]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * An async IO implementation for Linux | |
3 | * Written by Benjamin LaHaise <bcrl@kvack.org> | |
4 | * | |
5 | * Implements an efficient asynchronous io interface. | |
6 | * | |
7 | * Copyright 2000, 2001, 2002 Red Hat, Inc. All Rights Reserved. | |
8 | * | |
9 | * See ../COPYING for licensing terms. | |
10 | */ | |
caf4167a KO |
11 | #define pr_fmt(fmt) "%s: " fmt, __func__ |
12 | ||
1da177e4 LT |
13 | #include <linux/kernel.h> |
14 | #include <linux/init.h> | |
15 | #include <linux/errno.h> | |
16 | #include <linux/time.h> | |
17 | #include <linux/aio_abi.h> | |
630d9c47 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/syscalls.h> |
b9d128f1 | 20 | #include <linux/backing-dev.h> |
027445c3 | 21 | #include <linux/uio.h> |
1da177e4 | 22 | |
1da177e4 LT |
23 | #include <linux/sched.h> |
24 | #include <linux/fs.h> | |
25 | #include <linux/file.h> | |
26 | #include <linux/mm.h> | |
27 | #include <linux/mman.h> | |
3d2d827f | 28 | #include <linux/mmu_context.h> |
e1bdd5f2 | 29 | #include <linux/percpu.h> |
1da177e4 LT |
30 | #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> | |
9c3060be | 36 | #include <linux/eventfd.h> |
cfb1e33e | 37 | #include <linux/blkdev.h> |
9d85cba7 | 38 | #include <linux/compat.h> |
36bc08cc GZ |
39 | #include <linux/anon_inodes.h> |
40 | #include <linux/migrate.h> | |
41 | #include <linux/ramfs.h> | |
723be6e3 | 42 | #include <linux/percpu-refcount.h> |
1da177e4 LT |
43 | |
44 | #include <asm/kmap_types.h> | |
45 | #include <asm/uaccess.h> | |
1da177e4 | 46 | |
68d70d03 AV |
47 | #include "internal.h" |
48 | ||
4e179bca KO |
49 | #define AIO_RING_MAGIC 0xa10a10a1 |
50 | #define AIO_RING_COMPAT_FEATURES 1 | |
51 | #define AIO_RING_INCOMPAT_FEATURES 0 | |
52 | struct aio_ring { | |
53 | unsigned id; /* kernel internal index number */ | |
54 | unsigned nr; /* number of io_events */ | |
55 | unsigned head; | |
56 | unsigned tail; | |
57 | ||
58 | unsigned magic; | |
59 | unsigned compat_features; | |
60 | unsigned incompat_features; | |
61 | unsigned header_length; /* size of aio_ring */ | |
62 | ||
63 | ||
64 | struct io_event io_events[0]; | |
65 | }; /* 128 bytes + ring size */ | |
66 | ||
67 | #define AIO_RING_PAGES 8 | |
4e179bca | 68 | |
db446a08 BL |
69 | struct kioctx_table { |
70 | struct rcu_head rcu; | |
71 | unsigned nr; | |
72 | struct kioctx *table[]; | |
73 | }; | |
74 | ||
e1bdd5f2 KO |
75 | struct kioctx_cpu { |
76 | unsigned reqs_available; | |
77 | }; | |
78 | ||
4e179bca | 79 | struct kioctx { |
723be6e3 | 80 | struct percpu_ref users; |
36f55889 | 81 | atomic_t dead; |
4e179bca | 82 | |
4e179bca | 83 | unsigned long user_id; |
4e179bca | 84 | |
e1bdd5f2 KO |
85 | struct __percpu kioctx_cpu *cpu; |
86 | ||
87 | /* | |
88 | * For percpu reqs_available, number of slots we move to/from global | |
89 | * counter at a time: | |
90 | */ | |
91 | unsigned req_batch; | |
3e845ce0 KO |
92 | /* |
93 | * This is what userspace passed to io_setup(), it's not used for | |
94 | * anything but counting against the global max_reqs quota. | |
95 | * | |
58c85dc2 | 96 | * The real limit is nr_events - 1, which will be larger (see |
3e845ce0 KO |
97 | * aio_setup_ring()) |
98 | */ | |
4e179bca KO |
99 | unsigned max_reqs; |
100 | ||
58c85dc2 KO |
101 | /* Size of ringbuffer, in units of struct io_event */ |
102 | unsigned nr_events; | |
4e179bca | 103 | |
58c85dc2 KO |
104 | unsigned long mmap_base; |
105 | unsigned long mmap_size; | |
106 | ||
107 | struct page **ring_pages; | |
108 | long nr_pages; | |
109 | ||
4e23bcae | 110 | struct rcu_head rcu_head; |
723be6e3 | 111 | struct work_struct free_work; |
4e23bcae KO |
112 | |
113 | struct { | |
34e83fc6 KO |
114 | /* |
115 | * This counts the number of available slots in the ringbuffer, | |
116 | * so we avoid overflowing it: it's decremented (if positive) | |
117 | * when allocating a kiocb and incremented when the resulting | |
118 | * io_event is pulled off the ringbuffer. | |
e1bdd5f2 KO |
119 | * |
120 | * We batch accesses to it with a percpu version. | |
34e83fc6 KO |
121 | */ |
122 | atomic_t reqs_available; | |
4e23bcae KO |
123 | } ____cacheline_aligned_in_smp; |
124 | ||
125 | struct { | |
126 | spinlock_t ctx_lock; | |
127 | struct list_head active_reqs; /* used for cancellation */ | |
128 | } ____cacheline_aligned_in_smp; | |
129 | ||
58c85dc2 KO |
130 | struct { |
131 | struct mutex ring_lock; | |
4e23bcae KO |
132 | wait_queue_head_t wait; |
133 | } ____cacheline_aligned_in_smp; | |
58c85dc2 KO |
134 | |
135 | struct { | |
136 | unsigned tail; | |
137 | spinlock_t completion_lock; | |
4e23bcae | 138 | } ____cacheline_aligned_in_smp; |
58c85dc2 KO |
139 | |
140 | struct page *internal_pages[AIO_RING_PAGES]; | |
36bc08cc | 141 | struct file *aio_ring_file; |
db446a08 BL |
142 | |
143 | unsigned id; | |
4e179bca KO |
144 | }; |
145 | ||
1da177e4 | 146 | /*------ sysctl variables----*/ |
d55b5fda ZB |
147 | static DEFINE_SPINLOCK(aio_nr_lock); |
148 | unsigned long aio_nr; /* current system wide number of aio requests */ | |
149 | unsigned long aio_max_nr = 0x10000; /* system wide maximum number of aio requests */ | |
1da177e4 LT |
150 | /*----end sysctl variables---*/ |
151 | ||
e18b890b CL |
152 | static struct kmem_cache *kiocb_cachep; |
153 | static struct kmem_cache *kioctx_cachep; | |
1da177e4 | 154 | |
1da177e4 LT |
155 | /* aio_setup |
156 | * Creates the slab caches used by the aio routines, panic on | |
157 | * failure as this is done early during the boot sequence. | |
158 | */ | |
159 | static int __init aio_setup(void) | |
160 | { | |
0a31bd5f CL |
161 | kiocb_cachep = KMEM_CACHE(kiocb, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
162 | kioctx_cachep = KMEM_CACHE(kioctx,SLAB_HWCACHE_ALIGN|SLAB_PANIC); | |
1da177e4 | 163 | |
caf4167a | 164 | pr_debug("sizeof(struct page) = %zu\n", sizeof(struct page)); |
1da177e4 LT |
165 | |
166 | return 0; | |
167 | } | |
385773e0 | 168 | __initcall(aio_setup); |
1da177e4 LT |
169 | |
170 | static void aio_free_ring(struct kioctx *ctx) | |
171 | { | |
36bc08cc GZ |
172 | int i; |
173 | struct file *aio_ring_file = ctx->aio_ring_file; | |
1da177e4 | 174 | |
36bc08cc GZ |
175 | for (i = 0; i < ctx->nr_pages; i++) { |
176 | pr_debug("pid(%d) [%d] page->count=%d\n", current->pid, i, | |
177 | page_count(ctx->ring_pages[i])); | |
58c85dc2 | 178 | put_page(ctx->ring_pages[i]); |
36bc08cc | 179 | } |
1da177e4 | 180 | |
58c85dc2 KO |
181 | if (ctx->ring_pages && ctx->ring_pages != ctx->internal_pages) |
182 | kfree(ctx->ring_pages); | |
36bc08cc GZ |
183 | |
184 | if (aio_ring_file) { | |
185 | truncate_setsize(aio_ring_file->f_inode, 0); | |
186 | pr_debug("pid(%d) i_nlink=%u d_count=%d d_unhashed=%d i_count=%d\n", | |
187 | current->pid, aio_ring_file->f_inode->i_nlink, | |
188 | aio_ring_file->f_path.dentry->d_count, | |
189 | d_unhashed(aio_ring_file->f_path.dentry), | |
190 | atomic_read(&aio_ring_file->f_inode->i_count)); | |
191 | fput(aio_ring_file); | |
192 | ctx->aio_ring_file = NULL; | |
193 | } | |
194 | } | |
195 | ||
196 | static int aio_ring_mmap(struct file *file, struct vm_area_struct *vma) | |
197 | { | |
198 | vma->vm_ops = &generic_file_vm_ops; | |
199 | return 0; | |
200 | } | |
201 | ||
202 | static const struct file_operations aio_ring_fops = { | |
203 | .mmap = aio_ring_mmap, | |
204 | }; | |
205 | ||
206 | static int aio_set_page_dirty(struct page *page) | |
207 | { | |
208 | return 0; | |
209 | } | |
210 | ||
0c45355f | 211 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc GZ |
212 | static int aio_migratepage(struct address_space *mapping, struct page *new, |
213 | struct page *old, enum migrate_mode mode) | |
214 | { | |
215 | struct kioctx *ctx = mapping->private_data; | |
216 | unsigned long flags; | |
217 | unsigned idx = old->index; | |
218 | int rc; | |
219 | ||
220 | /* Writeback must be complete */ | |
221 | BUG_ON(PageWriteback(old)); | |
222 | put_page(old); | |
223 | ||
224 | rc = migrate_page_move_mapping(mapping, new, old, NULL, mode); | |
225 | if (rc != MIGRATEPAGE_SUCCESS) { | |
226 | get_page(old); | |
227 | return rc; | |
228 | } | |
229 | ||
230 | get_page(new); | |
231 | ||
232 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
233 | migrate_page_copy(new, old); | |
234 | ctx->ring_pages[idx] = new; | |
235 | spin_unlock_irqrestore(&ctx->completion_lock, flags); | |
236 | ||
237 | return rc; | |
1da177e4 | 238 | } |
0c45355f | 239 | #endif |
1da177e4 | 240 | |
36bc08cc GZ |
241 | static const struct address_space_operations aio_ctx_aops = { |
242 | .set_page_dirty = aio_set_page_dirty, | |
0c45355f | 243 | #if IS_ENABLED(CONFIG_MIGRATION) |
36bc08cc | 244 | .migratepage = aio_migratepage, |
0c45355f | 245 | #endif |
36bc08cc GZ |
246 | }; |
247 | ||
1da177e4 LT |
248 | static int aio_setup_ring(struct kioctx *ctx) |
249 | { | |
250 | struct aio_ring *ring; | |
1da177e4 | 251 | unsigned nr_events = ctx->max_reqs; |
41003a7b | 252 | struct mm_struct *mm = current->mm; |
41badc15 | 253 | unsigned long size, populate; |
1da177e4 | 254 | int nr_pages; |
36bc08cc GZ |
255 | int i; |
256 | struct file *file; | |
1da177e4 LT |
257 | |
258 | /* Compensate for the ring buffer's head/tail overlap entry */ | |
259 | nr_events += 2; /* 1 is required, 2 for good luck */ | |
260 | ||
261 | size = sizeof(struct aio_ring); | |
262 | size += sizeof(struct io_event) * nr_events; | |
1da177e4 | 263 | |
36bc08cc | 264 | nr_pages = PFN_UP(size); |
1da177e4 LT |
265 | if (nr_pages < 0) |
266 | return -EINVAL; | |
267 | ||
36bc08cc GZ |
268 | file = anon_inode_getfile_private("[aio]", &aio_ring_fops, ctx, O_RDWR); |
269 | if (IS_ERR(file)) { | |
270 | ctx->aio_ring_file = NULL; | |
271 | return -EAGAIN; | |
272 | } | |
273 | ||
274 | file->f_inode->i_mapping->a_ops = &aio_ctx_aops; | |
275 | file->f_inode->i_mapping->private_data = ctx; | |
276 | file->f_inode->i_size = PAGE_SIZE * (loff_t)nr_pages; | |
277 | ||
278 | for (i = 0; i < nr_pages; i++) { | |
279 | struct page *page; | |
280 | page = find_or_create_page(file->f_inode->i_mapping, | |
281 | i, GFP_HIGHUSER | __GFP_ZERO); | |
282 | if (!page) | |
283 | break; | |
284 | pr_debug("pid(%d) page[%d]->count=%d\n", | |
285 | current->pid, i, page_count(page)); | |
286 | SetPageUptodate(page); | |
287 | SetPageDirty(page); | |
288 | unlock_page(page); | |
289 | } | |
290 | ctx->aio_ring_file = file; | |
291 | nr_events = (PAGE_SIZE * nr_pages - sizeof(struct aio_ring)) | |
292 | / sizeof(struct io_event); | |
1da177e4 | 293 | |
58c85dc2 | 294 | ctx->ring_pages = ctx->internal_pages; |
1da177e4 | 295 | if (nr_pages > AIO_RING_PAGES) { |
58c85dc2 KO |
296 | ctx->ring_pages = kcalloc(nr_pages, sizeof(struct page *), |
297 | GFP_KERNEL); | |
298 | if (!ctx->ring_pages) | |
1da177e4 | 299 | return -ENOMEM; |
1da177e4 LT |
300 | } |
301 | ||
58c85dc2 KO |
302 | ctx->mmap_size = nr_pages * PAGE_SIZE; |
303 | pr_debug("attempting mmap of %lu bytes\n", ctx->mmap_size); | |
36bc08cc | 304 | |
41003a7b | 305 | down_write(&mm->mmap_sem); |
36bc08cc GZ |
306 | ctx->mmap_base = do_mmap_pgoff(ctx->aio_ring_file, 0, ctx->mmap_size, |
307 | PROT_READ | PROT_WRITE, | |
308 | MAP_SHARED | MAP_POPULATE, 0, &populate); | |
58c85dc2 | 309 | if (IS_ERR((void *)ctx->mmap_base)) { |
41003a7b | 310 | up_write(&mm->mmap_sem); |
58c85dc2 | 311 | ctx->mmap_size = 0; |
1da177e4 LT |
312 | aio_free_ring(ctx); |
313 | return -EAGAIN; | |
314 | } | |
36bc08cc GZ |
315 | up_write(&mm->mmap_sem); |
316 | ||
317 | mm_populate(ctx->mmap_base, populate); | |
1da177e4 | 318 | |
58c85dc2 KO |
319 | pr_debug("mmap address: 0x%08lx\n", ctx->mmap_base); |
320 | ctx->nr_pages = get_user_pages(current, mm, ctx->mmap_base, nr_pages, | |
321 | 1, 0, ctx->ring_pages, NULL); | |
36bc08cc GZ |
322 | for (i = 0; i < ctx->nr_pages; i++) |
323 | put_page(ctx->ring_pages[i]); | |
1da177e4 | 324 | |
58c85dc2 | 325 | if (unlikely(ctx->nr_pages != nr_pages)) { |
1da177e4 LT |
326 | aio_free_ring(ctx); |
327 | return -EAGAIN; | |
328 | } | |
329 | ||
58c85dc2 KO |
330 | ctx->user_id = ctx->mmap_base; |
331 | ctx->nr_events = nr_events; /* trusted copy */ | |
1da177e4 | 332 | |
58c85dc2 | 333 | ring = kmap_atomic(ctx->ring_pages[0]); |
1da177e4 | 334 | ring->nr = nr_events; /* user copy */ |
db446a08 | 335 | ring->id = ~0U; |
1da177e4 LT |
336 | ring->head = ring->tail = 0; |
337 | ring->magic = AIO_RING_MAGIC; | |
338 | ring->compat_features = AIO_RING_COMPAT_FEATURES; | |
339 | ring->incompat_features = AIO_RING_INCOMPAT_FEATURES; | |
340 | ring->header_length = sizeof(struct aio_ring); | |
e8e3c3d6 | 341 | kunmap_atomic(ring); |
58c85dc2 | 342 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 LT |
343 | |
344 | return 0; | |
345 | } | |
346 | ||
1da177e4 LT |
347 | #define AIO_EVENTS_PER_PAGE (PAGE_SIZE / sizeof(struct io_event)) |
348 | #define AIO_EVENTS_FIRST_PAGE ((PAGE_SIZE - sizeof(struct aio_ring)) / sizeof(struct io_event)) | |
349 | #define AIO_EVENTS_OFFSET (AIO_EVENTS_PER_PAGE - AIO_EVENTS_FIRST_PAGE) | |
350 | ||
0460fef2 KO |
351 | void kiocb_set_cancel_fn(struct kiocb *req, kiocb_cancel_fn *cancel) |
352 | { | |
353 | struct kioctx *ctx = req->ki_ctx; | |
354 | unsigned long flags; | |
355 | ||
356 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
357 | ||
358 | if (!req->ki_list.next) | |
359 | list_add(&req->ki_list, &ctx->active_reqs); | |
360 | ||
361 | req->ki_cancel = cancel; | |
362 | ||
363 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
364 | } | |
365 | EXPORT_SYMBOL(kiocb_set_cancel_fn); | |
366 | ||
bec68faa | 367 | static int kiocb_cancel(struct kioctx *ctx, struct kiocb *kiocb) |
906b973c | 368 | { |
0460fef2 | 369 | kiocb_cancel_fn *old, *cancel; |
906b973c | 370 | |
0460fef2 KO |
371 | /* |
372 | * Don't want to set kiocb->ki_cancel = KIOCB_CANCELLED unless it | |
373 | * actually has a cancel function, hence the cmpxchg() | |
374 | */ | |
375 | ||
376 | cancel = ACCESS_ONCE(kiocb->ki_cancel); | |
377 | do { | |
378 | if (!cancel || cancel == KIOCB_CANCELLED) | |
57282d8f | 379 | return -EINVAL; |
906b973c | 380 | |
0460fef2 KO |
381 | old = cancel; |
382 | cancel = cmpxchg(&kiocb->ki_cancel, old, KIOCB_CANCELLED); | |
383 | } while (cancel != old); | |
906b973c | 384 | |
57282d8f | 385 | return cancel(kiocb); |
906b973c KO |
386 | } |
387 | ||
36f55889 KO |
388 | static void free_ioctx_rcu(struct rcu_head *head) |
389 | { | |
390 | struct kioctx *ctx = container_of(head, struct kioctx, rcu_head); | |
e1bdd5f2 KO |
391 | |
392 | free_percpu(ctx->cpu); | |
36f55889 KO |
393 | kmem_cache_free(kioctx_cachep, ctx); |
394 | } | |
395 | ||
396 | /* | |
397 | * When this function runs, the kioctx has been removed from the "hash table" | |
398 | * and ctx->users has dropped to 0, so we know no more kiocbs can be submitted - | |
399 | * now it's safe to cancel any that need to be. | |
400 | */ | |
723be6e3 | 401 | static void free_ioctx(struct work_struct *work) |
36f55889 | 402 | { |
723be6e3 | 403 | struct kioctx *ctx = container_of(work, struct kioctx, free_work); |
3e845ce0 | 404 | struct aio_ring *ring; |
36f55889 | 405 | struct kiocb *req; |
5ffac122 KO |
406 | unsigned cpu, avail; |
407 | DEFINE_WAIT(wait); | |
36f55889 KO |
408 | |
409 | spin_lock_irq(&ctx->ctx_lock); | |
410 | ||
411 | while (!list_empty(&ctx->active_reqs)) { | |
412 | req = list_first_entry(&ctx->active_reqs, | |
413 | struct kiocb, ki_list); | |
414 | ||
415 | list_del_init(&req->ki_list); | |
bec68faa | 416 | kiocb_cancel(ctx, req); |
36f55889 KO |
417 | } |
418 | ||
419 | spin_unlock_irq(&ctx->ctx_lock); | |
420 | ||
e1bdd5f2 KO |
421 | for_each_possible_cpu(cpu) { |
422 | struct kioctx_cpu *kcpu = per_cpu_ptr(ctx->cpu, cpu); | |
423 | ||
424 | atomic_add(kcpu->reqs_available, &ctx->reqs_available); | |
425 | kcpu->reqs_available = 0; | |
426 | } | |
427 | ||
5ffac122 KO |
428 | while (1) { |
429 | prepare_to_wait(&ctx->wait, &wait, TASK_UNINTERRUPTIBLE); | |
3e845ce0 | 430 | |
5ffac122 KO |
431 | ring = kmap_atomic(ctx->ring_pages[0]); |
432 | avail = (ring->head <= ring->tail) | |
433 | ? ring->tail - ring->head | |
434 | : ctx->nr_events - ring->head + ring->tail; | |
3e845ce0 | 435 | |
34e83fc6 | 436 | atomic_add(avail, &ctx->reqs_available); |
5ffac122 KO |
437 | ring->head = ring->tail; |
438 | kunmap_atomic(ring); | |
439 | ||
440 | if (atomic_read(&ctx->reqs_available) >= ctx->nr_events - 1) | |
441 | break; | |
442 | ||
443 | schedule(); | |
3e845ce0 | 444 | } |
5ffac122 | 445 | finish_wait(&ctx->wait, &wait); |
3e845ce0 | 446 | |
34e83fc6 | 447 | WARN_ON(atomic_read(&ctx->reqs_available) > ctx->nr_events - 1); |
36f55889 KO |
448 | |
449 | aio_free_ring(ctx); | |
450 | ||
36f55889 KO |
451 | pr_debug("freeing %p\n", ctx); |
452 | ||
453 | /* | |
454 | * Here the call_rcu() is between the wait_event() for reqs_active to | |
455 | * hit 0, and freeing the ioctx. | |
456 | * | |
457 | * aio_complete() decrements reqs_active, but it has to touch the ioctx | |
458 | * after to issue a wakeup so we use rcu. | |
459 | */ | |
460 | call_rcu(&ctx->rcu_head, free_ioctx_rcu); | |
461 | } | |
462 | ||
723be6e3 | 463 | static void free_ioctx_ref(struct percpu_ref *ref) |
36f55889 | 464 | { |
723be6e3 KO |
465 | struct kioctx *ctx = container_of(ref, struct kioctx, users); |
466 | ||
467 | INIT_WORK(&ctx->free_work, free_ioctx); | |
468 | schedule_work(&ctx->free_work); | |
36f55889 KO |
469 | } |
470 | ||
db446a08 BL |
471 | static int ioctx_add_table(struct kioctx *ctx, struct mm_struct *mm) |
472 | { | |
473 | unsigned i, new_nr; | |
474 | struct kioctx_table *table, *old; | |
475 | struct aio_ring *ring; | |
476 | ||
477 | spin_lock(&mm->ioctx_lock); | |
da90382c | 478 | table = mm->ioctx_table; |
db446a08 BL |
479 | |
480 | while (1) { | |
481 | if (table) | |
482 | for (i = 0; i < table->nr; i++) | |
483 | if (!table->table[i]) { | |
484 | ctx->id = i; | |
485 | table->table[i] = ctx; | |
486 | spin_unlock(&mm->ioctx_lock); | |
487 | ||
488 | ring = kmap_atomic(ctx->ring_pages[0]); | |
489 | ring->id = ctx->id; | |
490 | kunmap_atomic(ring); | |
491 | return 0; | |
492 | } | |
493 | ||
494 | new_nr = (table ? table->nr : 1) * 4; | |
495 | ||
496 | spin_unlock(&mm->ioctx_lock); | |
497 | ||
498 | table = kzalloc(sizeof(*table) + sizeof(struct kioctx *) * | |
499 | new_nr, GFP_KERNEL); | |
500 | if (!table) | |
501 | return -ENOMEM; | |
502 | ||
503 | table->nr = new_nr; | |
504 | ||
505 | spin_lock(&mm->ioctx_lock); | |
da90382c | 506 | old = mm->ioctx_table; |
db446a08 BL |
507 | |
508 | if (!old) { | |
509 | rcu_assign_pointer(mm->ioctx_table, table); | |
510 | } else if (table->nr > old->nr) { | |
511 | memcpy(table->table, old->table, | |
512 | old->nr * sizeof(struct kioctx *)); | |
513 | ||
514 | rcu_assign_pointer(mm->ioctx_table, table); | |
515 | kfree_rcu(old, rcu); | |
516 | } else { | |
517 | kfree(table); | |
518 | table = old; | |
519 | } | |
520 | } | |
521 | } | |
522 | ||
1da177e4 LT |
523 | /* ioctx_alloc |
524 | * Allocates and initializes an ioctx. Returns an ERR_PTR if it failed. | |
525 | */ | |
526 | static struct kioctx *ioctx_alloc(unsigned nr_events) | |
527 | { | |
41003a7b | 528 | struct mm_struct *mm = current->mm; |
1da177e4 | 529 | struct kioctx *ctx; |
e23754f8 | 530 | int err = -ENOMEM; |
1da177e4 | 531 | |
e1bdd5f2 KO |
532 | /* |
533 | * We keep track of the number of available ringbuffer slots, to prevent | |
534 | * overflow (reqs_available), and we also use percpu counters for this. | |
535 | * | |
536 | * So since up to half the slots might be on other cpu's percpu counters | |
537 | * and unavailable, double nr_events so userspace sees what they | |
538 | * expected: additionally, we move req_batch slots to/from percpu | |
539 | * counters at a time, so make sure that isn't 0: | |
540 | */ | |
541 | nr_events = max(nr_events, num_possible_cpus() * 4); | |
542 | nr_events *= 2; | |
543 | ||
1da177e4 LT |
544 | /* Prevent overflows */ |
545 | if ((nr_events > (0x10000000U / sizeof(struct io_event))) || | |
546 | (nr_events > (0x10000000U / sizeof(struct kiocb)))) { | |
547 | pr_debug("ENOMEM: nr_events too high\n"); | |
548 | return ERR_PTR(-EINVAL); | |
549 | } | |
550 | ||
4cd81c3d | 551 | if (!nr_events || (unsigned long)nr_events > (aio_max_nr * 2UL)) |
1da177e4 LT |
552 | return ERR_PTR(-EAGAIN); |
553 | ||
c3762229 | 554 | ctx = kmem_cache_zalloc(kioctx_cachep, GFP_KERNEL); |
1da177e4 LT |
555 | if (!ctx) |
556 | return ERR_PTR(-ENOMEM); | |
557 | ||
1da177e4 | 558 | ctx->max_reqs = nr_events; |
1da177e4 | 559 | |
723be6e3 KO |
560 | if (percpu_ref_init(&ctx->users, free_ioctx_ref)) |
561 | goto out_freectx; | |
562 | ||
1da177e4 | 563 | spin_lock_init(&ctx->ctx_lock); |
0460fef2 | 564 | spin_lock_init(&ctx->completion_lock); |
58c85dc2 | 565 | mutex_init(&ctx->ring_lock); |
1da177e4 LT |
566 | init_waitqueue_head(&ctx->wait); |
567 | ||
568 | INIT_LIST_HEAD(&ctx->active_reqs); | |
1da177e4 | 569 | |
e1bdd5f2 KO |
570 | ctx->cpu = alloc_percpu(struct kioctx_cpu); |
571 | if (!ctx->cpu) | |
723be6e3 | 572 | goto out_freeref; |
1da177e4 | 573 | |
e1bdd5f2 KO |
574 | if (aio_setup_ring(ctx) < 0) |
575 | goto out_freepcpu; | |
576 | ||
34e83fc6 | 577 | atomic_set(&ctx->reqs_available, ctx->nr_events - 1); |
e1bdd5f2 | 578 | ctx->req_batch = (ctx->nr_events - 1) / (num_possible_cpus() * 4); |
6878ea72 BL |
579 | if (ctx->req_batch < 1) |
580 | ctx->req_batch = 1; | |
34e83fc6 | 581 | |
1da177e4 | 582 | /* limit the number of system wide aios */ |
9fa1cb39 | 583 | spin_lock(&aio_nr_lock); |
4cd81c3d | 584 | if (aio_nr + nr_events > (aio_max_nr * 2UL) || |
2dd542b7 | 585 | aio_nr + nr_events < aio_nr) { |
9fa1cb39 | 586 | spin_unlock(&aio_nr_lock); |
1da177e4 | 587 | goto out_cleanup; |
2dd542b7 AV |
588 | } |
589 | aio_nr += ctx->max_reqs; | |
9fa1cb39 | 590 | spin_unlock(&aio_nr_lock); |
1da177e4 | 591 | |
723be6e3 KO |
592 | percpu_ref_get(&ctx->users); /* io_setup() will drop this ref */ |
593 | ||
da90382c BL |
594 | err = ioctx_add_table(ctx, mm); |
595 | if (err) | |
596 | goto out_cleanup_put; | |
597 | ||
caf4167a | 598 | pr_debug("allocated ioctx %p[%ld]: mm=%p mask=0x%x\n", |
58c85dc2 | 599 | ctx, ctx->user_id, mm, ctx->nr_events); |
1da177e4 LT |
600 | return ctx; |
601 | ||
da90382c BL |
602 | out_cleanup_put: |
603 | percpu_ref_put(&ctx->users); | |
1da177e4 | 604 | out_cleanup: |
e23754f8 AV |
605 | err = -EAGAIN; |
606 | aio_free_ring(ctx); | |
e1bdd5f2 KO |
607 | out_freepcpu: |
608 | free_percpu(ctx->cpu); | |
723be6e3 KO |
609 | out_freeref: |
610 | free_percpu(ctx->users.pcpu_count); | |
1da177e4 | 611 | out_freectx: |
36bc08cc GZ |
612 | if (ctx->aio_ring_file) |
613 | fput(ctx->aio_ring_file); | |
1da177e4 | 614 | kmem_cache_free(kioctx_cachep, ctx); |
caf4167a | 615 | pr_debug("error allocating ioctx %d\n", err); |
e23754f8 | 616 | return ERR_PTR(err); |
1da177e4 LT |
617 | } |
618 | ||
36f55889 KO |
619 | /* kill_ioctx |
620 | * Cancels all outstanding aio requests on an aio context. Used | |
621 | * when the processes owning a context have all exited to encourage | |
622 | * the rapid destruction of the kioctx. | |
623 | */ | |
db446a08 | 624 | static void kill_ioctx(struct mm_struct *mm, struct kioctx *ctx) |
36f55889 KO |
625 | { |
626 | if (!atomic_xchg(&ctx->dead, 1)) { | |
db446a08 BL |
627 | struct kioctx_table *table; |
628 | ||
629 | spin_lock(&mm->ioctx_lock); | |
da90382c | 630 | table = mm->ioctx_table; |
db446a08 BL |
631 | |
632 | WARN_ON(ctx != table->table[ctx->id]); | |
633 | table->table[ctx->id] = NULL; | |
634 | spin_unlock(&mm->ioctx_lock); | |
635 | ||
723be6e3 KO |
636 | /* percpu_ref_kill() will do the necessary call_rcu() */ |
637 | wake_up_all(&ctx->wait); | |
dee11c23 | 638 | |
36f55889 | 639 | /* |
4fcc712f KO |
640 | * It'd be more correct to do this in free_ioctx(), after all |
641 | * the outstanding kiocbs have finished - but by then io_destroy | |
642 | * has already returned, so io_setup() could potentially return | |
643 | * -EAGAIN with no ioctxs actually in use (as far as userspace | |
644 | * could tell). | |
36f55889 | 645 | */ |
4fcc712f KO |
646 | spin_lock(&aio_nr_lock); |
647 | BUG_ON(aio_nr - ctx->max_reqs > aio_nr); | |
648 | aio_nr -= ctx->max_reqs; | |
649 | spin_unlock(&aio_nr_lock); | |
650 | ||
651 | if (ctx->mmap_size) | |
652 | vm_munmap(ctx->mmap_base, ctx->mmap_size); | |
653 | ||
723be6e3 | 654 | percpu_ref_kill(&ctx->users); |
36f55889 | 655 | } |
1da177e4 LT |
656 | } |
657 | ||
658 | /* wait_on_sync_kiocb: | |
659 | * Waits on the given sync kiocb to complete. | |
660 | */ | |
57282d8f | 661 | ssize_t wait_on_sync_kiocb(struct kiocb *req) |
1da177e4 | 662 | { |
57282d8f | 663 | while (!req->ki_ctx) { |
1da177e4 | 664 | set_current_state(TASK_UNINTERRUPTIBLE); |
57282d8f | 665 | if (req->ki_ctx) |
1da177e4 | 666 | break; |
41d10da3 | 667 | io_schedule(); |
1da177e4 LT |
668 | } |
669 | __set_current_state(TASK_RUNNING); | |
57282d8f | 670 | return req->ki_user_data; |
1da177e4 | 671 | } |
385773e0 | 672 | EXPORT_SYMBOL(wait_on_sync_kiocb); |
1da177e4 | 673 | |
36f55889 KO |
674 | /* |
675 | * exit_aio: called when the last user of mm goes away. At this point, there is | |
676 | * no way for any new requests to be submited or any of the io_* syscalls to be | |
677 | * called on the context. | |
678 | * | |
679 | * There may be outstanding kiocbs, but free_ioctx() will explicitly wait on | |
680 | * them. | |
1da177e4 | 681 | */ |
fc9b52cd | 682 | void exit_aio(struct mm_struct *mm) |
1da177e4 | 683 | { |
db446a08 | 684 | struct kioctx_table *table; |
abf137dd | 685 | struct kioctx *ctx; |
db446a08 BL |
686 | unsigned i = 0; |
687 | ||
688 | while (1) { | |
689 | rcu_read_lock(); | |
690 | table = rcu_dereference(mm->ioctx_table); | |
691 | ||
692 | do { | |
693 | if (!table || i >= table->nr) { | |
694 | rcu_read_unlock(); | |
695 | rcu_assign_pointer(mm->ioctx_table, NULL); | |
696 | if (table) | |
697 | kfree(table); | |
698 | return; | |
699 | } | |
700 | ||
701 | ctx = table->table[i++]; | |
702 | } while (!ctx); | |
703 | ||
704 | rcu_read_unlock(); | |
abf137dd | 705 | |
936af157 AV |
706 | /* |
707 | * We don't need to bother with munmap() here - | |
708 | * exit_mmap(mm) is coming and it'll unmap everything. | |
709 | * Since aio_free_ring() uses non-zero ->mmap_size | |
710 | * as indicator that it needs to unmap the area, | |
711 | * just set it to 0; aio_free_ring() is the only | |
712 | * place that uses ->mmap_size, so it's safe. | |
936af157 | 713 | */ |
58c85dc2 | 714 | ctx->mmap_size = 0; |
36f55889 | 715 | |
db446a08 | 716 | kill_ioctx(mm, ctx); |
1da177e4 LT |
717 | } |
718 | } | |
719 | ||
e1bdd5f2 KO |
720 | static void put_reqs_available(struct kioctx *ctx, unsigned nr) |
721 | { | |
722 | struct kioctx_cpu *kcpu; | |
723 | ||
724 | preempt_disable(); | |
725 | kcpu = this_cpu_ptr(ctx->cpu); | |
726 | ||
727 | kcpu->reqs_available += nr; | |
728 | while (kcpu->reqs_available >= ctx->req_batch * 2) { | |
729 | kcpu->reqs_available -= ctx->req_batch; | |
730 | atomic_add(ctx->req_batch, &ctx->reqs_available); | |
731 | } | |
732 | ||
733 | preempt_enable(); | |
734 | } | |
735 | ||
736 | static bool get_reqs_available(struct kioctx *ctx) | |
737 | { | |
738 | struct kioctx_cpu *kcpu; | |
739 | bool ret = false; | |
740 | ||
741 | preempt_disable(); | |
742 | kcpu = this_cpu_ptr(ctx->cpu); | |
743 | ||
744 | if (!kcpu->reqs_available) { | |
745 | int old, avail = atomic_read(&ctx->reqs_available); | |
746 | ||
747 | do { | |
748 | if (avail < ctx->req_batch) | |
749 | goto out; | |
750 | ||
751 | old = avail; | |
752 | avail = atomic_cmpxchg(&ctx->reqs_available, | |
753 | avail, avail - ctx->req_batch); | |
754 | } while (avail != old); | |
755 | ||
756 | kcpu->reqs_available += ctx->req_batch; | |
757 | } | |
758 | ||
759 | ret = true; | |
760 | kcpu->reqs_available--; | |
761 | out: | |
762 | preempt_enable(); | |
763 | return ret; | |
764 | } | |
765 | ||
1da177e4 | 766 | /* aio_get_req |
57282d8f KO |
767 | * Allocate a slot for an aio request. |
768 | * Returns NULL if no requests are free. | |
1da177e4 | 769 | */ |
a1c8eae7 | 770 | static inline struct kiocb *aio_get_req(struct kioctx *ctx) |
1da177e4 | 771 | { |
a1c8eae7 KO |
772 | struct kiocb *req; |
773 | ||
e1bdd5f2 | 774 | if (!get_reqs_available(ctx)) |
a1c8eae7 KO |
775 | return NULL; |
776 | ||
0460fef2 | 777 | req = kmem_cache_alloc(kiocb_cachep, GFP_KERNEL|__GFP_ZERO); |
1da177e4 | 778 | if (unlikely(!req)) |
a1c8eae7 | 779 | goto out_put; |
1da177e4 | 780 | |
1da177e4 | 781 | req->ki_ctx = ctx; |
080d676d | 782 | return req; |
a1c8eae7 | 783 | out_put: |
e1bdd5f2 | 784 | put_reqs_available(ctx, 1); |
a1c8eae7 | 785 | return NULL; |
1da177e4 LT |
786 | } |
787 | ||
11599eba | 788 | static void kiocb_free(struct kiocb *req) |
1da177e4 | 789 | { |
1d98ebfc KO |
790 | if (req->ki_filp) |
791 | fput(req->ki_filp); | |
13389010 DL |
792 | if (req->ki_eventfd != NULL) |
793 | eventfd_ctx_put(req->ki_eventfd); | |
1da177e4 | 794 | kmem_cache_free(kiocb_cachep, req); |
1da177e4 LT |
795 | } |
796 | ||
d5470b59 | 797 | static struct kioctx *lookup_ioctx(unsigned long ctx_id) |
1da177e4 | 798 | { |
db446a08 | 799 | struct aio_ring __user *ring = (void __user *)ctx_id; |
abf137dd | 800 | struct mm_struct *mm = current->mm; |
65c24491 | 801 | struct kioctx *ctx, *ret = NULL; |
db446a08 BL |
802 | struct kioctx_table *table; |
803 | unsigned id; | |
804 | ||
805 | if (get_user(id, &ring->id)) | |
806 | return NULL; | |
1da177e4 | 807 | |
abf137dd | 808 | rcu_read_lock(); |
db446a08 | 809 | table = rcu_dereference(mm->ioctx_table); |
abf137dd | 810 | |
db446a08 BL |
811 | if (!table || id >= table->nr) |
812 | goto out; | |
1da177e4 | 813 | |
db446a08 BL |
814 | ctx = table->table[id]; |
815 | if (ctx->user_id == ctx_id) { | |
816 | percpu_ref_get(&ctx->users); | |
817 | ret = ctx; | |
818 | } | |
819 | out: | |
abf137dd | 820 | rcu_read_unlock(); |
65c24491 | 821 | return ret; |
1da177e4 LT |
822 | } |
823 | ||
1da177e4 LT |
824 | /* aio_complete |
825 | * Called when the io request on the given iocb is complete. | |
1da177e4 | 826 | */ |
2d68449e | 827 | void aio_complete(struct kiocb *iocb, long res, long res2) |
1da177e4 LT |
828 | { |
829 | struct kioctx *ctx = iocb->ki_ctx; | |
1da177e4 | 830 | struct aio_ring *ring; |
21b40200 | 831 | struct io_event *ev_page, *event; |
1da177e4 | 832 | unsigned long flags; |
21b40200 | 833 | unsigned tail, pos; |
1da177e4 | 834 | |
20dcae32 ZB |
835 | /* |
836 | * Special case handling for sync iocbs: | |
837 | * - events go directly into the iocb for fast handling | |
838 | * - the sync task with the iocb in its stack holds the single iocb | |
839 | * ref, no other paths have a way to get another ref | |
840 | * - the sync task helpfully left a reference to itself in the iocb | |
1da177e4 LT |
841 | */ |
842 | if (is_sync_kiocb(iocb)) { | |
1da177e4 | 843 | iocb->ki_user_data = res; |
57282d8f KO |
844 | smp_wmb(); |
845 | iocb->ki_ctx = ERR_PTR(-EXDEV); | |
1da177e4 | 846 | wake_up_process(iocb->ki_obj.tsk); |
2d68449e | 847 | return; |
1da177e4 LT |
848 | } |
849 | ||
36f55889 | 850 | /* |
36f55889 | 851 | * Take rcu_read_lock() in case the kioctx is being destroyed, as we |
34e83fc6 | 852 | * need to issue a wakeup after incrementing reqs_available. |
1da177e4 | 853 | */ |
36f55889 | 854 | rcu_read_lock(); |
1da177e4 | 855 | |
0460fef2 KO |
856 | if (iocb->ki_list.next) { |
857 | unsigned long flags; | |
858 | ||
859 | spin_lock_irqsave(&ctx->ctx_lock, flags); | |
860 | list_del(&iocb->ki_list); | |
861 | spin_unlock_irqrestore(&ctx->ctx_lock, flags); | |
862 | } | |
11599eba | 863 | |
0460fef2 KO |
864 | /* |
865 | * Add a completion event to the ring buffer. Must be done holding | |
4b30f07e | 866 | * ctx->completion_lock to prevent other code from messing with the tail |
0460fef2 KO |
867 | * pointer since we might be called from irq context. |
868 | */ | |
869 | spin_lock_irqsave(&ctx->completion_lock, flags); | |
870 | ||
58c85dc2 | 871 | tail = ctx->tail; |
21b40200 KO |
872 | pos = tail + AIO_EVENTS_OFFSET; |
873 | ||
58c85dc2 | 874 | if (++tail >= ctx->nr_events) |
4bf69b2a | 875 | tail = 0; |
1da177e4 | 876 | |
58c85dc2 | 877 | ev_page = kmap_atomic(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
878 | event = ev_page + pos % AIO_EVENTS_PER_PAGE; |
879 | ||
1da177e4 LT |
880 | event->obj = (u64)(unsigned long)iocb->ki_obj.user; |
881 | event->data = iocb->ki_user_data; | |
882 | event->res = res; | |
883 | event->res2 = res2; | |
884 | ||
21b40200 | 885 | kunmap_atomic(ev_page); |
58c85dc2 | 886 | flush_dcache_page(ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]); |
21b40200 KO |
887 | |
888 | pr_debug("%p[%u]: %p: %p %Lx %lx %lx\n", | |
caf4167a KO |
889 | ctx, tail, iocb, iocb->ki_obj.user, iocb->ki_user_data, |
890 | res, res2); | |
1da177e4 LT |
891 | |
892 | /* after flagging the request as done, we | |
893 | * must never even look at it again | |
894 | */ | |
895 | smp_wmb(); /* make event visible before updating tail */ | |
896 | ||
58c85dc2 | 897 | ctx->tail = tail; |
1da177e4 | 898 | |
58c85dc2 | 899 | ring = kmap_atomic(ctx->ring_pages[0]); |
21b40200 | 900 | ring->tail = tail; |
e8e3c3d6 | 901 | kunmap_atomic(ring); |
58c85dc2 | 902 | flush_dcache_page(ctx->ring_pages[0]); |
1da177e4 | 903 | |
0460fef2 KO |
904 | spin_unlock_irqrestore(&ctx->completion_lock, flags); |
905 | ||
21b40200 | 906 | pr_debug("added to ring %p at [%u]\n", iocb, tail); |
8d1c98b0 DL |
907 | |
908 | /* | |
909 | * Check if the user asked us to deliver the result through an | |
910 | * eventfd. The eventfd_signal() function is safe to be called | |
911 | * from IRQ context. | |
912 | */ | |
87c3a86e | 913 | if (iocb->ki_eventfd != NULL) |
8d1c98b0 DL |
914 | eventfd_signal(iocb->ki_eventfd, 1); |
915 | ||
1da177e4 | 916 | /* everything turned out well, dispose of the aiocb. */ |
57282d8f | 917 | kiocb_free(iocb); |
1da177e4 | 918 | |
6cb2a210 QB |
919 | /* |
920 | * We have to order our ring_info tail store above and test | |
921 | * of the wait list below outside the wait lock. This is | |
922 | * like in wake_up_bit() where clearing a bit has to be | |
923 | * ordered with the unlocked test. | |
924 | */ | |
925 | smp_mb(); | |
926 | ||
1da177e4 LT |
927 | if (waitqueue_active(&ctx->wait)) |
928 | wake_up(&ctx->wait); | |
929 | ||
36f55889 | 930 | rcu_read_unlock(); |
1da177e4 | 931 | } |
385773e0 | 932 | EXPORT_SYMBOL(aio_complete); |
1da177e4 | 933 | |
a31ad380 KO |
934 | /* aio_read_events |
935 | * Pull an event off of the ioctx's event ring. Returns the number of | |
936 | * events fetched | |
1da177e4 | 937 | */ |
a31ad380 KO |
938 | static long aio_read_events_ring(struct kioctx *ctx, |
939 | struct io_event __user *event, long nr) | |
1da177e4 | 940 | { |
1da177e4 | 941 | struct aio_ring *ring; |
5ffac122 | 942 | unsigned head, tail, pos; |
a31ad380 KO |
943 | long ret = 0; |
944 | int copy_ret; | |
945 | ||
58c85dc2 | 946 | mutex_lock(&ctx->ring_lock); |
1da177e4 | 947 | |
58c85dc2 | 948 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 949 | head = ring->head; |
5ffac122 | 950 | tail = ring->tail; |
a31ad380 KO |
951 | kunmap_atomic(ring); |
952 | ||
5ffac122 | 953 | pr_debug("h%u t%u m%u\n", head, tail, ctx->nr_events); |
1da177e4 | 954 | |
5ffac122 | 955 | if (head == tail) |
1da177e4 LT |
956 | goto out; |
957 | ||
a31ad380 KO |
958 | while (ret < nr) { |
959 | long avail; | |
960 | struct io_event *ev; | |
961 | struct page *page; | |
962 | ||
5ffac122 KO |
963 | avail = (head <= tail ? tail : ctx->nr_events) - head; |
964 | if (head == tail) | |
a31ad380 KO |
965 | break; |
966 | ||
967 | avail = min(avail, nr - ret); | |
968 | avail = min_t(long, avail, AIO_EVENTS_PER_PAGE - | |
969 | ((head + AIO_EVENTS_OFFSET) % AIO_EVENTS_PER_PAGE)); | |
970 | ||
971 | pos = head + AIO_EVENTS_OFFSET; | |
58c85dc2 | 972 | page = ctx->ring_pages[pos / AIO_EVENTS_PER_PAGE]; |
a31ad380 KO |
973 | pos %= AIO_EVENTS_PER_PAGE; |
974 | ||
975 | ev = kmap(page); | |
976 | copy_ret = copy_to_user(event + ret, ev + pos, | |
977 | sizeof(*ev) * avail); | |
978 | kunmap(page); | |
979 | ||
980 | if (unlikely(copy_ret)) { | |
981 | ret = -EFAULT; | |
982 | goto out; | |
983 | } | |
984 | ||
985 | ret += avail; | |
986 | head += avail; | |
58c85dc2 | 987 | head %= ctx->nr_events; |
1da177e4 | 988 | } |
1da177e4 | 989 | |
58c85dc2 | 990 | ring = kmap_atomic(ctx->ring_pages[0]); |
a31ad380 | 991 | ring->head = head; |
91d80a84 | 992 | kunmap_atomic(ring); |
58c85dc2 | 993 | flush_dcache_page(ctx->ring_pages[0]); |
a31ad380 | 994 | |
5ffac122 | 995 | pr_debug("%li h%u t%u\n", ret, head, tail); |
3e845ce0 | 996 | |
e1bdd5f2 | 997 | put_reqs_available(ctx, ret); |
a31ad380 | 998 | out: |
58c85dc2 | 999 | mutex_unlock(&ctx->ring_lock); |
a31ad380 | 1000 | |
1da177e4 LT |
1001 | return ret; |
1002 | } | |
1003 | ||
a31ad380 KO |
1004 | static bool aio_read_events(struct kioctx *ctx, long min_nr, long nr, |
1005 | struct io_event __user *event, long *i) | |
1da177e4 | 1006 | { |
a31ad380 | 1007 | long ret = aio_read_events_ring(ctx, event + *i, nr - *i); |
1da177e4 | 1008 | |
a31ad380 KO |
1009 | if (ret > 0) |
1010 | *i += ret; | |
1da177e4 | 1011 | |
a31ad380 KO |
1012 | if (unlikely(atomic_read(&ctx->dead))) |
1013 | ret = -EINVAL; | |
1da177e4 | 1014 | |
a31ad380 KO |
1015 | if (!*i) |
1016 | *i = ret; | |
1da177e4 | 1017 | |
a31ad380 | 1018 | return ret < 0 || *i >= min_nr; |
1da177e4 LT |
1019 | } |
1020 | ||
a31ad380 | 1021 | static long read_events(struct kioctx *ctx, long min_nr, long nr, |
1da177e4 LT |
1022 | struct io_event __user *event, |
1023 | struct timespec __user *timeout) | |
1024 | { | |
a31ad380 KO |
1025 | ktime_t until = { .tv64 = KTIME_MAX }; |
1026 | long ret = 0; | |
1da177e4 | 1027 | |
1da177e4 LT |
1028 | if (timeout) { |
1029 | struct timespec ts; | |
a31ad380 | 1030 | |
1da177e4 | 1031 | if (unlikely(copy_from_user(&ts, timeout, sizeof(ts)))) |
a31ad380 | 1032 | return -EFAULT; |
1da177e4 | 1033 | |
a31ad380 | 1034 | until = timespec_to_ktime(ts); |
1da177e4 LT |
1035 | } |
1036 | ||
a31ad380 KO |
1037 | /* |
1038 | * Note that aio_read_events() is being called as the conditional - i.e. | |
1039 | * we're calling it after prepare_to_wait() has set task state to | |
1040 | * TASK_INTERRUPTIBLE. | |
1041 | * | |
1042 | * But aio_read_events() can block, and if it blocks it's going to flip | |
1043 | * the task state back to TASK_RUNNING. | |
1044 | * | |
1045 | * This should be ok, provided it doesn't flip the state back to | |
1046 | * TASK_RUNNING and return 0 too much - that causes us to spin. That | |
1047 | * will only happen if the mutex_lock() call blocks, and we then find | |
1048 | * the ringbuffer empty. So in practice we should be ok, but it's | |
1049 | * something to be aware of when touching this code. | |
1050 | */ | |
1051 | wait_event_interruptible_hrtimeout(ctx->wait, | |
1052 | aio_read_events(ctx, min_nr, nr, event, &ret), until); | |
1da177e4 | 1053 | |
a31ad380 KO |
1054 | if (!ret && signal_pending(current)) |
1055 | ret = -EINTR; | |
1da177e4 | 1056 | |
a31ad380 | 1057 | return ret; |
1da177e4 LT |
1058 | } |
1059 | ||
1da177e4 LT |
1060 | /* sys_io_setup: |
1061 | * Create an aio_context capable of receiving at least nr_events. | |
1062 | * ctxp must not point to an aio_context that already exists, and | |
1063 | * must be initialized to 0 prior to the call. On successful | |
1064 | * creation of the aio_context, *ctxp is filled in with the resulting | |
1065 | * handle. May fail with -EINVAL if *ctxp is not initialized, | |
1066 | * if the specified nr_events exceeds internal limits. May fail | |
1067 | * with -EAGAIN if the specified nr_events exceeds the user's limit | |
1068 | * of available events. May fail with -ENOMEM if insufficient kernel | |
1069 | * resources are available. May fail with -EFAULT if an invalid | |
1070 | * pointer is passed for ctxp. Will fail with -ENOSYS if not | |
1071 | * implemented. | |
1072 | */ | |
002c8976 | 1073 | SYSCALL_DEFINE2(io_setup, unsigned, nr_events, aio_context_t __user *, ctxp) |
1da177e4 LT |
1074 | { |
1075 | struct kioctx *ioctx = NULL; | |
1076 | unsigned long ctx; | |
1077 | long ret; | |
1078 | ||
1079 | ret = get_user(ctx, ctxp); | |
1080 | if (unlikely(ret)) | |
1081 | goto out; | |
1082 | ||
1083 | ret = -EINVAL; | |
d55b5fda ZB |
1084 | if (unlikely(ctx || nr_events == 0)) { |
1085 | pr_debug("EINVAL: io_setup: ctx %lu nr_events %u\n", | |
1086 | ctx, nr_events); | |
1da177e4 LT |
1087 | goto out; |
1088 | } | |
1089 | ||
1090 | ioctx = ioctx_alloc(nr_events); | |
1091 | ret = PTR_ERR(ioctx); | |
1092 | if (!IS_ERR(ioctx)) { | |
1093 | ret = put_user(ioctx->user_id, ctxp); | |
a2e1859a | 1094 | if (ret) |
db446a08 | 1095 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1096 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1097 | } |
1098 | ||
1099 | out: | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | /* sys_io_destroy: | |
1104 | * Destroy the aio_context specified. May cancel any outstanding | |
1105 | * AIOs and block on completion. Will fail with -ENOSYS if not | |
642b5123 | 1106 | * implemented. May fail with -EINVAL if the context pointed to |
1da177e4 LT |
1107 | * is invalid. |
1108 | */ | |
002c8976 | 1109 | SYSCALL_DEFINE1(io_destroy, aio_context_t, ctx) |
1da177e4 LT |
1110 | { |
1111 | struct kioctx *ioctx = lookup_ioctx(ctx); | |
1112 | if (likely(NULL != ioctx)) { | |
db446a08 | 1113 | kill_ioctx(current->mm, ioctx); |
723be6e3 | 1114 | percpu_ref_put(&ioctx->users); |
1da177e4 LT |
1115 | return 0; |
1116 | } | |
1117 | pr_debug("EINVAL: io_destroy: invalid context id\n"); | |
1118 | return -EINVAL; | |
1119 | } | |
1120 | ||
41ef4eb8 KO |
1121 | typedef ssize_t (aio_rw_op)(struct kiocb *, const struct iovec *, |
1122 | unsigned long, loff_t); | |
1123 | ||
8bc92afc KO |
1124 | static ssize_t aio_setup_vectored_rw(struct kiocb *kiocb, |
1125 | int rw, char __user *buf, | |
1126 | unsigned long *nr_segs, | |
1127 | struct iovec **iovec, | |
1128 | bool compat) | |
eed4e51f BP |
1129 | { |
1130 | ssize_t ret; | |
1131 | ||
8bc92afc | 1132 | *nr_segs = kiocb->ki_nbytes; |
41ef4eb8 | 1133 | |
9d85cba7 JM |
1134 | #ifdef CONFIG_COMPAT |
1135 | if (compat) | |
41ef4eb8 | 1136 | ret = compat_rw_copy_check_uvector(rw, |
8bc92afc KO |
1137 | (struct compat_iovec __user *)buf, |
1138 | *nr_segs, 1, *iovec, iovec); | |
9d85cba7 JM |
1139 | else |
1140 | #endif | |
41ef4eb8 | 1141 | ret = rw_copy_check_uvector(rw, |
8bc92afc KO |
1142 | (struct iovec __user *)buf, |
1143 | *nr_segs, 1, *iovec, iovec); | |
eed4e51f | 1144 | if (ret < 0) |
41ef4eb8 | 1145 | return ret; |
a70b52ec | 1146 | |
41ef4eb8 | 1147 | /* ki_nbytes now reflect bytes instead of segs */ |
eed4e51f | 1148 | kiocb->ki_nbytes = ret; |
41ef4eb8 | 1149 | return 0; |
eed4e51f BP |
1150 | } |
1151 | ||
8bc92afc KO |
1152 | static ssize_t aio_setup_single_vector(struct kiocb *kiocb, |
1153 | int rw, char __user *buf, | |
1154 | unsigned long *nr_segs, | |
1155 | struct iovec *iovec) | |
eed4e51f | 1156 | { |
8bc92afc | 1157 | if (unlikely(!access_ok(!rw, buf, kiocb->ki_nbytes))) |
41ef4eb8 | 1158 | return -EFAULT; |
a70b52ec | 1159 | |
8bc92afc KO |
1160 | iovec->iov_base = buf; |
1161 | iovec->iov_len = kiocb->ki_nbytes; | |
1162 | *nr_segs = 1; | |
eed4e51f BP |
1163 | return 0; |
1164 | } | |
1165 | ||
1da177e4 LT |
1166 | /* |
1167 | * aio_setup_iocb: | |
1168 | * Performs the initial checks and aio retry method | |
1169 | * setup for the kiocb at the time of io submission. | |
1170 | */ | |
8bc92afc KO |
1171 | static ssize_t aio_run_iocb(struct kiocb *req, unsigned opcode, |
1172 | char __user *buf, bool compat) | |
1da177e4 | 1173 | { |
41ef4eb8 KO |
1174 | struct file *file = req->ki_filp; |
1175 | ssize_t ret; | |
8bc92afc | 1176 | unsigned long nr_segs; |
41ef4eb8 KO |
1177 | int rw; |
1178 | fmode_t mode; | |
1179 | aio_rw_op *rw_op; | |
8bc92afc | 1180 | struct iovec inline_vec, *iovec = &inline_vec; |
1da177e4 | 1181 | |
8bc92afc | 1182 | switch (opcode) { |
1da177e4 | 1183 | case IOCB_CMD_PREAD: |
eed4e51f | 1184 | case IOCB_CMD_PREADV: |
41ef4eb8 KO |
1185 | mode = FMODE_READ; |
1186 | rw = READ; | |
1187 | rw_op = file->f_op->aio_read; | |
1188 | goto rw_common; | |
1189 | ||
1190 | case IOCB_CMD_PWRITE: | |
eed4e51f | 1191 | case IOCB_CMD_PWRITEV: |
41ef4eb8 KO |
1192 | mode = FMODE_WRITE; |
1193 | rw = WRITE; | |
1194 | rw_op = file->f_op->aio_write; | |
1195 | goto rw_common; | |
1196 | rw_common: | |
1197 | if (unlikely(!(file->f_mode & mode))) | |
1198 | return -EBADF; | |
1199 | ||
1200 | if (!rw_op) | |
1201 | return -EINVAL; | |
1202 | ||
8bc92afc KO |
1203 | ret = (opcode == IOCB_CMD_PREADV || |
1204 | opcode == IOCB_CMD_PWRITEV) | |
1205 | ? aio_setup_vectored_rw(req, rw, buf, &nr_segs, | |
1206 | &iovec, compat) | |
1207 | : aio_setup_single_vector(req, rw, buf, &nr_segs, | |
1208 | iovec); | |
eed4e51f | 1209 | if (ret) |
41ef4eb8 KO |
1210 | return ret; |
1211 | ||
1212 | ret = rw_verify_area(rw, file, &req->ki_pos, req->ki_nbytes); | |
8bc92afc KO |
1213 | if (ret < 0) { |
1214 | if (iovec != &inline_vec) | |
1215 | kfree(iovec); | |
41ef4eb8 | 1216 | return ret; |
8bc92afc | 1217 | } |
41ef4eb8 KO |
1218 | |
1219 | req->ki_nbytes = ret; | |
41ef4eb8 | 1220 | |
73a7075e KO |
1221 | /* XXX: move/kill - rw_verify_area()? */ |
1222 | /* This matches the pread()/pwrite() logic */ | |
1223 | if (req->ki_pos < 0) { | |
1224 | ret = -EINVAL; | |
1225 | break; | |
1226 | } | |
1227 | ||
1228 | if (rw == WRITE) | |
1229 | file_start_write(file); | |
1230 | ||
8bc92afc | 1231 | ret = rw_op(req, iovec, nr_segs, req->ki_pos); |
73a7075e KO |
1232 | |
1233 | if (rw == WRITE) | |
1234 | file_end_write(file); | |
1da177e4 | 1235 | break; |
41ef4eb8 | 1236 | |
1da177e4 | 1237 | case IOCB_CMD_FDSYNC: |
41ef4eb8 KO |
1238 | if (!file->f_op->aio_fsync) |
1239 | return -EINVAL; | |
1240 | ||
1241 | ret = file->f_op->aio_fsync(req, 1); | |
1da177e4 | 1242 | break; |
41ef4eb8 | 1243 | |
1da177e4 | 1244 | case IOCB_CMD_FSYNC: |
41ef4eb8 KO |
1245 | if (!file->f_op->aio_fsync) |
1246 | return -EINVAL; | |
1247 | ||
1248 | ret = file->f_op->aio_fsync(req, 0); | |
1da177e4 | 1249 | break; |
41ef4eb8 | 1250 | |
1da177e4 | 1251 | default: |
caf4167a | 1252 | pr_debug("EINVAL: no operation provided\n"); |
41ef4eb8 | 1253 | return -EINVAL; |
1da177e4 LT |
1254 | } |
1255 | ||
8bc92afc KO |
1256 | if (iovec != &inline_vec) |
1257 | kfree(iovec); | |
1258 | ||
41ef4eb8 KO |
1259 | if (ret != -EIOCBQUEUED) { |
1260 | /* | |
1261 | * There's no easy way to restart the syscall since other AIO's | |
1262 | * may be already running. Just fail this IO with EINTR. | |
1263 | */ | |
1264 | if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR || | |
1265 | ret == -ERESTARTNOHAND || | |
1266 | ret == -ERESTART_RESTARTBLOCK)) | |
1267 | ret = -EINTR; | |
1268 | aio_complete(req, ret, 0); | |
1269 | } | |
1da177e4 LT |
1270 | |
1271 | return 0; | |
1272 | } | |
1273 | ||
d5470b59 | 1274 | static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb, |
a1c8eae7 | 1275 | struct iocb *iocb, bool compat) |
1da177e4 LT |
1276 | { |
1277 | struct kiocb *req; | |
1da177e4 LT |
1278 | ssize_t ret; |
1279 | ||
1280 | /* enforce forwards compatibility on users */ | |
9c3060be | 1281 | if (unlikely(iocb->aio_reserved1 || iocb->aio_reserved2)) { |
caf4167a | 1282 | pr_debug("EINVAL: reserve field set\n"); |
1da177e4 LT |
1283 | return -EINVAL; |
1284 | } | |
1285 | ||
1286 | /* prevent overflows */ | |
1287 | if (unlikely( | |
1288 | (iocb->aio_buf != (unsigned long)iocb->aio_buf) || | |
1289 | (iocb->aio_nbytes != (size_t)iocb->aio_nbytes) || | |
1290 | ((ssize_t)iocb->aio_nbytes < 0) | |
1291 | )) { | |
1292 | pr_debug("EINVAL: io_submit: overflow check\n"); | |
1293 | return -EINVAL; | |
1294 | } | |
1295 | ||
41ef4eb8 | 1296 | req = aio_get_req(ctx); |
1d98ebfc | 1297 | if (unlikely(!req)) |
1da177e4 | 1298 | return -EAGAIN; |
1d98ebfc KO |
1299 | |
1300 | req->ki_filp = fget(iocb->aio_fildes); | |
1301 | if (unlikely(!req->ki_filp)) { | |
1302 | ret = -EBADF; | |
1303 | goto out_put_req; | |
1da177e4 | 1304 | } |
1d98ebfc | 1305 | |
9c3060be DL |
1306 | if (iocb->aio_flags & IOCB_FLAG_RESFD) { |
1307 | /* | |
1308 | * If the IOCB_FLAG_RESFD flag of aio_flags is set, get an | |
1309 | * instance of the file* now. The file descriptor must be | |
1310 | * an eventfd() fd, and will be signaled for each completed | |
1311 | * event using the eventfd_signal() function. | |
1312 | */ | |
13389010 | 1313 | req->ki_eventfd = eventfd_ctx_fdget((int) iocb->aio_resfd); |
801678c5 | 1314 | if (IS_ERR(req->ki_eventfd)) { |
9c3060be | 1315 | ret = PTR_ERR(req->ki_eventfd); |
87c3a86e | 1316 | req->ki_eventfd = NULL; |
9c3060be DL |
1317 | goto out_put_req; |
1318 | } | |
1319 | } | |
1da177e4 | 1320 | |
8a660890 | 1321 | ret = put_user(KIOCB_KEY, &user_iocb->aio_key); |
1da177e4 | 1322 | if (unlikely(ret)) { |
caf4167a | 1323 | pr_debug("EFAULT: aio_key\n"); |
1da177e4 LT |
1324 | goto out_put_req; |
1325 | } | |
1326 | ||
1327 | req->ki_obj.user = user_iocb; | |
1328 | req->ki_user_data = iocb->aio_data; | |
1329 | req->ki_pos = iocb->aio_offset; | |
73a7075e | 1330 | req->ki_nbytes = iocb->aio_nbytes; |
1da177e4 | 1331 | |
8bc92afc KO |
1332 | ret = aio_run_iocb(req, iocb->aio_lio_opcode, |
1333 | (char __user *)(unsigned long)iocb->aio_buf, | |
1334 | compat); | |
41003a7b | 1335 | if (ret) |
7137c6bd | 1336 | goto out_put_req; |
41003a7b | 1337 | |
1da177e4 | 1338 | return 0; |
1da177e4 | 1339 | out_put_req: |
e1bdd5f2 | 1340 | put_reqs_available(ctx, 1); |
57282d8f | 1341 | kiocb_free(req); |
1da177e4 LT |
1342 | return ret; |
1343 | } | |
1344 | ||
9d85cba7 JM |
1345 | long do_io_submit(aio_context_t ctx_id, long nr, |
1346 | struct iocb __user *__user *iocbpp, bool compat) | |
1da177e4 LT |
1347 | { |
1348 | struct kioctx *ctx; | |
1349 | long ret = 0; | |
080d676d | 1350 | int i = 0; |
9f5b9425 | 1351 | struct blk_plug plug; |
1da177e4 LT |
1352 | |
1353 | if (unlikely(nr < 0)) | |
1354 | return -EINVAL; | |
1355 | ||
75e1c70f JM |
1356 | if (unlikely(nr > LONG_MAX/sizeof(*iocbpp))) |
1357 | nr = LONG_MAX/sizeof(*iocbpp); | |
1358 | ||
1da177e4 LT |
1359 | if (unlikely(!access_ok(VERIFY_READ, iocbpp, (nr*sizeof(*iocbpp))))) |
1360 | return -EFAULT; | |
1361 | ||
1362 | ctx = lookup_ioctx(ctx_id); | |
1363 | if (unlikely(!ctx)) { | |
caf4167a | 1364 | pr_debug("EINVAL: invalid context id\n"); |
1da177e4 LT |
1365 | return -EINVAL; |
1366 | } | |
1367 | ||
9f5b9425 SL |
1368 | blk_start_plug(&plug); |
1369 | ||
1da177e4 LT |
1370 | /* |
1371 | * AKPM: should this return a partial result if some of the IOs were | |
1372 | * successfully submitted? | |
1373 | */ | |
1374 | for (i=0; i<nr; i++) { | |
1375 | struct iocb __user *user_iocb; | |
1376 | struct iocb tmp; | |
1377 | ||
1378 | if (unlikely(__get_user(user_iocb, iocbpp + i))) { | |
1379 | ret = -EFAULT; | |
1380 | break; | |
1381 | } | |
1382 | ||
1383 | if (unlikely(copy_from_user(&tmp, user_iocb, sizeof(tmp)))) { | |
1384 | ret = -EFAULT; | |
1385 | break; | |
1386 | } | |
1387 | ||
a1c8eae7 | 1388 | ret = io_submit_one(ctx, user_iocb, &tmp, compat); |
1da177e4 LT |
1389 | if (ret) |
1390 | break; | |
1391 | } | |
9f5b9425 | 1392 | blk_finish_plug(&plug); |
1da177e4 | 1393 | |
723be6e3 | 1394 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1395 | return i ? i : ret; |
1396 | } | |
1397 | ||
9d85cba7 JM |
1398 | /* sys_io_submit: |
1399 | * Queue the nr iocbs pointed to by iocbpp for processing. Returns | |
1400 | * the number of iocbs queued. May return -EINVAL if the aio_context | |
1401 | * specified by ctx_id is invalid, if nr is < 0, if the iocb at | |
1402 | * *iocbpp[0] is not properly initialized, if the operation specified | |
1403 | * is invalid for the file descriptor in the iocb. May fail with | |
1404 | * -EFAULT if any of the data structures point to invalid data. May | |
1405 | * fail with -EBADF if the file descriptor specified in the first | |
1406 | * iocb is invalid. May fail with -EAGAIN if insufficient resources | |
1407 | * are available to queue any iocbs. Will return 0 if nr is 0. Will | |
1408 | * fail with -ENOSYS if not implemented. | |
1409 | */ | |
1410 | SYSCALL_DEFINE3(io_submit, aio_context_t, ctx_id, long, nr, | |
1411 | struct iocb __user * __user *, iocbpp) | |
1412 | { | |
1413 | return do_io_submit(ctx_id, nr, iocbpp, 0); | |
1414 | } | |
1415 | ||
1da177e4 LT |
1416 | /* lookup_kiocb |
1417 | * Finds a given iocb for cancellation. | |
1da177e4 | 1418 | */ |
25ee7e38 AB |
1419 | static struct kiocb *lookup_kiocb(struct kioctx *ctx, struct iocb __user *iocb, |
1420 | u32 key) | |
1da177e4 LT |
1421 | { |
1422 | struct list_head *pos; | |
d00689af ZB |
1423 | |
1424 | assert_spin_locked(&ctx->ctx_lock); | |
1425 | ||
8a660890 KO |
1426 | if (key != KIOCB_KEY) |
1427 | return NULL; | |
1428 | ||
1da177e4 LT |
1429 | /* TODO: use a hash or array, this sucks. */ |
1430 | list_for_each(pos, &ctx->active_reqs) { | |
1431 | struct kiocb *kiocb = list_kiocb(pos); | |
8a660890 | 1432 | if (kiocb->ki_obj.user == iocb) |
1da177e4 LT |
1433 | return kiocb; |
1434 | } | |
1435 | return NULL; | |
1436 | } | |
1437 | ||
1438 | /* sys_io_cancel: | |
1439 | * Attempts to cancel an iocb previously passed to io_submit. If | |
1440 | * the operation is successfully cancelled, the resulting event is | |
1441 | * copied into the memory pointed to by result without being placed | |
1442 | * into the completion queue and 0 is returned. May fail with | |
1443 | * -EFAULT if any of the data structures pointed to are invalid. | |
1444 | * May fail with -EINVAL if aio_context specified by ctx_id is | |
1445 | * invalid. May fail with -EAGAIN if the iocb specified was not | |
1446 | * cancelled. Will fail with -ENOSYS if not implemented. | |
1447 | */ | |
002c8976 HC |
1448 | SYSCALL_DEFINE3(io_cancel, aio_context_t, ctx_id, struct iocb __user *, iocb, |
1449 | struct io_event __user *, result) | |
1da177e4 | 1450 | { |
1da177e4 LT |
1451 | struct kioctx *ctx; |
1452 | struct kiocb *kiocb; | |
1453 | u32 key; | |
1454 | int ret; | |
1455 | ||
1456 | ret = get_user(key, &iocb->aio_key); | |
1457 | if (unlikely(ret)) | |
1458 | return -EFAULT; | |
1459 | ||
1460 | ctx = lookup_ioctx(ctx_id); | |
1461 | if (unlikely(!ctx)) | |
1462 | return -EINVAL; | |
1463 | ||
1464 | spin_lock_irq(&ctx->ctx_lock); | |
906b973c | 1465 | |
1da177e4 | 1466 | kiocb = lookup_kiocb(ctx, iocb, key); |
906b973c | 1467 | if (kiocb) |
bec68faa | 1468 | ret = kiocb_cancel(ctx, kiocb); |
906b973c KO |
1469 | else |
1470 | ret = -EINVAL; | |
1471 | ||
1da177e4 LT |
1472 | spin_unlock_irq(&ctx->ctx_lock); |
1473 | ||
906b973c | 1474 | if (!ret) { |
bec68faa KO |
1475 | /* |
1476 | * The result argument is no longer used - the io_event is | |
1477 | * always delivered via the ring buffer. -EINPROGRESS indicates | |
1478 | * cancellation is progress: | |
906b973c | 1479 | */ |
bec68faa | 1480 | ret = -EINPROGRESS; |
906b973c | 1481 | } |
1da177e4 | 1482 | |
723be6e3 | 1483 | percpu_ref_put(&ctx->users); |
1da177e4 LT |
1484 | |
1485 | return ret; | |
1486 | } | |
1487 | ||
1488 | /* io_getevents: | |
1489 | * Attempts to read at least min_nr events and up to nr events from | |
642b5123 ST |
1490 | * the completion queue for the aio_context specified by ctx_id. If |
1491 | * it succeeds, the number of read events is returned. May fail with | |
1492 | * -EINVAL if ctx_id is invalid, if min_nr is out of range, if nr is | |
1493 | * out of range, if timeout is out of range. May fail with -EFAULT | |
1494 | * if any of the memory specified is invalid. May return 0 or | |
1495 | * < min_nr if the timeout specified by timeout has elapsed | |
1496 | * before sufficient events are available, where timeout == NULL | |
1497 | * specifies an infinite timeout. Note that the timeout pointed to by | |
6900807c | 1498 | * timeout is relative. Will fail with -ENOSYS if not implemented. |
1da177e4 | 1499 | */ |
002c8976 HC |
1500 | SYSCALL_DEFINE5(io_getevents, aio_context_t, ctx_id, |
1501 | long, min_nr, | |
1502 | long, nr, | |
1503 | struct io_event __user *, events, | |
1504 | struct timespec __user *, timeout) | |
1da177e4 LT |
1505 | { |
1506 | struct kioctx *ioctx = lookup_ioctx(ctx_id); | |
1507 | long ret = -EINVAL; | |
1508 | ||
1509 | if (likely(ioctx)) { | |
2e410255 | 1510 | if (likely(min_nr <= nr && min_nr >= 0)) |
1da177e4 | 1511 | ret = read_events(ioctx, min_nr, nr, events, timeout); |
723be6e3 | 1512 | percpu_ref_put(&ioctx->users); |
1da177e4 | 1513 | } |
1da177e4 LT |
1514 | return ret; |
1515 | } |