029 * ordered with the unlocked test.
1030 */
1031 smp_mb();
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 if (waitqueue_active(&ctx->wait))
1034 wake_up(&ctx->wait);
1035
Kent Overstreete34ecee2013-10-10 19:31:47 -07001036 percpu_ref_put(&ctx->reqs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037}
H Hartley Sweeten385773e2009-09-22 16:43:53 -07001038EXPORT_SYMBOL(aio_complete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Kent Overstreeta31ad382013-05-07 16:18:45 -07001040/* aio_read_events
1041 * Pull an event off of the ioctx's event ring. Returns the number of
1042 * events fetched
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Kent Overstreeta31ad382013-05-07 16:18:45 -07001044static long aio_read_events_ring(struct kioctx *ctx,
1045 struct io_event __user *event, long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 struct aio_ring *ring;
Kent Overstreet5ffac122013-05-09 15:36:07 -07001048 unsigned head, tail, pos;
Kent Overstreeta31ad382013-05-07 16:18:45 -07001049 long ret = 0;
1050 int copy_ret;
1051
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001052 mutex_lock(&ctx->ring_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Benjamin LaHaisefa8a53c2014-03-28 10:14:45 -04001054 /* Access to ->ring_pages here is protected by ctx->ring_lock. */
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001055 ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001056 head = ring->head;
Kent Overstreet5ffac122013-05-09 15:36:07 -07001057 tail = ring->tail;
Kent Overstreeta31ad382013-05-07 16:18:45 -07001058 kunmap_atomic(ring);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Kent Overstreet5ffac122013-05-09 15:36:07 -07001060 pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001061
Kent Overstreet5ffac122013-05-09 15:36:07 -07001062 if (head == tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 goto out;
1064
Kent Overstreeta31ad382013-05-07 16:18:45 -07001065 while (ret < nr) {
1066 long avail;
1067 struct io_event *ev;
1068 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
Kent Overstreet5ffac122013-05-09 15:36:07 -07001070 avail = (head <= tail ? tail : ctx->nr_events) - head;
1071 if (head == tail)
Kent Overstreeta31ad382013-05-07 16:18:45 -07001072 break;
1073
1074 avail = min(avail, nr - ret);
1075 avail = min_t(long, avail, AIO_EVENTS_PER_PAGE -
1076 ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE));
1077
1078 pos = head + AIO_EVENTS_OFFSET;
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001079 page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE];
Kent Overstreeta31ad382013-05-07 16:18:45 -07001080 pos %= AIO_EVENTS_PER_PAGE;
1081
1082 ev = kmap(page);
1083 copy_ret = copy_to_user(event + ret, ev + pos,
1084 sizeof(*ev) * avail);
1085 kunmap(page);
1086
1087 if (unlikely(copy_ret)) {
1088 ret = -EFAULT;
1089 goto out;
1090 }
1091
1092 ret += avail;
1093 head += avail;
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001094 head %= ctx->nr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001097 ring = kmap_atomic(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001098 ring->head = head;
Zhao Hongjiang91d80a82013-04-26 11:03:53 +08001099 kunmap_atomic(ring);
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001100 flush_dcache_page(ctx->ring_pages[0]);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001101
Kent Overstreet5ffac122013-05-09 15:36:07 -07001102 pr_debug("%li h%u t%u\n", ret, head, tail);
Kent Overstreet3e845ce2013-05-07 16:18:51 -07001103
Kent Overstreete1bdd5f2013-04-26 10:58:39 +10001104 put_reqs_available(ctx, ret);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001105out:
Kent Overstreet58c85dc2013-05-07 16:18:55 -07001106 mutex_unlock(&ctx->ring_lock);
Kent Overstreeta31ad382013-05-07 16:18:45 -07001107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return ret;
1109}
1110
Kent Overstreeta31ad382013-05-07 16:18:45 -07001111static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr,
1112 struct io_event __user *event, long *i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
Kent Overstreeta31ad382013-05-07 16:18:45 -07001114 long ret = aio_read_events_ring(ctx, event + *i, nr - *i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
Kent Overstreeta31ad382013-05-07 16:18:45 -07001116 if (ret > 0)
1117 *i += ret;
1118
1119 if (unlikely(atomic_read(&ctx->dead)))
1120 ret = -EINVAL;
1121
1122 if (!*i)
1123 *i = ret;
1124
1125 return ret < 0 || *i >= min_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
1127
Kent Overstreeta31ad382013-05-07 16:18:45 -07001128static long read_events(struct kioctx *ctx, long min_nr, long nr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 struct io_event __user *event,
1130 struct timespec __user *timeout)
1131{
Kent Overstreeta31ad382013-05-07 16:18:45 -07001132 ktime_t until = { .tv64 = KTIME_MAX };
1133 long ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (timeout) {
1136 struct timespec ts;
Kent Overstreeta31ad382013-05-07 16:18:45 -07001137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 if (unlikely(copy_from_user(&ts, timeout, sizeof(ts))))
Kent Overstreeta31ad382013-05-07 16:18:45 -07001139 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Kent Overstreeta31ad382013-05-07 16:18:45 -07001141 until = timespec_to_ktime(ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 }
1143
Kent Overstreeta31ad382013-05-07 16:18:45 -07001144 /*
1145 * Note that aio_read_events() is being called as the conditional - i.e.
1146 * we're calling it after prepare_to_wait() has set task state to
1147 * TASK_INTERRUPTIBLE.
1148 *
1149 * But aio_read_events() can block, and if it blocks it's going to flip
1150 * the task state back to TASK_RUNNING.
1151 *
1152 * This should be ok, provided it doesn't flip the state back to
1153 * TASK_RUNNING and return 0 too much - that causes us to spin. That
1154 * will only happen if the mutex_lock() call blocks, and we then find
1155 * the ringbuffer empty. So in practice we should be ok, but it's
1156 * something to be aware of when touching this code.
1157 */
1158 wait_event_interruptible_hrtimeout(ctx->wait,
1159 aio_read_events(ctx, min_nr, nr, event, &ret), until);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
Kent Overstreeta31ad382013-05-07 16:18:45 -07001161 if (!ret && signal_pending(current))
1162 ret = -EINTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Kent Overstreeta31ad382013-05-07 16:18:45 -07001164 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167/* sys_io_setup:
1168 * Create an aio_context capable of receiving at least nr_events.
1169 * ctxp must not point to an aio_context that already exists, and
1170 * must be initialized to 0 prior to the call. On successful
1171 * creation of the aio_context, *ctxp is filled in with the resulting
1172 * handle. May fail with -EINVAL if *ctxp is not initialized,
1173 * if the specified nr_events exceeds internal limits. May fail
1174 * with -EAGAIN if the specified nr_events exceeds the user's limit
1175 * of available events. May fail with -ENOMEM if insufficient kernel
1176 * resources are available. May fail with -EFAULT if an invalid
1177 * pointer is passed for ctxp. Will fail with -ENOSYS if not
1178 * implemented.
1179 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001180SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181{
1182 struct kioctx *ioctx = NULL;
1183 unsigned long ctx;
1184 long ret;
1185
1186 ret = get_user(ctx, ctxp);
1187 if (unlikely(ret))
1188 goto out;
1189
1190 ret = -EINVAL;
Zach Brownd55b5fd2005-11-07 00:59:31 -08001191 if (unlikely(ctx || nr_events == 0)) {
1192 pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n",
1193 ctx, nr_events);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 goto out;
1195 }
1196
1197 ioctx = ioctx_alloc(nr_events);
1198 ret = PTR_ERR(ioctx);
1199 if (!IS_ERR(ioctx)) {
1200 ret = put_user(ioctx->user_id, ctxp);
Al Viroa2e18592012-03-20 16:27:57 -04001201 if (ret)
Anatol Pomozove02ba722014-04-15 11:31:33 -07001202 kill_ioctx(current->mm, ioctx, NULL);
Kent Overstreet723be6e2013-05-28 15:14:48 -07001203 percpu_ref_put(&ioctx->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205
1206out:
1207 return ret;
1208}
1209
1210/* sys_io_destroy:
1211 * Destroy the aio_context specified. May cancel any outstanding
1212 * AIOs and block on completion. Will fail with -ENOSYS if not
Satoru Takeuchi642b5122010-08-05 11:23:11 -07001213 * implemented. May fail with -EINVAL if the context pointed to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 * is invalid.
1215 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001216SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217{
1218 struct kioctx *ioctx = lookup_ioctx(ctx);
1219 if (likely(NULL != ioctx)) {
Anatol Pomozove02ba722014-04-15 11:31:33 -07001220 struct completion requests_done =
1221 COMPLETION_INITIALIZER_ONSTACK(requests_done);
1222
1223 /* Pass requests_done to kill_ioctx() where it can be set
1224 * in a thread-safe way. If we try to set it here then we have
1225 * a race condition if two io_destroy() called simultaneously.
1226 */
1227 kill_ioctx(current->mm, ioctx, &requests_done);
Kent Overstreet723be6e2013-05-28 15:14:48 -07001228 percpu_ref_put(&ioctx->users);
Anatol Pomozove02ba722014-04-15 11:31:33 -07001229
1230 /* Wait until all IO for the context are done. Otherwise kernel
1231 * keep using user-space buffers even if user thinks the context
1232 * is destroyed.
1233 */
1234 wait_for_completion(&requests_done);
1235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 return 0;
1237 }
1238 pr_debug("EINVAL: io_destroy: invalid context id\n");
1239 return -EINVAL;
1240}
1241
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001242typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *,
1243 unsigned long, loff_t);
1244
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001245static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb,
1246 int rw, char __user *buf,
1247 unsigned long *nr_segs,
1248 struct iovec **iovec,
1249 bool compat)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001250{
1251 ssize_t ret;
1252
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001253 *nr_segs = kiocb->ki_nbytes;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001254
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001255#ifdef CONFIG_COMPAT
1256 if (compat)
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001257 ret = compat_rw_copy_check_uvector(rw,
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001258 (struct compat_iovec __user *)buf,
1259 *nr_segs, 1, *iovec, iovec);
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001260 else
1261#endif
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001262 ret = rw_copy_check_uvector(rw,
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001263 (struct iovec __user *)buf,
1264 *nr_segs, 1, *iovec, iovec);
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001265 if (ret < 0)
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001266 return ret;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001267
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001268 /* ki_nbytes now reflect bytes instead of segs */
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001269 kiocb->ki_nbytes = ret;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001270 return 0;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001271}
1272
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001273static ssize_t aio_setup_single_vector(struct kiocb *kiocb,
1274 int rw, char __user *buf,
1275 unsigned long *nr_segs,
1276 struct iovec *iovec)
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001277{
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001278 if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes)))
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001279 return -EFAULT;
Linus Torvaldsa70b52e2012-05-21 16:06:20 -07001280
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001281 iovec->iov_base = buf;
1282 iovec->iov_len = kiocb->ki_nbytes;
1283 *nr_segs = 1;
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001284 return 0;
1285}
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287/*
1288 * aio_setup_iocb:
1289 * Performs the initial checks and aio retry method
1290 * setup for the kiocb at the time of io submission.
1291 */
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001292static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode,
1293 char __user *buf, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001295 struct file *file = req->ki_filp;
1296 ssize_t ret;
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001297 unsigned long nr_segs;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001298 int rw;
1299 fmode_t mode;
1300 aio_rw_op *rw_op;
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001301 struct iovec inline_vec, *iovec = &inline_vec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001303 switch (opcode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 case IOCB_CMD_PREAD:
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001305 case IOCB_CMD_PREADV:
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001306 mode = FMODE_READ;
1307 rw = READ;
1308 rw_op = file->f_op->aio_read;
1309 goto rw_common;
1310
1311 case IOCB_CMD_PWRITE:
Badari Pulavartyeed4e512006-09-30 23:28:49 -07001312 case IOCB_CMD_PWRITEV:
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001313 mode = FMODE_WRITE;
1314 rw = WRITE;
1315 rw_op = file->f_op->aio_write;
1316 goto rw_common;
1317rw_common:
1318 if (unlikely(!(file->f_mode & mode)))
1319 return -EBADF;
1320
1321 if (!rw_op)
1322 return -EINVAL;
1323
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001324 ret = (opcode == IOCB_CMD_PREADV ||
1325 opcode == IOCB_CMD_PWRITEV)
1326 ? aio_setup_vectored_rw(req, rw, buf, &nr_segs,
1327 &iovec, compat)
1328 : aio_setup_single_vector(req, rw, buf, &nr_segs,
1329 iovec);
Leon Yu754320d2014-05-01 03:31:28 +00001330 if (!ret)
1331 ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes);
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001332 if (ret < 0) {
1333 if (iovec != &inline_vec)
1334 kfree(iovec);
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001335 return ret;
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001336 }
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001337
1338 req->ki_nbytes = ret;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001339
Kent Overstreet73a70752013-05-09 15:03:42 -07001340 /* XXX: move/kill - rw_verify_area()? */
1341 /* This matches the pread()/pwrite() logic */
1342 if (req->ki_pos < 0) {
1343 ret = -EINVAL;
1344 break;
1345 }
1346
1347 if (rw == WRITE)
1348 file_start_write(file);
1349
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001350 ret = rw_op(req, iovec, nr_segs, req->ki_pos);
Kent Overstreet73a70752013-05-09 15:03:42 -07001351
1352 if (rw == WRITE)
1353 file_end_write(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 break;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 case IOCB_CMD_FDSYNC:
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001357 if (!file->f_op->aio_fsync)
1358 return -EINVAL;
1359
1360 ret = file->f_op->aio_fsync(req, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 break;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 case IOCB_CMD_FSYNC:
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001364 if (!file->f_op->aio_fsync)
1365 return -EINVAL;
1366
1367 ret = file->f_op->aio_fsync(req, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 break;
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 default:
Kent Overstreetcaf41672013-05-07 16:18:35 -07001371 pr_debug("EINVAL: no operation provided\n");
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001372 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 }
1374
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001375 if (iovec != &inline_vec)
1376 kfree(iovec);
1377
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001378 if (ret != -EIOCBQUEUED) {
1379 /*
1380 * There's no easy way to restart the syscall since other AIO's
1381 * may be already running. Just fail this IO with EINTR.
1382 */
1383 if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
1384 ret == -ERESTARTNOHAND ||
1385 ret == -ERESTART_RESTARTBLOCK))
1386 ret = -EINTR;
1387 aio_complete(req, ret, 0);
1388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 return 0;
1391}
1392
Adrian Bunkd5470b52008-04-29 00:58:57 -07001393static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
Kent Overstreeta1c8eae2013-05-07 16:18:53 -07001394 struct iocb *iocb, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
1396 struct kiocb *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 ssize_t ret;
1398
1399 /* enforce forwards compatibility on users */
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001400 if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001401 pr_debug("EINVAL: reserve field set\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 return -EINVAL;
1403 }
1404
1405 /* prevent overflows */
1406 if (unlikely(
1407 (iocb->aio_buf != (unsigned long)iocb->aio_buf) ||
1408 (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) ||
1409 ((ssize_t)iocb->aio_nbytes < 0)
1410 )) {
1411 pr_debug("EINVAL: io_submit: overflow check\n");
1412 return -EINVAL;
1413 }
1414
Kent Overstreet41ef4eb2013-05-07 16:19:11 -07001415 req = aio_get_req(ctx);
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001416 if (unlikely(!req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 return -EAGAIN;
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001418
1419 req->ki_filp = fget(iocb->aio_fildes);
1420 if (unlikely(!req->ki_filp)) {
1421 ret = -EBADF;
1422 goto out_put_req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
Kent Overstreet1d98ebf2013-05-07 16:18:37 -07001424
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001425 if (iocb->aio_flags & IOCB_FLAG_RESFD) {
1426 /*
1427 * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an
1428 * instance of the file* now. The file descriptor must be
1429 * an eventfd() fd, and will be signaled for each completed
1430 * event using the eventfd_signal() function.
1431 */
Davide Libenzi13389012009-06-30 11:41:11 -07001432 req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd);
Hirofumi Nakagawa801678c2008-04-29 01:03:09 -07001433 if (IS_ERR(req->ki_eventfd)) {
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001434 ret = PTR_ERR(req->ki_eventfd);
Davide Libenzi87c3a862009-03-18 17:04:19 -07001435 req->ki_eventfd = NULL;
Davide Libenzi9c3060b2007-05-10 22:23:21 -07001436 goto out_put_req;
1437 }
1438 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Kent Overstreet8a660892013-05-07 16:19:10 -07001440 ret = put_user(KIOCB_KEY, &user_iocb->aio_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (unlikely(ret)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001442 pr_debug("EFAULT: aio_key\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 goto out_put_req;
1444 }
1445
1446 req->ki_obj.user = user_iocb;
1447 req->ki_user_data = iocb->aio_data;
1448 req->ki_pos = iocb->aio_offset;
Kent Overstreet73a70752013-05-09 15:03:42 -07001449 req->ki_nbytes = iocb->aio_nbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Kent Overstreet8bc92af2013-02-25 16:36:27 -08001451 ret = aio_run_iocb(req, iocb->aio_lio_opcode,
1452 (char __user *)(unsigned long)iocb->aio_buf,
1453 compat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 if (ret)
1455 goto out_put_req;
1456
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458out_put_req:
Kent Overstreete1bdd5f2013-04-26 10:58:39 +10001459 put_reqs_available(ctx, 1);
Kent Overstreete34ecee2013-10-10 19:31:47 -07001460 percpu_ref_put(&ctx->reqs);
Kent Overstreet57282d82013-05-13 13:42:52 -07001461 kiocb_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return ret;
1463}
1464
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001465long do_io_submit(aio_context_t ctx_id, long nr,
1466 struct iocb __user *__user *iocbpp, bool compat)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
1468 struct kioctx *ctx;
1469 long ret = 0;
Jeff Moyer080d6762011-11-02 13:40:10 -07001470 int i = 0;
Shaohua Li9f5b9422010-07-01 07:55:01 +02001471 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
1473 if (unlikely(nr < 0))
1474 return -EINVAL;
1475
Jeff Moyer75e1c702010-09-10 14:16:00 -07001476 if (unlikely(nr > LONG_MAX/sizeof(*iocbpp)))
1477 nr = LONG_MAX/sizeof(*iocbpp);
1478
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp)))))
1480 return -EFAULT;
1481
1482 ctx = lookup_ioctx(ctx_id);
1483 if (unlikely(!ctx)) {
Kent Overstreetcaf41672013-05-07 16:18:35 -07001484 pr_debug("EINVAL: invalid context id\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 return -EINVAL;
1486 }
1487
Shaohua Li9f5b9422010-07-01 07:55:01 +02001488 blk_start_plug(&plug);
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 /*
1491 * AKPM: should this return a partial result if some of the IOs were
1492 * successfully submitted?
1493 */
1494 for (i=0; i<nr; i++) {
1495 struct iocb __user *user_iocb;
1496 struct iocb tmp;
1497
1498 if (unlikely(__get_user(user_iocb, iocbpp + i))) {
1499 ret = -EFAULT;
1500 break;
1501 }
1502
1503 if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) {
1504 ret = -EFAULT;
1505 break;
1506 }
1507
Kent Overstreeta1c8eae2013-05-07 16:18:53 -07001508 ret = io_submit_one(ctx, user_iocb, &tmp, compat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 if (ret)
1510 break;
1511 }
Shaohua Li9f5b9422010-07-01 07:55:01 +02001512 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Kent Overstreet723be6e2013-05-28 15:14:48 -07001514 percpu_ref_put(&ctx->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return i ? i : ret;
1516}
1517
Jeff Moyer9d85cba2010-05-26 14:44:26 -07001518/* sys_io_submit:
1519 * Queue the nr iocbs pointed to by iocbpp for processing. Returns
1520 * the number of iocbs queued. May return -EINVAL if the aio_context
1521 * specified by ctx_id is invalid, if nr is < 0, if the iocb at
1522 * *iocbpp[0] is not properly initialized, if the operation specified
1523 * is invalid for the file descriptor in the iocb. May fail with
1524 * -EFAULT if any of the data structures point to invalid data. May
1525 * fail with -EBADF if the file descriptor specified in the first
1526 * iocb is invalid. May fail with -EAGAIN if insufficient resources
1527 * are available to queue any iocbs. Will return 0 if nr is 0. Will
1528 * fail with -ENOSYS if not implemented.
1529 */
1530SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr,
1531 struct iocb __user * __user *, iocbpp)
1532{
1533 return do_io_submit(ctx_id, nr, iocbpp, 0);
1534}
1535
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536/* lookup_kiocb
1537 * Finds a given iocb for cancellation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 */
Adrian Bunk25ee7e32005-04-25 08:18:14 -07001539static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb,
1540 u32 key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
1542 struct list_head *pos;
Zach Brownd00689a2005-11-13 16:07:34 -08001543
1544 assert_spin_locked(&ctx->ctx_lock);
1545
Kent Overstreet8a660892013-05-07 16:19:10 -07001546 if (key != KIOCB_KEY)
1547 return NULL;
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 /* TODO: use a hash or array, this sucks. */
1550 list_for_each(pos, &ctx->active_reqs) {
1551 struct kiocb *kiocb = list_kiocb(pos);
Kent Overstreet8a660892013-05-07 16:19:10 -07001552 if (kiocb->ki_obj.user == iocb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 return kiocb;
1554 }
1555 return NULL;
1556}
1557
1558/* sys_io_cancel:
1559 * Attempts to cancel an iocb previously passed to io_submit. If
1560 * the operation is successfully cancelled, the resulting event is
1561 * copied into the memory pointed to by result without being placed
1562 * into the completion queue and 0 is returned. May fail with
1563 * -EFAULT if any of the data structures pointed to are invalid.
1564 * May fail with -EINVAL if aio_context specified by ctx_id is
1565 * invalid. May fail with -EAGAIN if the iocb specified was not
1566 * cancelled. Will fail with -ENOSYS if not implemented.
1567 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001568SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb,
1569 struct io_event __user *, result)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 struct kioctx *ctx;
1572 struct kiocb *kiocb;
1573 u32 key;
1574 int ret;
1575
1576 ret = get_user(key, &iocb->aio_key);
1577 if (unlikely(ret))
1578 return -EFAULT;
1579
1580 ctx = lookup_ioctx(ctx_id);
1581 if (unlikely(!ctx))
1582 return -EINVAL;
1583
1584 spin_lock_irq(&ctx->ctx_lock);
Kent Overstreet906b9732013-05-07 16:18:31 -07001585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 kiocb = lookup_kiocb(ctx, iocb, key);
Kent Overstreet906b9732013-05-07 16:18:31 -07001587 if (kiocb)
Kent Overstreetbec68faa2013-05-13 14:45:08 -07001588 ret = kiocb_cancel(ctx, kiocb);
Kent Overstreet906b9732013-05-07 16:18:31 -07001589 else
1590 ret = -EINVAL;
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 spin_unlock_irq(&ctx->ctx_lock);
1593
Kent Overstreet906b9732013-05-07 16:18:31 -07001594 if (!ret) {
Kent Overstreetbec68faa2013-05-13 14:45:08 -07001595 /*
1596 * The result argument is no longer used - the io_event is
1597 * always delivered via the ring buffer. -EINPROGRESS indicates
1598 * cancellation is progress:
Kent Overstreet906b9732013-05-07 16:18:31 -07001599 */
Kent Overstreetbec68faa2013-05-13 14:45:08 -07001600 ret = -EINPROGRESS;
Kent Overstreet906b9732013-05-07 16:18:31 -07001601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
Kent Overstreet723be6e2013-05-28 15:14:48 -07001603 percpu_ref_put(&ctx->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605 return ret;
1606}
1607
1608/* io_getevents:
1609 * Attempts to read at least min_nr events and up to nr events from
Satoru Takeuchi642b5122010-08-05 11:23:11 -07001610 * the completion queue for the aio_context specified by ctx_id. If
1611 * it succeeds, the number of read events is returned. May fail with
1612 * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is
1613 * out of range, if timeout is out of range. May fail with -EFAULT
1614 * if any of the memory specified is invalid. May return 0 or
1615 * < min_nr if the timeout specified by timeout has elapsed
1616 * before sufficient events are available, where timeout == NULL
1617 * specifies an infinite timeout. Note that the timeout pointed to by
Jeff Moyer69008072013-05-24 15:55:24 -07001618 * timeout is relative. Will fail with -ENOSYS if not implemented.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001620SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id,
1621 long, min_nr,
1622 long, nr,
1623 struct io_event __user *, events,
1624 struct timespec __user *, timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625{
1626 struct kioctx *ioctx = lookup_ioctx(ctx_id);
1627 long ret = -EINVAL;
1628
1629 if (likely(ioctx)) {
Namhyung Kim2e410252011-01-12 17:01:08 -08001630 if (likely(min_nr <= nr && min_nr >= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 ret = read_events(ioctx, min_nr, nr, events, timeout);
Kent Overstreet723be6e2013-05-28 15:14:48 -07001632 percpu_ref_put(&ioctx->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 return ret;
1635}